AFRL-VA-WP-TR-2000-3046

DEVELOPMENT OF METHODS AND

DEVICES TO PREDICT AND PREVENT
PILOT-INDUCED OSCILLATIONS

DAVID G. MITCHELL
ROGER H. HOH

Hoh Aeronautics Inc

2075 Palos Verde Dr North
Lomita CA 90717

DECEMBER 2000

FINAL REPORT FOR PERIOD AUGUST 1995 – NOVEMBER 1999

THIS IS A SMALL BUSINESS INNOVATIVE RESEARCH (SBIR) PHASE II REPORT

Approved for public release; distribution unlimited

AIR VEHICLES DIRECTORATE

AIR FORCE RESEARCH LABORATORY
AIR FORCE MATERIEL COMMAND
WRIGHT-PATTERSON AIR FORCE BASE, OH 45433-7542
 HOH
AERONAUTICS
INC

Development of Methods and Devices to Predict and

Prevent Pilot-Induced Oscillations

David G. Mitchell
Roger H. Hoh
Hoh Aeronautics, Inc.

Technical Report No. 1056-1

10 December 1999

Prepared Under Contract No. F33615-95-C-3608

for US Air Force Research Laboratory
Wright-Patterson AFB OH 45433
 TABLE OF CONTENTS
Page

ACKNOWLEDGEMENTS..................................................................................................... xiii
I. TEN STEPS TO REDUCING THE RISK OF PIO................................................................. 1
A. Ten Definitions................................................................................................................. 1
B. Ten Steps.......................................................................................................................... 2
1. Be Prepared for PIO ...................................................................................................... 2
2. Design for PIO Resistance............................................................................................. 2
3. Apply Valid Prediction Criteria Early in the Design Process.......................................... 2
4. Continue to Apply Criteria as the Accuracy of the Model Improves............................... 4
5. Use High-Gain Maneuvers to Evaluate PIO Tendency in Piloted Simulations................ 4
6. Apply PIO Detection and Prevention Devices During Developmental Flight Testing..... 8
7. Extend Test Inputs and Application of Criteria to Large Input Amplitudes .................... 9
8. Update Ground Simulation Models With Flight Data .................................................... 9
9. Include PIO Recognition as a Part of the Training Syllabus for Pilots............................ 9
10. Be Prepared for PIO ................................................................................................... 9
II. INTRODUCTION ...............................................................................................................10
A. The Phenomenon .............................................................................................................10
B. The Elements of PIO .......................................................................................................11
1. The Airplane ................................................................................................................11
2. The Pilot ......................................................................................................................14
3. The Trigger ..................................................................................................................14
4. Other Contributors .......................................................................................................15
C. PIO in the Development Process......................................................................................15
D. Origins of this Research and Related Work......................................................................16
E. The Focus of this Research ..............................................................................................17
1. Prevention of PIOs by Design.......................................................................................17
2. PIOs and Handling Qualities ........................................................................................17
3. Pilot Opinion, Handling Qualities, and PIO ..................................................................18
F. Organization of the Report...............................................................................................20
III. DEFINING AND CATEGORIZING PIO...........................................................................22
A. Background .....................................................................................................................22
B. What is a PIO?.................................................................................................................22
C. Categorizing PIO.............................................................................................................23
1. The Occurrence of PIO.................................................................................................25
2. Characterization of the Oscillation................................................................................25
3. Defining PIOs ..............................................................................................................27
D. Pilot-Induced Oscillations, Pilot-in-the-Loop Oscillations, and Aircraft-Pilot Coupling ...28
E. The Three (or More) Categories of PIO ...........................................................................29
F. Some Examples ...............................................................................................................30
1. Low-Frequency (Phugoid) PIO.....................................................................................30
2. Medium-Frequency PIO ...............................................................................................30
3. High-Frequency, Small-Amplitude Pitch PIO (Pitch Bobble) .......................................30
4. High-Frequency, Small-Amplitude Roll PIO (Roll Ratchet) .........................................31

iii
 TABLE OF CONTENTS (Continued)
Page
IV. MEASUREMENTS OF PILOT BEHAVIOR IN PIO ........................................................33
A. The Pilot as a Part of the Process .....................................................................................33
B. Pilot Models ....................................................................................................................33
C. Characteristics from Category I PIOS in Landing (HAVE PIO) .......................................34
1. Quantitative Measures of PIO Frequency......................................................................37
2. Derived Phase Angles...................................................................................................39
3. Comparisons with Predictions of PIO Frequency..........................................................40
D. Characteristics of PIOs in Attitude Tracking (Rate-Limiting Simulation, Appendix A) ....43
1. Category I PIOs............................................................................................................44
2. Category II PIOs ..........................................................................................................46
V. PREVENTION OF PIO BY DESIGN: LONGITUDINAL CRITERIA FOR CATEGORY I
PIOS .........................................................................................................................................57
A. Background .....................................................................................................................57
1. Prevention by Design ...................................................................................................57
2. Flight Experience with PIOs.........................................................................................58
3. PIO Prediction Criteria .................................................................................................59
B. Prerequisites for Criteria ..................................................................................................59
C. Criteria Considered..........................................................................................................59
D. Criteria Not Considered ...................................................................................................60
E. Cockpit Controller Characteristics ...................................................................................60
F. The Criteria Selected .......................................................................................................60
1. Airplane Bandwidth/Pitch Rate Overshoot ...................................................................61
2. Neal-Smith (Original Definition) ..................................................................................65
3. Modified Neal-Smith Criteria (MAI) ............................................................................69
4. Smith-Geddes Criteria ..................................................................................................71
5. Gibson Phase Rate and Response Gain Criteria ............................................................75
6. Time-Domain Neal-Smith (TDNS) Criterion................................................................77
G. The Data Base Applied to the Criteria..............................................................................79
H. Application of the Criteria ...............................................................................................80
1. Some Important Assumptions.......................................................................................80
2. How the Criteria Fared .................................................................................................81
3. Ranking the Criteria .....................................................................................................82
4. Putting Correlation Rates in Perspective .......................................................................83
5. Examination of Graphical Trends .................................................................................83
I. Review of the Criteria in Light of the National Research Council’s Three Prerequisites...87
J. Application of the Criteria to Four Example Airplanes.....................................................88
1. Example 1 – The T-38A PIO........................................................................................89
2. Example 2 – The Space Shuttle in Display Tracking and Landing ................................90
3. Example 3 – PIOs on the B-2 During Flight Test..........................................................92
4. Example 4 – A PIO-Resistant Airplane: The F-15 in Landing......................................94
5. Recapitulation of the Criteria and their Predictions for the Four Airplanes....................95
K. Application to Other Documented PIOs...........................................................................96
1. Objectives ....................................................................................................................96

iv
 TABLE OF CONTENTS (Continued)
Page
2. Limitations ...................................................................................................................96
3. Data .............................................................................................................................97
4. The Bandwidth Criteria ................................................................................................97
5. Smith-Geddes Criteria ..................................................................................................97
L. The Effects of Control/Response Sensitivity ....................................................................99
M. Some Criteria for Low-Altitude, High-Speed Flight ...................................................101
1. STI High-Gain Asymptote Parameter .........................................................................101
2. A’Harrah-Siewert Criteria ..........................................................................................102
3. Dynamic Stick Force per g .........................................................................................103
VI. PREVENTION OF PIO BY DESIGN (CONTINUED): LONGITUDINAL CRITERIA
FOR CATEGORY II PIOs......................................................................................................105
A. Background ...................................................................................................................105
B. The Criteria for Category II PIOS..................................................................................105
1. OLOP Criterion..........................................................................................................105
2. Pilot-Vehicle Dynamics NonLinear (PVDNL)............................................................106
3. Time-Domain Neal-Smith (TDNS).............................................................................108
4. Airplane Bandwidth ...................................................................................................108
5. Application of the Criteria to HAVE LIMITS.............................................................109
C. Extension of Bandwidth Criteria in the Presence of Nonlinearities.................................112
1. Frequency Response Measures ...................................................................................113
2. Airplane Model ..........................................................................................................114
3. Sources of Nonlinearity..............................................................................................115
4. Application of Frequency Sweep Technique to HAVE LIMITS Configurations .........117
5. Impact on Bandwidth Parameters ...............................................................................122
6. Application of Bandwidth Criteria..............................................................................125
VII. PREVENTION OF PIO BY DESIGN (CONTINUED): LATERAL CRITERIA FOR
CATEGORY I AND II PIOS ..................................................................................................126
A. Background ...................................................................................................................126
B. Effects of Command/Response Sensitivity.....................................................................126
C. Bandwidth/Phase Delay Criteria for Category I PIOs.....................................................127
D. Smith-Geddes Criteria for Category I PIOs....................................................................130
VIII. GROUND SIMULATION AND FLIGHT TESTING FOR PIO ....................................132
A. Introduction ...................................................................................................................132
B. PIOs in the Development Process ..................................................................................132
1. Flight Test and PIOs...................................................................................................132
2. A Change in Philosophy .............................................................................................132
3. Selection of Pilot-in-the-Loop Tasks ..........................................................................133
C. Pilot Evaluation of PIOs ................................................................................................133
1. Pilot Evaluation: The PIO Tendency Rating Scale .....................................................133
2. PIO Rating Inflation ...................................................................................................138
D. The Need for a New Method for Evaluating PIO ...........................................................141
E. Some Possible Alternatives to the Decision-Tree Scale..................................................143

v
 TABLE OF CONTENTS (Continued)
Page
1. A True “PIO Classification” Method ..........................................................................143
2. A Modified PIO Tendency Rating Scale.....................................................................145
F. A Proposed Comprehensive PIO Classification Method.................................................146
G. PIO and Risk Assessment ..............................................................................................149
H. Maneuvers to Expose PIO .............................................................................................150
I. Recent Simulation Experience with Replicating PIOs ....................................................150
1. The HAVE PIO Truth Model .....................................................................................151
2. Simulator Replications of HAVE PIO ........................................................................151
3. Qualitative Comparisons ............................................................................................152
4. Quantitative Results ...................................................................................................154
5. Observations About Simulation ..................................................................................155
IX. PREVENTION OF PIO IN FLIGHT: A METHOD FOR REAL-TIME DETECTION....156
A. Reasons for Real-Time Detection ..................................................................................156
B. The Realities of Real-Time Detection ............................................................................157
C. Philosophies for PIO Detection......................................................................................158
D. Detection Method Developed for this Study...................................................................159
1. The Realtime Oscillation VERifier (ROVER) ............................................................161
2. PIO Severity...............................................................................................................162
E. Application of ROVER for Identification of PIOs..........................................................162
1. Application of ROVER for this Report .......................................................................162
2. Application of ROVER to the YF-22A PIO................................................................162
3. Application to HAVE PIO Project ..............................................................................167
4. Application to LAMARS Simulation Data..................................................................168
5. Effectiveness for the LAMARS Data..........................................................................170
F. The Next Steps ..............................................................................................................172
X. RECOMMENDATIONS FOR FUTURE WORK..............................................................174
A. PIO Prevention Criteria .................................................................................................174
B. Testing for PIO..............................................................................................................174
C. Real-Time PIO Detection and Prevention ......................................................................175
REFERENCES .......................................................................................................................176
APPENDIX A. PILOTED SIMULATION .............................................................................183
A. Introduction ...................................................................................................................183
B. Objectives......................................................................................................................183
C. Facility ..........................................................................................................................183
D. Tasks .............................................................................................................................184
E. Pitch Configurations ......................................................................................................185
F. Roll Configurations .......................................................................................................188
G. Rate Limiting.................................................................................................................188
H. Conduct of Simulation...................................................................................................189
I. Verification of Configurations .......................................................................................189
J. Run Log ........................................................................................................................189
K. References for Appendix A............................................................................................198

vi
 TABLE OF CONTENTS (Continued)
Page
APPENDIX B. AN EXAMINATION OF THE RESULTS OF HAVE CAP ..........................199
A. Background ...................................................................................................................199
B. Errata for the HAVE CAP Report ..................................................................................199
1. Summary of Corrections.............................................................................................199
2. Configuration A .........................................................................................................200
3. Configuration D .........................................................................................................203
4. Configurations H and I ...............................................................................................203
C. Measurements of Pitch Rate Overshoot and Pitch Attitude Dropback ............................204
D. Evaluation of Results Using Bandwidth and Revised Overshoot ....................................206
E. PIO Analysis .................................................................................................................207
1. Occurrence of PIOs ....................................................................................................207
2. Predictions Based on Pitch Attitude Bandwidth and Overshoot ..................................208
F. References for Appendix B............................................................................................209
APPENDIX C. ANALYSIS OF DATA FROM NAVY STUDY OF F-14 DUAL HYDRAULIC
FAILURES .............................................................................................................................211
A. F-14 Dual Hydraulic Failure Flight Test Program ..........................................................211
B. Analysis Techniques......................................................................................................211
C. Flight Test Data Analysis...............................................................................................211
D. Analysis Results ............................................................................................................212
1. Identification of Stick Dynamics ................................................................................212
2. Effects of Rate Limiting .............................................................................................213
E. Identification of PIO Frequency and Task Bandwidth....................................................217
F. Airplane Bandwidth Assessments ..................................................................................218
G. Observations from this Analysis ....................................................................................220
H. References for Appendix C............................................................................................220
Appendix D. Recommendations for Obtaining Airplane Longitudinal Bandwidth Parameters 221
A. Recommended Steps for Generating Parameters ............................................................221
B. Real-Time Monitoring of Frequency Sweeps .................................................................222

vii
 LIST OF FIGURES
Page
Figure 1. Definitions of Pitch Attitude Bandwidth and Phase Delay (Flight Path Bandwidth
ωBWγ is Measured from γ/Fes and is Defined as ωBWphase )................................................... 3
Figure 2. Definition of Pitch Rate Overshoot Parameter, ∆G(q)................................................. 4
Figure 3. PIO Criteria for Pitch Response When the Dynamics of the Cockpit Control Feel
System are Included............................................................................................................ 5
Figure 4. PIO Criteria for Pitch Response When the Dynamics of the Cockpit Control Feel
System are Excluded .......................................................................................................... 6
Figure 5. PIO Criteria for Roll Response, All Flight Conditions, All Aircraft Categories ........... 7
Figure 6. PIO Tendency Classification Method ......................................................................... 8
Figure 7. Example Time History of a PIO (YF-22)...................................................................11
Figure 8. Effect of High-Frequency Phase Rolloff on PIO ........................................................12
Figure 9. The Cooper-Harper Handling Qualities Rating (HQR) Scale .....................................19
Figure 10. The Current PIO Tendency Rating (PIOR) Scale, 1980 ...........................................19
Figure 11. Correspondence Between HQR and PIOR ...............................................................20
Figure 12. Representative Flight Conditions for Documented PIOs ..........................................25
Figure 13. PIO Frequency vs. Amplitude of Angular Rate........................................................26
Figure 14. PIO Frequency vs. Maximum Cockpit Control Force...............................................27
Figure 15. Documentation of a Phugoid PIO (Reproduced from Ref. 35)..................................31
Figure 16. Example of Pitch Bobble (HAVE PIO36 Configuration 2-B, Pilot A) .......................32
Figure 17. Example of Roll Ratchet (F-16XL)..........................................................................32
Figure 18. Landing Time History Segments from HAVE PIO36 (Configuration 2-5) ................36
Figure 19. Flight-Identified PIO Frequencies for HAVE PIO Configurations ...........................38
Figure 20. Identified Phase Angles Between Pitch Attitude and Stick Force for Severe-PIO
Configurations from HAVE PIO........................................................................................39
Figure 21. Comparison of Flight-Identified PIO Frequencies with Various Predictions.............42
Figure 22. PIO Frequencies for Category I Severe PIOs from HAI Simulation .........................45
Figure 23. Identified Phase Angles Between Pitch Attitude and Stick Force for Category I
Severe PIO Configurations ................................................................................................47
Figure 24. Examples of Pilot Variations in HQR and PIOR for Three Selected Configurations
from Simulation (Appendix A) ..........................................................................................48
Figure 25. Effect of Actuator Rate Limits on PIOR (Configuration 2DV).................................49
Figure 26. Sample Time History Comparison for Four Pilots (Configuration 2DUR20)............50
Figure 27. Comparison of Pilot-Vehicle Describing Functions (YpYc) for Configuration
2DUR20 (Yc is θ/Fes for 2DUR20 with No Rate Limiting).................................................52
Figure 28. Correlation of Approximate Pilot-Vehicle Crossover Frequency with Time Control
Was Lost During the Run (Configuration 2DUR20) ..........................................................52
Figure 29. Identified Frequencies and Phase Angles for Category II PIOs (Configuration
2DVR15, Moving-Base Simulation) ..................................................................................53
Figure 30. Identified Frequencies and Phase Angles for Category II PIOs (Configuration
2DVR10, Moving-Base Simulation) ..................................................................................55
Figure 31. PIO Tendency Rating as a Function of Percent of Time on Rate Limit for Selected
Configurations (Data for Pilot C).......................................................................................56
Figure 32. Neal-Smith Criteria .................................................................................................65

viii
 LIST OF FIGURES (Continued)
Page
Figure 33. PIO Boundaries for Neal-Smith Criteria Using Experimental Data (MAI Study;53
Figure Reproduced from NRC Report1) .............................................................................70
Figure 34. Smith-Geddes Criterion Frequency (Data from Fixed-Base Simulation69)................72
Figure 35. Identified Crossover Frequencies in Fixed-Base Compensatory Pitch Tracking
(McRuer et al.69 Data from Figure 34; “MFQ” is Minimum Flying Qualities Study67) .......74
Figure 36. Gibson PIO Criteria for Phase Rate and Maximum PIO Frequency Gain7 ................76
Figure 37. Summary Data Plots................................................................................................84
Figure 38. Problem With Applying Average Slope Method of Smith-Geddes to the B-2 (from
Ref. 73) .............................................................................................................................93
Figure 39. Bandwidth Criteria Applied to Real-World Data......................................................98
Figure 40. Smith-Geddes Criteria Applied to Real-World Data ................................................98
Figure 41. Effect of Pitch Command Sensitivity on PIO as a Function of Added Time Delay
(Ratings are PIORs with HQRs in Brackets) ....................................................................100
Figure 42. STI High-Gain Asymptote Parameter12 for A'Harrah-Siewert76 Data .....................102
Figure 43. A'Harrah-Siewert PIO Boundary76 ..........................................................................103
Figure 44. Dynamic Stick Force per g PIO Requirements4 ......................................................104
Figure 45. Example of PVDNL for Configuration 2D, 10 deg/sec Rate Limit84 ......................111
Figure 46. Application of OLOP to HAVE LIMITS Configurations79.....................................111
Figure 47. Modification to TDNS and HAVE LIMITS Data83 ................................................112
Figure 48. Example Output Frequency Response for a Rate-Limited Airplane........................114
Figure 49. Sources of Nonlinearity in Flight Control Systems.................................................115
Figure 50. Command Shaping ................................................................................................116
Figure 51. Effect of Rate Limiting on Pitch-Attitude-to-Stick-Force Frequency Responses of
Configurations 2D and 2DU (Sweep Amplitude Achieved ±1g Change in Load Factor) ..118
Figure 52. Effect of Input Amplitude on Frequency Response (Configuration 2DU, 60 deg/sec
Rate Limit) ......................................................................................................................120
Figure 53. Single-Sinewave Verification of Frequency Responses (Configuration 2DU, 20-
deg/sec Rate Limit, ±2g Load Factor Change) .................................................................121
Figure 54. Effect of Command Shaping on Frequency Response (Configuration 2DU, 20 deg/sec
Rate Limit, ±1g Load Factor Change)..............................................................................122
Figure 55. Effects of Rate Limit on Bandwidth Parameters (±1g Load Factor Change)...........123
Figure 56. Effects of Rate Limit on Bandwidth Parameters (±2g Load Factor Change)...........124
Figure 57. Effects of Roll Command Sensitivity on PIO Tendency as a Function of Added Time
Delay (Configuration with Roll Damping TR = 0.3 sec) ...................................................127
Figure 58. Comparison of Handling Qualities Ratings from Two Roll Experiments88,9 with Roll
Attitude Phase Delay (Reproduced from Ref. 45) ............................................................129
Figure 59. Comparison of Handling Qualities Ratings from Two Roll Experiments88,9 with Roll
Attitude Phase Angle from Smith-Geddes Criteria (Reproduced from Ref. 45) ................131
Figure 60. An Early PIO Tendency Rating Scale from Naval Air Test Center, 196393 ............134
Figure 61. Scale for Classification of PIO Tendency, from Cooper and Harper28 (Phrases in Bold
Italics are Missing from Decision-Tree Scale of Figure 10)..............................................135
Figure 62. FAA A-PC Rating Criteria77 ..................................................................................136

ix
 LIST OF FIGURES (Continued)
Page
Figure 63. Comparison of PIO Ratings and Handling Qualities Ratings from Selected Research
Experiments ....................................................................................................................139
Figure 64. Hybrid PIO Tendency Rating Scale21 .....................................................................142
Figure 65. A PIO Classification Scale Based on Demanding Desired Task Performance.........144
Figure 66. A Possible PIO Rating Scale Combining Decision-Tree and Adjectival Scales ......145
Figure 67. Departure Rating Scale97 .......................................................................................147
Figure 68. A Proposed Comprehensive PIO Classification Method ........................................148
Figure 69. Risk Assessment for PIO Severity versus Duration................................................149
Figure 70. Comparison of Mean Handling Qualities Ratings from Simulations and Flight......152
Figure 71. Comparison of Mean PIO Tendency Ratings from Simulations and Flight.............153
Figure 72. Comparison of Responses from HAVE PIO Flight and Simulation........................154
Figure 73. Time Histories of Longitudinal Stick Activity for Two Pilots from LAMARS
Simulation, PIO-Resistant Configuration (2DR30) ..........................................................160
Figure 74. Time Histories of Longitudinal Stick Activity for Two Pilots from LAMARS
Simulation, PIO-Prone Configuration (1DL3)..................................................................160
Figure 75. Outputs from ROVER Application to YF-22 Low-Approach Data ........................163
Figure 76. Expanded Time-Scale Plots for YF-22 PIO ...........................................................165
Figure 77. PIO Flags and Estimated PIO Severity for YF-22 on Expanded Time Scale...........166
Figure 78. Key for Interpreting ROVER Summary Data Plots ................................................169
Figure 79. Maximum “PIO Severity” Output from ROVER for Selected Configurations from
LAMARS Simulation (Configuration Number in Boxes on Charts) .................................171
Figure A-1. Heads-Up Display ...............................................................................................184
Figure A-2. Pitch and Roll Command Signals.........................................................................184
Figure A-3. Schematic of Model and Augmentation System Mechanizations .........................186
Figure A-4. Matlab®/Simulink Representation of Rate-Limited Actuator................................188
Figure B-1. Frequency-Response Data and LOES for Configuration A...................................200
Figure B-2. Time Histories of Frequency Sweeps for Configuration A ...................................202
Figure B-3. Time History of a Sweep with High-Frequency Inputs (Configuration G)............202
Figure B-4. Magnitude Bode Plot for Configuration D ...........................................................203
Figure B-5. Comparison of Pitch Attitude Magnitude Plots for Configuration H.....................204
Figure B-6. Corrected Bode Magnitude Plot for Configuration H (Ref. 1 Figure J62) .............205
Figure B-7. Corrected Bode Magnitude Plot for Configuration I (Ref. 1 Figure J74) ..............205
Figure B-8. Definition of Pitch Rate Overshoot in the Frequency Domain ..............................206
Figure B-9. Bandwidth Characteristics of HAVE CAP Configurations (Solid Symbols Denote
Configurations with PIO in Flight) ..................................................................................208
Figure C-1. Example Stick Force and Pitch Rate Time History Plots ......................................214
Figure C-2. Longitudinal Stick Dynamics (δLON/FLON) from Frequency Sweeps .....................215
Figure C-3. Frequency Responses of q/FLON for 35° Wing Sweep Cases.................................216
Figure C-4. q/FLON Frequency Responses for BUFCM-HIGH 35° Wing Sweep Cases ...........217
Figure C-5. PSD Comparisons for BUFCM-HIGH Drogue Tracking Run Segments ..............218
Figure C-6. q/FLON Frequency Responses for BUFCM-HIGH Drogue Tracking Run Segments
........................................................................................................................................219
Figure C-7. PIO Phase Delay Requirement for Hook-Up and Drogue Tracking Runs .............220

x
 LIST OF FIGURES (Continued)
Page
Figure D-1. Example Frequency Sweep and ROVER Output of Peak Values .........................224
Figure D-2. ROVER-Identified Frequency and Phase for Example Sweep..............................224
Figure D-3. Comparison of Frequency Responses of q/Fes for Example Sweep.......................225

xi
 LIST OF TABLES
Page
Table 1. Some Aircraft that Have Experienced PIOs.................................................................10
Table 2. Pilot Ratings for HAVE PIO Experiment36 (Severe PIO Cases are in Italic Font) .......35
Table 3. Measured and Predicted PIO Frequencies for HAVE PIO Configurations...................41
Table 4. Configurations Exhibiting Category I PIO in HAI Simulation (Appendix A) ..............44
Table 5. Pitch PIO Susceptibility Criteria Evaluated.................................................................61
Table 6. Some Examples of Application of the Neal-Smith Criteria..........................................67
Table 7. Pitch Data Base Applied to the PIO Criteria................................................................79
Table 8. Overall Correlation Rate for the PIO Prediction Criteria .............................................82
Table 9. Application of the PIO Prediction Criteria to the T-38A..............................................89
Table 10. Application of the PIO Prediction Criteria to Three Shuttle PIOs72............................91
Table 11. Application of the PIO Prediction Criteria to Two B-2 PIOs .....................................93
Table 12. Application of PIO Prediction Criteria to the F-15A in Landing................................94
Table 13. Summary of PIO Predictions for the Category I Criteria............................................95
Table 14. Effectiveness of the Criteria for HAVE LIMITS Data (Discrete Task) ....................109
Table 15. Transfer Functions of Analytical Models ................................................................115
Table 16. Bandwidth Parameters for Configuration 2DU with Different Command Shaping
Curves .............................................................................................................................125
Table 17. Configurations from Neal-Smith Experiment51 with HQRs 6 or Better, and PIORs 3.5
or Better, and Pilot Comments of PIO Tendencies ...........................................................140
Table 18. Application of ROVER to HAVE PIO36 (Severe PIO Cases are in Italic Font) .......167
Table A-1. Longitudinal Configurations .................................................................................187
Table A-2. Run Log for Piloted Simulation ............................................................................190
Table B-1. Bandwidth/Overshoot Values for Have Cap Configurations..................................207
Table B-2. Application of Bandwidth/Overshoot to HAVE CAP for PIO ...............................209
Table C-1. Run Log of Frequency Sweep Cases .....................................................................212
Table C-2. Run Log of Aerial Refueling Cases.......................................................................213
Table C-3. Run Log of Field Offset Landing Run...................................................................213

xii
 ACKNOWLEDGEMENTS

The authors wish to acknowledge the contributions of the following individuals. For the piloted
simulations at Wright-Patterson AFB we gratefully acknowledge the efforts of Mr. Brian Stadler
from AFRL/VACD and Mr. Jeff Slutz from Halifax Corporation. We also recognize the
diligence of the nine test pilots: Mr. Michael Stortz; Col Harold Steck, Jr., USAF; Mr. Dave
Noah; Cdr John O’Connor, USN; and Maj John Hunnell, LtCol David Hunter, LtCol Joe Arvai,
Maj Karl Major, and Maj William Mason, all USAF. Their availability, commitment, and
patience throughout the two simulation sessions made possible the generation of a valuable data
base. Finally, Mr. Dave Hoyman and the staff of RealTime Reporters are to be commended for
their excellent job in transcribing stacks of audiotaped pilot comments.
From our subcontractors, we wish to acknowledge the technical contributions of Mr. John
Bessolo and Mr. Dave Bodden of Lockheed Martin Tactical Aircraft Systems; Mr. Frank George
of Northrop Grumman; and Mr. John Hodgkinson and Mr. Terry von Klein of McDonnell
Douglas Corporation (now Boeing Phantom Works). Other individuals deserving recognition
are Mr. Dave Klyde and Dr. Vineet Sahasrabudhe of Systems Technology, Inc., Mr. Bob Heffley
of Robert Heffley Engineering, and Mr. James McWha (now retired) and Mr. Brian Lee of
Boeing Commercial Airplane Group.
Throughout the contract we were provided the opportunity to interact with students of the US Air
Force Test Pilot School, Edwards AFB, CA, and to actively participate in several student-led
Test Management Projects (TMPs). These TMPs have produced results of the highest quality
and have contributed significantly to our objectives. The authors wish to acknowledge the
project teams, identified by the project names and the names of the team leaders, all USAF
officers. Project HAVE GAS, Maj Michael Taschner; HAVE CAP, Capt Dave Kivioja; HAVE
LIMITS, Capt Brian Kish; HAVE GAS II, Capt Kelly Latimer.
The authors participated in the planning for a workshop on PIO hosted by NASA Dryden Flight
Research Center in April 1999. We acknowledge the leadership of Mary Shafer for arranging
the workshop and for tolerating us as we attempted to usurp control.
Finally, the authors wish to acknowledge the very special contributions of the engineers at the
now-disbanded Flight Dynamics and Controls Branch at Wright Laboratory, Wright-Patterson
AFB, OH. The opportunities to participate in their simulations, flight research projects, and
general technical discussions benefited this effort immeasurably. For this we especially
acknowledge Messrs. Tom Cord, Dave Leggett, Ba Nguyen, and Wayne Thor. Special thanks go
to Mr. Cord for his many contributions to, and careful reading of, this technical report.

xiii
 I. TEN STEPS TO REDUCING THE RISK OF PIO

A. TEN DEFINITIONS*
DEFINITION #1: A PIO is a sustained or uncontrollable unintentional oscillation resulting
from the efforts of the pilot to control the aircraft.
DEFINITION #2: PIO is an event that results from faulty aircraft design, extension of the
airplane’s operational usage into an area for which it was not intended, or following a failure,†
and is not the fault of the pilot.
DEFINITION #3: PIO is commonly found to be related to deficiencies in basic flying qualities
characteristics, though the phenomenon should be treated independently from flying qualities.
Most PIOs outside of the research world are related to rate limiting of a control effector or
software element upstream of a control effector, but rate limiting can be both the cause of PIO
and the result of it.
DEFINITION #4: PIO may be either constant-amplitude, convergent, or divergent with time.
DEFINITION #5: PIO may be any number of cycles of oscillation; there is no minimum
number to declare it a PIO.
DEFINITION #6: PIO occurs when a response state of the airplane is approximately 180
degrees out of phase with the pilot. This could be any response state of the airplane, though the
most common are pitch attitude, load factor, and roll attitude.
DEFINITION #7: PIO may occur at very low frequencies – near the phugoid mode in pitch –
up to frequencies of around 3 Hz (“roll ratchet”). The most common frequency is in the range
for pilot closed-loop control, typically 1/6 Hz to slightly above 1 Hz (1 rad/sec to 8 rad/sec), but
frequency alone does not determine whether an oscillation is a PIO.
DEFINITION #8: High-frequency, small-amplitude oscillations in pitch (sometimes referred to
as “pitch bobble”), and in roll (“roll ratchet”), may be considered a “mild” form of PIO, and may
not even be judged as PIO in all cases. If their amplitudes become intrusive on the piloting task,
they are PIOs.
DEFINITION #9: A PIO that interferes with, but does not prevent, performance of a primary
mission task is a “moderate” PIO; if a Cooper-Harper Handling Qualities Rating is obtained, it is
usually in the range of 4-6 (Level 2 by handling qualities specifications). “Moderate” PIO
requires corrective action for normal operation of the airplane, but if it occurs in developmental
testing the flight test program can continue.
DEFINITION #10: A PIO that prevents performance of the task, or that requires the pilot to
abandon the task to try to stop the oscillation, is a “severe” PIO; if a Cooper-Harper Handling
*
This technical report details the development of methods and devices intended to predict and
prevent the occurrence of pilot-induced (or pilot-in-the-loop) oscillations. This section is a
checklist summarizing the entire effort documented in the remainder of the report. It is presented
here for immediate access and to establish the direction that the rest of this report will take.
†
Specifics of failure transients and their effect on PIO are not covered in this report, except to the
extent that such transients may be considered triggers to the occurrence of PIO.

1
Qualities Rating is obtained, it is usually 7 or worse (Level 3 or unflyable by handling qualities
specifications). “Severe” PIO requires immediate changes to the airplane, and if it occurs in
developmental testing the flight test program should be postponed or redirected until the
corrections are made.

B. TEN STEPS

1. Be Prepared for PIO

Experience has clearly demonstrated that it is almost impossible to avoid PIO in some form
during the development process for any airplane. It is safe to say that at least one representative
of just about every flying vehicle has experienced some form of PIO, from the 1903 Wright
Flyer to the most modern transport airplane. Fortunately, the vast majority of these events are
mild in nature and the cures are easily found. Given the wide variety of possible conditions,
airplane designs, and triggers, it is practically impossible to make an airplane absolutely PIO-
proof for its entire lifetime. If PIO is possible, the best defense against “surprise” encounters is
to be prepared for the eventuality. This is especially important in a success-oriented
development program, where the unexpected occurrence of PIO can threaten to cripple the entire
project. Exploration for PIO should become a routine element in all phases of the development
of a new aircraft. (Section III.)

2. Design for PIO Resistance

This may seem like motherhood – after all, who is going to design for PIO susceptibility? – but
the goal is to assure that the aerodynamics, flight control system, effector sizes and actuators, and
cockpit control inceptors, are all specified with the prospects of PIO in mind. The bandwidth
requirements for a surface actuator, for example, should include provision for the rare pilot input
that can drive the actuator to its rate limit for an extended period of time: such an input can and
will occur, someday. If the program cannot afford the protection provided by an added margin
of safety on the actuator, a redesign of the flight control system should be explored to relieve
demands on the actuator. Otherwise, provision must be made early on to incorporate some other
form of protection, such as software rate limiting or a real-time detector. (Section IX.)

3. Apply Valid Prediction Criteria Early in the Design Process

As soon as the first set of aerodynamic derivatives is estimated, it should be possible to begin to
apply criteria. Full application will require knowledge not only of the unaugmented airframe,
but also of expected types and levels of augmentation, including, as early in the process as
possible, reasonable models of surface actuators and cockpit inceptor dynamics.
Because a fundamental hypothesis of this study was that the most effective method for
preventing PIOs is by design, a significant portion of the technical effort was devoted to finding
valid prediction criteria. Criteria recommended for application are based on pitch attitude and
flight path Bandwidth and pitch rate overshoot, using the parameters defined in Figure 1 and
Figure 2. It is preferable that the parameters always be measured with the dynamics of the
cockpit control feel system included. Since this is sometimes impractical – such as during
preliminary design, where the cockpit configuration has not been fully defined – the parameters
may also be measured with the feel system excluded. Two sets of criteria plots are presented.

2
The core of the criteria is a crossplot of angular attitude Bandwidth frequency versus Phase
Delay, as defined in Figure 1. Bandwidth measures the basic stability of the airplane and
determines the frequency range over which piloted control is possible with a minimum of pilot
equalization. Phase Delay measures the high-frequency phase loss if the pilot operates at high
frequencies.
For the pitch requirements, there are regions where PIO is unlikely on the basis of the attitude
Bandwidth characteristics alone. In some instances high pitch rate overshoot is a contributor,
and limits are placed on the frequency-domain-based metric, ∆G(q) (Figure 2). In others
inadequate flight path control is the culprit, so limits are placed on flight path Bandwidth
frequency, ωBWγ .

Phase delay:
∆Φ2ω180
τp =
57.3 (2ω180)

Rate response-types: ωBW is lesser of ωBWgain and ωBWphase

Attitude response-types: ωBW = ωBWphase

ωBWgain

θ GM = 6 dB
Fes dB

-90

Φ (deg)
ΦM = 45°
-180
ωBW
phase
ω180 ∆Φ2ω
180

2ω180
-270
Frequency (rad/sec)
(log scale)

Figure 1. Definitions of Pitch Attitude Bandwidth and Phase Delay (Flight Path
Bandwidth ωBWγ is Measured from γ/Fes and is Defined as ωBWphase )

3
 ∆G (q )
q
F es

(d B )

0. 0 1 0. 1 1 10

F re qu en cy, ω (rad /sec)

Figure 2. Definition of Pitch Rate Overshoot Parameter, ∆G(q)

Requirements on pitch attitude Bandwidth versus Phase Delay are presented in Figure 3 (feel
system included in the aircraft model) and in Figure 4 (feel system excluded).*
Roll requirements are given in Figure 5. They are not as comprehensive as those in pitch, in part
because of a lack of data and in part because of a lack of concern about severe lateral PIOs. No
requirements could be established for other axes at this time. (Sections V, VI, VII.)

4. Continue to Apply Criteria as the Accuracy of the Model Improves

There will be a natural increase in sophistication for the aerodynamics and control system
models; there should be a system in place for immediate application of the PIO prevention
criteria every time a certain milestone is met.

5. Use High-Gain Maneuvers to Evaluate PIO Tendency in Piloted Simulations

If ground-based simulation is used to evaluate the new vehicle’s characteristics, a minimum set
of pilot-in-the-loop, high-gain maneuvers must be evaluated. At this stage, any warnings of PIO
tendency by any pilot should be investigated. A minimum of five qualified evaluation pilots
should be used, if at all possible. The pilots must be aware of the concern for PIO and familiar
with the maneuvers to be flown. Typically such pilots are trained experimental test pilots from
one of the military pilot schools. Test pilots have been provided the opportunity to fly a wide
variety of aircraft and should be more capable of relating their exposure to the test airplane to
previous experiences in similar aircraft.

*
The lines dividng PIO boundaries on the figures are intentionally very wide. There is no clear
division between “no-PIO” and “PIO” and we want to emphasize this fact.

4
 Level 3
"S e v e re " P IO
P itch 0.2
A ttitude 0 .1 9 Leve l 2 "M ode rate" PIO if Flight Path B andw idth
P hase ωB W γ < 0.7 rad/sec
D elay, 0 .14
τp θ Level 2
L eve l 2
(sec) 0.1 "M o d era te " N o P IO Le ve l 1
P IO
[B o b b le N o P IO
if ∆G (q)
>12 dB if ∆G (q ) [L e v e l 2 /P itc h B o b b le if P itch R a te
>9 dB] O v e rsh o o t R a tio ∆G (q ) > 9 d B ]
0
0 1 2 3 4 5 6 7
P itch A ttitude B andw idth, ωB W θ (rad/sec)
a) F ig h ters, L an d in g

Level 3 L e ve l 2
P itch 0.2 "S e ve re " [L e v e l 3 /"S e v e re " P IO Level 1
if ωB W γ < 0 .5 8 ra d /s e c ]
A ttitude P IO [L e ve l 2 /"M o d e ra te " P IO if
P hase ωB W γ < 0 .5 8 ra d /s e c ]
D elay, 0 .14
τp θ
Level 2
(sec) 0.1 L eve l 2 L e ve l 1
"M o d e ra te "
N o P IO N o P IO
P IO [B o b ble
if ∆G (q ) if ∆G (q) [L e ve l 2 /B o b b le
>12 dB >9 d B ] if ∆G (q ) > 9 d B ]
0
0 1 2
P itch A ttitude B andw idth, ωB W θ (rad/sec)
b ) T ran sp o rts, L a nd in g

Level 3
"S e v e re " P IO
P itch 0.2
A ttitude 0 .1 9 L e v e l 2 "M o d e ra te " P IO if F lig h t P a th B a n d w id th
P hase ωB W γ < 0 .7 ra d /s e c
D elay, 0 .14
τp θ
Level 2
(sec) 0.1 L e vel 2
"M o de rate " N o P IO Level 1
P IO [B o b b le N o P IO
if ∆G (q )
> 12 dB
if ∆G (q ) [L e v e l 2 /B o b b le
>9 dB] if ∆G (q ) > 9 d B ]
0
0 1 2 3 4 5 6 7
P itch A ttitude B andw idth, ωB W θ (rad/sec)
c ) A ll C lasses, P itch Track ing

Figure 3. PIO Criteria for Pitch Response When the Dynamics of the Cockpit Control Feel
System are Included

5
 P itch 0.2 Level 3
A ttitude "S e v e re " P IO
P hase
D elay, 0 .14
τp θ Le vel 2 "M ode rate" P IO if Fligh t Pa th B and wid th
ωB W γ < 0.75 ra d/sec
(sec) 0.1
0 .09
Le vel 2 Level 2 L e v el 1
"M o d erate " N o PIO N o P IO
P IO if
∆G (q )
[Bo bble if [L e v e l 2 /P itch B o b b le if Pitc h R a te
> 12 d B ∆G (q )>9 dB ] O ve rs h o o t R a tio ∆G (q ) > 9 d B ]
0
0 1 2 3 4 5 6 7
P itch A ttitude B andw idth, ωB W θ (rad/sec)
a) F ig h ters, L an d in g

Level 3
P itch 0.2 "S e ve re " L e ve l 2
A ttitude P IO [L e ve l 3 /"S e v e re " P IO
P hase if ωB W γ < 0 .6 ra d /se c ]
D elay, Level 1
[L e v e l 2 /"M o d e ra te " P IO
τp θ
if ωB W γ < 0 .6 ra d /s e c ]
(sec) 0.1
0 .09
Le vel 2
Level 2 N o P IO L e ve l 1
"M o d e ra te " P IO [B o bb le N o P IO [L e v e l 2 /B o b b le
if ∆G (q )
if ∆G (q )> 1 2 d B >9 dB ] if ∆G (q ) > 9 d B ]
0
0 1 2
P itch A ttitude B andw idth, ωB W θ (rad/sec)
b ) T ran sp o rts, L a nd in g

P itch 0.2
A ttitude Level 3
P hase "S e ve re " P IO
D elay, 0 .14
τp θ
Level 2 ["M oderate" P IO if ωB W γ < 0 .75 rad/sec]
(sec) 0.1
0 .0 9
Le vel 2
"M o de rate " Level 2
P IO if Level 1
N o P IO [B obble if No PIO [Level 2/B obble
∆G (q )
> 12 dB
∆G (q)>9 dB] if ∆G (q) > 9 dB ]
0
0 1 2 3 4 5 6 7
P itch A ttitude B andw idth, ωB W θ (rad/sec)
b ) A ll C lasses, U p -an d-Aw ay P itc h T rack in g

Figure 4. PIO Criteria for Pitch Response When the Dynamics of the Cockpit Control Feel
System are Excluded

6
 Level 2 [Level 3/"S evere" P IO if
R oll 0.2 control/response sensitivity is too high]
A ttitud e
Ph ase 0.17 L eve l 2 ["M od erate" o r "S e ve re " PIO if co n tro l/resp on se sen sitivity is too h igh ]
D elay, 0 .14
τp φ
(se c) 0 .1 Level 2
[Lack of P IO data Level 1
in this region] N o P IO

0
0 1 2
R oll Attitu de B an dw id th, ωBW φ (ra d/s ec)
a) Feel S y stem In clu d ed

R oll 0.2
Attitude
Level 2 [Level 3/"S evere" P IO if
P has e
D e lay,
control/response sensitivity is too high]
τp φ 0.12
L eve l 2 ["M o de rate" or "Se vere " P IO if con tro l/res po nse se nsitivity is to o high ]
(sec) 0.1
0.0 9
Level 2
Level 1
[Lack of P IO data
N o P IO
in this region]
0
0 1 2
R oll A ttitude Ba nd w id th, ωB W φ (rad /s ec )

b ) F eel S y stem E x clu d ed

Figure 5. PIO Criteria for Roll Response, All Flight Conditions, All Aircraft Categories

There is always an initial reluctance to fly certain maneuvers because they are not “operationally
relevant,” or because “no experienced, trained pilot would ever do that in the air.” It must be
stressed that PIOs are also not operationally relevant, and unusually high-gain or large-amplitude
tasks are used in simulations because it is simply not possible to replicate every possible scenario
for PIO. Tasks should include, but not be limited to, attitude captures, precision landings, aerial
refueling (or close formation flight), and command tracking. Pilots should always be aware of
the potential for PIO when any task is flown, even if it is not intended to look specifically for
PIO. Engineers and pilots must work together in simulations, because it is possible that PIOs can
go undetected by the pilot.
Pilot opinion is critical. Use of a valid PIO tendency classification scale will greatly assist the
task of converting pilot opinion into engineering actions. A method for classifying pilot opinion
is presented in Figure 6.

7
 S everity D uration D em ands O verall
on pilot assessm ent
D a ngero us N ever sto pped C ouldn 't preven t it W hat airplan e?
(ab andon airplan e) (ab andon airplan e) (D e lta S ierra)

S eve re C ouldn 't preven t it Intole rable for

(aban don tas k) M o st of th e tim e (aban don ta sk) the ta sk
(fix the airplan e)

A llevia ted
M odera te b y techniq ue O bje ctiona ble
(can't ignore it) O cca sional (ta sk perform an ce (w arra nts
com p rom ise d) im provem en t)

P reven ted or elim ina ted Tolera ble

M ild O nly a v ery short tim e b y techniq ue (s atisfactory w itho ut
(can ignore it) (ta sk perform an ce im provem en t)
no t com p rom ise d)

N o ne N ever sa w one N o tenden cy to W hat P IO ?

induce oscillatio ns (S ie rra H otel)

Figure 6. PIO Tendency Classification Method

In general, based on recent experience with PIOs in piloted simulation, the following
observations can be made. For an airplane evaluated by several qualified experimental test pilots
in a particular flight condition/configuration/loading combination, using a well-defined high-gain
PIO assessment task:
• If the simulated airplane is judged by all of the pilots to have Level 1 or near Level 1
handling qualities (Cooper-Harper Handling Qualities Ratings (HQRs of 4 or better), and
is judged to have no PIO tendencies, PIO is not likely in the real airplane.
• If the simulated airplane is judged by one or more of the pilots to be Level 3 or worse
(HQR 7 or worse), and to have tendencies toward PIO, severe PIO is highly likely in the
real airplane.
• If the simulated airplane is judged by most or all of the pilots to be in between in HQR (4,
5, or 6), and to possibly exhibit some tendency to PIO, the simulation cannot be
considered conclusive. In this case, engineering expertise and quantitative analysis using
PIO-detection methods must be applied to the simulation data. Absence of evidence of
PIO will indicate a reasonably safe airplane for flight testing.
It must be emphasized that the extrapolation from simulation to flight will be only as good as the
simulation model. Significant errors in the model, or significant nonlinearities in the airplane not
incorporated into the model, can eliminate any value to the simulation results. (Sections VIII
and IX.)

6. Apply PIO Detection and Prevention Devices During Developmental Flight Testing
As with simulation, flight testing must include maneuvers intended specifically to look for PIO.
In addition, reliable real-time PIO detection devices – either onboard or operated remotely via
telemetry – will help monitor the occurrence of PIOs. Such devices may include active
intervention to prevent or recover from PIOs. (Section IX.)

8
7. Extend Test Inputs and Application of Criteria to Large Input Amplitudes
The fundamental theory behind Bandwidth is that it is a measure of piloted closed-loop activity,
and hence is most effective for describing small-amplitude control inputs. There is a natural
reduction in Bandwidth for any physical system as input amplitude increases beyond a certain
value, resulting from limitations of the airplane, limiting on actuator rates and positions, etc.
Still, experience has shown that the Bandwidth criteria defined above are very effective at
predicting PIO susceptibility for quite large inputs. If any of the PIO-susceptible regions is
reached for a reasonable input size, PIO is likely. Frequency sweeping should emphasize input
amplitudes that result in aircraft responses at and above the Bandwidth frequency that approach
the operational limits for the aircraft. The data obtained in such sweeps will both enhance the
fidelity of simulation models and help prevent large-amplitude PIOs. (Section VI.)

8. Update Ground Simulation Models With Flight Data

This is a step that is always desired in a developmental program, but experience has shown that it
is not always done, or at least not in a timely manner. It should be possible to continue to make
use of ground simulation to search for PIO, but the simulator is only as good as the model. A
regular process must be implemented to keep the simulation model as up-to-date as possible, and
regular sessions should be scheduled to look for PIO tendencies with the updated model.

9. Include PIO Recognition as a Part of the Training Syllabus for Pilots

Whether the aircraft is commercial or military, there is always a potential for the occurrence of
PIO in follow-on flight testing or operational use. This may be as a result of a design flaw, an
excursion into untested flight conditions or loadings, or following a failure. It is not likely that
the typical fleet pilot will encounter PIO very often, and perhaps never. Pilots who are aware of
the characteristics of PIO, however, are much more prepared for dealing with the event, and for
accurately reporting it to cognizant agencies. (Section V.)

10. Be Prepared for PIO

See step 1. If there is one overwhelming recommendation that can be made, it is that all parties
involved in the development of a new aircraft must always be prepared for the occurrence of
PIO. It should not come as a complete surprise.

9
 II. INTRODUCTION

A. THE PHENOMENON
Pilot-induced oscillations (PIO) are “rare, unexpected, and unintended excursions in aircraft
attitude and flight path caused by anomalous interactions between the aircraft and pilot.”1 Most
commonly, PIO occurs as a result of inappropriate flight control system (FCS) design, such as
excessive filtering or lags, or from a nonlinear event, such as saturation of rate or position limits
on a surface actuator. The nature of most PIOs is such that the airplane up until the onset of the
oscillation is stable and seemingly well-behaved; encounter with some form of “trigger” leads
the pilot into a situation where the closed-loop, pilot-vehicle system is neutrally damped or
unstable (Figure 7).
PIOs have occurred during the development process for almost every new airplane (see Table 1).
Frequently the severity of the oscillations is sufficiently low that the PIO is detected and fixes
applied to the airplane’s FCS with little or no public acknowledgment of the event.
Occasionally, however, either the severity, the frequency of occurrence, or the consequences of
the PIO are such that it becomes headline news (for example, the YF-22 crash2).
In the case of the YF-22, the pilot was making low approaches over the runway at Edwards AFB.
The event occurred on the second approach; the first approach, performed with partial power,
was uneventful (Figure 7a), and the conditions were almost identical for the second, except the
pilot applied full power shortly before retracting the landing gear (Figure 7b). With gear
retraction, thrust-vectoring nozzles became active, and a gain changer, in combination with the
application of full power, resulted in excessive control sensitivity. Both the stabilizer and
nozzles became rate-limited, and the ensuing PIO led to a controlled crash on the runway about
eight seconds after the event began.

Table 1. Some Aircraft that Have Experienced PIOs

Transports,
Experimental
Fighters, Small Aircraft Bombers, Large
Aircraft
Aircraft
XF-89A F-111 C-97 XS-1
Mirage YF-16 B-52G X-5
A4D-2 MRCA KC-135A X-15
F8U-1 Tornado B-58 M2-F2
F-101B F/A-18A YF-12A Shuttle
F-104B F-14A A-320 AD-1
YF-5A JAS-39 MD-11
T-38A YF-22A C-17
F-4 V-22 B-2
Boeing 777

10
 20 20

pitch rate (deg/sec)

pitch rate (deg/sec)

10 10
0 0
-10 25 30 35 40 -10 30 35 40 45
-20 -20
-30 -30

30 30
long. stick (lbs)

long. stick (lbs)

20 20
10 10
0 0
-10 25 30 35 40 -10 30 35 40 45
-20 -20

100 100
80 80
power (%)

power (%)
60 60
40 40
20 20
0 0
25 30 35 40 30 35 40 45

30 Stab. 30 Stab.

Sfc angle (deg)

sfc angle (deg)

20 Nozzle 20 Nozzle
10 10
0 0
-10 25 30 35 40 -10 30 35 40 45
-20 time (sec) -20 time (sec)
-30 -30
Gear up cmd Gear up cmd Impact w/ rwy

a) First pass (partial power) b) PIO (full power)

Figure 7. Example Time History of a PIO (YF-22)

Despite the ubiquitous nature of PIOs, it is also true that they always come as a surprise to the
pilot and to the developers of the airplane. Typically, after a PIO has been encountered in flight,
an intense research effort is undertaken to determine the causes of the event and to understand
why the tendency to PIO could have gone undiscovered for so long. In the case of the YF-22,
among the findings of an accident review board were the need for application of analytical
criteria throughout the development process, and the requirement for high-gain, closed-loop
tracking tasks for evaluation of PIO susceptibility 3.

B. THE ELEMENTS OF PIO

In a global sense, the players in the development of PIOs may be divided into three elements:1
the airplane, the pilot, and the trigger. In this report it is assumed that prevention of PIOs
requires a focus on the airplane. We can train pilots to recognize certain characteristics of PIO
onset, and we can attempt to discover and find ways to defuse possible triggers, but neither
element is as easy to reliably measure, control, and correct than the airplane.

1. The Airplane
For the purposes of this discussion, consider the airplane as consisting of three separate
elements: the linear augmented airplane, nonlinear components such as control surface
actuators, and the cockpit control feel system.

11
a. The Dynamics of the Linear Airplane
Experience has shown that the dynamics of the augmented linear airplane are usually major
players in the occurrence of PIOs. Experimental data shows that very low actuator rates on an
otherwise good airplane can lead to PIOs, but in practice, we have seldom had PIOs occur on
“good” airplanes. The dynamics of the linear airplane are what the flying qualities
specifications4 are intended to address. Whether the cockpit feel system should be included in
these dynamics is examined below.
With full-authority augmentation systems, it has become relatively straightforward to provide
modern airplanes with a reasonable level of basic short-term dynamics, in the absence of
significant time delays or lags. With advanced control systems, there are three forms of added
complexity that serve to degrade the flying qualities of the augmented airplane: 1) time delays
from computer frame time and sample rate, resulting in a rolloff in phase angle between control
input and airplane response at high frequencies; 2) filters on the stick, for bending modes, etc.,
that also contribute to phase rolloff; and 3) unavoidable lags, such as those due to the actuator,
that also introduce phase rolloff.
The conclusion to be drawn here is that, for the vast majority of modern airplanes, a key
contributor to PIOs is high-frequency phase rolloff, from both linear and nonlinear elements. An
example of the effects of this phase rolloff is shown in Figure 8, where two aircraft with very
similar pitch attitude responses up to the Bandwidth frequency received very different Handling
Qualities Ratings (HQRs) because of the high-frequency differences.

Kp
1
6 dB Kp
2

C onfig . 1D :
φM = 4 5° τp = 0.01 se c
φM = 30°
H Q R = 4.5,5/3,4
φM = 0° PIO R = 2 ,2 .5/1,2

C onfig . 2I:
τp = 0.17 se c
H Q R = 8/8
PIO R = 4 .5 /4

Figure 8. Effect of High-Frequency Phase Rolloff on PIO

12
To be successful, any criterion used to predict PIOs must be sensitive to this high-frequency
phase rolloff in aircraft response. It was the motivation for developing the "phase delay"
parameter in the Bandwidth criteria,5 and it is the phenomenon that is captured (with varying
degrees of success) by equivalent time delay when equivalent-systems criteria are applied.6 The
phase rate parameter developed by Gibson in Great Britain7 also defines the effects of phase
rolloff; when measured over the same frequencies, phase rate is identical in form to phase delay.
Any criterion that does not directly address the high-frequency rolloff of phase angle must be
considered incomplete. (It is important to note that phase delay is not a measure of time delay
per se. It is a measure of the shape of the phase response of angular attitude to control inputs
above the neutral-stability frequency, of which time delay is only one factor.)
When properly defined and applied, conventional flying qualities criteria are sufficient to
identify the expected Level of flying qualities for the linear augmented airplane. Further
determination of PIO potential, and the type and severity of the PIO, requires new criteria, or a
refinement to existing ones. This is the topic of the majority of this report.

b. The Control Surface Actuator

Any physical object has mass and inertia. As a consequence there must be some limit to the
capabilities of the system to generate accelerations, rates, and positions, no matter how hard the
system is pushed. With the use of high gains and small control surfaces, great demands are
placed on the airplane's surface actuator to achieve as rapid a response as possible. Sometimes,
there will be an incompatibility between the demands of the control system and the capability of
the actuator, and a rate limit will result.
It seems to be true that most recent PIOs have exhibited actuator rate limiting. It is also true that
many of these PIOs have been with aircraft that have marginal to poor flying qualities to begin
with. Therefore, it is not possible to determine if rate limiting of the actuator is the cause of the
PIO, or simply an effect of it.
Clearly more work is justified in this area. It should be possible to define the response
requirements for the actuator, in terms of handling qualities, that will protect against PIO. For
now, their linear characteristics are included as a part of the linear airplane dynamics.

c. The Cockpit Control Feel System

Considerable work has been performed to determine the impact of the dynamics of the cockpit
control feel (force/deflection) system on handling qualities. For handling qualities purposes, the
assumption has historically been that the feel system is simply one of the elements in the
dynamics of the linear airplane. Most handling-qualities requirements4 are written in terms of
responses to control force inputs. Therefore, it is assumed that when the command sensing for
the aircraft is based on control position (feel system in series), the feel system should be included
as a part of the airplane; when the airplane responds to control force (feel system in parallel), it
should be ignored.
In a brief NASA experiment, Smith and Sarrafian8 found that the effects on handling qualities of
the feel system were not equivalent to the effects of an added pure time delay. The researchers
concluded that the feel system is a unique dynamic element that is not addressed by time-delay
requirements in the military specifications. This sparked considerable debate on how to deal
with the feel system in future handling qualities criteria and requirements.

13
A follow-on study9 was performed by Calspan using the variable-stability NT-33A with a greatly
expanded matrix of configurations, tasks, and pilots. The results of this study do not directly
refute the conclusions drawn by Smith and Sarrafian. Instead, they augment the observation that
the feel system is a unique dynamic element and does not impact handling qualities to the extent
of an equivalent time delay.
The conclusions from a detailed analysis 10 of the Calspan flight test data, augmented with other
data from helicopters and ground based simulators, may be summarized as follows. The feel
system lags usually result in a degradation in pilot ratings. Some pilots, for some tasks, are more
sensitive to the feel system dynamics than others, and run-to-run variability has even been
observed for a single pilot. The mechanism of how a pilot copes with feel systems lags is not
well understood. It has been suggested that the pilot can close a separate, high gain loop around
the cockpit manipulator (if it is position sensing), and therefore override feel system lags. The
ability of the pilot to override the feel system is enhanced if the inertia of the system is low. The
above noted variability suggests that overriding the feel system requires a conscious effort that
may not always be applied, especially in conditions of high workload. Hence we may
hypothesize that in a PIO situation, it is unlikely that the pilot will override feel system lags.
Clearly more experimental data is required to better understand the ability of the pilot to override
the feel system. Until such data becomes available, it seems prudent to include the feel system
lags in calculating the parameters for PIO criteria. For force sensing feel systems (e.g., the F-16
and the MD-11), the pilot cannot override the feel system and it must be included.
Some researchers have concluded that the feel system should not be included in calculating PIO
criteria, based primarily on empirical observations. That is, the data fit the criterion boundaries
better if the feel system is ignored. This result is in conflict with the analysis discussed above,
and suggests that pilots always override the feel system lags. It is more likely that the criterion
boundaries need to be refined.

2. The Pilot
The dynamics of the human operator in manual closed-loop control are well-known from the
works of researchers at Systems Technology, Inc.11 There is much less certainty about the pilot's
characteristics during a PIO, however. A model based on the "synchronous" pilot has been
proposed.12 Essentially this model assumes that the pilot develops no compensation and exhibits
no effective lags or delays, instead reacting synchronously with visual information to generate a
sinusoidal response. This model has been disputed by R. H. Smith,13,14 among others, who
usually suggest that the pilot reacts more to the sensation of linear accelerations, or to a
combination of visual and motion information depending upon a number of factors.
While it is not possible to conclusively determine the pilot’s behavior in a PIO, the results of
research documented in this report (Section IV) will show that there may be times when the pilot
is nearly synchronous, and times when the pilot is compensatory. This issue was one focus of
the present study.

3. The Trigger
A key catalyst in the occurrence of a PIO is a trigger mechanism. Typically this is a requirement
for a rapid change in the pilot's control strategy during an otherwise normal task. For the
precision landing, for example, it can be a sudden gust, a wind shear, or the demands of the
landing flare itself, as the pilot attempts to touch down within a target zone. It is important to

14
recognize that the trigger does not cause the PIO: bad flying qualities, or rate saturation, or mode
switching, are to blame. For airplanes that have flying qualities that are acceptable for otherwise
benign maneuvering, a trigger is necessary simply to force the pilot into the necessary control
behavior to induce an oscillation.
The form of the trigger necessary to induce PIO has proven difficult to define. What works in a
real-world situation might never have the same effect in a constrained environment. For certain,
we know that triggers in flight will be different from those in a simulator. There is a natural
change in piloting technique in the simulator that can make an obvious trigger ineffective at
initiating PIO. Research into the characteristics of PIO triggers is basically nil. It is probable
that the triggers introduced into any development program will be unique in format to that
program. This almost assures that we will never be able to quantify and classify every possible
PIO-inducing trigger.
It has been assumed in this study that triggering mechanisms will always be present, and that
attempts to document them will be futile. Instead, our goal has been to determine ways to
prevent the airplane from being susceptible to PIO when the trigger is encountered.

4. Other Contributors
For most PIOs encountered by modern aircraft, there are several other contributing factors that
increased the propensity for handling qualities problems. These factors, therefore, are
contributors to the flying qualities of the airplane, but their effects are not generally sufficiently
well known to include them as part of the airplane. Included in this category are unconventional
inceptors (sidesticks, for example) with unusual control/deflection characteristics, force sensing,
etc.; and control/response gradients that are excessive or excessively nonlinear with input
amplitude. Many of these contributors are addressed by MIL-STD-1797A, but what constitutes
good and bad characteristics are not thoroughly defined by the standard.
There is a small amount of data to at least investigate the significance of some of these
contributors to PIO. For example, stick force per g is addressed through a limited data base and
a handful of criteria. In general, it is suspected that non-optimum controller and control/response
characteristics have played a part in many PIOs, but our primary source of data – flight research
with variable-stability airplanes – is minimal. This is because most such experiments keep the
controller characteristics fixed, so if those characteristics are not ideal they will have an
immeasurable influence on all piloted evaluations.
In Sections V and VI, we will look at the sparse data in an attempt to at least identify general
trends. Effects of controller characteristics and control/response gearing (including both
sensitivity and linearity of response) on PIO remain poorly-documented elements that deserve
further study.

C. PIO IN THE DEVELOPMENT PROCESS

Success-oriented development programs are inherently structured to produce PIO-free, mission-
effective airplanes starting with the first test article. Flight test organizations, such as those at
Edwards AFB and NAS Patuxent River, have come to expect that every new prototype will
exhibit some tendency to PIO. Clearly, there is a disconnect.
There is an obvious need for a fundamental shift in philosophy about the occurrence of PIOs.
Rather than react with surprise if a PIO is encountered late in full-scale development, we should

15
be prepared for the eventuality and plan for it from the initial design stages. This includes
guidance (prior to the current Acquisition Reform, these would have been requirements) in the
first request for proposals (RFPs), through preliminary design, and into flight testing. Flying
qualities guide specifications contain criteria that will help reduce the propensity for PIO, even if
they would not prevent it altogether.

D. ORIGINS OF THIS RESEARCH AND RELATED WORK

All of the recent research into the causes of PIO can be traced directly to several well-publicized
events in the early 1990s – the crash of the prototype YF-22A, two crashes of Sweden’s SAAB
JAS 39 Gripen, and reported PIOs on the Air Force C-17 during developmental testing.
PIO became a politically charged issue. The atmosphere within and between certain government
and contractor organizations sometimes led to distrust. There are factions that believe that
almost every airplane flying today is susceptible to PIO, and experiences them on a regular basis.
There are others who feel that real “PIOs” are extremely uncommon events.
Several parallel thrusts began in the mid-1990s. In 1995, the Aeronautics and Space Engineering
Board of the National Research Council formed a Committee on the Effects of Aircraft-Pilot
Coupling on Flight Safety. The Committee’s report was published in book form in 1997 and is
excellent reading.1 The focus of the Committee, however, was to understand the phenomenon
and provide an overview of possible solutions. As such, there is little in the way of detailed
analysis of such events, or any strong efforts to define the best methods and devices for
predicting and preventing PIO.
Also in the mid-1990s, the Air Force began an effort to develop a “Unified PIO Theory.” Under
Air Force funding, four U.S. contractors conducted independent research studies for the Unified
PIO Theory effort. They consisted of Systems Technology, Inc.;15 McDonnell Douglas
Aerospace (now Boeing Phantom Works), with Hoh Aeronautics, Inc., as a subcontractor;16
Virginia Tech;17 and Calspan Corp. (now Veridian). 18 The U.S. Air Force also provided funding
to several Russian research organizations – the Moscow Aviation Institute, Central
Aerohydrodynamic Institute, and Gromov Flight Research Institute – to conduct PIO research.19
As a parallel in-house effort, researchers at Wright-Patterson AFB performed several simulation
experiments to investigate the effectiveness of ground-based simulation for predicting PIO.20
New flight-based PIO data have been generated under Air Force funding by both the Air Force
Test Pilot School21 and Veridian.22
Smaller-scale research programs have been performed by such U.S. organizations as NASA
Ames Research Center23 and the University of California, Davis. 24
The authors of this report have been able to take full advantage of the work done by the
foregoing, along with many others in both the U.S. and abroad. Elements of this report can be
traced to many of the referenced sources.
This report summarizes the work performed by Hoh Aeronautics, Inc. (HAI) under the
sponsorship of the US Air Force Research Laboratory through a Small Business Innovation
Research contract. It culminates a total of five years’ worth of effort (including the initial Phase
I effort, through the end of Phase II) toward developing a unified method for predicting and
preventing the occurrence of pilot-induced oscillations (PIO).

16
Throughout the last five years, HAI has actively participated in numerous flight and ground
simulation experiments, research studies, incident investigations, and training sessions, in an
attempt to obtain the most concise and comprehensive data base possible. We have been able to
design experiments, interview pilots and engineers, access proprietary data, and perform indepth
analysis of related data. This report is the culmination of that work.

E. THE FOCUS OF THIS RESEARCH

We have operated under two assumptions: 1) the best way to prevent PIOs is from the first
preliminary design, rather than through fixes after first flight; and 2) PIOs can often be related to
poor handling qualities, and hence are amenable to similar forms of analysis.

1. Prevention of PIOs by Design

It is safe to say that the designers of every modern airplane in use today have paid attention to
the potential for PIO. PIOs seem to occur with alarming regularity during the early
developmental flight testing for just about any class, size, or type of aircraft. Fortunately, most
such events are mild and the fixes easily found.
It has also become clear that there is a lack of trustworthy PIO assessment criteria, and that the
users of these criteria have (rightly or wrongly) come to doubt the veracity of existing handling-
qualities criteria for preventing PIO. The consequence has typically been that, rather than
attempting to use several criteria in early design, none are applied, or at least with any weight.
In truth, most classical handling-qualities criteria4 have not explicitly addressed PIO, and
especially the most common form of PIO in modern operational aircraft – PIO due to some form
of rate limiting.
The first defense against catastrophic PIO is criteria. But these criteria must be valid and must
apply to both linear and nonlinear phenomena. There is always a risk that design-oriented
criteria that are conservative will lead to overdesign, resulting in increased weight, cost, or
schedule, or in reduced performance. Any criterion, no matter how valid, can be misused or
misinterpreted as well.

2. PIOs and Handling Qualities

We have tried to keep the focus of this research on PIO, but it has of necessity included aircraft
handling qualities as well. If it is hypothesized that PIOs can be a consequence of deficiencies in
handling qualities, it must be concluded that the mere existence of a handling qualities
specification is not sufficient to prevent PIO.
Experience has shown that schedule and cost constraints nearly always result in compromises
during the development of the flight control systems of modern aircraft. These compromises
typically result in degradations in aircraft handling qualities, and therefore failure to meet the
stringent Level 1 limits. Most (if not all) current highly-augmented aircraft do not meet all of the
Level 1 limits of MIL-STD-1797A for all flight conditions. Some modern airplanes are not
considered to be Level 1 when evaluated in the context of high-workload, high-demand
handling-qualities tasks,25,26,27 even though these airplanes may have proven to be perfectly
acceptable for performing their intended missions. Typical problems are slow actuators, stick
filters necessitated by the use of force transducers, bending mode filters, anti-aliasing filters,

17
computational time delay, excessive parallel integrator gains, and inappropriate flight control
system architecture.
The problem is amplified by the fact that the use of active control technology often results in a
significant modification of bare airframe response characteristics. For example, the
unaugmented X-29A doubled pitch amplitude in well under one second, and was completely
uncontrollable by the pilot. The augmented aircraft was well-damped with a crisp response
characteristic. This requires control power. Control power has classically been achieved with
increased control surface sizing. This is, however, in direct conflict with the primary
performance objectives that led to active control design in the first place. The result is smaller
surfaces that must move very rapidly, leading to actuator rate limiting as an inherent problem in
the design. In that context, it is not surprising that we have seen a strong tendency for PIO in
highly-augmented aircraft. On the other hand, strict adherence to the flying qualities
specification could result in excessively large control surfaces. Clearly, we need criteria that
allow the tradeoff between performance and handling qualities to be made with a firm
understanding of the inviolable PIO limits.
With modern flight control systems it is possible to achieve Level 1 handling qualities
throughout more of the flight envelope. The augmented airplane can perform better, provide
more carefree handling, exhibit completely docile characteristics – until a limit is reached
somewhere in the flight control system. The airplane can almost instantly become
uncontrollable, entering a rapidly divergent PIO. Thus, what was moments before a Level 1
airplane has degenerated into a Level 3 (or worse) airplane.
With little or no augmentation, and with no inherent nonlinearity from surface or software rate or
position limiting, the pilot would have direct feedback of the shortcomings of the design. If
short-period damping or stick force per g were low, for example, the experienced pilot will
recognize the potential for PIO. But if the airplane seems to behave exceedingly well, the
occurrence of a sudden, large-amplitude PIO can come as a complete surprise. This form of
event is not easily addressed by classical flying qualities requirements, since the augmented
airplane may easily surpass those requirements. This is one reason why we must look at PIO as
a phenomenon separate from handling qualities alone.

3. Pilot Opinion, Handling Qualities, and PIO

Like it or not, we are dependent upon the opinion of pilots to determine the propensity for PIO
for any particular airplane. Ideally we will find a method to quantify all elements of PIO – their
likelihood to occur, their severity if they do occur, methods for recovery – independently of the
pilot. So far, however, the best source for information about PIOs is the pilot.
In flight research, especially, we are accustomed to receiving piloted evaluations of aircraft
handling qualities entirely by pilot opinion. In truth, the best pilot opinion is personal comments
and impressions. Unfortunately, engineers always seek to quantify everything. The most
common method for quantifying pilot opinion is by use of the 10-point Cooper-Harper Handling
Qualities Rating (HQR) scale, Figure 9.
Similarly, we rely on PIO information from pilot opinion expressed through a rating scale such
as the six-point PIO Tendency Classification Rating (PIOR) scale, shown in Figure 10. Again,
the most important information the pilot can give is comments, but the scale (discussed in detail
in Section VIII) provides a method for quantifying the severity of the PIO.

18
 Excellent Pilot com pensation not a factor for
Highly desirable desired performance 1
G ood Pilot compensation not a factor for
Negligible deficiencies desired perform ance 2
Fair - Som e mildly M inim al pilot com pensation required for
unpleasant deficiencies desired performance 3

Minor but annoying Desired performance requires moderate

deficiencies pilot compensation 4
Is it Deficiencies
satisfactory without M oderately objectionable Adequate performance requires
improvement ?
warrant deficiencies considerable pilot compensation 5
im provement
Very objectionable but Adequate performance requires
tolerable deficiencies extensive pilot com pensation 6

Adequate performance not attainable with

Major deficiencies m aximum tolerable pilot compensation. 7
Is adequate Controllability not in question
performance Deficiencies
Considerable pilot com pensation is required
attainable with a tolerable require M ajor deficiencies
for control 8
pilot workload? improvement
Intense pilot compensation is required to
Major deficiencies
retain control 9

Is it Improvement Control will be lost during some portion

Major deficiencies of required operation 10
controllable? mandatory

pilot decisions

Figure 9. The Cooper-Harper Handling Qualities Rating (HQR) Scale28


1
No
2
No
Is
Do
Ye s Tas k
U n d e sira b le
P e rfo rm a n ce
M o tion s Ten d to
C o m pro m is e d?
O ccu r?

Ye s
No 3
No 4
Ye s
C au se s
O scillatio n s D ive rg en t
Ye s
5

P ilot In itiate d
A bru p t M a ne uv ers
or
Tigh t C o ntrol
No

C au se s Ye s
D ive rge nt 6
O scilla tion

P ilo t A tte m p ts
to E nte r C o ntro l
Lo o p

Figure 10. The Current PIO Tendency Rating (PIOR) Scale, 1980

19
If we’re lucky, we can glean information about the occurrence and severity of PIO based solely
on the HQR and PIOR numbers. But these ratings are not the whole story, and in fact, the piece
of the story told by the combination of HQR and PIOR might not even be the right answer. As
an example, consider the general mapping of HQRs and PIORs suggested by Figure 11.
The interpretation of combinations of HQR and PIOR, at least as far as PIO is concerned, allows
for considerable ambiguity. As suggested by the unfilled areas of Figure 11, there are only a few
combinations of ratings that can be clearly interpreted: any HQR with a PIOR of 1 (no PIO, no
tendency for undesirable motions of any kind), and HQRs of 7 or worse with PIORs of 5 or
worse (severe PIO, dangerous to unflyable airplane). Ambiguities arise, by contrast, when
PIORs indicate some tendency to PIO, but the HQRs are only moderately bad (suggesting PIO
that could be tolerable) or very bad (suggesting the PIO itself could be only mild, and something
else is driving the airplane to have such poor handling qualities). Ultimately, we must always
consult pilot comments – a rational move, but one that is not as satisfying, or as easy to present
to upper management, as a series of numbers.
This ambiguity is just one more signal of the need to address PIO separately from basic flying
qualities – recognizing that there is still a high probability that the two are related nonetheless.

F. ORGANIZATION OF THE REPORT

The layout of this report is intended to make it of value to the manager, the novice engineer, and
the highly experienced professional. The sections increase in technical complexity with
extensive discussions of the key issues involved in prediction and prevention of PIO.

N o H Q R /P IO R com bin atio n a m b iguity

H Q R /P IO R com bina tion no t likely
K ey to shad ing:
H Q R /P IO R com bina tion no t possib le
H Q R /P IO R com bina tion am big uous
6
R e g io n n o t like ly
(L e ve l 2 w ith "S e ve re " P IO "S e ve re " P IO
R egion d ive rg e n t P IO ) (L e ve l 3 ) (U n co n tro lla b le )
5 no t possib le "M ild ,"
"M ild ," "M o d e ra te ."
(Level 1 + P IO ) "M ild " o r "M o d e ra te " P IO
o r "S e ve re " P IO
"M o d e ra te ."
Pilot- (L e ve l 2 ) o r "S e ve re " P IO
(L e v e l 3 ) (U n c o n tro lla b le )
In duced 4
O scillation R e g io n
P itch b o b b le
n o t lik e ly P itc h b o b b le o r "M ild " P IO
Tende ncy 3 (L e ve l 1 w ith (L e v e l 2 ,u n d e s ira b le o r "M ild " P IO U n co n -
R atin g u n d e sira b le m o tio n s) (L e ve l 3 ,u n d e s ira b le tro lla b le
m o tio n s) m o tio n s)
(P IO R )
2 D e sire d A d e q u a te U n a cc e p ta b le
(L e ve l 1 , (L e ve l 2 , (L e ve l 3 , Uncon-
n o u n d e s ira b le n o u n d e s ira b le n o u n d e sira b le tro lla b le
m o tio n s) m o tio n s) m o tio n s )
1
1 2 3 4 5 6 7 8 9 10
C oope r-H arper H and lin g Q ualities R ating (H Q R )
Figure 11. Correspondence Between HQR and PIOR

20
Section I was effectively an executive summary, consisting of definitions to isolate what we
mean by “PIO,” and steps to reduce its occurrence. It should be read by both the casual reader
and the serious researcher, as it sets the theme for the rest of this report.
Section III defines more precisely what we mean by PIO. There are many differing definitions
of “PIO;’ in Section III summary data from many documented occurrences are analyzed to help
narrow the focus of this report.
Section IV gets into technical details. It presents the results of analysis of pilot/vehicle
characteristics in PIO. Fundamental questions of pilot behavior in the PIO are addressed.
Section V reviews the different well-established PIO assessment criteria for linear (Category I1)
events and analyzes the PIO data base with these criteria.
Section VI extends pitch criteria to PIOs resulting from rate limiting (Category II PIOs).
Section VII briefly describes some criteria for roll. Because both the flight research data base,
and our real-world experiences, with lateral PIOs is quite limited, this section is quite short.
Section VIII discusses all aspects of pilot-in-the-loop testing for PIO, including maneuvers and
tasks for evaluating the potential for PIO. These maneuvers are not always the same as those
applied for assessment of handling qualities,29 though there are occasional overlaps. Issues with
the gathering of pilot opinion, through pilot rating scales, are addressed in some detail. A
proposed new method for classifying PIOs is introduced.
Section IX makes the case for on-board detection and prevention devices even if the new
airplane has been designed with every effort to avoid PIO. The truth is that PIO will never be
prevented entirely – there will always be a possibility for some combination of factors, such as
flight outside the normal operating envelope, flight with failures, inadvertent pilot inputs, etc., to
lead to an unexpected PIO. Several methods for looking for PIO have been devised and are
reviewed in Section VIII, including a method developed under this contract.
Section X wraps up the report with final observations and recommendations (there are no
“conclusions” per se in Section X – they essentially comprise Section I).
Appendices contain supporting data generated under this contract by HAI and by the US
government. Included in this research was a piloted simulation on the Air Force’s five-degree-
of-freedom Large Amplitude Multimode Aerospace Research Simulator (LAMARS), at Wright-
Patterson AFB, Ohio. The results of the simulation, except for transcribed pilot comments, are
documented in Appendix A; in the interest of size, the condensed pilot comments (comprising
over 100 printed pages!) have not been included but are available on request from either the Air
Force or the authors of this report. Anyone with an interest in analyzing the results of the
simulation is strongly encouraged to request an electronic copy of the pilot comments.
During this effort, three primary subcontractors were employed to assist with data collation and
analysis. Lockheed Martin Tactical Aircraft Systems, Ft. Worth, TX, provided analysis of PIO
tendencies for the F-16 with nonstandard US loadings. Data for the YF-22A and roll ratchet data
for the F-16 were also provided by LMTAS. Northrop Grumman Military Aircraft Systems,
Pico Rivera, CA, provided an analysis of the PIO susceptibility of the YF-23A and the B-2,
along with flight data for the B-2. Finally, McDonnell Douglas Aerospace (now The Boeing
Company), Phantom Works, Long Beach, CA, performed an independent review of the
applicability of equivalent-systems flying qualities criteria for the prediction of PIOs.

21
 III. DEFINING AND CATEGORIZING PIO

A. BACKGROUND
Pilot-induced oscillations are not new. Even the original 1903 Wright Flyer was susceptible to
PIOs, and it seems that at least once in every new aircraft development program a report of PIO
problems surfaces.
At the start of this research effort, a fundamental question was asked: “What is a PIO?” That is,
what are the fundamental characteristics that define a PIO and that differentiate it from any other
phenomenon? The truth is, after more than five years’ research, and following almost a century
of exposure to PIO, we still do not have a single, concise definition of PIO. There are no
universally accepted rules that identify a specific set of time traces as PIO.
Despite this, we certainly have a much clearer picture of the general nature of PIOs – their
frequency of occurrence, their measurable parameters such as frequency and amplitude of
oscillation, etc. From the assembly of an extensive catalog of PIOs, we can draw certain
commonalities that help us isolate what PIO is as well as what it is not.
This section of the report summarizes the characteristics of PIOs that have been documented and
reviews the different types (or Categories) of PIO. The section concludes with some
recommended definitions for PIO.

B. WHAT IS A PIO?
The Department of Defense Interface Standard for Flying Qualities of Piloted Airplanes, MIL-
STD-1797A,4 contains a concise definition of PIO: it consists of “sustained or uncontrollable
oscillations resulting from efforts of the pilot to control the aircraft.” It has been suggested that
the word “unintentional” be added before “sustained,” to distinguish from intentional oscillatory
behavior.
Taken literally, the MIL-STD-1797A definition means that any oscillation that occurs during
manual, piloted control may be classified as a PIO. Yet many times this oscillation is nothing
more than a result of pilot overcontrol in an otherwise normal circumstance. For example, to the
outsider the typical ballooning in flight path that any student pilot encounters during landing
training may appear to be a PIO. Yet this ballooning is simply part of standard pilot
compensation and is usually no more than one or two cycles, with no threat of developing into a
real PIO. Indeed, visual inspection of the time history records from even the experienced pilot in
the landing flare with a known good airplane will reveal small corrections that might appear to
be signs of a PIO. These are not what MIL-STD-1797A is referring to.
Residual oscillations that continue even if the pilot is no longer making an effort to control the
aircraft are also not PIO. This distinction can become fuzzy if the cause of the residual
oscillations can lead to a PIO. For example, failure of a pitch damper could result in an effective
short period mode with zero damping; this will most definitely be difficult to control and will
lead to PIOs, and the airplane will continue to oscillate after the pilot has released the stick.
Technically, the airplane experiences PIO when the pilot is flying and undamped motions when
the pilot is not flying.

22
Many of the PIOs recorded in older (1950s and earlier vintage) aircraft are traceable directly to
low inherent damping of the short period or dutch roll. We therefore need to exercise the
definition of the military standard and emphasize that we are only interested in such PIOs when
there is clear evidence that they result from "efforts of the pilot to control the aircraft." In the
search for PIO time histories, numerous references to PIO were found where the time traces
themselves showed residual oscillations (hands off controls) and not true PIOs.*
Since the PIO is evidence of an undamped closed-loop, pilot-vehicle oscillation, then there must
exist during the PIO at least one measurable aircraft state that is 180 degrees out of phase with at
least one pilot control. This leads to the following proposed definition: A PIO exists when the
airplane attitude, angular rate, normal acceleration, or other quantity derived from these states,
is approximately 180 degrees out of phase with the pilot's control inputs.
Because of the critical importance of distinguishing between a potentially catastrophic PIO and
nuisance oscillations, one solution is to change the definition of PIO. The primary emphasis is to
make a distinction between closed-loop pilot/aircraft oscillations that are a side effect of the
pilot's tracking effort and those that have a potential for loss of control. These oscillations may
look identical on recorded data, and only the pilot can properly make this crucial distinction.
One way of viewing the crucial distinction between oscillations resulting from degraded
handling and those that can result in a divergent PIO is to note that in the former case the pilot
drives the oscillation, whereas in a “real” PIO (as defined here) the pilot is driven by the
oscillation. If the oscillation requires that the pilot redirect efforts away from the primary task by
a noticeable amount, we say that a new task has been created (stop the oscillation). In such cases
the pilot is being driven by the oscillation (forced to do a new task). In extreme cases (e.g., YF-
22, JAS 39), the pilots thought that they had experienced a flight control system failure, and that
the new task was to cope with that failure. This is the phenomenon that we must quantify if we
are to achieve clarity on the difference between degraded handling qualities and PIO.

C. CATEGORIZING PIO
Understanding what we are facing requires a thorough understanding of what we have
experienced. For this reason, it is paramount that the history of PIOs be carefully reviewed. For
those events where a reasonably complete record exists, a detailed analysis may prove useful as
we try to develop methods to prevent PIOs from happening again.
To this end an effort was begun to document every recorded PIO available in the published
literature. The effort was not meant to repeat or replace past sources of information,12 but
instead to present PIO information graphically. It was recognized that a comprehensive data
collection, including not only a record of the event but a complete description of the aircraft,
flight condition, etc., would require a great amount of time. It was decided, therefore, that the
collation process would be limited. Documentation consisted of time histories of the PIOs and
an associated reference list. An analysis of the PIO events themselves (frequencies and
amplitudes primarily) was also performed. The primary goal was, and continues to be, the
collation of a "PIO catalog" that can be used for future study by anyone interested in the subject.

*
In most instances these oscillations were the result of low modal damping (short period or
dutch roll), and, while explicit evidence of PIO could not always be located, it is recognized that
low damping of these modes will lead to PIO in closed-loop piloted control.

23
Because this literature search closely paralleled an objective of a separate Air Force contract to
Systems Technology Inc. (STI), and because the STI library has an extensive collection of such
events, the search was coordinated with Mr. David H. Klyde of STI. The foundation of the
search was a list of PIO events assembled by Mr. Duane McRuer of STI and updated by Mr.
David B. Leggett of the U.S. Air Force Research Laboratory at Wright-Patterson AFB, OH. This
catalog is not considered complete by any means; it is, however, a significant step in the
development of a PIO knowledge base. A subset of the PIO time histories – those with no
special security classification or limitations on distribution – is contained in an appendix to a
Wright Laboratory report.15
Aircraft that have experienced PIOs can be divided into two basic classifications: 1) operational,
flight test, or prototype aircraft, including experimental designs; and 2) flight research aircraft.
The latter specifically refers here to the variable-stability aircraft owned by the US Air Force and
operated by Calspan Corporation: the NT-33A, TIFS, and VISTA. This distinction is somewhat
fuzzy in some cases, e.g., when the NT-33A was used to simulate the prototype YF-17 30
(counted as a "flight research" case) or when the X-15 was used to investigate dutch roll
oscillations31 (counted among the "flight test" events).
Both classifications of PIO data are of immense value. Events that occur with operational
aircraft, especially, deserve scrutiny since they are the best examples of a true "surprise" event,
where the pilot was neither expecting the PIO nor protected by a safety pilot or ground
monitoring equipment. On the other hand, there are very few published records of PIOs from
fleet aircraft, and these few are either under close protection by the manufacturers or in dispute
as to their causes, and even whether they were really PIOs. The "operational" aircraft are more
typically modifications or proposed modifications to fleet vehicles, operated by test pilots, under
the aegis of a military activity at one of the test centers. Still, since PIOs even in these cases
were not always expected, and since the pilots most likely reacted as they would if the event
occurred in operational use, the data from these PIOs are extremely valuable.
The bad news in the case of the "operational" PIOs is that published documentation of the
dynamics of the airplane is not always available. Hence, if more detailed analysis of these events
is to be performed, we are left with creating math models of what we hope is a reasonable
representation of the airplane. For prototypes there is usually a more thorough documentation of
the airplane, though this information can be hard to come by at times since it is considered
proprietary information by the manufacturer or it has been discarded by the testing activity.
The second classification – flight research aircraft – holds the most hope for performing
extensive analysis of PIOs. Most, if not all, flight research reports published by Calspan contain
comprehensive information on the dynamics of the simulated airplane. In addition, there are
always pilot comments and pilot ratings to help in the analysis. On the other hand, these PIOs
may not be considered "real" since they occurred in a highly controlled environment, with a
safety pilot, and typically are due entirely to linear causes (most often high time delays or phase
lags).
Our best hope for fully understanding and preventing PIOs is a reasoned assembly of information
from all recorded events.
The PIO catalog contains a total of 76 time histories for the first classification and 85 for the
second. Both include multiple events for the same airplane or variable-stability configuration.
Certainly the first number seems small, given the wealth of anecdotal information about the

24
prevalence of PIO in modern airplanes. Remember, however, that we are counting only
documented events for which some form of time history is available. So while the number
includes some famous PIOs (for example, the YF-22A2), other equally famous ones (for
example, two separate crashes of the Saab JAS 39 Gripen) are not included simply because no
detailed time history (other than video!) of relevant parameters could be located or was made
available.

1. The Occurrence of PIO

Not surprisingly, PIOs have occurred at just about every flight condition imaginable. Figure 12
illustrates the Mach Number and altitude combinations for some of the events cataloged for
operational, flight test, and prototype aircraft. (Other events, at Mach Numbers as high as 2.6
and at altitudes much higher than the maximum in Figure 12, have also been noted.)
Figure 12 reflects the general makeup of the catalog of events, which is dominated by pitch-axis
PIOs with a few roll cases and only three yaw events. (The events have been divided into three
axes for convenience only; there are numerous multi-axis PIOs, and others in which flight path
appears to be the dominant axis.)

2. Characterization of the Oscillation

A wide range of PIOs has been experienced, from unstable closed-loop phugoid oscillations to
high-frequency "ratchet." The atmosphere surrounding the general topic of PIOs has been so
highly charged that there has been a tendency by some to classify just about any oscillatory
response as a PIO (even in cases where the pilot has let go of the stick!), while others try to avoid
the stigma by suggesting that the pilot-aircraft coupling is not a PIO at all. Later in this section
we will make an effort to define such distinctions based on the PIO catalog. The PIO catalog
allows us to begin a systematic process of defining and classifying PIOs.
25,000

20,000

Pitch
Roll
Yaw
15,000

Altitude (ft)

10,000

5,000

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Mach Number

Figure 12. Representative Flight Conditions for Documented PIOs

25
The first step has been to attempt to group PIOs by common parameters such as frequency and
amplitude of oscillation. Figure 13 shows a crossplot (from the flight test and operational
events) of PIO frequency versus maximum angular rate achieved during the oscillation. PIO
frequency has been measured from time histories of angular rate as a time average of the
oscillation, which may itself show slight variations in frequency. The maximum angular rate is
measured from the approximate neutral value (which may or may not be zero). For example, a
pitch PIO with oscillation peaks in pitch rate of +30 and -10 deg/sec (a 40-deg/sec change about
a 10-deg/sec neutral value) would be plotted in Figure 13 at 20 deg/sec. Obviously, Figure 13
contains data only from those time histories for which angular rate is available.
Figure 14 is similar to Figure 13, except in place of angular rate the amplitude of cockpit control
force has been plotted. Maximum control force falls rapidly as PIO frequency increases. This
simply reflects the fact that it is difficult for the pilot to apply large forces at high frequencies.
Some of the specific data points may differ from those in Figure 13 since, in some cases, control
force information was available from the time histories where angular rate was not, and vice
versa.
Both Figure 13 and Figure 14 indicate that PIOs have been experienced at frequencies as high as
25 rad/sec and as low as 1.4 rad/sec (lower-frequency events have been recorded, but angular-
rate information is not available for them). PIOs between 2 and 5 rad/sec are most prevalent.
60
Pitch, roll, Pitch PIO Roll ratchet
yaw PIO (bobble if
Pitch
∆q < 10)
Roll
Yaw

Solid symbol: Severe PIO, or HQR 7 or

40 worse, or PIOR 4 or worse
Flagged symbol: Pitch bobble, or Mild
Amplitude PIO, or PIOR 3 or better
of
angular
rate
(deg/sec)

20

0
0 5 10 15 20 25
Frequency of oscillation (rad/sec)

Figure 13. PIO Frequency vs. Amplitude of Angular Rate

26
 60

Pitch
40
Roll
Amplitude Yaw
of cockpit
control
force
inputs
(lb)

20

0
0 5 10 15 20 25
Frequency of oscillation (rad/sec)

Figure 14. PIO Frequency vs. Maximum Cockpit Control Force

As with Figure 12, the events in Figure 13 and Figure 14 are separated by axis with three
symbols. In addition, those events shown as solid symbols had specific commentary indicating
that they were considered severe or violent, or, if in a flight test program, they received pilot
Cooper-Harper Handling Qualities Ratings (HQRs)28 of 7 or worse, or PIO Tendency Ratings
(PIORs) of 4 or worse. Most of the data points with low angular rates and low control forces, in
the frequency range of 5 to 10 rad/sec, received pilot comments explicitly related to pitch bobble,
mild oscillations, or PIORs of 3 or better.

3. Defining PIOs
Inspection of PIO time histories shows some common elements. The most obvious is that there
is an oscillation. It may differ in the details (convergent, divergent, sustained; cyclical at
constant frequency or intermittent with changing frequency), but a clear oscillation is always
present.
The second common element is that some state of the airplane is out of phase with the pilot
during the PIO. That is, there is a single quantity that can be measured that is, for the initiation
of the event at least, roughly 180 degrees out of phase with pilot inputs. For most events this
quantity is pitch attitude, but there is evidence that flight path, normal acceleration, and even
derived visual signals may be out of phase as well.
Four types of PIO can be identified based on frequency, amplitude of response in Figure 13, and
amplitude of pilot input in Figure 14:
• Roll ratchet/yaw chatter characterized by oscillations at frequencies above about 2 Hz
(12 rad/sec). Maximum angular rates normally about 10 deg/sec or less, though ratchets
with roll rates as high as 50 deg/sec have occurred. Control forces are always relatively

27
 low, however, generally below about 5 lb. Can be considered violent if accompanied by
large side accelerations. Occurrence in roll usually related to excessive control
sensitivity; in yaw related to dutch roll damping.
• Pitch bobble can occur at any frequency between about 1 and 9 rad/sec, but most
common at frequencies above 5 rad/sec. Characterized by maximum pitch rates of 10
deg/sec or less and maximum control forces of 5 lb or less.
• High-frequency pitch PIOs have frequencies above 5 rad/sec and severe PIOs can exhibit
pitch rates from less than 10 deg/sec to at least 40 deg/sec (the event corresponding to
this point on Figure 2 is the well-known T-38 PIO12). Control forces are usually above
10 lb but severe PIOs can have forces well below 10 lb.
• Pitch, roll, and yaw PIOs that are most often destructive occur below about 5 rad/sec.
Angular rates are usually below 25 deg/sec, though at least one case (the YF-16 "flight
zero" incident32) reached roll rates of 50 deg/sec. There is no obvious separation between
severe and non-severe PIOs in this frequency range (Figure 13). Many of the low-pitch-
rate severe events in Figure 13 resulted from extreme actuator rate limiting; very high
control forces (solid circles between 35 and 45 lb in Figure 14) generated only small
pitch rates and the pilots assigned these cases HQRs of 8, 9, or 10.

D. PILOT-INDUCED OSCILLATIONS, PILOT-IN-THE-LOOP OSCILLATIONS, AND

AIRCRAFT-PILOT COUPLING
There have been recent efforts at modifying the label for what have been called "PIOs." As of
the 1995 revision to MIL-STD-1797A,4 the preferred term is "pilot-in-the-loop oscillation,"
suggesting the pilot does not induce the event (which is, essentially, true, since the event is due
to a bad airplane and not a bad pilot), but is just in the loop when the event occurs. Because
"PIO" is the familiar three-letter acronym, the 1797A revision continues to use this abbreviation.
The newest term for PIO is “Aircraft-Pilot Coupling,” or APC.1 The National Research
Council’s Committee on the Effects of Aircraft-Pilot Coupling on Flight Safety adopted a rather
conflicting mix of terms for defining APC on page one of their report:
“If the PVS [pilot-vehicle system] instability takes the form of an oscillation, the
APC event is called a `pilot-involved oscillation’ (PIO)…. If the unstable
motions of the closed-loop PVS are divergent rather than oscillatory in nature,
they are referred to as either APC events or as non-oscillatory APC events.”
Taken literally, this set of definitions says that there can be convergent but unstable motions of
the PVS – a contradiction in itself. In addition, there is no allowance for divergent but
oscillatory motions – the most common form of severe PIO. All PIOs are APC but not all APC
are PIOs. All PIOs are oscillatory and unstable but not divergent.
Which of these jargons to use to describe “PIO” is not nearly as important as solving the problem
in the first place. It is highly unlikely that the use of the term APC will enhance the ability of a
pilot to communicate whether a PIO occurred. There is some danger that the addition of a new
term may further complicate an already confused issue and defocus efforts away from more
serious searches at a solution. As a part of this effort, however, we have conducted informal
polls of experts, including both engineers and test pilots, on their preferences. The following
represents the collected thoughts of these experts (including the authors of this report).

28
Certainly it is noble to revise the label for “PIOs” to relieve the pilot of any feeling of guilt in the
occurrence of the event. The NRC alternative, Aircraft-Pilot Coupling, seems too be too far
removed, however. First, since the word “oscillation” does not appear, it is not descriptive of the
event we are concerned about in the first place. Second, taken literally, there may be many
forms of “coupling,” most of which are desirable: Webster's defines coupling as “the act of
bringing or coming together.” The pilot “couples” with the airplane any time closed-loop control
is exerted, and therefore we might say that one form of “aircraft-pilot coupling” is a perfect
landing, or a successful refueling hookup, etc. This is definitely not the type of “coupling” we
are trying to regulate against. The bottom line, then, is that “APC” seems to be too loose a term.
While there may be a real concern for the increasing occurrence of non-oscillatory APC, there is
a dearth of evidence of its existence as a separate, unique phenomenon. The NRC Committee’s
report1 is devoted almost entirely to oscillatory adverse APC – in other words, PIO. There are
only two examples of non-oscillatory APC in the committee’s report. The first involved a
description of a terrain-avoidance run by a Tornado (for which there are no graphics to disprove
the existence of a PIO, and the text says “APC was fully developed” – suggesting PIO). The
second was the famous SAAB JAS 39 Gripen airshow accident,33 which was a divergent PIO.
“Pilot-in-the-loop oscillation” is not as objectionable, since it means that at least “PIO” can be
retained. Experienced test pilots, including instructors for the U.S. test pilot schools, expressed a
strong allegiance to the traditional “pilot-induced” term. This is the way all students of the U.S.
military test pilot schools are taught; it is well-established in the literature; and most importantly,
no pilot with whom we have spoken has expressed any personal feeling of blame because of the
possibility that he “induced” the oscillation.
On this basis, we recommend retaining the term “pilot-induced oscillations.”

E. THE THREE (OR MORE) CATEGORIES OF PIO

The NRC Committee separates PIOs by Category, depending essentially on the degree of
nonlinearity in the event:1
• Category I: Linear pilot-vehicle system oscillations. These PIOs result from identifiable
phenomena such as excessive time delay, excessive phase loss due to filters, improper
control/response sensitivity, etc. They are the simplest to model, understand, and
prevent.
• Category II: Quasi-linear events with some nonlinear contributions, such as rate or
position limiting. For the most part, these PIOs can be modeled as linear events, with an
identifiable nonlinear contribution that may be treated separately.
• Category III: Nonlinear PIOs with transients. Such events are difficult to recognize and
rarely occur, but are always severe. Mode switching, that cannot be represented by a
quasi-linear equivalent, is the common culprit.
These three Categories will be used throughout this report, though the most common distinction
– as it is for most recorded PIOs – is whether rate-saturation of a surface actuator was a major
player in the event. In this report, “Category II” will almost always refer to rate limiting.
Since the publication of the NRC’s findings, several researchers have suggested that there may
be other Categories that are distinct from the three defined above. “Category IV” PIO sometimes

29
refers to events that are caused by, or have as a major contributor, structural modes and their
interactions with the pilot. Because this report focuses entirely on rigid-body modal responses,
such a distinction is not needed here. This seems to be a critical topic for future study, especially
in light of the recent interest in super-large commercial transport airplanes.

F. SOME EXAMPLES
There are several recently-published reports that contain numerous examples of PIOs,1,15,34 so
this report documents primarily those events not covered elsewhere. It is useful, however, to
examine a few examples to illustrate the types of PIOs we have chosen to focus on, and the types
we have not examined in any detail in this study.

1. Low-Frequency (Phugoid) PIO

Longitudinal PIOs that occur at frequencies well below the short-period mode are rare. In fact,
the only recorded and documented instance of such a PIO occurred during a ground-controlled
approach (GCA) of a C-97 cargo airplane in 1954. 35 Figure 15 is a reproduction of the event
from Ref. 35, showing a 128-sec sequence with roughly 2-1/2 cycles of the divergent oscillation.
This corresponds to a PIO frequency of approximately 0.12 rad/sec.
It is clear that the PIO in Figure 15 is a result of a combination of the visual conditions and
aircraft dynamics that led to an unstable phugoid oscillation. Such an occurrence, at least based
on the documented evidence, is rare, and although the PIO led to a crash, this type of event is not
what the present study has focused on.

2. Medium-Frequency PIO
This is the most common form of PIO, occurring near the short-period mode in pitch and the roll
damping or dutch roll mode in roll. It is the focus of this entire study.

3. High-Frequency, Small-Amplitude Pitch PIO (Pitch Bobble)

Pitch “bobble” appears to be quite common in operational aircraft. The primary characteristic
that distinguishes the “bobble” is amplitude: pilot control inputs tend to be small (on the order of
±5 lbs or less on centerstick controllers) and pitch rates tend to be less than ±10 deg/sec (see
Figure 13). There is even some disagreement over whether pitch bobble should be classified as a
PIO, but it certainly meets the definition of MIL-STD-1797A, so it should be considered to be a
“nuisance” or “mild” form of PIO.
There is no clear evidence of a bobble tendency that has directly grown into a full-blown severe
PIO. Bobble has sometimes occurred on an airplane that also exhibits PIO, but the two events
have not been correlated. When pitch bobble occurs in a flight research program, it is not
uncommon for pilots to assign a PIOR that indicates the presence of an undesirable motion, if
not a PIO – typically a PIOR of 3 or 4 – but with an associated HQR that suggests the airplane’s
overall response is adequate – typically an HQR of 4 or 5.
An example of stick force input and pitch rate response of a bobble-prone airplane is shown in
Figure 16. These data come from the USAF Test Pilot School’s HAVE PIO project,36,37 the
results of which are applied throughout this report. The HAVE PIO project is described in detail
in the next section. Figure 16 is included here merely to illustrate the nature of pitch bobble as a
high-frequency nuisance mode.

30
 Figure 15. Documentation of a Phugoid PIO (Reproduced from Ref. 35)

4. High-Frequency, Small-Amplitude Roll PIO (Roll Ratchet)

Roll ratchet is very similar in form to pitch bobble, except it occurs in the roll axis. Because roll
inertia of most airplanes is lower than pitch, the tendency for roll ratchet, and its normal
frequency of occurrence, are both higher than for pitch bobble. In most cases roll ratchet is no
more than a nuisance, impacting ride qualities but not necessarily task performance. On
occasion, however, the magnitude of the roll oscillations can become sufficiently high that
ratchet prevents task performance entirely, and then it may be classified as a severe PIO, as with
the example in Figure 17, experienced on the experimental F-16XL airplane.38
Roll ratchet appears to be a result of the biomechanical coupling between pilot and control stick,
as discussed in Ref. 38, among others, though there is some disagreement about the specifics of
the phenomenon. For our purposes, we simply define ratchet as a “mild” PIO unless it becomes
intrusive. Because ratchet is a much less common problem in terms of PIO than more short-
period oscillations, we chose to not perform any detailed analysis of the phenomenon.

31
 15

stick force (lb)

10

0
50 60 time (sec) 70 80

5
Evidence of
pitch rate (deg/sec)

pitch bobble

0
50 60 70 80

time (sec)
-5

Figure 16. Example of Pitch Bobble (HAVE PIO36 Configuration 2-B, Pilot A)

Figure 17. Example of Roll Ratchet (F-16XL)

32
 IV. MEASUREMENTS OF PILOT BEHAVIOR IN PIO

A. THE PILOT AS A PART OF THE PROCESS

Historically, the attention paid to the dynamics of the pilot and the pilot’s actions in PIOs has
been almost as great as the attention given the airplane itself. This is certainly not surprising
given the pilot’s critical involvement in the event. If we can understand the behavior of the pilot
immediately before and during a PIO, we can strive to achieve airplane designs that prevent the
need for such pilot behavior.
But there is a caveat to be stated here. A recurring theme in this report is that PIOs are not the
fault of the pilot. While some pilots may have a greater predisposition toward PIO, this does not
equate to poor piloting skills or inappropriate behavior. So the focus on pilot behavior is
certainly appropriate, but only so that we can determine those traits that make it possible for a
particular airplane to be susceptible to PIO.
Unfortunately, there have been, and there continue to be, strong inferences that a key contributor
to PIOs is incorrect pilot behavior. For example, MIL-STD-1797A4 quotes a venerable reference
document on PIOs12 that lists pilot-centered causes of PIO and places “incomplete pilot
equalization” at the top. The existence of the quoted material in MIL-STD-1797A can be traced
to work done by the authors of this report39 in writing the draft version, and we have joined
others in suggesting the text be removed from MIL-STD-1797A. The truth is that it is difficult
to find documented evidence of any PIO that has occurred for any other reason than that the
airplane was deficient in its design. Certain pilots are more apt to discover PIO than others, as
will be demonstrated in this section. Limitations of the pilot to adapt to poor design do not
absolve the airplane itself of all blame, and the only pilot-centered “cause” of PIO is the fact that
the pilot is in the loop in the first place.
That said, there are also instances for which the occurrence of PIO has been attributed to poor
pilot training. The authors of this report take the position that inadequate pilot training can lead
to degraded performance, bad technique, loss of control – but PIO requires the active
participation of the airplane, and the sort of PIO we are trying to prevent is due not to the training
level of the pilot but to the design of the airplane. A PIO-proof airplane will not enter a severe,
life-threatening oscillation no matter how bad the pilot’s training or skills.

B. PILOT MODELS
Models for pilot behavior during a PIO have been developed and expounded by many other
researchers12,34,40 and will not be described in any great detail in this report. Rather, we have a
luxury not available to the previous researchers: a wealth of real PIO time-history data with
which to test the various theories. From this data we can extract measures of pilot compensation,
including PIO frequencies, phase angles between pilot input and aircraft response, and
magnitudes of input and output variables.
For all of the following data analysis, two sets of software have been applied extensively. First,
the ROVER tools developed under this contract and described in Section VI were applied to
extract the relevant information on the dynamics of the PIO itself. Second, pilot model
information was obtained via fast-Fourier transform (FFT) methods using a software package
called FREquency Domain Analysis (FREDA), from Systems Technology, Inc.

33
There are two fundamental theories for pilot models in a PIO.
Synchronous behavior: As the PIO develops, the pilot responds to perceived oscillations
(whether in attitude, rate, or acceleration, and whether sensed visually or through motion) by
applying inputs that are at the oscillation frequency, with no attendant pilot delay or
compensation. Past studies of severe PIOs have suggested that this is a common pilot
behavior.12,34 If this behavior is normally adopted by the pilot, we should find that the
characteristics of severe PIOs – their frequencies and amplitudes – are relatively consistent
across the pilot population. In addition, we would expect that the PIO frequency will be near the
neutral-stability frequency of some measured aircraft state in response to pilot control inputs.
For example, several analytical criteria (described in this report) assert that the neutral-stability
frequency between pitch attitude and stick input (either force41 or position42) will be the most
common PIO frequency for Category I PIOs.*
Compensatory behavior: This theory assumes that the pilot will apply some form of
compensation, even during a fully-developed, severe PIO. In this case we would expect to see
some variation in the identifiable characteristics of PIOs as each pilot adopts different behavior,
and as that behavior might change even during the PIO. We would not expect the PIO frequency
to correspond in any precise manner with the airplane’s neutral-stability frequency. Rather, PIO
frequency should be reasonably constant for any one pilot, as each pilot will adopt different
compensation strategies in order to achieve some level of internally-defined level of
performance. The predicted PIO frequency in the Smith-Geddes criteria,40 for example, is based
on a relationship derived from experimental studies of compensatory control.

C. CHARACTERISTICS FROM CATEGORY I PIOS IN LANDING (HAVE PIO)

In 1986, students of the US Air Force Test Pilot School at Edwards AFB, CA, conducted an in-
flight simulation study of PIOs.36 This program, HAVE PIO, investigated the effects on PIO
susceptibility of variations in longitudinal short-period dynamics and added filters for the
precision offset landing task. The results of the study have been analyzed, scrutinized, and
applied many times during the last decade.
Until recently, the only PIO information openly available from HAVE PIO has been that which
was published in a TPS report36 and a more familiar Master’s thesis from the Air Force Institute
of Technology.37 In 1997, as a result of the intense interest in the HAVE PIO data, Calspan
Corporation processed the original data tapes from the 1986 experiment. The time histories of
all of the data runs are now available for detailed analysis.
The HAVE PIO program used the USAF variable-stability NT-33A with three TPS students
serving as evaluation pilots. Precision offset landings were performed, with desired and
adequate performance limits similar to those defined for other experiments.43,44 Typically three
landings were flown, starting with a straight-in, followed by an offset from one side and then
from the other. In a number of cases – especially those configurations with severe PIOs – only
two landings were conducted. The dynamics of the 18 configurations were similar to several of
those in the Calspan-conducted Landing Approach High-Order Systems (LAHOS) experiment. 43

*
Nonlinearities will clearly complicate the definition of a “PIO frequency,” and the
measurement of PIO frequency for Category II events is discussed separately in this section.

34
The results of the HAVE PIO experiment must be viewed with some caution. First, as a TPS
project, it is possible that the pilots were not sufficiently experienced in evaluating flying
qualities and assigning HQRs and PIORs. Second, since the project was intended to look at
PIOs, the pilots may have been more conscious of any hint of a potential PIO than might
normally be expected in operational flight. Third, encountering a PIO on a variable-stability
airplane with a safety pilot may be very different from the operational environment, and the
pilots’ opinions may have been influenced by the artificial scenario. Fourth, considering the
relatively limited number of evaluations for any one configuration – theoretically consisting of a
maximum of nine landings, three from each of the three pilots – those configurations with severe
PIOs clearly had extremely poor dynamics, but it is possible that some of the other cases simply
were not evaluated enough to expose a PIO tendency. Nevertheless, this is an important data
source because it provides not only handling qualities and PIO information, but PIO time-history
data as well.
In the report,36 it is indicated that 12 of the 18 configurations exhibited PIOs on landing. As is
shown in analysis documented elsewhere,45 however, in three of the cases the “PIO” was actually
a pitch bobble; PIORs suggest PIO tendency, but when the pilots answered a pilot comment
questionnaire on “PIO tendency,” the answers reflected bobble with little or no PIO tendency.
The remaining nine configurations were considered to exhibit severe PIO according to at least
one of the three evaluation pilots. In this case, “severe” PIO is considered to have occurred if
both of the following are true: 1) at least one pilot assigned a PIO tendency rating of 4, 5, or 6,
and an HQR of 7, 8, 9, or 10; and 2) the oscillation frequencies, input/output phase angles, and
input and output amplitudes all met the definitions of severe PIO as described in Section III.
For reference, the HQRs and PIORs from the HAVE PIO flight program are listed in Table 2.
The nine cases with severe PIO are noted in the table.

Table 2. Pilot Ratings for HAVE PIO Experiment36 (Severe PIO Cases are in Italic Font)
Handling Qualities Ratings PIO Ratings
Conf. PIO in flight? Bobble in flight?
A B C A B C
2-B 7 3 3 3 3 2 2 1 No Yes
2-1 2 2 3 1 1 1 No
2-5 10 7 10 4 4 5 Yes
2-7 7 4 4 4 3 2 Yes
2-8 8 10 8 4 4 4 Yes
3-D 2 2 1 1 No
3-1 5 3 4 3 2 2 No Yes
3-3 7 2 3 3 1 1 No
3-6 5 4 2 2 No Yes
3-8 8 5 8 4 3 4 Yes
3-12 7 9 4 5 Yes
3-13 10 10 4 5 Yes
4-1 3 2 3 1 1 1 No
4-2 3 3 4 1 1 2 No
5-1 2 5 1 1 No
5-9 7 8 7 4 5 4 Yes
5-10 10 10 5 5 Yes
5-11 7 7 5 2 4 3 Yes

35
While there are occasional large differences in ratings (for example, the very first case, 2-B, with
an HQR of 7 from Pilot A and 3’s from the other two pilots), for the most part the ratings for the
severe-PIO cases are quite consistent between the pilots. In general, for all of the HAVE PIO
configurations, the worst (highest numerical) ratings were given by Pilot A. This suggests that
Pilot A was either more aware of the limitations of the configurations, or that he was a more
harsh evaluator. In the former case there should be some method for identifying differences in
piloting technique.
Some hint of piloting differences both during normal flying and during PIO can be seen simply
by comparing time histories for the three HAVE PIO pilots. Example thirty-second segments,
for Configuration 2-5, are shown in Figure 18. This configuration was one of the few flown by
all three pilots and considered by all three to exhibit PIO (as evidenced by the HQRs and PIORs
noted on Figure 18).
NOTE: For all plots, Thick Line is Stick Force (lb), Thin Line is Pitch Rate (deg/sec)
30 30
Straight in Straight in
20 20

10 10

0 0
50 60 70 80 55 65 75 85
time (sec) time (sec)
-10 -10

30 30
Right offset
Right offset
20 20

10 10

0 0
40 50 60 70 40 50 60 70
time (sec) time (sec)
-10 -10

30 30
Left offset
20 Left offset 20

10 10

0 0
40 50 60 70 45 55 65 75
time (sec) time (sec)
-10 -10

a) Pilot A (HQR = 10, PIOR = 4) b) Pilot B (HQR = 7, PIOR = 4)

30
Straight in
20

10

0
10 20 30 40
time (sec)
-10
30

Left offset
20

10

0
10 20 30 40
time (sec)
-10

c) Pilot C (HQR = 10, PIOR = 5)

Figure 18. Landing Time History Segments from HAVE PIO36 (Configuration 2-5)

36
Each segment was selected from the full time-history data file because it best illustrated the
character of the PIOs encountered by the pilots. The order of the traces for each pilot is as it was
flown: straight-in, right offset, left offset for Pilots A and B; straight-in and left offset for Pilot C
(who flew only two landings for this configuration).
The stick force traces for Pilot A seem to show more high-frequency control activity and shorter
episodes of PIO (fewer cycles of oscillation) than for the other two pilots. The two segments for
Pilot C appear to show the longest continuous bursts of PIO, but with the smallest peak-to-peak
stick force magnitudes. In general, there does appear to be some qualitative difference between
the three pilots in the character of the time traces, though all have evidence of a similar severe
PIO.

1. Quantitative Measures of PIO Frequency

Section III used oscillation frequency as a primary discriminator in characterizing PIOs. There
are two underlying assumptions not stated in Section III: 1) that the oscillations occur at some
relatively well-defined and consistent frequency; and 2) that there are no significant changes in
vehicle dynamics that change the oscillation frequency for a given PIO. The latter assumption is
not easy to prove with analysis or experience, since it is always possible for the dynamics of the
airplane to be changed with such factors as rate limiting. The first assumption comes from an
assertion that there is a unique PIO frequency, that is repeatable and relatively constant, for each
PIO-prone airplane. This assumption is testable with the HAVE PIO data.
PIO frequencies were measured for the nine HAVE PIO configurations that exhibited severe
PIOs. Frequencies were defined as the time between peaks in the pitch rate response (either
minimum-to-maximum or vice versa) when at least three of the four measures of severe PIO
(frequency, phase angle, stick force magnitude, and pitch rate magnitude) met the definitions of
Section III. Hence multiple readings of PIO frequency were made for each episode of PIO, with
one reading for each half-cycle of oscillation. Average frequencies were then computed for each
pilot/configuration combination by averaging all individual numbers, from all landings; overall
averages for each configuration were computed by performing the same averaging but with the
data for all the pilots.
The flight-identified PIO frequencies for the nine HAVE PIO configurations that had severe
PIOs are shown in Figure 19. The configurations are separated by their identifiers (2-5, etc.),
and the different symbols show the average frequencies for Pilots A (circles), B (squares), and C
(triangles), respectively. The overall average frequencies for all pilots are represented by X’s.
The small solid dots are discussed below.
It is obvious from Figure 19 that there is no single “PIO frequency” for all of the pilots. In some
cases the spread in frequencies is greater than 1 rad/sec (the data for Configuration 5-11 are an
exception and are discussed below). In every case, Pilot A had the highest average PIO
frequency (note that Pilot A did not fly Configuration 3-12, and he was the only pilot to
experience severe PIO on 2-7), and in most cases Pilot C had the lowest frequency.
The largest spread in frequencies is for Configuration 5-11, but this case takes some explaining.
Only Pilot B reported a PIO (HQR 7, PIOR 4, Table 2); for Pilot A (HQR 7, PIOR 2) there is
only one cycle of possible PIO out of his three landings, and for Pilot C (HQR 5, PIOR 3) there
is a single half-cycle on one landing. As a result, only the data for Pilot B should be considered
representative of a severe PIO for 5-11.

37
 8
Model (force)
7 Pilot A
Pilot B
Pilot C
6
Average

5
PIO
Frequency 4
(rad/sec)
3

0
2-5 2-7 2-8 3-8 3-12 3-13 5-9 5-10 5-11
Configuration

Figure 19. Flight-Identified PIO Frequencies for HAVE PIO Configurations

The small solid dots on Figure 19 represent the neutral-stability frequency for the transfer
function of θ/Fes , that is, the frequency where the phase angle between stick force and pitch
attitude is -180 degrees (written as ω180θ ). If the pilots were to adopt synchronous behavior in
the PIO, this would be the expected PIO frequency. In truth, the values of ω180θ plotted in Figure
19 are obtained from analytical models of the configurations as documented in Ref. 36 and are
only as accurate as the models. Based on more recent experiences with the variable-stability NT-
33A airplane,44 there is every reason to believe that the models are generally quite accurate,
except the airplane may have had more overall phase loss (due to either greater delays or
undocumented lags) than has been reported. Based on the analysis in Ref. 44, the effective time
delay of the phase loss was on the order of 50 msec. In this case, the ω180θ values plotted in
Figure 19 are a bit too high. Nevertheless, we can use these analytical values as references to
evaluate the likely pilot behavior during the PIOs.
In almost all cases, the flight-identified PIO frequencies are higher than the frequencies for
ω180θ . This is clear evidence that the pilots did not adopt synchronous behavior in any case.
There is, however, a general correlation between PIO frequency and ω180θ , suggesting pilot
behavior is still influenced by the neutral-stability frequency of the airplane. In other words, it
does not appear as if the three pilots attempted to adopt purely compensatory behavior either,
since we would expect PIO frequencies to be more constant across configurations.
Since the PIO frequencies for Pilot C are closest to ω180θ among the three pilots, this may
indicate that this pilot was closest to becoming synchronous. The generally higher frequencies

38
for Pilot A could indicate more lead compensation (to increase the phase characteristics of the
pilot-vehicle system), or it may simply indicate that this pilot was slower to react to the
oscillations in pitch rate. That is, Pilot A may have allowed the deviation in pitch rate to
continue for a longer time before applying an input. In any case, Pilot A clearly applied a
different level of compensation than the other pilots – perhaps explaining this pilot’s greater
likelihood of finding PIO and poor handling qualities, as reflected in the ratings in Table 2.

2. Derived Phase Angles

While the ROVER software used the phase angle between pitch rate and stick force to identify
PIOs in the HAVE PIO data base, it is insightful to convert this phase angle to pitch attitude. To
do so, we simply subtract 90 degrees (assuming the original phase angle is expressed as a lag,
that is, a number less than zero to begin with) for each identified half-cycle of PIO. Then we can
crossplot this phase angle with PIO frequency on a semi-log scale, and what we should obtain is
a fairly faithful Bode-plot representation of the airplane at the PIO frequencies.
Such a series of crossplots, for the nine severe-PIO cases, is shown in Figure 20. The different
symbols again represent the different pilots, and the ω180θ frequencies from the analytical models
are shown as solid circles. Each data point is one-half cycle of an identified oscillation, so the
number of data points on each plot signifies the amount of data available for that configuration.
For example, there are only four points on the plot for Configuration 5-11 (lower right corner),
compared with 37 for 5-9 (upper right corner). There are, invariably, occasional wild points that
do not appear on the plots in Figure 20 (though these points were included in the computation of
the averages plotted in Figure 19).
-90
Conf. 2-5 Conf. 5-9
Conf. 3-8
phase (deg)

-180
A A A
B B B
-270
C C C
Model Model Model
-360
1 frequency (rad/sec) 10 1 frequency (rad/sec) 10 1 frequency (rad/sec) 10

-90
Conf. 2-7 Conf. 3-12 Conf. 5-10
phase (deg)

-180
B
A C A
-270
C Model C
Model Model
-360
1 frequency (rad/sec) 10 1 frequency (rad/sec) 10 1 frequency (rad/sec) 10

-90
Conf. 2-8 Conf. 3-13 Conf. 5-11
phase (deg)

-180
A
A A B
-270
C C C
Model Model Model
-360
1 frequency (rad/sec) 10 1 frequency (rad/sec) 10 1 frequency (rad/sec) 10

Figure 20. Identified Phase Angles Between Pitch Attitude and Stick Force for Severe-PIO
Configurations from HAVE PIO

39
These plots confirm the observation made above, that the PIO frequencies were almost always
above the neutral-stability frequencies. They also show that each pilot, for each configuration,
was relatively consistent in his adoption of a PIO frequency, with Pilot A at higher frequencies
than the other two pilots. The data trends follow the expected phase loss with frequency for the
particular aircraft configuration: a line sketched through the data defines the phase curve of
pitch-attitude-to-stick-force for the airplane configuration quite accurately. In fact, the overall
very consistent data trends confirm the effectiveness of the time-domain approach used to
identify these data in the first place.

3. Comparisons with Predictions of PIO Frequency

Along with the differing theories of pilot dynamics in a PIO, there are differing methods of
predicting the likely PIO frequency in the event of a severe PIO. Because some of these methods
are an integral part of PIO prediction criteria, they are discussed in more detail in the next section
of this report. Here we will make use of the HAVE PIO data to evaluate the efficacy of these
methods at predicting the average PIO frequency.
Remember, as was shown above, that there is no single PIO frequency for any particular PIO-
prone airplane; differences in pilot compensation can have a significant impact on the actual
frequency of the PIO. For this analysis, we will use the overall average frequencies as plotted in
Figure 19 and consider them to be the representative frequencies for each of the nine severe-PIO
configurations from HAVE PIO.
Five methods for predicting PIO frequency will be evaluated. They are as follows:
1) Neutral-stability frequency of pitch attitude to stick force, ω180θ ( θ/Fes ): As mentioned
earlier in this section, if the pilot adopts synchronous behavior in a PIO this should be the
PIO frequency. It was shown to be effective for the HAVE PIO data, using PIO frequencies
published in the HAVE PIO report – with a minor adjustment – and by adding a factor of 0.5
rad/sec to the frequency ω180θ ( θ/Fes ).41,45

2) Neutral-stability frequency of pitch attitude to stick position, ω180θ ( θ/δ es ): This is an

42
element of Gibson’s phase-rate PIO criterion and is the predicted PIO frequency if a PIO is
to occur. The only difference from the frequency above is removal of the dynamics of the
feel system, resulting in slightly higher predicted frequencies. As was shown in Figure 19,
the actual PIO frequencies are higher than those predicted by ω180θ based on force, so it
would be expected that any increase in predicted frequency should be an improvement.
3) Smith-Geddes criterion frequency, ωc: A complete definition is given in the next section;
essentially, this is the frequency predicted for Smith-Geddes Type III (attitude-dominant)
PIOs40 and is a linear function of the slope of the magnitude of θ/Fes in the frequency range
of 1 to 6 rad/sec. The steeper the slope, the lower the criterion frequency. There is some
difference of opinion on whether the dynamics of the force-feel system should be included
when applying the Smith-Geddes PIO criteria, but because ωc is measured from the
magnitude, and not the phase angle, of the airplane’s attitude response, the feel system has
only a very small effect on ωc. With the high-frequency feel system dynamics used in HAVE
PIO (damping ratio of 0.6, natural frequency of 26 rad/sec), there is essentially no difference
in ωc when the feel-system dynamics are removed.

40
4) Neutral-stability frequency assuming a compensatory pilot model, ω u cm : This is the
frequency at which the phase angle of the pilot-vehicle system is –180 degrees. In Ref. 34,
approximate values of ω u cm are estimated for six of the nine HAVE PIO configurations. The
values reported in Ref. 34 are used here.
5) Resonance frequency for Smith Type I (acceleration-dominant) PIO, ωR(az): The analysis of
the HAVE PIO data36 includes estimates of this frequency. It is the predicted PIO frequency
if the pilot were responding to normal acceleration cues only, not attitude cues.14 Hence it is
fundamentally different from the previous four predictions. The values used here were taken
directly from Ref. 36.
A summary of all the predicted PIO frequencies, along with the flight-identified average
frequencies, is given in Table 3.
Linear regression was performed on all of the different prediction methods, using the formula
ω PIO = a + bω PRED , with the results shown graphically in Figure 21. The overall correlation with
actual PIO frequencies, based on the regression coefficient r, is generally good for all of the
methods. The value of r is slightly higher for the acceleration resonance frequency prediction
(Figure 21e), at 0.95, and slightly lower for the compensatory-pilot method (Figure 21d), at 0.87,
compared to r = 0.93 for the other three approaches.
All but the compensatory-pilot estimate (where n = 6, not 9) are correlated at a 1% level of
significance, and the compensatory-pilot estimate is at a 5% level of significance. On the basis
of correlation alone, therefore, we cannot conclude a clear superiority for one method over the
others.

Table 3. Measured and Predicted PIO Frequencies for HAVE PIO Configurations
Predicted PIO frequencies (rad/sec) based on –

Measured Smith-Geddes Neutral

Neutral Neutral Smith
average PIO criterion stability
Config. stability stability (acceleration-
frequency, ωPIO (attitude- (compens-
(force), (position), dominant),
dominant), atory pilot),
(rad/sec) ω180 θ ω180 θ ω u cm ωR(az)
ωc

2-5 2.8 2.3 2.5 3.2 2.0 2.7

2-7 4.1 3.9 4.4 4.4 4.1
2-8 4.3 3.5 3.9 4.3 3.9 3.8
3-8 6.0 5.2 5.9 5.0 5.1
3-12 2.2 2.2 2.3 3.3 2.2 2.6
3-13 3.2 2.9 3.1 4.0 2.2 3.2
5-9 3.5 2.5 2.6 3.6 3.3 2.9
5-10 2.6 2.1 2.2 3.1 2.6 2.5
5-11 3.6 2.8 3.1 3.8 3.3

41
 ωPIO = 0.80 + 1.05ω180θ, r = 0.93 ωPIO = 1.09 + 0.88ω180θ, r = 0.93
6
Measured frequency (rad/sec)

Measured frequency (rad/sec)

6

4 4

2 2

0 0
0 2 4 6 0 2 4 6
Predicted frequency (rad/sec) Predicted frequency (rad/sec)

a) Prediction based on ω180θ b) Prediction based on ω180θ

(including feel system) (excluding feel system)

ωPIO = -2.46 + 1.68ωc, r = 0.93 ωPIO = 1.01 + 0.90ωu, r = 0.87

6 6
Measured frequency (rad/sec)

Measured frequency (rad/sec)

4 4

2 2

0 0
0 2 4 6 0 2 4 6
Predicted frequency (rad/sec) Predicted frequency (rad/sec)

c) Prediction based on ωc d) Prediction based on ωu

(Smith-Geddes Type III PIO) (compensatory model: n = 6)

ωPIO = -0.26 + 1.27ωR, r = 0.95

6
Measured frequency (rad/sec)

0
0 2 4 6
Predicted frequency (rad/sec)

e) Prediction based on ωR(az)

(Smith Type I PIO)

Figure 21. Comparison of Flight-Identified PIO Frequencies with Various Predictions

42
There is, however, a clear difference when we look at the constants of the linear regression
equation, and ask: which method produces a prediction that is closest in value to the measured
frequency? The dashed lines in Figure 21 represent the regression equations and demonstrate
that the best predictor of trends is the neutral-stability frequency with the feel system included,
Figure 21a. The slope parameter b = 1.05 is closest to unity for all the methods, with an offset, a,
of 0.8 rad/sec. This regression formula says that the PIO frequency is essentially the neutral-
stability frequency, with an addition of 0.8 rad/sec – slightly higher than the 0.5 rad/sec proposed
in the past,41,45 but still very effective.
Predicted PIO frequencies based on the neutral-stability frequency, but with the feel system
removed (Figure 21b), and predicted frequencies based on the compensatory pilot model (Figure
21d), are similar, with slopes about 90% of unity and with a constant of 1 rad/sec.
The Smith-Geddes criterion frequency predictions (Figure 21c) are farthest off, with a very large
slope and negative constant. Similar characteristics are exhibited by the Smith acceleration-
dominant resonance peak predictions (Figure 21e).
Of all the prediction methods applied to the HAVE PIO data, only those based on Smith-Geddes
criterion frequency and acceleration resonance frequency do a noticeably poor job of capturing
the change in frequencies with change in dynamics, though all need an additional offset
correction to improve their effectiveness.
It must be emphasized that all of these predictions are reliant on the analytical models of the
HAVE PIO configurations, and are only as good as those models. As stated previously, it
appears that the models were quite accurate representations of the airplane configurations, except
the NT-33A variable-stability airplane may have had more high-frequency phase loss than
previously reported. If this were true, the predicted frequencies for four of the five methods
would have been even lower, and the most likely effect would have been an increase in the
regression constant, a, for all methods, with a slight change in the slope parameter, b. The
exception is the Smith-Geddes criterion frequency, which is entirely a function of magnitude
slope and is unaffected by changes in phase angle.

D. CHARACTERISTICS OF PIOS IN ATTITUDE TRACKING (RATE-LIMITING

SIMULATION, APPENDIX A)
As a part of this study, a piloted simulation was conducted on the US Air Force’s Large
Amplitude Multimode Aerospace Research Simulator (LAMARS). The simulation is
documented in Appendix A. Data from this simulation, which used nine pilots and investigated
both Category I and II (rate-limiting) PIOs, provides a wealth of information about pilot
characteristics during PIOs. Several of the configurations that exhibited severe PIOs were
analyzed in a manner similar to that applied to HAVE PIO. The ROVER software developed on
this contract was used to search for oscillations meeting the definitions of PIO, and the
FREquency Domain Analysis (FREDA) software from Systems Technology, Inc., was used to
identify pilot-vehicle characteristics.
The majority of configurations were designed to look at the causes of Category II PIO. As such,
the dynamics of these configurations were intended to be resistant to PIO in the absence of rate
limiting, and to place different levels of demand on the actuators. A portion of the cases looked
specifically at Category I PIO, primarily by adding lag filters to increase the loss of phase with

43
increasing control frequency. A subset of the Category I cases will be reviewed before
proceeding to Category II.

1. Category I PIOs
Not all configurations were evaluated by all nine pilots, making an assessment of pilot
differences less complete. Further complicating matters is the fact that some pilots did not see
PIOs for some configurations, even though the conditions for PIO might have been found when
the time-history data were examined by ROVER. Ten longitudinal configurations were judged
by at least one pilot to have a tendency for severe PIO – evidenced by assigning an HQR of 7 or
worse with a PIOR of 4 or worse.* The configurations varied from the nominal by adding lag
filters or by reducing short-period frequency or damping.
Summary PIO information for these cases is listed in Table 4, including the letter identifier for
the pilots who evaluated each of the cases (A through I), and of those, the pilots who assigned
ratings indicating severe PIO. Neutral-stability frequencies for the configurations, including and
excluding the dynamics of the force feel system, and identified average PIO frequencies are also
listed.
Identified PIO frequencies for each pilot for these configurations are plotted in Figure 22. As
with the HAVE PIO data analyzed above, each data point represents an averaged frequency, and
in the case of individual pilots, it may be as little as one-half cycle or as many as 20 or more half-
cycles of identified oscillations.

Table 4. Configurations Exhibiting Category I PIO in HAI Simulation (Appendix A)

Config- Pilots ω180 ω180 ωPIO
Variable Evaluation
uration indicating (force) (position) (average)
pilots:
severe PIO: (rad/sec) (rad/sec) (rad/sec)
1D All H, I 3.5 5.4 3.8
Low ωsp, added
1DL2 A,B,C,D,G,H A 2.7 3.6 4.4
lag filters
1DL3 All A,C,D,E,F,H 2.5 2.9 3.9
2DL2 A,B,C,D,I B 4.8 6 4.1
Added lag filters 2DL3 A,B,C,D,G D 3.9 4.5 4.2
2DL4 All except I C, E 2.9 3.3 4.1
Low ωsp 18 All A,B,C,G,H,I 2.4 3.7 3.8
15 All C, H 5.3 7.7 3.9
Reduced ζsp 5A A,B,C,D,E,F,G A,B,C,E,F 5.0 6.5 5.6
16 All except I All except I 4.9 5.3 5.4

*
Several marginal configurations also were considered to have tendencies toward severe PIO,
but changes in control command/response sensitivity (stick force per g) were sufficient to
alleviate the PIO tendencies. These cases are not included in the “severe-PIO” count, but they
are evaluated later in this report.

44
 7

4
PIO
frequency
A
(rad/sec) 3
B
C
D
2 E
F
G
1 H
Average
Model

0
1D 1DL2 1DL3 2DL2 2DL3 2DL4 18 15 5A 16
Configuration

Figure 22. PIO Frequencies for Category I Severe PIOs from HAI Simulation

While some trends in Figure 22 are similar to those for the HAVE PIO landing data (Figure 19),
there are several very distinct differences as well. The most striking difference is the apparent
lack of correlation between neutral-stability frequency (feel-system included, identified as
“Model” on Figure 22) and identified average PIO frequency. Case 1D has a low short-period
frequency, and 1DL2 and 1DL3 have progressively lower-frequency added lags, resulting in a
reduction in ω180θ , but the average PIO frequency is almost unchanged across these three
configurations. Likewise, the sequence 2DL2 – 2DL3 – 2DL4 has increasing phase loss due to
lags, as evidenced by ω180θ , but the PIO frequencies are almost constant.

The short-period frequency for Configuration 18 is lower than that for 1D, and the spread in
frequencies is greater, but the average is almost the same as for 1D. The configuration sequence
15 – 5A – 16 represents a decrease in short-period damping from 0.4 to 0.18 to 0.01, all with a
natural frequency of 4.6 rad/sec. As the damping decreases, the oscillation frequencies tend to
be closer to the neutral-stability frequency – as one would expect, since the neutral-stability
frequency is also near the short-period frequency, and the slightest pilot input results in an
excitation of the short-period mode.
The low-damping configurations posed a challenge for the evaluation pilots, since it was possible
to initiate a sustained oscillation with very little effort, and in the case of Configuration 16, the
oscillations persisted even if the pilots stopped flying the airplane. Every pilot considered this to
be PIO, but there are long stretches of tracking runs for all pilots where there is no stick input
correlated to the oscillatory output. Several pilots expressed some discomfort with the

45
limitations of the PIO tendency rating scale in trying to describe the oscillatory motions: they
were easily induced and they interfered with task performance, but they could not be stopped by
any reasonable pilot technique. Some pilots considered this to be more of a residual oscillation
than a true PIO.
One possible explanation for the lack of correlation between PIO frequency and neutral-stability
frequency in the Figure 22 data could be the fact that data for different pilots are used for
different configurations. Yet this trend is true for each pilot individually as well – the clearest
example is Pilot A (open circles in Figure 22), who experienced PIO in nine of the ten cases. For
the first six cases, especially, his identified PIO frequencies are not even consistent in trend with
the neutral-stability frequencies.
The data of Figure 22 provide very strong support for the theory that, for the combination of
simulator, airplane dynamics, and task, the pilots tended to be much more compensatory than
synchronous in their behavior. Across the sampling of pilots and changes in airplane dynamics,
the overall trend was to adopt a closed-loop strategy, even in the PIO, that produced an average
PIO frequency of about 4 rad/sec (slightly higher for the low-damped cases). For the most part,
this frequency was above the neutral-stability frequency of the airplane alone (the most obvious
exception, for Configuration 2DL2, has data for only one pilot – Pilot A – and only six half-
cycles of oscillation for that pilot).
One clear similarity with the HAVE PIO data is the intra-pilot trend: Pilots A and C tended to
attain the highest oscillation frequencies, and Pilots D and H the lowest. It is significant to note
that Pilots A and C tended to identify PIO more often than the other pilots, and Pilots D and H
least often (though the entries in Table 4 may suggest otherwise, but remember that Table 4
identifies only those pilots who reported severe PIOs, not any PIOs). As it turns out, this was
especially true for Category II events, as discussed below.
Figure 23 shows the identified frequency-response data for the severe-PIO cases. Trends here
are similar to those for the HAVE PIO data shown in Figure 20. The data confirm the
observation that in PIO the airplane is approximately 180 degrees out of phase with the pilot.

2. Category II PIOs
a. Pilot Variability
With nine evaluation pilots, some differences in piloting technique and aggressiveness were
certain to appear. Generally, it was possible during the simulation to categorize the pilots as
either highly aggressive, moderately aggressive, or relatively unaggressive in their approach to
flying the task. This categorization of pilots was entirely qualitative and was based on observing
each pilot’s maximum input rates and amplitudes, and the resulting response rates and
amplitudes. It was intended only to verify that all configurations were evaluated by at least one
pilot from each category.
The most common separator among pilots was the amount of rate limiting each pilot could
tolerate for one of the highly augmented configurations with an unstable unaugmented aircraft.
At some point, all of the pilots hit the rate limiter and experienced divergent PIO. Of the nine
pilots, however, one pilot was especially capable of performing the task with even the most
severely rate-limited configurations.

46
 -90
Conf. 2DL2 Conf. 15
Conf. 1D
phase (deg)

-180

-270 A G
A H
G
Model Model Model
-360
1 frequency (rad/sec) 10 1 frequency (rad/sec) 10 1 frequency (rad/sec) 10

-90
Conf. 1DL2 Conf. 5A
phase (deg)

-180
Conf. 2DL3

A A
-270 D B E
H D F
Model Model Model
-360
1 frequency (rad/sec) 10 1 frequency (rad/sec) 10 1 frequency (rad/sec) 10

-90
A Conf. 2DL4 A Conf. 16
Conf. 1DL3
B C A
C D C
phase (deg)

-180 D E D
E F E
F G F
G H
-270 G
H Model
H
Model
Model
-360
1 10 1 10 1 10
frequency (rad/sec) frequency (rad/sec) frequency (rad/sec)

-90
Conf. 18
phase (deg)

-180 A
B
D
E
-270
F
Model

-360
1 10
frequency (rad/sec)

Figure 23. Identified Phase Angles Between Pitch Attitude and Stick Force for Category I
Severe PIO Configurations

An example of pilot variability in ratings is shown in Figure 24. This figure shows the mean
HQRs and PIORs (in several cases each pilot evaluated a configuration more than once) for three
selected configurations. One was not intended to exhibit PIOs, the second was a Category I PIO
case, and the third was a Category II case.
Configuration 2DR30 was the baseline for the simulation, and was flown multiple times by
almost every pilot. This configuration sometimes received Level 2 HQRs (ratings of 4 or 5) and
elevated PIORs (2 and 3) because of a tendency to bobble in pitch, but it did not PIO. Pilot C
considered it almost ideal, and pilot D did not indicate any objectionable bobble tendencies.
The most PIO-prone configuration, for Category I PIOs, was 1DL3, described above. Only pilot
B assigned a PIOR better than 4, and his HQR of 8 still indicates a very poor airplane. Two
pilots (A and H) thought it exhibited a tendency for divergent PIO (PIOR of 5).

47
 10
9 2D U R 20
8
7
M e a n 65
HQR 4
3 2D R 30
2
1
A B C D E F G H I
P ilot

6 2D U R 20
5
M ean 4
P IO R 3
2 2DR 30
1
A B C D E F G H I
P ilot
Figure 24. Examples of Pilot Variations in HQR and PIOR for Three Selected
Configurations from Simulation (Appendix A)

The third configuration for Figure 24, 2DUR20, was a highly PIO-susceptible airplane for
Category II PIOs. Only pilot D was able to fly this configuration through the tracking task
without losing control, and he flew the case twice. In the absence of rate limiting, Configuration
2DUR20 looked like 2DR30, and the ratings from Pilot D almost overlay for these two cases.
Once rate limiting was reached on 2DUR20, however, the airplane quickly became completely
uncontrollable with a rapidly divergent oscillation.
The results shown in Figure 24 are typical of those from the entire simulation (Appendix A).
The PIO-resistant configurations show some variation in pilot ratings, but not unusually large
when compared with past experience on ground simulation. The differences in ratings are
somewhat larger for configurations susceptible to Category I PIO, as some pilots might
occasionally complete the task without seeing any obvious PIO. But for those configurations
intended to exhibit Category II PIO with severe rate limiting, apparent pilot scatter is sometimes
very large, as with configuration 2DUR20.
As further evidence of the differences in pilot aggressiveness, Figure 25 shows the PIORs for
configuration 2DV (see Appendix A) as rate limit is reduced from 20 to 10 deg/sec.
Configuration 2DV was slightly less unstable than 2DU, and hence required lower feedback
gains to stabilize, and placed a lower demand on the elevator actuator. Even when the elevator
was fully rate-saturated, it was sometimes possible for the pilots to recover and continue flying.

48
 6

A
5 B
C
D
E
4 F
G
PIOR H

1
10 15 20 25
Rate Limit

Figure 25. Effect of Actuator Rate Limits on PIOR (Configuration 2DV)

With no saturation, this configuration looked like the baseline 2DR30, but with full saturation it
became a highly unstable airplane. The three sets of solid symbols in Figure 25, for pilots B, D,
and E, indicate the ratings from those pilots judged (entirely qualitatively by the experimenters)
as being the least aggressive during the simulation.
For the highest and middle rate limits (Configurations 2DVR20 and 2DVR15), PIORs ranged
from 1 (indicating no undesirable motions) to 6 (indicating divergent PIO simply from
attempting to enter the control loop). For both values the three “less-aggressive” pilots assigned
PIORs of 1 or 2. When rate limiting was reduced to 10 deg/sec (Configuration 2DVR10), all
pilots encountered divergent PIO (there are six ratings of 5 on Figure 25, including one from
each of pilots B, D, and E). This plot of apparent scatter in ratings is actually a sign of the highly
nonlinear nature of the problem of Category II PIO.
Historically, in classical flying qualities experiments a large variation in pilot opinion is taken as
a sign of problems with setup, task, configurations, etc. This is true only as long as the variables
are all linear in form; the very nature of rate limiting, and the potentially catastrophic
consequences it can lead to, make it a very nonlinear phenomenon. As a result, occasional large
scatter in pilot opinion and ratings (e.g., the single HQR of 2, PIOR of 1 for Configuration
2DUR20 in Figure 24) should be expected, and not come as a surprise. The fact that one pilot
out of nine was able to fly this very PIO-prone configuration is good news, not bad: we must
seek to understand how pilot D was able to adapt to such a potentially terrible configuration.

49
Some information can be drawn from examining the time history data. As an example, Figure 26
shows a 10-sec segment of selected time traces for Configuration 2DUR20. Traces are shown
for longitudinal stick deflection, pitch rate, and elevator rate. The data for four pilots are shown:
pilots A and C (two of the most aggressive pilots) and B and D (two of the least aggressive). In
this segment, pilot C encounters divergent PIO following the pull at about 45 sec; his run was
stopped at 49 sec due to the rapidly diverging response. Pilot A has just reached the same
condition at the end of the segment, and his evaluation is stopped at about 56 sec. Pilot B
managed to almost complete the run, with his run ending at 129 sec, while pilot D completed the
task without encountering divergent PIO.

2.5 Pilot A
2 Pilot B
Pilot C
1.5
Pilot D
1
Long.0.5
Stick
(in.) 0
-0.5 44 46 48 50 52 54

-1
-1.5 time (sec)
-2

40

30

20
pitch
rate 10
(d/s)
0
44 46 48 50 52 54
-10
time (sec)
-20

25
20
15
10

Elev. 5
rate 0
(d/s) -5 44 46 48 50 52 54
-10 time (sec)
-15
-20
-25

Figure 26. Sample Time History Comparison for Four Pilots (Configuration 2DUR20)

50
There are some noticeable differences in Figure 26. For example, on the pull at approximately
45 sec pilot C applies the largest input, generating the highest pitch rate, and leading to rate
saturation of the elevator. Pilot A is second in input size, resulting in a momentary saturation of
the elevator, but he is able to recover by about 46 sec. Pilot B is third in aggressiveness, and his
inputs at around 45 sec do not quite command the full 20 deg/sec of elevator rate. On several
occasions during this segment pilot B almost reaches rate saturation. Well below the other pilots
is pilot D, whose input magnitudes are almost always lowest, and his use of elevator rate is the
lowest of all the pilots.
The time for peak input follows the same progression, with pilot C generally applying the most
rapid and D the least rapid (the pull at about 45 sec is the best example). These differences
continue to show along the sequence for pilots A, B, and D, even after pilot C’s run has ended.
The data plotted in Figure 26 suggest that we can characterize the apparently less aggressive
behavior of pilot D by analyzing his closed-loop tracking activity. One way to characterize
closed-loop tracking behavior, especially in a compensatory tracking task, is to identify
describing-function models of the pilot-plus-vehicle system. 46 Fast-Fourier transforms (FFTs) of
the time-response data for θ/θerror were generated for the seven pilots who flew Configuration
2DUR20.
A measure of pilot-vehicle closed-loop tracking bandwidth is the crossover frequency from the
describing function for θ/θerror. For compensatory tracking, the crossover model46 states that the
pilot will attempt to achieve near-k/s behavior of the total pilot-vehicle system, YpYc. The
crossover frequency, where |YpYc| = 0 dB, is therefore a measure of the pilot’s tracking
bandwidth. An example of the magnitudes of YpYc for pilots A and D is shown in Figure 27.
For comparison, the magnitude of Yc alone (|θ/Fes|, units of deg/lb, converted to dB) is also
plotted in Figure 27.
This crossover frequency was found to have a remarkable correlation with the pilots’ ability to
perform the tracking task, measured in this case by the time at which the run had to be stopped
because of the divergent PIO. (For pilot D, of course, this time is total task time, since he never
encountered the PIO.) For example, the crossover frequency for high-gain pilot A in Figure 27 is
approximately 2.3 rad/sec, and for low-gain pilot D it is 0.7 rad/sec. A crossplot of crossover
frequency vs. total runtime for all seven pilots is shown in Figure 28.
The discrete tracking task is not intended to generate continuous, steady closed-loop tracking
behavior from the pilots. In general, pilots demonstrate a combination of high-gain and low-gain
tracking during a typical run with this task. As a result, generation of the describing functions
for θ/θerror is quite difficult, with large scatter and poor overall correlation. Even with these
challenges, the correlation shown in Figure 28 is excellent. Pilot D does indeed exhibit the
lowest crossover, and pilot A the highest.

b. Characteristics of Category II PIOs

It is much more difficult to identify trends for configurations prone to Category II PIO, and
especially for several of the configurations evaluated in the piloted simulation documented in
Appendix A. Simply imposing an extremely low rate limit on an otherwise good airplane did not
consistently cause PIOs to occur. Post-simulation analysis showed this was because the pilots
simply did not require high actuator rates to perform the attitude tracking task, so they spent very
little time on the rate limit.

51
 30
Pilot A
20 Pilot D
Yc
10

Magnitude
0
(dB)

-10

-20

-30
0.1 1 10
frequency (rad/sec)
Figure 27. Comparison of Pilot-Vehicle Describing Functions (YpYc) for Configuration
2DUR20 (Yc is θ/Fes for 2DUR20 with No Rate Limiting)

2.5
Pilot A

2
C
Approx. 1.5 F
crossover
frequency
(rad/sec) 1 E H
B

0.5
Pilot D
(Completed Task)
0
40 60 80 100 120 140
Run length (sec)

Figure 28. Correlation of Approximate Pilot-Vehicle Crossover Frequency with Time

Control Was Lost During the Run (Configuration 2DUR20)

52
The most consistent Category II PIOs were created by closing a SAS around a highly unstable
set of dynamics. Saturation of the rate limiter led to both a phase loss due to the limiter and total
loss of stability from the SAS feedbacks. Unfortunately, one set of dynamics (designated 2DU)
was so unstable that the pilots typically had no more than perhaps one cycle of a rapidly
divergent PIO before the simulation was stopped (a good example is the time history for Pilot C
in Figure 26). The consequence of this is that very little pilot model information can be gleaned
from the PIO itself.
The second unstable configuration (2DV) was designed specifically to be less unstable and
provide some chance at recovery following initial rate saturation. This was precisely what
happened, so information about pilot and airplane can be extracted. The best data come from
2DVR15 and 2DVR10. As with the Category I data discussed above, ROVER was applied to
the time-history files to extract information about the PIOs.
For Configuration 2DVR15, seven of the eight pilots who flew it experienced divergent PIOs
(see PIORs in Figure 25). For three pilots, the PIO occurred so quickly that there is not enough
information for ROVER to identify details of the event, or the information is a single half-cycle
with questionable accuracy. This leaves data for four pilots. Oscillation frequencies are plotted
against computed pitch attitude-to-stick force phase angles in Figure 29.

-90
C
F
G
H
small-input sweep
max input sweep
no feedbacks
-180 no FB, saturated actuator

phase
angle
(deg)

-270

-360
1 frequency (rad/sec) 10

Figure 29. Identified Frequencies and Phase Angles for Category II PIOs (Configuration
2DVR15, Moving-Base Simulation)

53
Figure 29 shows a prevalence for PIO at frequencies around 2-4 rad/sec. When the actuator is
not rate-limited, the phase angle response (shown as long dashed lines in Figure 29) is well-
behaved with a neutral-stability frequency of 6 rad/sec.
Further information on the phase angles for Configuration 2DVR15 was obtained by applying
frequency sweeps of larger and larger amplitude to the Matlab/Simulink model until the airplane
diverged during the sweep. Phase angle for the largest possible sweep is plotted in Figure 29 as
well (short dashed lines with dot symbols). This phase angle shows a more uneven slope as a
result of the combination of rate limiting and loss of feedback.
The phase angle for an analytical model of the unstable bare airframe is also shown on Figure 29
(dotted line with “x” symbols). The worst-case scenario is this unstable airplane with the phase
lag caused by full rate saturation; the phase curve for that condition (obtained by subtracting 90
deg from the phase for the unstable airplane alone) is the lowest dotted line on the figure.
While there are not many points on Figure 29, it is significant to note that the measured data are
clustered at PIO frequencies between about 2 and 4 rad/sec, and phase angles between –180 and
–270 deg. This places them midway between the phase curves for the case with no feedbacks
and the case with no feedbacks and maximum saturation. Most importantly, the data are far off
in phase angle from the full augmented airplane (top phase curve). This is what we might expect
in a PIO: the effective airplane will look much more like the unaugmented than the augmented
configuration.
The identified PIO frequencies are similar to, but slightly lower than, those for Category I PIOs
from the same simulation (compare Figure 23).
When the rate limit was further reduced to 10 deg/sec, Configuration 2DVR10, the four pilots
who flew it all experienced divergent PIOs. One, Pilot C, encountered a run-ending divergence
only a few seconds after the start of the simulation when he applied a control pulse to test the
airplane! Clearly, such a highly PIO-prone airplane is far-fetched and would never be built
(except, possibly, in the event of an unusual failure), but it is interesting as a study of pilot
behavior in a repeatable Category II situation. Identified PIO information for Configuration
2DVR10 is shown in Figure 30. Some amount of PIO information was obtained for all four of
the evaluation pilots.
The four model phase angles are shown in Figure 30. The “small-input sweep” and both
responses with no feedbacks are identical to those in Figure 29, while the “max-input sweep”
was the largest possible frequency sweep on the Matlab/Simulink model with the rate limit set to
10 deg/sec. As with the data in Figure 29, the phase angles are generally between –180 and –270
deg (the points far from this range probably indicate questionable readings from ROVER), and
the frequencies are clustered between about 2 and 3 rad/sec. There is a large amount of data for
Pilot C; following his divergent PIO during initial test inputs, mentioned above, Pilot C flew the
task and consciously tried to maintain control. He was able to complete the task, but with the
actuator rate-limited much of the time, and assigned an HQR of 9 and a PIOR of 5 for this case.
From this limited amount of data for Category II PIOs, it is clear that – as we might suspect – the
task of identifying a likely PIO-prone airplane and a likely PIO frequency is almost impossible.
By its nature, rate limiting is highly sensitive to several factors, including pilot input bandwidth,
the amount of rate limiting experienced, and the consequences of reaching the rate limit.

54
 -90 B
C
D
E
small-input sweep
max-input sweep
no feedbacks
no FB, saturated actuator
-180

phase
angle
(deg)

-270

-360
1 10
frequency (rad/sec)

Figure 30. Identified Frequencies and Phase Angles for Category II PIOs (Configuration
2DVR10, Moving-Base Simulation)

As a graphical example, consider the data plotted in Figure 31. Percent of time on the rate limit
was computed for several selected configurations, all flown by Pilot C, and the numbers are
plotted against Pilot C’s assigned PIO Tendency Rating. HQRs are noted next to each data
point. It is clear from these data that rate limiting, alone, is not the culprit in Category II PIOs.
Lines connect individual data points in Figure 31 and progress from higher to lower actuator
rates in all cases. For example, the circles are Configuration 2D with rate limits of 20 deg/sec
(PIOR = 1) and 10 deg/sec (PIOR = 3). At the lower limit, the actuator was rate-saturated for
25% of the run – a significant amount given the periods of either little or no pitch command (see
Appendix A). Still, Pilot C did not consider this airplane to have tendencies to PIO, nor did any
of the other pilots who evaluated it.
By contrast, for the sluggish Configuration 2P (squares on Figure 31), reductions in rate limit
from 180 (at PIOR = 1) to 10 deg/sec (PIOR = 5) resulted in an apparent tendency for divergent
PIO, even though the percent of time on the rate limiter increased to only 4% at the worst case.
With augmentation and no rate saturation, Configuration 2DU (diamonds on Figure 31) flew like
2D and received a PIOR of 1. A reduction in rate limit to 30 deg/sec led to an occasional
“nibble,” resulting in saturation for only 1% of the run and no degradation in ratings. When the
rate limit was decreased to 20 deg/sec, the overall percentage of time in saturation was only 10%,
but it occurred all at once and resulted in a divergent PIO and a stoppage of the simulation. The
PIOR of 6 and HQR of 10 reflect the extreme susceptibility to PIO for this configuration.

55
 6
10

5
9

6
4
7
PIOR

2D
3 5 6 2P
HQR
2DU
2DV
2 5

1
1
0 20 40 60 80
% of time on rate limit

Figure 31. PIO Tendency Rating as a Function of Percent of Time on Rate Limit for
Selected Configurations (Data for Pilot C)

Because the bare-airframe dynamics for 2DV were not quite as unstable as those for 2DU, the
trends are more gradual (triangles in Figure 31). The point with PIOR = 3 was for a 20-deg/sec
rate limit, and it resulted in saturation for only 4% of the run, but that was enough to degrade the
handling qualities (with no saturation, 2DV looked like 2D). At a rate limit of 15 deg/sec, Pilot
C was able to complete the task without crashing, with saturation 17% of the time and an
assigned PIOR of 4. At the lowest rate limit of 10 deg/sec, Pilot C completed one run only with
intense concentration, knowing that he was flying a highly PIO-susceptible configuration (this is
the run discussed above, with identified PIO information plotted in Figure 30). He managed to
maintain control of the airplane despite rate saturation for 66% of the run. His PIOR of 5 and
HQR of 9 indicate the extremely poor characteristics of this configuration.
The data shown in Figure 31 simply serve to reinforce the observations that 1) hitting a rate limit,
alone, and 2) spending considerable time on the rate limit, are not necessarily causes of PIO.
The consequences of reaching the rate limiting, and the unaugmented and augmented dynamics
of the airplane to begin with, are the key elements.

56
 V. PREVENTION OF PIO BY DESIGN: LONGITUDINAL CRITERIA FOR
CATEGORY I PIOS

A. BACKGROUND
This section reviews several alternative PIO prediction criteria that are aimed specifically at
Category I longitudinal PIOs. Similar criteria for Category II PIOs are left for the next section,
followed by a separate discussion of criteria for lateral PIOs in the following section. The focus
in this section, and in this report, is on prevention of PIOs for any airplane, with any level of
augmentation, any response type. In the mid-1990’s there was a brief flurry of activity with
some modal criteria developed three decades ago, and because of this interest, a brief discussion
of those criteria is warranted. The discussion is left for the end of this section, after we review
criteria intended for all airplanes.

1. Prevention by Design
The best method for prevention of PIO is by designing the airplane to be resistant to PIOs. In
truth, of course, no airplane is ever intended to be susceptible to PIO, and it is usually through
design compromises and shortcomings that an airplane comes to be susceptible to PIO once it
has been built. Still, much can be done to minimize the risk if well-developed analytical criteria
are applied as early in the design process as possible.
Until recently, there have been few true PIO prediction criteria, at least in the published
literature. Traditional flying qualities criteria, such as those in MIL-STD-1797A, may include
explicit mention of PIO, but most do not. It is well-known to the experienced flying qualities
engineer that a common manifestation of poor handling qualities* is, in fact, PIO: almost every
flight research airplane ever flown with Level 3 handling qualities has exhibited a tendency to
PIO. Many Level 2 aircraft have, as well. The traditional view, from the flying qualities
engineer’s perspective: make the airplane Level 1, and the likelihood for PIO is reduced.
Unfortunately, there has been an air of distrust of the traditional flying qualities criteria in the
design community.2,3,47 This is not totally surprising, given frequent conflicts in predicted flying
qualities Levels between the criteria, and the perceived difficulty with application of some of the
criteria.45 A conclusion drawn by engineers working on both the YF-22 and the C-17, however,
is that, even with their conflicting predictions, all criteria have something to say, and mature
criteria can indeed be useful in a development program.3,47
More recently – in the 1990s especially – there has been a more dedicated focus on development
of PIO-specific criteria. This is both in response to the perceived lack of applicability of the
traditional flying qualities criteria, and the recognized need to avoid PIO, even if Level 1 flying
qualities are not obtained. Now it has further become apparent that most PIOs experienced by

*
Here we distinguish “handling qualities” from “flying qualities.” By the former we mean the
characteristics of the airplane with the pilot in the loop, encompassing both the dynamics
described by analytical criteria and the overall handling of the aircraft. “Flying qualities” are
interpreted as those characteristics measured by analytical criteria and parameters, specifically
directed toward objective quantities – even though the criteria themselves are based on
subjective pilot opinion. Occasionally in this report the terms are used interchangeably.

57
“real” airplanes – prototype and operational aircraft, as contrasted with variable-stability research
aircraft – are not linear in nature, involving some degree of control surface or software rate
limiting. None of the traditional criteria of MIL-STD-1797A and its ilk are aimed at such events.
The focus of the latter half of the 1990s has been on criteria to predict such Category II events.
This section presents a comparative evaluation of Category I PIO prediction criteria. The entire
data base of flight research on handling qualities can be applied to such criteria as well. By
contrast, the few criteria for Category II are much newer, of necessity more difficult to apply,
and have a much smaller data base for evaluation. A set of criteria developed under this
contract, based on pitch attitude Bandwidth,5 are considered to be sufficiently mature to be
applied to both Category I and II events (see next section for application to Category II PIOs).

2. Flight Experience with PIOs

Before the introduction of artificial augmentation and boosted controls, almost all PIOs were
Category I in nature; that is, they were due to some deficiency in the aerodynamic design of the
airplane, such as low short-period damping or low control force per g. As fly-by-wire flight
control systems were introduced, PIOs were related to excessive delays coming from filter lags
and computational delays. Force-sensing controllers were introduced to high-performance
airplanes, and excessive filtering was used to minimize the adverse effect of force commands.
Recently, most PIOs on highly-augmented, fly-by-wire airplanes have been Category II or III
events. This includes the YF-22 at Edwards AFB,2,3 two occurrences with the Swedish SAAB
JAS-39 Gripen,33 and early flight tests of both the C-17 and the Boeing 777 transport aircraft.47,48
There is a very large data base for Category I PIOs – much larger than that for Categories II and
III combined. This Category I PIO data base comes from flight research, most of it conducted by
Calspan Corporation under sponsorship of the Air Force or NASA. In fact, any handling-
qualities research experiment performed on a variable-stability aircraft serves to augment the
data base. As discussed later in this paper, there is a subset of this data base that is considered
valid for PIO criteria assessment and development, and it is this subset that will be used in this
section for comparative evaluation of the criteria.
The flight research data base for Category II PIOs is relatively very small. The lack of an
extensive set of research data is a consequence of the focus on handling qualities issues in
general, and not PIO specifically, for the past 40 years or more. PIO is one (but not the only)
result of an airplane with deficient handling qualities, and it commonly occurs during flight
research. The recent concern about PIOs and other aircraft-pilot coupling (APC1) has led
directly to several flight research studies of rate limiting.49,21,22
That most of our experience with PIO comes from flight research is both good and bad. Since it
occurs in a constrained environment, on an airplane with known dynamic characteristics, for a
specific flight task, it is possible to analyze the PIO in great detail. On the other hand, the
constrained environment also means that the pilot may not react as one might when encountering
a PIO in an experimental or operational airplane, with no safety pilot, and with no expectation of
such an event. The research test pilot may be pre-loaded to expect the PIO and hence adapt to
avoid it – a luxury the operational pilot does not have. Certainly the occurrence of PIO for a
particular variable-stability configuration indicates a serious problem, but the lack of such
occurrence does not necessarily indicate that the airplane is PIO-free. After all, the typical
variable-stability “airplane” in a flight research program may have an existence of only a matter
of minutes, seen by as few as two (and occasionally only one) skilled test pilots. The purpose for

58
using an experienced test pilot, in a highly constrained and demanding task, is to attempt to make
up for the lack of exposure time, but it is by no means guaranteed.
By contrast, the little data we have from “real” (operational or experimental) PIOs is extremely
valuable, but there is always some question: Do we really know the dynamics of the airplane
just before the PIO? Do we know what triggered the event? Etc.

3. PIO Prediction Criteria

The lack of a real-world data base, and the relative wealth of research data, has, not surprisingly,
led PIO researchers to focus on the latter for all analytical work in the development of criteria for
the prediction of PIO susceptibility. The result is that there are many criteria in current use, but
none of the most well-known directly address nonlinear phenomena.
As this section shows, the use of the best current PIO susceptibility prediction criteria will
greatly reduce the chances of encountering a Category I PIO. This is a natural fallout of the data
base used for all of the criteria.

B. PREREQUISITES FOR CRITERIA

The National Research Council’s Committee on the Effects of Aircraft-Pilot Coupling on Flight
Safety1 define three prerequisites for PIO criteria. These prerequisites should be kept in mind
through the rest of this section, and for that matter, the two sections that follow. The NRC
Council’s prerequisites are as follows:
Validity: “implies that a criterion embodies properties and characteristics that define the
environment of interest and are associated with parameter spaces covering the vast majority of
known cases…. [T]he criterion must relate to closed-loop, high-gain, aggressive, urgent, and
precise pilot-control behavior.”
Selectivity: “demands that the criterion differentiate sharply between ‘good’ and ‘bad’
systems…. [T]he most important selectivity feature is the capability of distinguishing
configurations that may be susceptible to severe PIOs from those that are not.”
Ready applicability: “requires that the criterion be easily and conveniently applied. Expression
of the criterion in terms of readily available system parameters should be compact.”

C. CRITERIA CONSIDERED
Few criteria in the published literature deal explicitly with PIO, and most of these address pitch
PIOs (as opposed to roll or yaw). This is, perhaps, not surprising since the most spectacular and
destructive PIOs have almost always been longitudinal. Longitudinal PIOs also usually result in
the largest acceleration excursions, and the number of contributing factors (including control and
response dynamics) is naturally larger longitudinally than laterally or directionally.
The review of criteria presented here is not meant to be exhaustive. The intent has been to
evaluate those criteria that are readily available from published literature, or that have been
considered by experts to show some promise, using selected subsets of the minimum data base.
There are some known criteria and evaluation tools not discussed here, but they are either not
regularly applied or not well documented.
The goal is not to find criteria that will successfully predict the occurrence of PIOs; this is an
unreasonable objective given the sporadic nature of PIOs. Because of the large number of

59
variables that must all come together to result in a PIO (including a trigger to initiate the
oscillation), it is impossible to ever devise a criterion that will accurately predict the occurrence
of a PIO. The goal, instead, is to find those criteria that will, with reasonable confidence, predict
the susceptibility to PIO for a particular aircraft. It is not expected that all aircraft that fail the
best criteria will have experienced a PIO, or that all that pass will never have experienced a PIO.
But it should be true that those that fail have a strong susceptibility to PIO, while those that pass
are regarded as being resistant to PIO.

D. CRITERIA NOT CONSIDERED

Many existing handling-qualities criteria may be applicable to prediction of PIO susceptibility
with some modification. No attempt has been made to either assess the effectiveness of these
criteria for prediction of PIOs or adapt them for this purpose. Criteria that may be considered
very effective at predicting handling qualities Level were not evaluated here unless they
explicitly also addressed PIO.
As an example, the classical requirements on short-period damping and frequency are, in effect,
requirements on PIO prevention, to the extent that low damping or low frequency can lead to
PIO. Low short-period damping is, in fact, a common contributor to PIO. But the classical
short-period damping and Control Anticipation Parameter criteria were developed to specifically
deal with handling qualities, not PIO. It may be possible, with new research, to extend these
criteria (along with limits on equivalent time delay) to PIOs. But the known shortcomings in
applying these "classical" criteria to highly-augmented airplanes, 50 and the requirement to obtain
equivalent-system models of high-order responses in order to apply them, make such an effort
not worthwhile.
One set of criteria evaluated refers to PIOs but without a clear “Yes/No” delineation: the Neal-
Smith criteria, 51 with regions of handling qualities Levels defined but with PIOs mentioned only
in annotations on the handling-qualities plot. This text is commonly not shown when the criteria
are published in the literature. In fact, in their most recent incarnation (MIL-STD-1797A4), there
is no mention of PIO whatsoever.
Several other criteria have recently been proposed, primarily for the prediction of Category II
PIOs, that may also be applicable to Category I events. Some are simply too new to have been
critically analyzed in any detail, while others appear to be too complex to justify their application
for Category I. These criteria are discussed separately later in this section.

E. COCKPIT CONTROLLER CHARACTERISTICS

Most criteria assume good cockpit controller characteristics and control/response gradients.
Only one, the Gibson criteria, directly incorporates the force/response effects of the cockpit
manipulator.

F. THE CRITERIA SELECTED

Six criteria for the prediction of susceptibility to Category I pitch PIOs were selected for
evaluation. The criteria, and the input data and output parameters required for their application,
are listed in Table 5. Supplemental descriptions for the criteria are given in the following.

60
 Table 5. Pitch PIO Susceptibility Criteria Evaluated
Criteria Input Data Output Parameters
Bandwidth/Pitch Rate Frequency response plots of ω BWθ , τ pθ , ω BWγ , ∆G (q), ω180θ
Overshoot52 θ/Fes , q/Fes , γ/Fes

Neal-Smith (original Transfer function of θ/Fes θ/θ c , ∠pc

max
definition51)
Neal-Smith (modified by Transfer function of θ/Fes (plus θ/θ c max
, ∠pc (as applied
Moscow Aviation optimal control pilot model if here)
Institute53) applied as intended by MAI)
Smith-Geddes54 Frequency response plots of S, ωc , ∠θ/Fes ( jωc ), Φ( jω c )
θ/Fes , a z p /Fes

Gibson7 Frequency response plot of Limit on θ/δ es , phase rate,

θ/δ es (gain corrected for force
ω180θ
feel system)
Time-Domain Neal- Transfer function or Simulink 2nd derivative of rms tracking
Smith18 model of θ/Fes d 2 (rmsθ e | t >D )
error,
dD 2

1. Airplane Bandwidth/Pitch Rate Overshoot

a. Theory Behind the Bandwidth Criteria
Pitch attitude Bandwidth criteria, measured from a frequency response (Bode plot) of attitude to
control force as sketched in Figure 1 (Section I), were developed for evaluation of the handling
qualities of highly-augmented airplanes where more conventional criteria cannot be easily
applied.5 The criteria are included in the Flying Qualities Interface Standard4 MIL-STD-1797A.
(The limits in MIL-STD-1797A have been found to be much too stringent and have been
adjusted significantly, especially with the addition of a requirement on pitch rate overshoot.41)
They have been adapted to the prediction of PIO susceptibility as shown Figure 3 (also in
Section I).
The fundamental theory behind “airplane Bandwidth” – which is not the “bandwidth” that is
used in other control systems applications – is that the principal stability characteristics of the
aircraft are described by the frequency response of angular attitude for control inputs. This is
true, at least, for continuous closed-loop control of attitude by the pilot, and when attitude is used
as an inner loop to generate changes in load factor or flight path. The concept of “airplane
Bandwidth” is that the aircraft should have good inherent stability, whether from basic design or
by augmentation with SAS. The lower this inherent stability, the more stability the pilot must
provide to perform required tasks, resulting in increasing
workload÷½÷½÷½÷½÷½÷½÷½÷½rmance, and poor flying qualities.
There are three measures in the criteria that capture the basic pitch attitude characteristics of the
airplane. The first is the “phase margin Bandwidth frequency,” the lowest frequency for which

61
there is a phase margin of 45 degrees. The higher this frequency, the better attitude follows
control inputs: if phase margin is 180 degrees, that is, phase angle is zero, then output follows
input exactly. At the frequency for 0 degrees phase margin – the “neutral-stability” or -180-
degree frequency – attitude is exactly out of phase with inputs. If the phase margin Bandwidth
frequency is very low, the pilot must generate lead to improve the overall response of the pilot-
plus-aircraft system in order to do a task.
The second measure is the “gain margin Bandwidth frequency,” and it is basically the same type
of measure, except it determines the change in effective-aircraft dynamics the pilot will
encounter if closed-loop gain is increased by a factor of 2 (6 decibels).
The third measure, incorrectly named “Phase Delay,” is really a measure of how rapidly the
phase angle of attitude/control inputs degrades at high frequencies. The assumption is that, if the
pilot should find it necessary to operate at higher frequencies – which can be done with closed-
loop stability only if the pilot generates lead compensation – a gradually-degrading phase curve
is much better than a rapidly-degrading one. Phase delay has been mistakenly equated with pure
time delay, which it is not. In fact, for most modern aircraft the high-frequency
loss÷½÷½÷½÷½÷½÷½÷½÷½ much more to filters and lags (such as actuators) than to real time
delay. The units are (deg of phase)/(rad/sec of frequency), and with a unit change from deg to
rad, become rad/rad/sec, or seconds – hence the “delay” in the name of this parameter.
The relative importance of frequency-domain, as opposed to time-domain, criteria for flying
qualities has its roots in the fact that precise piloting is fundamentally a closed-loop
phenomenon. This assertion takes into account the fact that pilots are highly adaptable and
develop precognitive control strategies to overcome handling deficiencies. Such precognitive
control is useful for most flying tasks, but fails the pilot when faced with critical scenarios such
as landing on a short runway in strong turbulence and windshear. In such cases, desirable
responses arise from the pilot having the ability to continuously monitor and effect changes in
the aircraft state based on immediately evident cues.
For this to happen, the closed-loop system must have adequate response magnitudes at
frequencies encompassing the input disturbance spectrum, and the open-loop frequency response
behavior must allow the operator to effectively close the loop at a variety of gains, depending on
the urgency of the task. The former characteristic is determined by the Bandwidth frequency
ωBW, and the latter is related to how rapidly the phase curve rolls off above the crossover
frequency. Phase Delay (τp), a measure of the shape of the high-frequency phase curve of the
pitch attitude frequency response, has been established as a
usef÷½÷½÷½÷½÷½÷½÷½÷½describe this phase rolloff.
Pilot complaints of “abruptness,” overshoot, and settling when attempting precise attitude
control, led to the incorporation of a frequency-domain measure of pitch rate overshoot as a
measure of mid-frequency pitch command-following ability. Some deficiencies in basic flying
qualities are not captured by the Bandwidth parameters alone, and for a time, a modified version
of Gibson’s Dropback versus pitch rate overshoot criterion55 was adopted to supplement the
basic Bandwidth criteria. Proper application of Dropback requires that the airplane exhibit no
noticeable phugoid motion; attempts to apply the criterion in a flight test program demonstrated
that a time-domain metric based on steady-state pitch rate simply could not be applied in the real
world.56 A frequency-domain measure of pitch rate overshoot has instead been developed to
supplement the Bandwidth criteria.

62
b. Parameters for the Bandwidth criteria
1. Pitch attitude Bandwidth frequency, ω BWθ (Figure 1), is a measure of the operating range a
pure-gain pilot has for controlling the airplane without threatening stability, that is, with good
phase and gain margins.
2. Pitch attitude Phase Delay, τpθ (Figure 1), describes the rate of phase rolloff beyond the
neutral-stability frequency, ω180θ . If the pilot finds it necessary to apply lead compensation,
because of either a low airplane Bandwidth or high task demands, a low value of Phase
Delay indicates relative insensitivity to small changes in pilot gain. High values of Phase
Delay limit the pilot’s ability to operate at frequencies well above the Bandwidth frequency,
and PIO often results.
3. Flight path Bandwidth, ω BWγ , is defined as the phase-margin Bandwidth frequency of the
flight-path-to-stick-force response. A low flight path Bandwidth indicates a need for pilot
lead generation to improve the response; in combination with a moderate value of Phase
Delay, PIO is possible.
4. Pitch rate overshoot, ∆G(q) (Figure 2), is a frequency-domain measure of overshoot,
normally defined in the time domain. High pitch rate overshoot has been known to be
undesirable, resulting in bobble tendencies and excessively abrupt short-term response.
Unfortunately, measures of overshoot in the time domain typically use a ratio of peak rate to
steady-state rate, and the latter has proven to be difficult to determine in flight.56 The
frequency-domain parameter defined in Figure 2 does not suffer from this shortcoming. In
addition, it correctly references overshoot to the mid-frequency response of the airplane,
where the pilot is normally applying closed-loop control, rather than to the steady-state
response.
5. Neutral-stability frequency, ω180θ , has been shown to be close to PIO frequency41 for
Category I PIOs and is used by these criteria for that purpose. Correlation with flight-
identified frequency was improved if a factor of 0.5 rad/sec is added to estimate the
frequency of the PIO.
The Bandwidth/Phase Delay criteria in Figure 3 separate PIOs by severity, and include a metric
for predicting the possibility of pitch bobble (which, to some pilots, may be viewed as a very
minor PIO). PIO severity, in keeping with the discussions in Section III, is defined as follows:
1. “Bobble” should be considered separately from PIO, though we also characterize it as “mild”
PIO. Pitch bobble is related to excessive pitch rate overshoot, and as long as the basic
Bandwidth characteristics of the airplane are good (Bandwidth frequency above 1 rad/sec and
Phase Delay below 0.14 sec) it should not develop into a full-blown PIO. A limit on pitch
rate overshoot of 9 dB is equivalent to a limit of 3 on a linear scale, meaning an overshoot
ratio limit of 3 in the time domain definition of qpk/qss.
2. “Moderate” PIO corresponds to an event that normally would elicit a PIO tendency rating no
worse than 4 (possibly even a 3, depending upon the specific rating scale and the pilot’s
internal definition of PIO) and a Cooper-Harper Handling Qualities Rating of no worse than
6.

63
3. “Severe” PIO is expected to receive both degraded PIORs of 4, 5, or 6, and HQRs of 7, 8, 9,
or 10. Note that severe PIO is predicted almost entirely on the basis of the value of Phase
Delay, τp θ : the limits for PIO in Figure 3 are generally either vertical or horizontal lines.
Recently published data from a series of flight experiments, performed by Boeing (formerly
McDonnell Douglas) Phantom Works with a simulated transport airplane, 57 suggests the
severe-PIO line should look more like the thin diagonal Level 2 handling qualities lines in
Figure 3.

c. Evolution of the Bandwidth criteria

As with almost every short-term pitch response criterion, the Bandwidth criteria owe their
origins to the Neal-Smith51 data base. Because of the selection of added lead and lag filters,
there were obvious problems in attempting to apply classical (equivalent systems) short-period
requirements to the Neal-Smith data.51 The Bandwidth concept was originally developed to
account for unconventional response modes in highly advanced aircraft with control of six
degrees of freedom.58 This was later extended to include Phase Delay to address the unusual
nature of the configurations in the Neal-Smith study. Using pitch attitude alone, requirements
for both up-and-away tracking and precision landing were developed for the military handling
qualities standard.39 There was no attempt at that time to extend the criteria to PIOs.
It was recognized by the authors of the criteria that the pitch attitude Bandwidth requirements
were quite stringent.59 But such stringency was considered to be justified by a number of
configurations in the Neal-Smith data base with Level 2 handling qualities and moderately high
Bandwidth frequencies. More recently, these cases have been shown to have high overshoot
characteristics, and the addition of the pitch rate overshoot criterion has allowed a considerable
relaxation in the Bandwidth requirements.45
Efforts to develop a flight path Bandwidth requirement for landing have been extremely
successful,45 while such requirements for other flight conditions simply lack a solid data base.

d. Strengths of the Bandwidth criteria

The Bandwidth criteria are based on requirements for piloted control. As a result, they should
apply to any set of response dynamics (i.e., response-type). And unlike the Neal-Smith criteria
(described below), they require only frequency and time responses, not transfer functions, and
hence can be applied at any stage of the development process.
The criteria combine well-recognized elements of requirements from other researchers. The
Phase Delay parameter was developed by engineers at McDonnell Aircraft Co., in an effort to
define an alternative to equivalent time delay that was based on phase rolloff and not on
mathematical algorithms. Pitch rate overshoot has long been proposed as a time-domain flying
qualities parameter, so its adaptation to the frequency domain seems only logical. The Phase
Delay parameter is related to Gibson’s phase rate,7 with similar limits for the prevention of PIO.

e. Shortcomings of the Bandwidth criteria

Parameters for the Bandwidth criteria must be measured from frequency responses derived from
analytical models, simulation models, or in-flight data. Therefore, the criteria are only as good
as the data from which the measurements are made. (This is, of course, true of every PIO
criterion). The criteria assume linear response and a more or less linear transfer function. There
have been instances in which response modes at low frequencies (below 2ω180, see Figure 1)

64
introduce nonlinearities that require some special care. This typically occurs for structural
modes on helicopters, however, and is relatively rare for fixed-wing airplanes. Nonlinear
phenomena such as rate limiting reduce data quality and can have an impact on the measurement
of the Bandwidth parameters (a subject left for Section VI.)

2. Neal-Smith (Original Definition)

a. Theory Behind the Neal-Smith Criteria
As with airplane Bandwidth, the Neal-Smith criteria were developed as handling-qualities
requirements for airplanes with nonconventional responses. And as with Bandwidth, the
underlying hypothesis behind the criteria is that good attitude control is essential for any closed-
loop operations.
The original Neal-Smith requirements explicitly referred to handling-qualities Levels but only
indirectly addressed PIOs (Figure 32). While there is a region (corresponding to handling
qualities Level 3) where PIO tendencies are mentioned, there is no clear “no PIO/PIO” dividing
line on Figure 32. “Strong PIO tendencies” are indicated throughout the Level 3 region, so this
is clearly a region where PIO is predicted, but “tendencies to oscillate” in the middle of Level 2
might also signify a milder PIO tendency. For this analysis it was assumed that the Level 3
region corresponds to a prediction of susceptibility to PIO.
The Neal-Smith criteria were originally developed in the early 1970s.51 They have undergone
more or less continuous application and evolution since then. The Neal-Smith criteria are based
on piloted closed-loop attitude control during tight tracking. The data base was generated for
general maneuvering and compensatory attitude tracking in up-and-away flight, where control of
flight path was of secondary importance (though the overall normal acceleration response may
still have been important, at least as far as it affected ride qualities).

Figure 32. Neal-Smith Criteria

65
As originally developed by Neal and Smith, the parameters in the criteria are obtained by finding
the dynamics of the pilot model given by
(TI s + 1) -τ p s
Yp = K p e
(TL s + 1)
for the pitch attitude dynamics of the airplane described by Yc = θ/Fes. The pilot time delay, τp,
is a fixed value (Neal and Smith assumed 0.3 sec) and Kp, TI, and TL are varied to meet specific
performance constraints. In the original Neal-Smith analysis, the performance criteria were 1) a
specified closed-loop bandwidth, BW (where the phase of the closed-loop system, ∠(θ/θc), is -90
degrees), of either 3.0 or 3.5 rad/sec; and 2) closed-loop droop of exactly -3dB.
Determination of the best pilot model is not a trivial task when performed manually. Most users
of the Neal-Smith criteria have developed software that will perform the loop closure process
automatically.

b. Parameters of the Neal-Smith Criteria

1. Closed-loop resonance, |θ/θc|max, is the peak amplitude in the closed-loop frequency response.
2. Phase angle of pilot compensation, ∠pc, at the bandwidth frequency.
Figure 32 shows the Neal-Smith criteria from the original report.51

c. Evolution of the Neal-Smith criteria

In the three decades since the Neal-Smith criteria were introduced, numerous researchers have
applied and modified the criteria as shortcomings have surfaced. Primary among these
modifications have been adjustments in the criteria to apply them to the landing task. As is the
case with most such criteria, there has, unfortunately, been a tendency to adapt the criteria to fit
the most recently generated data base, with no attempt to determine the effects of the adaptations
on the effectiveness of the criteria for prior data. More importantly, as mentioned above, this
evolution has also served to eliminate any reference whatsoever to PIOs. The focus has been on
handling qualities only.
In almost every published reference where the Neal-Smith criteria are used there is some
modification. Indeed, it is difficult to find more than two references where the criteria were
applied in exactly the same way.
As an example, Table 6 summarizes the application of the criteria as reported in several recent
references. As this table indicates, the most common changes to the methodology of the original
Neal-Smith criteria (the first reference in Table 6) have included: 1) reduction in assumed pilot
time delay from 0.3 to 0.25 sec, and sometimes 0.20 sec; 2) computation of the parameters for a
range of bandwidths to look for trends as well as correlations; 3) addition of a pilot proportional-
plus-integral filter to improve the ability to achieve reasonable closures; 4) relaxation in the
droop requirement, from a rigid requirement of exactly -3 dB to a general goal of no more
negative than -3 dB; 5) removal of the pilot compensation parameter entirely as a measure of
handling qualities Level. There has also been some effort devoted to development of an
alternative that captures the same basic vehicle characteristics, but that does not require any loop
closures. This alternative was first proposed by Neal and Smith.

66
 Table 6. Some Examples of Application of the Neal-Smith Criteria
Pilot
BW Delay Performance
Ref. Aircraft Task (rad/sec) (sec) Criteria Comments

AFFDL- Fighter Up-&- 3.0 and 3.5 0.30 Droop ≡ -3 dB Original development of
TR-70- (NT-33A) away criteria; recommend using BW
7451 man- = 3.5 rad/sec, but used lower
euver, BW for lower n/α (resulted in
pitch 81% correlation for data based
track on HQRs)

AFFDL- All All 3.0 (Cat. A, N/A N/A “Simplified” Neal-Smith (no
TR-72- Class I,IV); loop closures); used previous
4160 2.0 (Cat. A, data to refine approach
Class II,III); originally published in Neal-
1.2 (Cat. B,C, Smith report
all Classes)

AFFDL- Fighter Ldg. 1.5, 2, 2.5, 3 0.30 Droop ≡ -3 dB No BW yielded measurable

TR-78- (NT-33A) correlation: “It is clear that
12243 correlation cannot be achieved
without some modification to
the correlating parameters”

NASA Transport Ldg. 2.5 0.25 Droop ≤ -3 dB Applied to only four

CR- (SCR) configurations in report
15923661
NASA Fighter Ldg. 3.0 0.20 None Reanalyzed AFFDL-TR-78-
CR- (NT-33A) 122 data, found good
16309762 correlation with slightly
revised Level 2 limits

AFWAL Fighter Ldg. 0.5, 1.5, 2.5, 3.5 0.30 Not stated Applied to both pitch and roll;
-TR-81- (NT-33A) BW of 2.5 worked best in
311663 pitch, but no correlation
possible in roll

NLR TR Transport Ldg. 1.2 0.30 Droop ≤ -3 dB Shows good correlation, but
83037 (NLR reported pilot models do not fit
U64 Ground & with reported aircraft
Flight dynamics; reanalysis shows
Sim.) very different results

AFWAL Large Ldg. 1.5 0.25 Droop < -3dB; Added (5s+1)/s to pilot model;
-TR-81- (107 lb) Resonance < +3dB not possible to obtain good
311865 (TIFS) closure for some
configurations

NASA Trans- Ldg. 1, 1.5, 2, 2.5, 3, 0.25 Droop ≤ -3dB BW of 2.0 appeared to work
CR- port 3.5 best
17249166 (TIFS)

MIL- All All 3.5 (Cat. A); 2.5 0.25 For Level 1: Allows addition of (5s+1)/s to
STD- (Ldg.); 1.5 (All Droop ≤ -3dB; pilot model
1797A4 other Cat. C and Resonance ≤ +3dB
all Cat. B)

67
d. Strengths of the Neal-Smith criteria
It is clear that there is a large user community for the Neal-Smith criteria. In their original form,
they were as effective as any criteria ever developed at explaining the results of the Neal-Smith
data.
The Neal-Smith criteria focus on pitch attitude control, which is essential as an inner loop for
any tight closed-loop control task. They are based on well-founded principles of pilot control in
a compensatory task and hence are physically insightful. Evaluation of the effects of changes in
closed-loop bandwidth frequency provides an indication of the sensitivity to changes in pilot
gain.

e. Shortcomings of the Neal-Smith criteria

It is apparent from Table 6 that there are many different variants of the Neal-Smith criteria.
Even in the original report51 it was necessary to use two different values of bandwidth for the
two different flight conditions evaluated (i.e., different n/α). Yet most researchers have used
only a single value, usually 3.5 rad/sec, for all subsequent analysis of aircraft handling qualities
for the up-and-away condition.
In the most recent version listed in Table 6, taken from MIL-STD-1797A, the criteria bear little
resemblance to the original. All requirements on pilot compensation are gone, as is any mention
of PIO tendency. Several different bandwidths are specified, depending upon the flight phase
Category. Yet in the course of this review it was not possible to find any detailed data
correlations that supported any of the numbers given. In fact, since the application of the criteria
is different from that originally used by Neal and Smith (i.e., time delay reduced from 0.3 to 0.25
sec, droop requirement relaxed, pilot integration allowed), it would be surprising if a bandwidth
of 3.5 rad/sec were still appropriate for the Neal-Smith data base.
The criteria are based on fundamental theories of compensatory attitude control. As was
demonstrated in Section IV, this may be a reasonable hypothesis for pilot behavior in a PIO if it
is encountered during attitude tracking, but for PIO in a task such as landing the pilot may be
adopting a control strategy that is more closely aligned with synchronous behavior, or a
combination of compensatory and synchronous. In applying the criteria (original definition) to
the LAHOS landing data base,43 the author of that report concluded that “correlation cannot be
achieved without some modification to the correlating parameters.” The complete reliance on
attitude control may be the reason for this lack of correlation.
Determination of the appropriate pilot time delay appears to be very much a function of
individual preference. Neal and Smith refer to early work by Systems Technology, Inc., that
indicates a pilot delay between 0.2 and 0.4 sec. For simplicity, they took 0.3 sec based on
examination of the time records in their flight experiment. In a study of flying qualities
requirements for Supersonic Cruise Research (SCR) aircraft,61 Chalk recommends a time delay
of 0.25 sec based on “delays observed in records for the discrete tracking task” and “empirical
correlation of data” from the Neal-Smith51 and LAHOS43 reports. For compensatory tracking
with good (k/s-like) vehicle dynamics and a reasonably high forcing function bandwidth, the data
presented by McRuer and Krendel11 suggest a pilot time delay closer to 0.1 sec. This latter
number has been supported by pilot model analysis of data from ground-based simulations.67
While the specific delay to use may not be critical, and in fact probably should vary with
bandwidth frequency, this serves to illustrate one of the shortcomings of the criteria as different

68
researchers try to determine the “best” pilot delay for use in the analysis. A similar discussion
could be presented for the selection of bandwidth frequency, as well as for the limits on closed-
loop droop and resonance. Droop, in particular, has been addressed several different ways, from
the original rigid enforcement of a droop of -3 dB, to a requirement that the droop be no more
negative than -3 dB, to elimination of the droop constraint entirely. (It is important to recognize
that a requirement to achieve a specified droop is the only way to assure a unique answer for the
pilot model in the Neal-Smith analysis. Without this requirement a wide range of pilot models is
possible, each with different values of closed-loop resonance and pilot compensation.)
There are occasionally combinations of aircraft dynamics for which the Neal-Smith criteria, at
least as originally defined by Neal and Smith, simply will not work. Typically this indicates a
requirement to reduce the target bandwidth frequency until a solution is found, and fortunately
such aircraft will have poor handling qualities and PIO tendencies to begin with. (Examples of
this, in the Neal-Smith data base, are Configurations 1G, 4E, and 5E.) In some cases (such as
Neal-Smith Configurations 6B, 6C, and 6D), the bandwidth must be increased to obtain a
solution. And finally, if there is a pronounced phugoid mode it may be impossible to obtain a
reasonable solution at any bandwidth, simply because either low-frequency error or mid-
frequency droop will be excessive. This has occurred even for aircraft with otherwise good
short-term handling qualities, and even when the pilot integrator term (5s+1)/s is added. There
are, in some instances, cases for which the criteria cannot be applied, such as for attitude
command systems.
The Neal-Smith criteria have difficulty with unconventional airplane response-types, such as
attitude command systems. They will almost always predict closed-loop control problems and
PIO potential for such systems.
The Neal-Smith criteria are not suited for use with flight data. Most computer-based algorithms
use transfer-function models for application of the criteria. Flight-determined frequency
responses may be applied, but only if the graphical approach adopted by Neal and Smith is used
– or if some clever method can be devised to get the required information from the flight data.
Equivalent systems, for example, may be used to get the necessary transfer-function models, but
the use of equivalent systems itself opens the door for problems.50

3. Modified Neal-Smith Criteria (MAI)

a. Theory Behind the Modified Neal-Smith Criteria
Researchers at the Moscow Aviation Institute in Russia, under U.S. Air Force sponsorship,
investigated the parameters of the Neal-Smith criteria in a ground-based simulation experiment.53
Selected configurations from the Neal-Smith51 and LAHOS43 experiments were evaluated for a
pitch tracking task and PIO ratings were obtained. The closed-loop resonance and pilot
compensation parameters were measured directly from pilot describing functions, and from these
parameters boundaries for handling qualities and PIO were developed. The PIO boundaries are
shown in Figure 33, along with the data for the configurations evaluated in the simulation.
The correlation between PIO ratings and boundaries in Figure 33 is very good. Since the data
come from experimentally-derived results, however, there is still no clear method for applying
the Neal-Smith criteria analytically. Questions about appropriate bandwidth frequencies and
pilot delay still must be addressed if the Figure 33 boundaries are to be put to practical
application.

69
 Figure 33. PIO Boundaries for Neal-Smith Criteria Using Experimental Data (MAI
Study;53 Figure Reproduced from NRC Report1)

The approach taken in this report does not follow the recommendations of MAI. The MAI
approach involves the use of modified optimal-control pilot models (MOCM) to obtain the
parameters for Figure 33, but attempts to replicate their results have not been totally successful.
For the purposes of this report, we have not attempted to replicate the MAI results or follow their
procedures. Instead, we are using the modified boundaries of Figure 33, with the original Neal-
Smith parameters described above. This mix of approaches certainly could contribute to a
reduction in overall effectiveness, but the effort required to adhere entirely to the MAI approach,
and resolve the problems encountered by other researchers, is not justified.

b. Parameters of the Modified Neal-Smith Criteria

As indicated above, strict application of the MAI version of the criteria requires use of modified
optimal control pilot models to obtain the overshoot and pilot compensation parameters. For
Figure 33 these parameters were measured from describing-function data taken in a ground
simulation. For the purposes of the following analysis, the parameters that will be applied are
those from the application of the Neal-Smith criteria as described above.

c. Evolution of the Modified Neal-Smith Criteria

The MAI-developed methods are relatively new (the Air Force technical report53 was issued in
1996).

d. Strengths of the Modified Neal-Smith Criteria

This approach allows for application of MOCM methods for pilot-vehicle analysis. Direct
indicators of pilot workload and performance can be predicted. Taken as new boundaries only,
the MAI limits are supported by actual pilot performance measurements rather than analysis
alone.

70
e. Shortcomings of the Modified Neal-Smith Criteria
MOCM techniques are by nature complex and subject to large variations in the models with
relatively small changes in input parameters. Some researchers (results unpublished as of this
writing) have had difficulty following the MAI procedures and obtaining their results.
As with the original Neal-Smith criteria, these criteria are not well suited for application to flight
data.

4. Smith-Geddes Criteria
a. Theory Behind the Smith-Geddes Criteria
The Smith-Geddes PIO criteria (sometimes referred to as the R. Smith criteria) were developed
from basic principles of closed-loop piloted control of pitch attitude and normal acceleration.14
They were initially aimed toward handling qualities, and they owe as their foundation the Neal-
Smith data base. The criteria have undergone some revisions54 as well as extension into the roll
axis.68
A fundamental hypothesis of the criteria is that the pilot desires the frequency response of
angular attitude to control inputs to be close to k/s around the frequencies of closed-loop control.
This hypothesis is a foundation for the Bandwidth and Neal-Smith criteria as well, but in the case
of the Smith-Geddes criteria the focus is on the shape of the amplitude ratio curve, not on phase
angle. Only after determining the “k/s-ness” of the amplitude ratio curve is there any attention
paid to phase angle.
As the basis for their phase slope formula, defined below, Smith and Geddes used a set of fixed-
base simulation data generated by McRuer et al.69 for idealized controlled elements: dynamics
of Yc = Kc, Kc/s, and Kc/s2. Several subjects performed sum-of-sines tracking with each of these
controlled elements and the crossover frequencies (frequencies at which the amplitude ratio of
the open-loop pilot-vehicle system is 0 dB) were measured. A weak variation in crossover
frequency was found with forcing function bandwidth, but the greater effect by far was the
controlled element itself. Smith and Geddes used the average crossover frequencies for the three
controlled elements to establish a formula relating amplitude ratio slope to crossover frequency,
as shown in Figure 34. The original formula recommended by Smith and Geddes,40 and the
current formula, are shown on Figure 34 as well.
The fixed-base simulation data were gathered using an RC-model airplane control stick. These
data are the basis for the Smith-Geddes criteria, with ultimate derivation of the requirements
coming from analysis of the Neal-Smith data base.
Smith divides PIOs into three types for the application of the criteria:
• Type I: Occurs when pilot switches from attitude control to acceleration control at
resonant frequency.
• Type II: Possible when any open-loop mode has a damping ratio less than 0.2.
• Type III: Initiated by piloted control of attitude, regardless of acceleration dynamics.
Regions for expected Type I and III PIOs are as defined by the parameters listed below. Type II
PIOs require only the determination of damping ratios.

71
 7

Yc = Kc McRuer et al.
6

Original formula:
ωc = 6.27 + 0.27S
Crossover
frequency 5
(rad/sec)
Current formula:
ωc = 6.0 + 0.24S Yc = Kc/s

Yc = Kc/s2

3
0 -2 -4 -6 -8 -10 -12 -14
Slope (dB/octave)

Figure 34. Smith-Geddes Criterion Frequency (Data from Fixed-Base Simulation69)

b. Parameters for the Smith-Geddes Criteria

The parameters of the Smith-Geddes criteria, as currently applied,54 are as follows:
1. Slope of the pitch attitude-to-stick-force transfer function, S, defined between 1 and 6
rad/sec, in units of dB/octave. No limit is given for PIO prediction.
2. Criterion frequency, ωc, defined as 6 + 0.24S. If a PIO is predicted this is the expected PIO
frequency.
3. Phase angle of the θ/Fes transfer function measured at ωc. If the phase angle ∠θ/Fes(jωc) is
more negative than -180 deg, PIO is predicted. If the phase is between -160 and -180 deg,
PIO is predicted if, in addition, the normal acceleration response at the pilot's station is such
that Φ(jωc) = ∠nzp/Fes(jωc) - 14.3ωc ≤ -180 deg. In other words, given no phase margin in
pitch attitude at the criterion frequency, the aircraft is definitely susceptible to PIOs; if there
is 15 deg or less phase margin, PIO will occur if there is no phase margin in adjusted normal
acceleration.
4. Rise time of the step response of pitch rate, tq, defined as the time to first peak or, if there is
no peak, time to 90% of steady state response. This is a handling-qualities parameter only,
with no specific limit for PIO susceptibility.

c. Evolution of the Smith-Geddes Criteria

The initial development of these criteria followed fundamental pilot-vehicle modeling rules.11
Application of the original criteria14 required closed-loop analysis using a simple pilot model,
with an outcome of an expected attitude phase margin at the predicted crossover frequency. The

72
selection of pilot model forms and crossover frequencies was not clearly called out; such analysis
is almost an art form, as the user must carefully determine the most likely form of pilot model
from existing rules and then decide on the requirements for a reasonable closed-loop response.
As such, the initial Smith-Geddes criteria were unwieldy and were certainly not in a form
amenable to incorporation into a military specification format. As the criteria evolved,40 ,68 a set
of rules was established that require only the aircraft-alone frequency response of pitch or roll
attitude to stick force.

d. Strengths of the Smith-Geddes Criteria

The Smith-Geddes criteria address both pitch attitude and normal acceleration as contributors to
PIO. Since longitudinal PIOs occur often as a result of control of normal acceleration, this
makes the criteria attractive. They are relatively easy to apply, since no transfer function models
are required (all parameters can be obtained from frequency responses). The criteria are very
effective at correlating the HQRs and PIORs from the Neal-Smith data base.
There is a strong user community for the Smith-Geddes criteria at the Air Force Flight Test
Center (AFFTC) at Edwards AFB. It has become standard for these criteria to be applied in the
flight test program for every new Air Force aircraft. A subset of the criteria has been adopted for
inclusion into MIL-STD-1797A.
A feature of the criteria that is considered to be especially strong is the ability to estimate the
expected PIO frequency if a PIO is predicted. This is considered by the AFFTC to be an
extremely valuable tool. Its effectiveness was evaluated in Section IV of this report and is
reviewed again, briefly, later in this section.

e. Shortcomings of the Smith-Geddes Criteria

The Smith-Geddes criteria do not explicitly account for the shape of the phase curve above the
pilot's normal closed-loop frequency. This is an important metric for PIO susceptibility, because
it is a measure of how fast the phase decreases as the pilot gain and crossover frequency increase
when approaching neutral stability.
As with most “post-Neal-Smith” criteria – criteria based on the Neal-Smith51 flight research data
– the Smith-Geddes criteria do not extend to other data very well. Smith has written68 that the
criteria “apply regardless of piloting task, flight condition, or aircraft type or size.” Because they
are based on the results of an experiment conducted in up-and-away conditions at indicated
airspeeds of 250 and 350 knots, the criteria effectively require a comparable dynamic response
from aircraft in other flight conditions, such as landing at an indicated airspeed of 120 knots.
There is some justification for questioning the foundation for the criterion frequency parameter
as well. While there is no refuting the data upon which the formula for ωc is based (Figure 34),
the dynamics of the configurations used for the formula are far removed from the short-term
response dynamics of a real airplane. For a conventional airplane, in the absence of extensive
augmentation or added filters, the short-period approximation for pitch attitude response to stick
inputs is
θ
=
(
M δe s + 1/Tθ2 )
δ e s [s + 2ζ sp ωsp s + ωsp
2 2
]

73
The values of the numerator zero 1/Tθ2 , the short-period damping ζsp, and the short-period
frequency ωsp together define the entire frequency response. (The gain M δe only shifts the
amplitude ratio of the response; in reality, higher-frequency dynamics, such as elevator actuators,
also contribute some effective time delay.) If Smith and Geddes had used data from tracking
experiments where the controlled elements looked like real airplanes, their formula for criterion
frequency would look very different.
As a first example, Figure 35 shows a small set of data from a fixed-base simulation flown for
the Minimum Flying Qualities study.67 The setup, task, and simulation facilities were almost
identical to those of McRuer et al.69 – including the use of an analog simulator to minimize time
delays – with one exception: a hydraulically actuated sidestick was used for the Minimum Flying
Qualities simulation, rather than an RC-model airplane controller. Crossover frequencies for the
idealized controlled elements Kc/s and Kc/s2 were about one rad/sec lower than in the McRuer et
al. experiment (compare open and solid circles in Figure 35), but the trends were the same.
When more realistic airplane dynamics are introduced as the controlled element, however, the
trends for crossover frequency with slope are very different. For all configurations ωsp = 5
rad/sec, with ζsp = 0.8 for the baseline and 0.2 for the low-damping cases. Addition of almost
200 msec of time delay reduces crossover frequency, but the trend with slope is similar.
The decrease in crossover frequency as amplitude ratio slope approaches zero is to be expected:
for a real airplane, the only ways to achieve such a configuration are to increase short-period
frequency well above the frequency of the zero 1/Tθ2 , and produce an overly abrupt airplane, or
reduce short-period damping, and produce a very oscillatory airplane. In either case, pilots are
not likely to track at very high frequencies, as they would with a controlled element of Kc.
7

Kc
6 McRuer et al.
MFQ (idealized)
Crossover MFQ (airplane)
frequency 5
(rad/sec)

4 Kc/s

3
Low damping Baseline
K c/s2

2 Baseline plus
time delay
Low damping
plus time delay
1

0
2 0 -2 -4 -6 -8 -10 -12 -14
Slope (dB/octave)

Figure 35. Identified Crossover Frequencies in Fixed-Base Compensatory Pitch Tracking

(McRuer et al.69 Data from Figure 34; “MFQ” is Minimum Flying Qualities Study67)

74
For real airplanes whose short-term dynamics are well-defined by the short-period
approximation, it is reasonable to suspect that crossover frequency would vary nonlinearly with
amplitude ratio slope: steep slopes suggest low short-period frequency, requiring more lead than
the pilot can generate to perform the tracking task to desired standards. By contrast, very
shallow slopes would be the high-frequency or low-damped configurations discussed above.
Then, of course, the addition of time delays and high-frequency lags – not captured at all when
one measures only the amplitude ratio curve at frequencies between 1 and 6 rad/sec – will place
further demands on the pilot, and even the best of dynamics will suffer from a reduction in
crossover frequency.
All of the foregoing is meant to investigate the basic hypothesis of the Smith-Geddes criteria.
No matter what might be said about the theory behind the criteria, however, we still want to
evaluate their effectiveness as PIO prediction criteria. Ultimately, what we want to know is, do
these criteria perform better than the others?

5. Gibson Phase Rate and Response Gain Criteria

a. Theory Behind the Gibson Criteria
Numerous PIO design guides and criteria have been developed by Gibson. One version,
centered around Nichols chart envelopes, is included in MIL-STD-1797A as a short-term pitch
requirement.4 The version considered here7 consists of a combination of limits on pitch attitude
phase rate and control/response sensitivity beyond the neutral-stability frequency (Figure 36).
As with the other criteria examined here, Gibson’s PIO criteria are based on the assumption that
pitch attitude control is primary, and that tendencies to PIO can be detected through measures of
attitude control. Deficiencies in flight path response are addressed through other requirements,
such as dropback. Unlike the other criteria analyzed in this report, Gibson’s include limits on the
magnitude of the transfer function of pitch attitude to control force.
There is one significant difference between Gibson’s criteria and all the others examined in this
report. The Neal-Smith and Smith-Geddes criteria are referenced to control force, and therefore
control feel system dynamics are included as part of the system when the feel system is in series.
For Bandwidth, the feel system should always be included. For Gibson’s criteria, the feel system
dynamics are always excluded. Application does require, however, that the gearing resulting
from the feel system be included, i.e., the frequency response plotted on the Nichols chart in
Figure 36 is referenced to control force, even though the force feel dynamics are excluded.
The “Level” limits in Figure 36 actually deal explicitly with PIO susceptibility and severity, not
handling qualities as is traditionally the case. This is not clear in Gibson’s writing, where he
states7 that “Although Level values are ascribed to the amplitude boundaries, they are strictly to
be interpreted as design aids only. This applies equally to the phase rate/frequency
boundaries…. No example of a significant high order PIO has been found which did not violate
one or more of the criteria.”
In their report on Aircraft-Pilot Coupling,1 the NRC Committee observed the following: “The
boundaries… were originally intended to indicate flying qualities in general rather than PIOs in
particular. A comparison of [the Gibson and Bandwidth requirements] shows that flight data [for
PIOs] fall along the L2-L3 boundary. This boundary is currently used as a dividing line for
assessing PIO potential.”

75
 Figure 36. Gibson PIO Criteria for Phase Rate and Maximum PIO Frequency Gain7

In personal communications, Gibson has indicated that the NRC Committee’s interpretation is in
fact not correct. “L1” in Figure 36 defines the region where no Category I PIO will occur. If a
design falls in “L2” some PIO tendency might be expected, but it is unlikely to be dangerous.
For a design in “L3” PIO susceptibility makes it dangerous.

b. Parameters for the Gibson Criteria

1. Phase rate is a measure of the slope of the attitude phase curve around ω180θ . Because it is
defined between ω180θ and 2 ω180θ , it is identical to the Phase Delay parameter used in the
Bandwidth criteria, as illustrated by the sketch in Figure 36.

76
2. Frequency at 180° lag is the neutral-stability frequency, ω180θ . It is not the same as the
ω180θ used by the Bandwidth criteria since the dynamics of the feel system are excluded here.
As with the Bandwidth criteria, Gibson observes that if a PIO occurs, the likely PIO
frequency is ω180θ . This was examined in Section III of this report.

c. Evolution of the Gibson Criteria

While there are several different parts to the Gibson criteria, the most significant portions as far
as handling qualities and PIO have been dropback, Nichols charts for optimum design, and
finally, the requirements of Figure 36. Gibson adopted phase rate as a measure of phase loss
with frequency, and with an adjustment for units, it is identical to Phase Delay from the
Bandwidth criteria.
The Nichols chart envelopes included in MIL-STD-1797A were not meant to be “criteria” as
such. Instead, they were design guides for good handling qualities. As such, there were no clear
flying qualities Level limits on the envelopes. The envelopes suffered from a problem common
to any such criteria: there is no way to know how significant a small violation of the envelopes
might be.
Requirements on phase rate and control/response sensitivity, such as those shown in Figure 36,
are much more tenable as standards for the evaluation of PIO susceptibilities.

d. Strengths of the Gibson Criteria

The Gibson PIO criteria have as their core a measure of the rate of phase loss with frequency. It
is clear that loss of phase is a major player in Category I PIOs, so the more directly a criterion
judges the phase loss, the stronger it should be at correlating data.
The Gibson criteria are significantly more powerful than the others because of the gain
requirements at the PIO frequency. No other criteria have a direct measurement of
control/response gain. They also include an explicit estimate of the PIO frequency, if one is
predicted to occur – the frequency for 180 deg phase lag, or neutral-stability frequency.

e. Shortcomings of the Gibson Criteria

The biggest strength of these criteria may also be their biggest weakness: the response gain
requirements were developed for one particular class of aircraft (fighters) and one cockpit
manipulator (center stick). They may not be correct for other aircraft sizes or other controllers.

6. Time-Domain Neal-Smith (TDNS) Criterion

a. Theory Behind the TDNS Criterion
The Time-Domain Neal-Smith (TDNS) criterion developed by Bailey and Bidlack18 owes its
origins to a combination of the Neal-Smith criteria described above, and the Step Target
Tracking (STT) criterion developed by Onstott and Faulkner.70
A simple pilot model (similar to the Neal-Smith model) is used to find the closed-loop response
to a step pitch command of 5 deg. The time to acquire the target attitude within a defined error
limit is used to drive aggressiveness. The theory is that PIO-prone configurations will have
trouble with closed-loop stability long before PIO-resistant configurations as the acquisition

77
time, D, is systematically reduced. The task is run for five seconds and rms tracking error is
computed starting at the acquisition time.

b. Parameters for the TDNS Criterion

Bailey and Bidlack evaluated several different parameters that are an outcome of their approach.
The primary parameter that they found to be most effective for evaluation of PIO susceptibility
was the 2nd derivative of the rms tracking error, d 2 (rmsθ e | t > D ) dD 2 , measured after the desired
attitude was acquired, within the predefined error bound. This is the acceleration function for
rms tracking error, and large values indicate a rapid increase in tracking error rate.

c. Evolution of the TDNS Criterion

The TDNS criterion was developed as a part of the Air Force’s Unified PIO Theory program,
and hence is relatively new.

d. Strengths of the TDNS Criterion

This is the only criterion in this report that is time-domain-based. Advocates of time-domain
analysis will find it attractive for that reason. When applied to purely linear systems, as is done
for this section, there is no advantage to time-domain over frequency-domain methods – they
should give the same answer. The difference is in the details of the approach, not in the selection
of the realm.
The methodology was developed to address PIO due to nonlinear causes, especially rate limiting.
At the time their work was performed (1994-95), there were no flight data sources for evaluation
of the criteria for Category II PIOs. The effectiveness of these criteria for such events is
discussed later in this report.
This criterion introduces a parameter that is missing from the frequency-domain counterpart: a
sensitivity to the rate of change of a performance parameter. The Neal-Smith criteria have seen
many different versions, each successful for its own subset of data but usually not very effective
when applied to a larger data base. Experienced users apply the frequency-domain criteria as a
way to see the relative change in the parameters as pilot bandwidth is increased. Acquisition
time in the time-domain version is equivalent to pilot bandwidth, and the 2nd-derivative factor is
a measure of rates of change, not absolutes – akin to the common application of the frequency-
domain criteria.

e. Shortcomings of the TDNS Criterion

Any methodology that employs closed-loop analysis is subject to criticism as assumptions are
made. In this case, a shortcoming in the closed-loop analysis is the complication involved in
obtaining the closed-loop response in the first place – the selection of pilot model,
implementation of the methods, etc. Such a criterion can be applied only through analysis, since
it is not possible to implement a closed-loop criterion in flight. In addition, the criterion is
established using step input in commanded attitude, so there are issues about the appropriateness
of a pure step, the size of the change, the possible need for a separate load-factor-based criterion,
etc.
In part because of a lack of familiarity with the TDNS approach, implementation of the
technique was considered to be too complex for this evaluation. Only those data published by

78
Bailey and Bidlack18 have been used, and Bailey and Bidlack applied the method to more than
half of the total data base used for this analysis (see below). Criterion values were taken from
the report with no attempt at independent confirmation.
The most significant shortcoming is in the calculation of the 2nd-derivative parameter. Bailey
and Bidlack18 developed an approximation that worked reasonably well for their analytical
approach, but accurate determination of a 2nd-derivative function requires that the initial closed-
loop analysis be performed for very small changes in acquisition time – large changes lead to
errors when the first derivative is taken, and they are further magnified when the second
derivative is computed.

G. THE DATA BASE APPLIED TO THE CRITERIA

Seven sets of flight research data, listed in Table 7, were selected for application to the criteria.
These seven references were selected because they provide a well-documented, complete set of
data using stringent and well-defined flight tasks. They also cover fighters in up-and-away
maneuvering and landing, and transports in landing. A total of 207 configurations are available
from these references, including several where incorrect control/response gearing may have had
an influence on PIO. Because it is not possible to confirm this, all relevant configurations have
been gleaned from all of the reports listed. The possible errors introduced by using a handful of
questionable configurations should not have an impact on the overall results, and they should
have similar impact for all of the criteria.
At least two different PIO tendency rating scales were used – singly and in combination – for the
different flight research programs. While the differences are, for the most part, quite small, there
is still evidence of distinct changes in pilot interpretation between projects and between scales
(more is written about this elsewhere in this report).
Table 7. Pitch Data Base Applied to the PIO Criteria
Number of configurations
Ref. Aircraft (Class) Task(s) Total No PIO PIO PIO
reported tendency reported
reported
Up-&-away maneuvering;
Neal-Smith51 NT-33A (fighter) 62 38 9 15
pitch tracking
Straight-in & offset
LAHOS43 NT-33A (fighter) 49 19 10 20
landings
Pitch rate landing TIFS (small
Offset landings 26 16 1 9
study65 transport)
Flared landing TIFS (small
Offset landings 29 20 0 9
study66 transport)
HAVE PIO36 NT-33A (fighter) Offset landings 18 9 0 9
HAVE
NT-33A (fighter) Offset landings 13 7 1 5
CONTROL71
HAVE CAP56 VISTA (fighter) Offset landings 10 7 0 3

TOTALS 207 116 21 70

79
As a result, it is not reasonable to differentiate the PIO tendencies of the various configurations
solely on the basis of PIOR and HQR. Rather, the pilot comments must be scrutinized and
specific mention of any PIO (or lack thereof) noted and compared with the ratings. While the
correlation with PIO rating should be very good, it is not necessarily perfect because of the
different rating scales and different interpretations of those scales. Separation of the
configurations based on pilot comments rather than ratings is considered to be both a more
accurate accounting of the occurrence of PIO, and more in keeping with the intent of the process
in the first place. This was done for every evaluation in all seven of the references listed in Table
7.
The configurations are broken out in Table 7 by three categories: 1) Those for which no PIOs
were reported, by any pilot, during the evaluations. 2) Those for which at least one pilot
reported a tendency to PIO, but none reported an actual PIO. 3) Those for which at least one
pilot reported encountering a PIO.
As Table 7 indicates, over half of the 207 configurations selected (116, or 56%) did not indicate
a reported PIO or tendency to PIO. Of the remaining 91, 70 (34% of all cases) had PIOs. For
this study, the 21 aircraft with a reported PIO tendency were grouped with those that had a PIO.
The assumption is that, in a limited flight program, with stringent tasks and experienced test
pilots, observation of a tendency to PIO should be weighed as a warning sign that, given enough
flying time, a real PIO would have occurred. Ideally, we would hope that an effective PIO
prediction method would be able to discern a region for PIO tendencies as well as real PIOs.

H. APPLICATION OF THE CRITERIA

1. Some Important Assumptions

The six PIO prediction criteria were applied, to the extent possible, to all 207 configurations in
the selected data base. For all but the Gibson criteria, the reference control input was assumed to
be force; for Gibson, position was used, but with adjustments to transfer-function gain to reflect
the force/displacement gearing of the controller. Because most of the Neal-Smith51
configurations were force-sensing, the dynamics of the controller were excluded for all but the
Bandwidth criteria, where it is recommended that force feel dynamics always be accounted for,
regardless of command sensing.
For the Neal-Smith criteria, the bandwidth frequency originally selected by Neal and Smith (3
rad/sec) has been found to be excessive for airplanes in landing,43 and no single bandwidth value
has been found to be effective without changing some of the groundrules for application of the
criteria.62 Retention of the original Neal-Smith boundaries (Figure 32), however, requires
retention of the original analysis techniques as well, so proposed modifications were not
attempted. Instead, all configurations were analyzed with two values of bandwidth frequency: 3
and 3.5 (for the Neal-Smith data); and 2 and 3 (for all other data).
In addition to the effect of bandwidth frequency on the Neal-Smith criteria, there were
configurations for which a closed-loop solution could not be found at all. In this case the
bandwidth frequency was varied slightly, both above and below the nominal value, until a
solution was found. Often the solution had exceedingly high resonance or pilot compensation
near the maximum value of 90 degrees, or both.

80
The gain portion of the Gibson criteria (Figure 36) was applied to all configurations, including
the TIFS configurations with column controllers. The limits were not developed for large
airplanes and yoke/wheel manipulators, so some loss in correlation would be expected to occur.
For the application of the Time-Domain Neal-Smith criterion, all information was obtained from
the reference source,18 with no independent analysis or confirmation. Fortunately, the three
experiments selected for analysis in Ref. 18 – Neal-Smith, LAHOS, and TIFS rate landing study
– comprise over half the total data base. Reference 18 has data for 51 cases from Neal-Smith, 45
from LAHOS, and 25 from TIFS.
The TDNS report includes analysis of the results for two of these data bases (Neal-Smith and
LAHOS), with explicit values of the PIO criterion d 2 (rmsθ e | t > D ) dD 2 listed in tables. There is
no analysis and no listing of the numbers for the TIFS data. Plots of the criterion as a function of
acquisition time D are printed in an appendix, but with a very crude logarithmic scale, making
accurate readings difficult. For most of the 25 TIFS cases the maximum value of
d 2 (rmsθ e | t > D ) dD 2 was either well below or well above 100, making determination of
predicted PIO susceptibility easy.
It was clear that the criterion of d 2 (rmsθ e | t > D ) dD 2 > 100 for PIO was not effective for the
TIFS configurations: all 25 cases are predicted to PIO, while only nine actually experienced
PIO. It was noted that, with a slight change in the threshold of d 2 (rmsθ e | t > D ) dD 2 to around
150 or so, the TIFS configurations correlated better. An evaluation of TDNS was made using
this threshold for the TIFS data only; the original limit of 100 was retained for the other data.

2. How the Criteria Fared

While correlation percentages are not as important as trends, they are still of value here because
the primary goal is to judge the effectiveness of the selected PIO susceptibility prediction
criteria. It may be possible (as was done for the TDNS criterion) to improve percentages with
slight adjustments to parameters, but this was not the intent of this comparison.
Table 8 summarizes the correlation rates for the criteria with the data base.
For both versions of the Neal-Smith criteria, separate listings are given for application of
bandwidths of 2 and 3 rad/sec (the latter also includes some cases in the Neal-Smith
experiment51 where a bandwidth frequency of 3.5 rad/sec was used). A bandwidth frequency of
2 rad/sec was too low for the data from the Neal-Smith experiment,51 so these data are excluded
for the low-bandwidth numbers (the result is that all low-bandwidth numbers are for the landing
task only).

It is unreasonable to expect a correlation rate of 100%. In many instances among the 207
configurations, only one or two pilots actually performed evaluations, so it is entirely possible
that some PIO-prone cases escaped detection. Conversely, several of the PIO cases may very
well have suffered from unrelated problems that led to the PIO, including inability of the test
airplane’s model-following system to produce the desired response, control sensitivity
inadvertently set incorrectly, etc.

81
 Table 8. Overall Correlation Rate for the PIO Prediction Criteria
Successful correlation for --
Overall success rate
Criteria No PIO PIO
n No. (%) n No. (%) N No. (%) Ranking
Bandwidth/Pitch Rate Overshoot 116 111 (96%) 91 78 (86%) 207 188 (91%) 1
Neal-Smith (Bandwidth=2) 78 69 (88%) 67 8 (12%) 145 77 (53%) 8
Neal-Smith (Bandwidth=3) 116 83 (72%) 91 75 (82%) 207 158 (76%) 4
MAI Neal-Smith (Bandwidth=2) 78 45 (58%) 67 22 (33%) 145 67 (46%) 9
MAI Neal-Smith (Bandwidth=3) 116 58 (50%) 91 79 (87%) 207 137 (66%) 6
Smith-Geddes 116 51 (44%) 91 82 (90%) 207 133 (64%) 7
Gibson 116 100 (86%) 91 66 (73%) 207 166 (80%) 2
Time-Domain Neal-Smith
58 41 (71%) 62 48 (77%) 120 89 (74%) 5
(threshold at 100 for all data)
Time-Domain Neal-Smith
58 51 (88%) 62 44 (71%) 120 95 (79%) 3
(threshold at 150 for TIFS data)

3. Ranking the Criteria

Overall, highest correlation was found with the Bandwidth/Pitch Rate Overshoot criteria. These
criteria correctly predicted PIO tendency for 78 of the 91 PIO cases (a success rate of 86%). The
Smith-Geddes criteria were most effective at predicting PIO (90%) – but only because they
predict most of the configurations to PIO: of the 74 cases incorrectly correlated, the Smith-
Geddes criteria predict PIO for 66 that had no reports of PIO.
Second were the criteria of Gibson, with an overall success rate of 80%, but only 66 of the 91
PIOs were identified (a 73% correlation). Third was the Time-Domain Neal-Smith criterion with
adjustments to the PIO threshold for the TIFS data, at 79% overall. This was the only criterion
for which such an adjustment was made, and it comes with a smaller data base, so it is difficult
to make a direct comparison of the success of the criterion. Given the poor correlation with the
TIFS data, addition of the data from the second TIFS experiment would probably only make the
correlation worse.
Perhaps surprisingly, the results suggest that the original Neal-Smith boundaries, with a
bandwidth frequency of 3 rad/sec, are reasonably effective at predicting PIO tendencies (they
were fourth overall, at a success rate of 76%). Some combination of the MAI-modified
boundaries and the original limits might be even more effective. This relatively high success rate
may reflect the high-bandwidth nature of PIOs – they are not events that occur near the normal
operating frequency for closed-loop piloted control. There may, in fact, be two relevant
bandwidth frequencies for the Neal-Smith criteria: the frequency for normal closed-loop control
(for assessment of handling qualities), and the frequency for urgent control (for PIOs).
Without adjusting the PIO threshold, the Time-Domain Neal-Smith criterion did not fare very
well overall (a 74% success rate for the 120 configurations, fifth overall). Changing the
threshold for the TIFS cases increased the overall effectiveness to 79%, but reduced the
effectiveness at predicting PIO, from 77% (48 of 62 cases) to 71% (44 cases). If the TIFS data
were excluded entirely, the TDNS criterion would be among the best performers in the table –
but correlation rates for most of the criteria would jump considerably if the TIFS data were

82
removed, reflecting the strong reliance on fighter data – and especially the Neal-Smith data base
– for development of the PIO criteria.

4. Putting Correlation Rates in Perspective

The Air Force has established a goal of preventing the occurrence of 80% of all PIOs by design.
With an 80% threshold on PIOs alone, four of the nine criteria and variants listed in Table 8
would serve. Unfortunately, three (Neal-Smith with Bandwidth = 3, MAI-modified Neal-Smith
boundaries with Bandwidth = 3, and Smith-Geddes) do so by throwing the proverbial baby out
with the bathwater: all obtain their high success rate by overpredicting PIO. In the case of
Smith-Geddes, the criteria predict PIO for over half the configurations for which no PIO or
tendencies to PIO were reported. The Bandwidth/Phase Delay criteria are effective at predicting
both PIOs and non-PIOs – the obvious objective of the Air Force’s goal.
To further assess the numbers in Table 8, consider two rather facetious assertions. Suppose we
said the following:
• “Every airplane PIOs” – since there were 91 PIO configurations, we would be correct
91/207 = 44% of the time. For the configurations applied to the Neal-Smith criteria with
a pilot bandwidth of 2 rad/sec, the rate would be 67/145 = 46% -- almost exactly the
success rate of both versions of the Neal-Smith criteria with the low bandwidth! In other
words, because they underpredicted the occurrence of PIO, the criteria were not much
better than pure luck.
• “No airplane PIOs” – since there were 116 such cases, the success rate would be 116/207
= 56%, not much worse than the Smith-Geddes criteria, which overpredicted the
occurrence of PIO.
It is clear, then, that the numbers in Table 8 must be looked at with some sense of perspective.
If, for example, our data base consisted only of airplanes that exhibited PIO, a criterion such as
“every airplane PIOs” would be phenomenally successful and totally useless. The best criterion
(or criteria) must be able to account for both good and bad airplanes or it is not to be trusted.

5. Examination of Graphical Trends

It must be stressed again that more important than correlation percentages is correlation trends.
It is possible, for example, that correlation for a particular set of criteria could be improved with
a minor adjustment in boundaries. Rather than rely entirely on the tabulated results in Table 8,
we must look at graphical results as well.
A summary of the graphical results is given for all criteria except TDNS in Figure 37.
No graphical presentation is suitable for a single criterion such as is used by TDNS.
Consideration was given to plotting the criterion against median PIO tendency rating, but
specific values of the criterion were not reported for the TIFS data. In addition, because of the
different interpretations of the PIO tendency rating scales among the different references,
extensive discussion would be required in order to make any interpretation of such a plot. It was
finally decided that no graphical representation of the TDNS criterion would be reported here.

83
 0.4

0.35

0.3
No PIO
Tendency
PIO
0.25
Pitch
attitude PIO
Phase 0.2
Delay PIO possible
(sec) 0.15 (if flt path BW low)

No PIO
0.1

0.05

0
0 1 2 3 4 5 6 7 8

Pitch attitude Bandwidth (rad/sec)

a) Pitch Attitude Bandwidth

250

No PIO
Tendency
200
PIO

L3 (PIO)
150
Average
phase
L2 (PIO)
rate
(deg/Hz)
100

L1 (No PIO)

50

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

Frequency at 180 deg lag (Hz)

b) Gibson Phase Rate

Figure 37. Summary Data Plots

84
 55
Closed-loop
50 resonance (dB)

45

No PIO 40
Tendency
35
PIO
30

25

Original
20 Neal-
Smith
15 Level 3
Limit
10
MAI
(PIOR=3.5)
MAI 5
(PIOR=2)
0
-60 -40 -20 0 20 40 60 80 100
-5

Pilot compensation (deg)

c) Neal-Smith (3-rad/sec Bandwidth)

55

50
Closed-loop
45 resonance (dB)

40

35
No PIO
30
Tendency
25 PIO

20
Original Neal-Smith
15 Level 3 Limit

10
MAI (PIOR=3.5)

MAI (PIOR=2) 5

0
-60 -40 -20 0 20 40 60 80 100
-5

Pilot compensation (deg)

d) Neal-Smith (2-rad/sec Bandwidth)

Figure 37. Summary Data Plots (continued)

85
 6

PIO
5

No PIO
4

No PIO
Tendency 3
PIO

Type I 2 Criterion
PIO Frequency
Poss-
ible
(rad/sec)

1
-40 -60 -80 -100 -120 -140 -160 -180 -200 -220 -240 -260

Pitch Attitude Phase Angle at Criterion Frequency (deg)

e) Smith-Geddes
Figure 37. Summary Data Plots (concluded)
Caution must be used in drawing any conclusions from Figure 37: not all requirements for all
criteria are shown on these plots. For example, for the Bandwidth criteria there appears to be a
mix of non-PIO and PIO cases in the region of Phase Delays between 0.14 and 0.2 sec. Many of
these cases, however, are properly accounted for when the auxiliary requirement on flight path
Bandwidth is applied. Similarly, several of the PIO cases in the no-PIO region of Gibson’s
phase-rate criterion actually fail the maximum gain criterion (Figure 36), and hence should be
prone to PIO, as indicated.
Still, such exceptions are very few – and there are no such additional requirements for the Neal-
Smith criteria. Only a handful of points on the Smith-Geddes criteria are accounted for by limits
on normal acceleration phase angle (Type I PIOs) or on mode damping ratio (Type II).
With a bandwidth frequency of 2 rad/sec, the Neal-Smith criteria do not differentiate between
non-PIO and PIO cases at all. With a bandwidth of 3 rad/sec, the PIO cases are almost entirely
located above the original Level 3 limit. But so are many non-PIO cases, most of which received
PIO tendency ratings of 1 or 2 and were considered to have Level 1 handling qualities. In
addition, many of the configurations – both PIO-prone and not – resulted in almost ridiculously
high values of closed-loop resonance just to get a solution.
The Smith-Geddes plot in Figure 37 reflects the conservatism of these criteria: a very large
number of configurations is located in the PIO region. Almost all of these are for the landing
data, and most for the transport experiments. It was noted by Calspan 62 that there was a lack of
correlation for the criteria when the Handling Qualities Ratings from the LAHOS landing

86
experiment43 were compared. This was later noted by Smith as well,68 who suggested that the
primary cause of this lack of correlation may be the fact that the landing task is relatively open-
loop, and therefore it simply does not expose the tendency to PIO in landing as well as in pitch
tracking up-and-away. *
It is also possible that pilots simply do not need the same level of response for landing as for air
combat. This is the basis for the division of requirements in MIL-STD-1797A by mission phases
and categories. It is the basis for allowing a lower short-period frequency or lower pitch attitude
Bandwidth for landing. Pilots simply do not need the same level of Bandwidth to land.

I. REVIEW OF THE CRITERIA IN LIGHT OF THE NATIONAL RESEARCH

COUNCIL’S THREE PREREQUISITES
Earlier in this section, the three prerequisites for PIO prediction criteria, as defined by the
National Research Council’s Committee on the Effects of Aircraft-Pilot Coupling on Flight
Safety,1 were outlined. They were validity, selectivity, and ready applicability. These
prerequisites may now be applied to each of the criteria.
In terms of validity, all of the candidate criteria certainly exhibit characteristics that are pilot-
centered. Given that PIO is a high-gain phenomenon, only indirectly related to the overall
handling qualities of the airplane, those criteria that are rooted in handling-qualities metrics may
suffer somewhat in validity. For example, the concept behind pitch attitude Bandwidth is that
this is the frequency for closed-loop piloted control, for a pure-gain pilot, providing no
compensation. As shown in Section IV, however, during a PIO the pilot operates at a different
frequency, and that frequency may be dependent upon the task or the severity of the PIO. This
may explain why the “Bandwidth/Phase Delay” requirements of Figure 3 are all straight lines,
primarily limits on Phase Delay alone. Even the Gibson phase rate limits could be applied
reasonably well without the complication of the ω180θ frequency. Similar criticism may be laid
on the Neal-Smith criteria when the bandwidth frequency is selected to be typical of closed-loop
control, not PIO.
The Smith-Geddes criteria suffer in validity to an extent as well. A basic premise of the Smith-
Geddes criteria is that the magnitude of the airplane’s pitch attitude response, in the frequency
range of typical piloted control (1 to 6 rad/sec), may be considered to be a straight line. The
more “k/s-like” this line, the better. While this premise is well-founded in experimental data and
theory,11 there is an underlying assumption that all of the dynamics typical of a short-period-
controlled airplane (i.e., the pitch attitude zero 1/Tθ2 and the short-period mode described by
ζ sp and ωsp ) can all be approximated by a single slope.

A major test of validity must be the sensitivity of the criteria to changes in pilot gain or operating
frequency. This is why the two most successful criteria are Bandwidth and Gibson. Both are
fundamentally based on a measure of the rate of phase loss in the pitch attitude response (Phase

*
Given the current state of maturity of evaluation tasks for handling-qualities and PIO
assessment, the relatively undefined maneuvers of the Neal-Smith experiment seem downright
crude. There were no clear delineations of desired and adequate performance, for example.
Later replications of Neal-Smith configurations, with more tightly defined tasks, have served
only to support the overall findings of the original experiment.

87
Delay and phase rate, respectively). Conversely, the Neal-Smith criteria account for phase
effects indirectly, to the extent that they influence required pilot compensation, and the Smith-
Geddes criteria use only a “snapshot” of the phase rolloff, with a check of the phase angle at one
frequency. Neither of these criteria explicitly address rate of phase loss with pilot gain, a key
Category I effect.
Bandwidth and Gibson’s criteria are centered around frequency responses, which are by
definition sinusoidal in nature, as are PIOs (the Bode plot is referred in some control systems
texts as the “sinusoidal frequency response”). Criteria that are centered around a fundamentally
different response form, such as Time-Domain Neal-Smith, suffer in terms of validity. On the
other hand, to the extent that PIOs typically are initiated by some sudden “trigger,” the step
command change applied in TDNS might be a reasonable surrogate.
Finally, some credit for validity must be given to the Gibson criteria for making a first effort to
account for the effect of response gain on PIO susceptibility. The net effect on the total
correlation rate in Table 8 is close to zero for all data, however, as almost as many PIO cases in
the non-PIO region were caught as were non-PIO cases in the PIO region.
Selectivity can be judged from the results in Table 8 and Figure 37. The Bandwidth criteria are
most effective, followed by Gibson, with Neal-Smith (with a bandwidth frequency of 3 rad/sec) a
distant third.
For ready applicability, lowest on the list must be Time-Domain Neal-Smith, which requires a
computer program to apply. Slight variations in the parameters of the original Neal-Smith
criteria can make differences in the answer obtained. The attitude portions of the Bandwidth,
Gibson, and Smith-Geddes criteria are all easily applied. Then some form of complication sets
in: for Bandwidth, a Bode plot of flight path angle is required; for Gibson, a gain-phase plot of
the airplane’s attitude to control position response – but with the gain set to reflect stick force –
must be used, and for Smith-Geddes, a Bode plot of normal acceleration and a measure of the
damping ratio of all longitudinal modes is needed.

J. APPLICATION OF THE CRITERIA TO FOUR EXAMPLE AIRPLANES

Up to this point, the focus of this section has been on application of the PIO prediction criteria to
a uniform data base. Given the nature of the experimental data used for the analysis, there is
some justification for criticizing the overall results. After all, what should matter is how the
criteria fare when applied to real airplanes, not variable-stability simulations of real airplanes.
We have applied as many of the criteria as possible to four airplanes as a method for comparison
of the practical effectiveness of the criteria. In the following, computer software was not
available to apply the TDNS criterion, but all other criteria were used.
While the focus in this study has been on the prediction and prevention of PIOs, most of the
criteria examined in this section also contain requirements on handling qualities. These
requirements are not discussed in detail in this report, but for the applicable criteria the predicted
flying qualities Levels were determined and are listed with minimal comment in the following.
The estimated Levels of flying qualities were determined for all but the Gibson criteria (which
are not specifically geared toward flying qualities).
The Bandwidth, Smith-Geddes, and Gibson criteria provide predictions of PIO frequency as
well, allowing us the opportunity to check the effectiveness of this portion of the respective
criteria.

88
1. Example 1 – The T-38A PIO
In 1960, a T-38A suffered a violent pitch PIO in low-altitude transonic flight, with oscillations in
load factor as high as –9g to +8g, at a PIO frequency of approximately 7.8 rad/sec. Because of
the severity of the event, an investigation was conducted to identify the causes and cures. As a
result of the documentation that has resulted, this incident has been thoroughly analyzed by
several researchers;14,12 a major contributor was found to be a nonlinear bobweight in the pilot’s
control loop. Transfer functions are available for the T-38A with the bobweight loop closed (the
PIO condition) and open (for assessment of the impact of the bobweight). Results of application
of the criteria to both conditions are listed in Table 9.
Table 9. Application of the PIO Prediction Criteria to the T-38A
Predict PIO? (Predicted
Criteria Parameters
FQ Level)
a) Bobweight Loop Open

ωBWθ = 1.9 rad/sec; τ pθ = 0.118 sec;

No, but pitch bobble
Bandwidth/Phase Delay ∆G(q) = 10 dB; ω BWγ N/A; ω180 = 9.0 (Level 2)
rad/sec
Orig. limits BW = 2 rad/sec: N/A (no solution N/A
MAI limits possible below BW = 3.5) N/A
Neal-Smith No, but “tendencies to
Orig. limits BW = 3.5 rad/sec: |θ/θc|max = 6.7 dB; oscillate” (Level 2)
∠pc = 15 deg
MAI limits No (Level 2)
ωc = 5.4 rad/sec; ∠θ/Fes(jωc) = -109
Yes – Type II (All criteria
Smith-Geddes deg; Φ(jωc) = -152 deg; lowest ζ =
Level 1)
0.18 ( feel system at ω = 18 rad/sec)
ph. rate = 22 deg/Hz; ω180 = 1.7 Hz; Yes – L2 based on gain at
Gibson Phase Rate/Gain
Gain at ω180 = -15 dB PIO frequency (N/A)
b) Bobweight Loop Closed (PIO Condition) – PIO Frequency = 7.8 rad/sec

ωBWθ = 0.4 rad/sec; τ pθ = 0.143 sec; Yes – In “moderate”

Bandwidth/Phase Delay ∆G(q) = 23 dB; ω BWγ N/A; ω180 = 10.5 region but fails both τ p
θ
rad/sec and ∆G(q) (Level 3)

Orig. limits BW = 2.5 rad/sec (lowest possible): No (Level 1-2)

MAI limits |θ/θc|max = 1.6 dB; ∠pc = -16 deg No (Level 1)
Neal-Smith
Orig. limits BW = 3 rad/sec: |θ/θc|max = 5.2 dB; No (Level 2)
MAI limits ∠pc = -10 deg No (Level 2)
ωc = 5.5 rad/sec; ∠θ/Fes(jωc) = -65
Yes – Type II (All criteria
Smith-Geddes deg; Φ(jωc) = -106 deg; lowest ζ =
Level 1)
0.14 (ω = 10 rad/sec)
Yes – Worse than L3
Gibson Phase Rate/Gain (Feel system ph. rate = 103 deg/Hz; ω180 = 1.7 Hz;
based on gain at PIO
included – see text) Gain at ω180 = -4 dB frequency (N/A)

89
With the bobweight loop open (Table 9a), the Bandwidth criteria predict pitch bobble, but not
PIO. The Neal-Smith criteria also predict no PIO, and the T-38A lies in a region identified as
exhibiting “tendencies to oscillate,” perhaps referring to pitch bobble (no Neal-Smith solution
was possible for pilot bandwidths below 3.5 rad/sec). The Smith-Geddes predict Type II PIO
because of the damping of the feel system, and the Gibson criteria predict a “L2” PIO on the
basis of gain at the PIO frequency.
With the bobweight loop closed (Table 9b), the T-38A lies in a region of “moderate” PIO on the
Bandwidth/Phase Delay limits of Figure 3, but it also has very high pitch rate overshoot, so the
combination should certainly be expected to lead to more than a moderate oscillation. The Neal-
Smith criteria do not predict PIO in any case, possibly because the pilot bandwidths of 2 and 3
rad/sec are well below the response modes of the airplane. Smith-Geddes predicts Type II PIO
again, but otherwise suggests the airplane would be Level 1. And Gibson’s gain criterion at the
PIO frequency predicts a severe PIO. For the Gibson criteria with the bobweight loop closed, the
dynamics of the feel system were retained, even though the criteria are to be applied with them
removed. Because of the interaction of the bobweight and feel system, it was felt that the
airplane’s response was better represented by retaining the effects of the feel system. In
addition, since the bobweight is an effective feedback of load factor around the feel system, it
was not obvious how the dynamics of the feel system could justifiably be ignored.
The T-38A PIO occurred at relatively high frequencies. Measurement of approximate PIO
frequencies from the available time history indicates a PIO frequency of almost 8 rad/sec.
Bandwidth and Gibson criteria both predict a frequency of around 10.5 rad/sec, so they are not
very close. Smith-Geddes does not predict PIO in attitude or load factor, but does predict a Type
II as a result of a low-damped mode at 10 rad/sec, or about the same frequency as the other
criteria. In this case, it must be concluded that none of the criteria does a very good job at
predicting PIO frequency.

2. Example 2 – The Space Shuttle in Display Tracking and Landing

Twisdale and Kirsten72 analyzed three PIOs that occurred during flight testing of a “high-
performance aircraft” at Edwards AFB. While it was not identified as such at the time, this
airplane was the Space Shuttle. The first PIO analyzed occurred on landing and the second and
third during maneuvering up-and-away. The third PIO involved pilot tracking of a flight path
error indication on a cockpit display, so the model includes the dynamics of the display.
Transfer functions are provided in Ref. 72 so all criteria analyzed here, with the exception of
TDNS, were applied to the three cases, as summarized in Table 10. The dynamics of the
Shuttle’s Rotational Hand Controller (RHC) are not known, but should be minimal and were
ignored.
Flight path Bandwidth is an important parameter for application of the Bandwidth/Phase Delay
criteria. The transfer functions in Ref. 72 include pitch rate and normal acceleration at the center
of rotation and at the pilot’s station, but not at the center of gravity. A reasonable estimation of
flight path response can be made from normal acceleration, but the criteria require the response
referenced to the center of gravity. For the Shuttle, the flight path Bandwidth at the c.g. should
be similar to, and probably less than, that at the pilot’s station. In all three cases, the estimated
flight path Bandwidth based on normal acceleration at the pilot’s station is well below the
criterion limit of 0.7 rad/sec. Hence it is expected that the flight path Bandwidth at the c.g.
would be as well.

90
 Table 10. Application of the PIO Prediction Criteria to Three Shuttle PIOs72
Predict PIO? (Predicted
Criteria Parameters
FQ Level)
a) Landing Approach – Reported PIO Frequency = 3.6 rad/sec
ωBWθ = 1.59 rad/sec; τ pθ = 0.141 Yes – In “moderate”
Bandwidth/Phase Delay sec; ∆G(q) = 3.0 dB; ω BWγ < 0.7 region but fails both τ p
θ
rad/sec; ω180 = 3.4 rad/sec and ω BWγ (Level 2)
Orig. limits BW = 2 rad/sec: |θ/θc|max = -2.0 dB; No (Level 2)
MAI limits ∠pc = 55 deg No (Level 2)
Neal-Smith No, but “tendencies to
Orig. limits BW = 3 rad/sec: |θ/θc|max = 7.2 dB; oscillate” (Level 2)
∠pc = 77 deg
MAI limits No (Level 2)
ωc = 3.8 rad/sec; ∠θ/Fes(jωc) = -187 Yes – Type I & III (Level
Smith-Geddes
deg; Φ(jωc) = -222 deg 3)
ph. rate = 101 deg/Hz; ω180 = 0.54 Hz;
Gibson Phase Rate/Gain Yes – L2 (N/A)
Gain at ω180 = N/A
b) Up-and-Away – Reported PIO Frequency = 3.1 rad/sec
ωBWθ = 1.9 rad/sec; τ pθ = 0.155 sec; Yes – In “moderate”
Bandwidth/Phase Delay ∆G(q) = 2.6 dB; ω BWγ < 0.7 rad/sec; region but fails both τ pθ
ω180 = 3.3 rad/sec and ∆G(q) (Level 3)
Orig. limits BW = 2 rad/sec: |θ/θc|max = -1.2 dB; No (Level 1)
MAI limits ∠pc = 45 deg No (Level 2)
Neal-Smith
Orig. limits BW = 3 rad/sec: |θ/θc|max = 8.8 dB; Yes (Level 3)
MAI limits ∠pc = 72 deg Yes (Level 3)
ωc = 4.0 rad/sec; ∠θ/Fes(jωc) = -194 Yes – Type I & III (Level
Smith-Geddes
deg; Φ(jωc) = -237 deg 3)
ph. rate = 112 deg/Hz; ω180 = 0.53 Hz;
Gibson Phase Rate/Gain Yes – L2 (N/A)
Gain at ω180 N/A
c) Up-and-Away (Display Tracking; Display Dynamics Included) – Reported PIO Frequency = 2 rad/sec
ωBWθ = 1.1 rad/sec; τ pθ = 0.338 sec;
Bandwidth/Phase Delay ∆G(q) = 0.8 dB; ω BWγ < 0.7 rad/sec; Yes – Severe (Level 3)
ω180 = 1.9 rad/sec
Orig. limits BW = 2 rad/sec: |θ/θc|max = 6.3 dB; No (Level 2)
MAI limits ∠pc = 73 deg Yes (Level 3)
Neal-Smith
Orig. limits BW = 2.7 rad/sec (highest possible): Yes (Level 3)
MAI limits |θ/θc|max = 42 dB; ∠pc = 84 deg Yes (Level 3)
ωc = 3.1 rad/sec; ∠θ/Fes(jωc) = -232 Yes – Type I & III (Level
Smith-Geddes
deg; Φ(jωc) = -237 deg 3)
ph. rate = 243 deg/Hz; ω180 = 0.30 Hz; Yes – worse than L3
Gibson Phase Rate/Gain
Gain at ω180 N/A (N/A)

91
The Bandwidth/Phase Delay, Smith-Geddes, and Gibson criteria all predict PIO for the three
conditions. The Neal-Smith criteria predict no PIO for landing approach, and, with the lower
value of pilot bandwidth, for the up-and-away condition.
As for PIO frequency, the landing PIO was reported to be at 3.6 rad/sec. All three criteria came
close to predicting this frequency: Bandwidth and Gibson (both at 3.4 rad/sec) and Smith-
Geddes (at 3.8 rad/sec). The frequency for the up-and-away event was reported to be 3.1 rad/sec,
in very good agreement with Bandwidth and Gibson (3.3 rad/sec), but not very close to Smith-
Geddes (4.0 rad/sec). The display tracking PIO was reportedly at 2 rad/sec, in almost perfect
agreement with Bandwidth and Gibson (1.9 rad/sec), but not with Smith-Geddes (3.1 rad/sec).

3. Example 3 – PIOs on the B-2 During Flight Test

The B-2 experienced two reported PIOs during developmental flight test.73 The author of Ref.
73 provided reproductions of figures from the reference with larger scales. In addition, Northrop
Grumman provided time-history data to Hoh Aeronautics, Inc. through a subcontract for this
project, and performed an independent analysis of the B-2 (see Appendix E). There are some
significant differences in the frequency responses, and it has not been possible to resolve the
differences. As explained in Appendix E we have chosen to use information from the original
white paper, as is done here.
One B-2 PIO occurred during landing approach in an off-nominal configuration. The pilot
assigned a PIOR of 4 for the event. The other PIO happened during aerial refueling with an
interim flight control system modification, and again was rated a PIOR of 4.
Because the only information available in Ref. 73 is in the form of Bode plots and time histories,
the Neal-Smith and Time-Domain Neal-Smith criteria could not be applied to the B-2. In all
other cases, relevant parameters had to be taken from Bode plots, so the values are not precise.
In addition, the dynamics of the feel system were not included in the Bode plots and are not
included here. This also precludes application of the gain portion of the Gibson criteria, since
feel system gearing is to be included (units for the criterion are deg/lb, Figure 36). Results are
listed in Table 11.
All three criteria predict PIO for the off-nominal landing approach, but none predicts the aerial
refueling case to be susceptible to PIO. The Smith-Geddes criteria are very close – close enough
to raise concerns about the possibility for PIO.
The Smith-Geddes criteria require the fitting of a frequency response magnitude plot with a
straight line (to determine slope S), or an average of straight lines, depending upon how the
criteria are applied. A straight line fit of the Bode magnitude plot is inappropriate for most
aircraft. For example, Givens73 found in applying the criteria to the B-2 that a very poor fit, and
therefore poor agreement with flight experience of PIOs, was obtained; this is not surprising
since the B-2 has a very classical response including a shelf in the frequency range between 1/Tθ2
and ωsp (Figure 38). To resolve this, Northrop Grumman revised the region over which the
straight line fit was accomplished, extending it beyond the short period. This is not consistent
with the original derivation of the Smith-Geddes criteria, nor is it reasonable to arbitrarily revise
the fitting region to make the answer agree with a PIO that has already occurred.

92
 Table 11. Application of the PIO Prediction Criteria to Two B-2 PIOs
Predict PIO? (Predicted
Criteria Parameters
FQ Level)
a) Off-Nominal Landing Approach – PIO Frequency = 2.7 rad/sec

ωBWθ = 1.0 rad/sec; τ pθ = 0.20 sec;

Bandwidth/Phase Delay ∆G(q) ≈ 7 dB; ω BWγ N/A; ω180 = 2.6 Yes – Severe (Level 3)
rad/sec
ωc = 2.6 rad/sec; ∠θ/δes(jωc) = -180 Yes – Type I & III (Level
Smith-Geddes
deg; Φ(jωc) = -217 deg 3)
ph. rate = 144 deg/Hz; ω180 = 0.41 Hz;
Gibson Phase Rate/Gain Yes – L3 (N/A)
Gain at ω180 N/A
b) Aerial Refueling – PIO Frequency = 3.8 rad/sec

ωBWθ = 4.0 rad/sec; τ pθ = 0.07 sec;

Bandwidth/Phase Delay ∆G(q) ≈ 7 dB; ω BWγ N/A; ω180 = 6.0 No (Level 1)
rad/sec
ωc = 5.0 rad/sec; ∠θ/Fes(jωc) = -158 Marginal – Type I (Level
Smith-Geddes
deg; Φ(jωc) = -250 deg 2)
ph. rate = 52 deg/Hz; ω180 = 1.0 Hz;
Gibson Phase Rate/Gain No (N/A)
Gain at ω180 N/A

Figure 38. Problem With Applying Average Slope Method of Smith-Geddes to the B-2
(from Ref. 73)

93
The average PIO frequency for the B-2 landing event was 2.7 rad/sec, in almost perfect
agreement with all three criteria (2.6 rad/sec predicted). The frequency for the aerial refueling
event was 3.8 rad/sec, and none of the criteria was very good, with Smith-Geddes closest at 5.0
rad/sec, and Bandwidth and Gibson predicting 6.0 rad/sec.

4. Example 4 – A PIO-Resistant Airplane: The F-15 in Landing

It may not be possible to define precisely what a “PIO-proof” airplane is, but based on
discussions with Air Force pilots, it has been suggested that most models of the F-15, with
augmentation functioning properly, are a good representative. Reportedly there is no tendency
for PIOs or bobble in landing, so PIO should not be predicted for this airplane. Models of A and
E versions of the F-15 in landing configuration were analyzed in a report by McDonnell Douglas
and Hoh Aeronautics, Inc.16 Transfer functions are available only for the F-15A, so it is
analyzed here.
Parameters for the PIO prediction criteria and their corresponding values for the F-15A are listed
in Table 12. Only the MAI-modified Neal-Smith limits (with a pilot bandwidth of 3 rad/sec) and
the Smith-Geddes criteria predict PIO for this airplane. Significantly, the F-15A fails all of the
tests of the Smith-Geddes criteria, indicating susceptibility to Type I, II, and III PIO. The
conservatism of the Smith-Geddes criteria for airplanes operating at low airspeeds can be clearly
seen.
None of the flying qualities criteria estimate the F-15A to be a Level 1 airplane in landing. The
MAI-modified Neal-Smith limits and the Smith-Geddes criteria both indicate Level 3 (or worse)
flying qualities. Comparison of the Neal-Smith parameters for high pilot bandwidth with the
limits in Figure 32 suggests that the F-15A would be expected to be close to Level 3.
As a point of reference, in a flight experiment conducted as a student project at the Air Force
Test Pilot School, for the precision offset landing task the F-15 received an average HQR of 3.7
– just barely Level 2. (The results of this experiment have never been formally published.)
Table 12. Application of PIO Prediction Criteria to the F-15A in Landing
Predict PIO? (Predicted
Criteria Parameters
FQ Level)

Bandwidth/Phase Delay
ωBWθ = 1.45 rad/sec; τ pθ = 0.076 No (Level 2)
sec; ∆G(q) = 5.9 dB; ω180 = 3.1 rad/sec
Orig. limits BW = 2 rad/sec: |θ/θc|max = -0.6 dB; No (Level 2)
MAI limits ∠pc = 58 deg No (Level 2)
Neal-Smith
Orig. limits BW = 3 rad/sec: |θ/θc|max = 6.1 dB; No (Level 2)
MAI limits ∠pc = 80 deg Yes (Level 3)
ωc = 3.2 rad/sec; ∠θ/Fes(jωc) = -182
Yes – Type I, II, and III
Smith-Geddes deg; Φ(jωc) = -228 deg; lowest ζ =
(HQR 10)
0.12 (ω = 26 rad/sec)
ph. rate = 40 deg/Hz; ω180 = 20 rad/sec;
Gibson Phase Rate/Gain No (N/A)
Gain at ω180 = -44 dB

94
5. Recapitulation of the Criteria and their Predictions for the Four Airplanes
Table 13 summarizes the results of application of the criteria to the four example airplanes. With
multiple conditions, there are a total of eight separate combinations of airplane/flight condition in
Table 13. Of these eight, two are considered to be non-PIO cases: the T-38A with bobweight
loop open and the F-15A in landing. The other six cases are PIOs. The Bandwidth, Smith-
Geddes, and Gibson criteria include estimates of the PIO frequency, if PIO is predicted, and
these estimates are also given in Table 13. Successful predictions of PIO are indicated by bold
print. Where PIO is predicted and actually occurred, the estimate of PIO frequency is considered
correct if it is within ½ rad/sec of the measured PIO frequency, and such estimates are also
indicated by bold type.
The Bandwidth/Phase Delay criteria are most effective, making correct predictions about the
expectation of PIO on seven of the eight cases. The B-2 refueling condition is the only one not
predicted by the Bandwidth/Phase Delay criteria.
Second in effectiveness are the Gibson criteria, missing on two of the eight cases. As with
Bandwidth, one of the cases the Gibson criteria do not pick up is the B-2 refueling case.
The Smith-Geddes criteria are third, correctly predicting PIO for five cases. These criteria,
however, predict PIO for seven of the eight cases, missing on the B-2 refueling condition, and
predicting PIO for the F-15A in landing. For the two T-38A cases, however, the prediction of
PIO is based on damping of low-damped modes, not because of the attitude or normal
acceleration parameters; these parameters all predict no PIO.
None of the Neal-Smith criteria work very well. The success rate runs from four of six for the
high-bandwidth cases with the original Neal-Smith limits to only one correct for the MAI limits
at lower bandwidth.

Table 13. Summary of PIO Predictions for the Category I Criteria

Neal-Smith Predictions –
Bandwidth BW = 2 BW = 3 Smith-Geddes Gibson
Airplane/PIO Information
Predictions – Predictions – Predictions –
N-S MAI N-S MAI
limits limits limits limits
PIO
Airplane in ωPIO PIO? ωPIO PIO? PIO? PIO? PIO? PIO? ωPIO PIO? ωPIO
flight?
T-38A (B.W. open) No -- No -- N/A N/A No No Yes ?* Yes 10.5
T-38A (B.W. closed) Yes 7.8 Yes 10.5 No No No No Yes ?* Yes 10.5
Shuttle (landing) Yes 3.6 Yes 3.4 No No No No Yes 3.8 Yes 3.4
Shuttle (up-&-away) Yes 3.1 Yes 3.3 No No Yes Yes Yes 4.0 Yes 3.3
Shuttle (tracking) Yes 2.0 Yes 1.9 No Yes Yes Yes Yes 3.1 Yes 1.9
B-2 (landing) Yes 2.7 Yes 2.6 N/A N/A N/A N/A Yes 2.6 Yes 2.6
B-2 (refueling) Yes 3.8 No N/A N/A N/A N/A N/A No N/A No N/A
F-15A (landing) No N/A No N/A No No No Yes Yes 3.2 No N/A
*ωc ≈ ωPIO ≈ 5.4-5.5 rad/sec, but no PIO predicted based on criteria; Type II PIO predicted based on low-damped modes

95
The Gibson criteria are slightly more effective than Bandwidth at predicting PIO frequency,
getting four correct (within ±½ rad/sec of actual PIO frequency), compared to three with
Bandwidth. For the Shuttle and B-2 cases, the only difference between the criteria in prediction
of PIO frequency is the factor of 0.5 that is added to the Bandwidth-criteria estimates. The
Smith-Geddes criteria, favored by many because they are reported to correctly predict PIO
frequency, are correct only twice. *
In conclusion, as with the flight research data examined earlier in this section, the
Bandwidth/Phase Delay criteria are most effective at predicting the susceptibility to – and
resistance to – PIO.

K. APPLICATION TO OTHER DOCUMENTED PIOS

1. Objectives
With the exception of the four airplanes discussed in the previous subsection, this section has
until now focused on PIO data generated in flight research. Yet there is a sizable data base for
events that have occurred in operational, test, or applied research flying. Application of the
criteria to such events is much more difficult. First, there rarely is sufficient documentation of
the airplane dynamics to apply most, if any, of the criteria. Second, many of these PIOs were
Category II in nature, involving some degree of nonlinearity (commonly rate limiting of a
surface or software limiter). Third, we are not always able to get direct access to detailed
information about the incident, such as time histories, pilot comments, etc.
Equally important in this category are those airplanes that are considered to be resistant to PIO.
Based on the information presented in this report, it should be obvious that there is probably no
such thing as a PIO-proof airplane. Every airplane probably can be made to PIO simply by
changing mass, inertia, center-of-gravity, or flight control system properties. Still, it is safe to
say that just about every airplane in operational use today is judged by its pilots to be generally
resistant to PIO in its nominal configuration. So we need to look at some of these airplanes to
make sure we are not unfairly labeling them as PIO-prone.

2. Limitations
To the extent possible, data were obtained only from flight-generated models, or at least from
models based on flight data. In some cases, it was simply not possible, and analytical transfer-
function models had to be applied. In addition, it was not always possible to separate the effects
of the feel system dynamics, or in some cases to verify if the feel system were included in the
aircraft models. To the extent possible the feel system was always included for this analysis. In
many of the known events, some nonlinearity – typically rate limiting – was a contributor, but
none of the models included the effects of such nonlinearities. It is possible here only to
examine the dynamics of the linear airplane, to see if there was a tendency for PIO even without
rate limiting.

*
Ref. 72, which analyzed the three Shuttle PIOs and applied several criteria, lists different values
for Smith-Geddes criterion frequency than those in this report. Presumably the differences are
due to differing definitions of the criterion frequency. The values in Ref. 72 are much closer to
the measured PIO frequencies.

96
Because of a scarcity of data, it was not possible to apply all of the criteria discussed above to all
of the events. Rather than attempt to assemble a meager data base for the Neal-Smith criteria, it
was decided that these criteria would not be applied at all. No data are available for the Time-
Domain Neal-Smith criterion. This leaves us with Bandwidth, Gibson, and Smith-Geddes. But
since the primary parameter of the Gibson criteria – pitch attitude phase rate – is simply Phase
Delay multiplied by 720, and since the neutral-stability frequency used by Gibson’s criteria is
related to Bandwidth frequency, the Gibson criteria also were not applied to the data base. So
we are left with two criteria – Bandwidth and Smith-Geddes.

3. Data
A total of 45 airplane/flight condition sets could be identified. In some cases multiple PIOs
occurred for an airplane at different flight conditions or different tasks, and each event counts as
one set of data. In others, it was possible to identify both the condition for PIO and the condition
for the improved airplane. As a result, the data consist of fewer than 45 different airplanes. Of
these, 21 are considered to be real PIOs, and 24 are not.
All of the data sets examined here are for pitch/longitudinal control. A much smaller data base
exists for roll/lateral control.
Those data provided by our subcontractors, or data obtained from recent open literature, are
documented in Appendix E of this report. Appendix E is therefore a valuable adjunct to the
catalog published in Ref. 15. F-14 data come from Appendix C of this report. In addition, much
of the basic data can be found in a handful of references, primarily Refs. 15, 16, 48, 72, and 73.

4. The Bandwidth Criteria

Figure 39 presents the data base applied to the Bandwidth criteria. Correlation is excellent,
given the limitations of the data discussed above: only six of the 45 cases seem to be in the
wrong region, and four are PIOs in the non-PIO area. Of these four, the X-15 flight 1-1-5 event
was examined by Klyde et al.,15 who determined that severe rate limiting was present, and that
inclusion of rate limiting in the model places it in the PIO-possible region on Figure 39. The B-2
and YF-12 PIOs, both during refueling, were very likely flightpath or load-factor events, not
accounted for by pitch attitude Bandwidth. The fourth “PIO” was never a real PIO, but a
condition reported by Boeing 777 pilots early in the flight program as having the possibility for
one. Boeing redesigned the control laws to eliminate the condition.
In terms of percentages, 17 of the 21 PIOs are correctly correlated in Figure 39, as are 22 of the
24 non-PIOs. The overall effectiveness is therefore 39/45 = 87%, with a success rate of 17/21 =
81% at finding PIOs, and 22/24 = 92% at finding non-PIO cases.
Besides this overall effectiveness, the criteria show the right trends as well. For example, every
C-17 event plotted shows later software to have reduced the possibility for PIO. The 777 event
during landing derotation was fixed by a software modification, and the final FCS model lies
outside the PIO region in Figure 39.

5. Smith-Geddes Criteria
The conservative nature of the Smith-Geddes criteria again shows up, as Figure 40 clearly
indicates. The B-2 refueling and YF-12 events, both missed by Bandwidth, also are outside the
PIO region for the Smith-Geddes criteria.

97
 0.35
S hu ttle
D is p la y trk . N o P IO
D irec t+ 10 0 m s e c 3 6 0 kt/1 8k '
0 .3 P IO (N o t P re d icted )
C -1 7
F -1 4 H QDT
2 5 0k t/17 k '/35 ° Λ S /W A P IO (P re d icted )
B U F C M H ig h C -17
0 .2 5 AR
S /W F C -1 7 3 -e n g. L dg . S /W F
P ha s e E x pe r. 7 7 7 L dg . D e rot. F -8 D F B W
F -1 6 E a rly F lt. Te st
B -2 A p p roa c h C A S +1 0 0 m s ec PIO
D e la y, 1 0k /.9M
R at e- D ire ct
0 .2
τp θ L im ite d
S A S O ff C -17
E x pe r. F -1 6
(se c) T-3 8 2 0k /.95 M S h u ttle M e ch . PIO Pos sible
B o b w eig h t A LT -5 B-1
L oo p C lo se d M D -11 C -1 4 1 E xp e r. F -16
2 0 k /.9 M L d g. B -1
S /W G
S /W G (Fligh tp ath B and w id th)
0 .1 5 C ru is e Ld g .
SAS
Lo w -A lt. S /W G

SAS O n F in al
S hu ttle 777 C ruise B-1
S TS -4 T -3 8 E arly 7 77 V m o/M m o
AR 777
B .W . L o op 777 Flaps 20
S ec o nd a ry 250kt/15k' F-117A
0 .1 F ina l
Open
3k/.48M B -2 R e fu elin g
D C -10
L a nd in g F -1 5 A
La n d in g F -117A
F -1 5 E Ldg, Aft cg
0 .0 5 PIO P ossib le La n d in g Y F -1 2
L a nd in g
F-117A
Ldg, Fwd cg Y F -1 2 R igid B o dy
(Pitch R ate
N o P IO
O ve rsho ot) X -1 5 F lt. 1 -1 -5

0
0 1 2 3 4 5
P itc h A ttitud e B a n dw idth , ωB W θ (rad /s e c)

Figure 39. Bandwidth Criteria Applied to Real-World Data

6
T -3 8 P IO
B o b w eig h t T -3 8 No PIO P ossible PIO
B ob w e igh t
L oo p C lo se d
L o o p O p en

B -2 E x pe r. F -16
S /W G C-1 7 A R 1 0k /.9 M
R e fu e lin g 77 7 C ru is e
S/W F
5 YF -1 2 R igid B od y E arly SA S O ff

F-117A Ex pe r. F -16 S AS
Ldg , F wd cg 2 0k /.9 M On
F-11 7A C -1 7 3 -en g . L dg .
S /W G
3k /.4 8 M S /W F
X -15 S /W G Fina l
Flt. 1-1 -5 F -8 DF B W
Y F -12 Ex pe r. F-1 6 F-14
C A S +100 m s ec
L a nd in g SAS 20 k/.95M C-1 7 250 kt/
4 F -117 A M ech . 17k '/3 5°L
C riterion No P IO Ld g, A ft c g
Shu ttle
BU F CM H ig h
Direc t
F requency, A LT-5 +100 m sec
PIO (Not P redicte d) S huttle S TS -4
D ire ct
ωc θ F -15 E
C -1 41
P IO (P redicted) 77 7 La nding
Ldg . C -17 H Q D T S/W A
(rad/sec) V m o/M m o Se c.
F-1 5A 77 7
77 7
3 Flap s 20
L and in g D C -1 0
L dg . Lan ding D erot.
777 Ea rly F lt. Tes t
Final
2 50k t/1 5k' B-2 M D -11
Ap proa ch C ru is e

2
-40 -60 -80 -100 -120 -140 -160 -1 80 -200 -220 -240
P hase A ngle at C riterion F requency, ∠θ/F es (jω c ) (deg)

Figure 40. Smith-Geddes Criteria Applied to Real-World Data

98
While the Smith-Geddes criteria correctly predict many of the PIOs, they also predict PIO for
airplanes that did not exhibit them. For example, all versions of software for the Boeing 777 in
landing derotation are in the PIO region, as are almost all conditions for the C-17.
Overall, 25 of the 45 cases are correctly correlated with the Smith-Geddes criteria, for an
effectiveness rate of 56%. PIO was correctly found for 18/21 = 86% of the PIO cases, but of the
24 non-PIO cases only eight are correctly located, resulting in a success rate of 8/24 = 33%.

L. THE EFFECTS OF CONTROL/RESPONSE SENSITIVITY

With the exception of Gibson’s PIO Gain limits, the criteria reviewed in this section have all
been generated with an assumption that airplane design parameters not addressed by the criteria
are otherwise optimized. This includes, especially the mechanical characteristics (forces,
deflections, breakouts, friction, etc.) of the cockpit pitch controller, and the control/response
sensitivity. Historically, control/response sensitivity is defined in terms of standard parameters
such as control force per g, Fs/n, or initial acceleration per pound, q /Fs . The next subsection
looks at some relatively old criteria that deal in some way with control force per g.
There is little documented information to even attempt to define design limits on the mechanical
characteristics of the pitch controller. This is an area that justifies some research, since the
interplay between breakout/friction/deflection forces, response linearity, and response dynamics
certainly can have an influence on the occurrence of PIOs.
In the flight research studies of handling qualities or PIO reviewed for this report, the
control/response sensitivity has been established in one of two ways. In older experiments, such
as Neal-Smith, LAHOS, and HAVE PIO, the evaluation pilots were allowed to select values
during sampling of the configurations before starting formal evaluations. In more recent studies,
including the two TIFS experiments, sensitivity has been defined by the researchers, usually as a
constant value of q /Fs .
We have only a little evidence of the effect of pitch command sensitivity, primarily from the
1986 TIFS experiment. In that study, the majority of the pitch configurations had a “nominal”
value of pitch command sensitivity, q /Fs = 0.42 deg/sec2/lb. A portion of the study investigated
the effects of changes in sensitivity, repeating some configurations with a “high” value of 0.63
deg/sec2/lb and others with a “low” value of 0.25 seg/sec2/lb. The three sets of sensitivity were
applied to two good configurations, to which time delay values of 0.1 and 0.2 sec were added.
Results for this mini-experiment are shown in Figure 41. The plot shows added time delay
versus pitch command sensitivity, with PIO Tendency Ratings noted next to each data point.
HQRs are shown in brackets below the PIORs. Several pilots participated in the study; slashes
separate ratings from different pilots and commas separate repeats by the same pilot. The trends
show that, with no added time delay, any value of sensitivity is acceptable as far as PIO
tendency. There are Level 2 HQRs (4 and 5) for the high and low sensitivity values, suggesting
the nominal value of 0.42 is appropriate.
As time delay is added, however, the data in Figure 41 suggest a preference for lower values of
pitch command sensitivity. When 0.1 sec of delay is added, there are reports of “moderate” PIO
(HQRs better than 7) for the nominal sensitivity; with the high value, severe PIOs are reported;
but for the low value, not only is there no indication of PIO, but the PIORs from five different
pilots are all 1’s, suggesting there was not even a hint of undesirable motions, either.

99
 0.7 4/6
[5/10]
4/5/4/5/4
0.6 1/1/1 [5/8/6/8/4.5] 2,6/6/4/3/3 B
[2.5/4/3] [3,8/10/7/8/5
17L/L
Open -- No PIO
0.5 Solid -- PIO reported
1/1/1/1 3/5
Pilots separated by slash
[3/3/3/3] [4.5/9]
Repeats separated by comma
Pitch 4/4/1/3
0.4 1/1/1/1/1,1/1 [4/6/2/5] 4/6/2/4
command
[3/3/3/2/3,2/2] [7/9/5/6]
sensitivity,
MFES
2 0.3 1/1
(deg/sec /lb) [3.5/5]
1/1/1/1/1
PIO
1/1/1 [4/3/5/3/3.5] 1,1/3/3/1/4
0.2 [3/3/5] [3,4/5/8/4/7]
No PIO
0.1

0
0 0.1 0.2 0.3
Added time delay (sec)

Figure 41. Effect of Pitch Command Sensitivity on PIO as a Function of Added Time
Delay (Ratings are PIORs with HQRs in Brackets)

With 0.2 sec of added delay, only one pilot out of seven (counting both configurations)
considered the low-sensitivity cases to exhibit PIO, while the nominal and high cases exhibited
severe PIOs.
These data clearly suggest that 1) the “nominal” value of pitch sensitivity, q /Fs = 0.42
deg/sec2/lb, used in this study as well as the earlier TIFS experiment, was too high when time
delay was added, and 2) it is possible to minimize the risk of PIO by adjusting pitch sensitivity to
suit the dynamics of the airplane. Support for the first observation can be gleaned from the pilot
comments for the experiment; several pilots complained of excessive sensitivity and very light
control forces, many of these associated with PIOs. The first observation has also been
confirmed by a series of flight experiments conducted by McDonnell Douglas Aerospace (now
Boeing) and the Air Force.57 These experiments, also performed with the TIFS airplane, used a
lower (but still fixed) value of sensitivity, and the results suggest different Bandwidth/Phase
Delay limits for both Level 2 flying qualities and PIO susceptibility than those discussed in this
report.
Unfortunately, if we accept the hypothesis that the “nominal” pitch sensitivity was too high for
both the 1984 and 1986 TIFS experiments, we must eliminate virtually all of the data from these
two studies in our development of PIO criteria. Instead, in this report we have used the data
without judgment, assuming that PIO limits based on the data will be slightly conservative.

100
M. SOME CRITERIA FOR LOW-ALTITUDE, HIGH-SPEED FLIGHT
Beginning in the early 1960’s, largely as a result of the T-38A PIO, there was an intense effort to
develop criteria for prediction of PIOs primarily in low-altitude, high-speed flight. Several
candidate criteria that showed considerable promise were proposed by researchers. For the most
part, these criteria were modal – that is, they placed explicit limits on the short-period mode, or
some combination of the short-period and stick force per g or control system phase lags.
These criteria drew renewed interest in the mid-1990’s when the Federal Aviation
Administration (FAA), in reaction to a reported PIO on an MD-11,74 issued proposed revisions to
Advisory Circular 25-7, aimed specifically at protecting against PIOs on commercial airliners in
cruise. One of the criteria from the 1960’s was included in early versions of the revised text.
The consequence of the FAA’s text was a brief flurry of attention to the modal criteria, because
they seemed to work very well for predicting the PIO susceptibility for simple (unaugmented or
lightly-augmented) airplanes in low-altitude, high-speed flight. And in truth, one of the criteria
still shows value for such a limited application. For more complex modern airplanes, with high-
gain augmentation systems and unusual response modes, a more sophisticated requirement such
as Bandwidth/Phase Delay must be applied.
Following is a review of three of the modal criteria, using a limited set of data to compare their
effectiveness.

1. STI High-Gain Asymptote Parameter

Ashkenas and associates12 at Systems Technology, Inc. (STI) observed that pitch control at high
speeds and low altitude, using pitch-attitude cues, can be described by the relationship between
the total damping of the short-period mode, 2ζ sp ωsp , and the pitch attitude zero 1/Tθ2 . For a
pure-gain pilot, the limiting value of short-period total damping is dictated by the “High-Gain
Asymptote” parameter
1 1
σ a = ζ sp ωsp −
2 Tθ 2

The closer this value is to zero, the more likely a PIO will develop. The STI researchers point
out further, however, that this High-Gain Asymptote parameter will be greater than zero for all
but the most unusual airframe configurations. They conclude that “for airplanes with negligible
control system dynamics (including nonlinear elements or bobweight effects), longitudinal PIO's
involving only attitude control are essentially impossible.”
Obviously, the initial assumption that the pilot is a pure-gain operator with no input shaping or
time delay is key to this conclusion. Nonetheless, the High-Gain Asymptote parameter, while
useful for illustration, has never been formally adopted as a criterion for predicting PIO. STI
developed a requirement that incorporated a measure of the phase lag of the flight control system
at the short-period frequency, 75 with a lower limit on σ a of 0.5 for most flight tasks.
Examination of a single example data set will illustrate the overall lack of correlation for the
High-Gain Asymptote parameter. Around the same time that Ashkenas and STI were doing their
PIO work, two researchers at North American Aviation, A’Harrah and Siewert, were pursuing
their own criteria. The criteria will be discussed below; for now we are interested only in the
data collated by A’Harrah and Siewert.76 They compiled a list of 43 flight data points for low-

101
altitude, high-speed flight involving fighter and attack aircraft. For 41 of these points the basic
dynamics (short-period characteristics plus stick force and position per g) were noted. PIO was
observed for 14 of the 41 cases. Data sources for many of the cases are rather obscure and
difficult to locate after so many years; the few references that could be located had little
additional information about the dynamics of the airplanes. Hence any attempt to apply the
criteria described earlier in this section would be futile.
Figure 42 shows the High-Gain Asymptote for the 41 cases. The crossplotted parameter, stick
force per g, was selected simply for convenience, not because it is necessarily expected to help
correlate the data. There is some indication that σ a is important, since all PIO cases lie below a
value of about 0.9. On the other hand, it is clearly not the only parameter of importance, since
there are many non-PIO points at low σ a values as well. The second parameter selected by STI
for their criteria,75 control system phase lag at the short-period frequency, is unavailable for these
cases. As will be seen below, however, there are other parameters – especially stick force per g
– that better correlate the Figure 42 data, so we may conclude that σ a is not the answer to the
PIO problem for traditional airplanes.

2. A’Harrah-Siewert Criteria
A’Harrah and Siewert proposed their own criteria for prevention of PIO, especially for low-
altitude, high-speed flight, in the 1960’s. The most mature of these recently resurfaced as the
Federal Aviation Administration searched for ways to regulate against PIOs in commercial
airliners. Early draft revisions to Advisory Circular AC 25-7, Flight Test Guide for Certification
of Transport Category Airplanes, proposed the A’Harrah-Siewert criteria, but later versions,
including the 1998 release of AC 25-7A, 77 did not.
16
F s/n
(lb /g)
14
PIO O ccu rre n ce
12 N o P IO Te n d e n cie s

10

0
-2 -1 0 1 2 3 4

H igh -G a in A sym p to te ( σa = ζsp ωsp - 1 1 )

2 T θ2
Figure 42. STI High-Gain Asymptote Parameter12 for A'Harrah-Siewert76 Data

102
The A’Harrah-Siewert PIO criteria76 use an airplane response metric and a control metric. The
response metric is defined as the time to one-tenth amplitude of the short-period response,
computed as T1/10 = ln(0.1)/(ζ sp ωsp ) , and the control metric is a combination of stick force and
position per g, (Fs /n L ) 3 × (δ/n L ) , in units of in.-lb3/g4. The PIO boundary for their criteria is
shown in Figure 43, along with their flight data introduced in Figure 42.
Correlation is very good for the flight data, as there are only two PIO cases clearly on the non-
PIO side of the boundary (and both are at least near the boundary), and two non-PIO cases on the
PIO side. This makes the A’Harrah-Siewert criteria appear to be very effective. (One significant
shortcoming, noted by large-airplane manufacturers, has been that the control metric for wheel-
and-yoke transports is orders of magnitude larger than the highest value in Figure 43, and this
has understandably raised some concern.)

3. Dynamic Stick Force per g

The developers of the U.S. military specifications in the 1960’s were aware of the work of both
STI and A’Harrah and Siewert. Based on research at Cornell Aeronautical Laboratory (later
Calspan and now flight operations for Veridian), however, the choice for the military
requirements was a dynamic stick force per g parameter that is primarily a function of short-
period damping ratio. The requirements on stick force per g from the military standard4 are
shown on Figure 44, along with the now-familiar data collected by A’Harrah and Siewert.

10 0.0

P IO O ccu rren ce
No P IO Tend encies

10.0
R e sp o n se
M e tric ,
T 1 /1 0
(s e c )

1.0

0 .1
1 10 1 00 1 000 10000

C o n tro l M e tric , (F s /n )3 x ( δ/n ) (in .-lb 3 /g 4 )

Figure 43. A'Harrah-Siewert PIO Boundary76

103
 100

P IO Occurre nce
N o P IO Te n d e n cie s
M IL-S T D -1 79 7 A
Le v el 2 L im it
on (F s /n )m in
M an e u ve rin g
F o rce
G ra d ie n t, 1 0
F s /n
(lb /g )

M IL-S T D -1 797 A
Le ve l 2 Lim it

1
0 .0 1 0.1 0 1 .0 0

S h o rt-P e rio d D a m p in g R a tio, ζsp

Figure 44. Dynamic Stick Force per g PIO Requirements4

The requirements of Figure 44 are effective at correlating the data, better than either the High-
Gain Asymptote or A’Harrah-Siewert criteria. One PIO case lies on the non-PIO side, and one
non-PIO case is on the PIO side of the boundary. The A’Harrah-Siewert criterion had four
incorrectly located points. Thus, we may conclude that, for conventional airplanes where short-
period damping and control response are the primary causes of PIO, dynamic stick force per g is
a very effective PIO criterion.
It may be conjectured, in fact, that dynamic stick force per g is a more appropriate parameter
than any of the attitude-dominant criteria reviewed earlier at very high speeds – where small
attitude changes produce large normal acceleration changes. The lack of flight research data
showing a compelling need for the criteria may be more a function of the capabilities of variable-
stability airplanes than proof that such criteria are redundant.

104
 VI. PREVENTION OF PIO BY DESIGN (CONTINUED):
LONGITUDINAL CRITERIA FOR CATEGORY II PIOS

A. BACKGROUND
The previous section of this report presented several alternative analytical criteria whose purpose
is to prevent the occurrence of Category I (effectively linear) PIOs. In truth, most recent PIOs –
and perhaps most PIOs since the early days of flight – have involved some nonlinearity, usually
limiting of the surface actuator rates or positions. So while it is critically important to start with
analytical criteria that prevent linear PIOs, from early in the design stage through full-scale
development the possibility of a nonlinear, Category II PIO must be accounted for. This means a
new form of criteria that somehow allows for the existence of one or more nonlinear elements.
Because the importance for such criteria has only recently been exposed, work on their
development lags far behind that for the Category I criteria in the last section. In addition, of
course, even the addition of a “simple” nonlinear element such as actuator rate limiting greatly
complicates the task of criteria development: such nonlinearities are, by nature, functions of both
frequency and amplitude.
As of this writing, there are only four known criteria that are intended specifically to deal with
the causes of Category II PIOs. One of them (Bandwidth-based criteria that extend the criteria of
the last section to cover nonlinearities) was developed under this research effort and hence is
discussed here in some detail. The other three, however, are too complex to investigate in detail
in this report, and besides, the data base for their assessment is very limited (consisting, from
flight research at least, of a single reference, the HAVE LIMITS experiment conducted by
students of the Air Force Test Pilot School21). Instead, a very brief description is given for each
of them, including known strengths and weaknesses. A more detailed description is given for the
Bandwidth criteria.

B. THE CRITERIA FOR CATEGORY II PIOS

The number of PIO susceptibility prediction criteria for Category II PIOs is quite small. All
have been developed and proposed in only the last few years, and all are, of necessity, much
more complex to apply than those for Category I PIOs. Four criteria were identified in this study
based on publications in the open literature, and of these four, one (based on Bandwidth) was
developed under this research effort.
Following is a synopsis of the four criteria, following the format of Section V for Category I
criteria. For the first three, the information given is based entirely on published papers and
reports. Because of their relative complexity, no attempt was made in this contract to implement
and exercise any of the three. Recently, detailed documentation of the methods, and in some
cases accompanying software, has become available, and all of the approaches now be
implemented with relatively little effort.

1. OLOP Criterion
a. Theory Behind the OLOP criterion
The Open Loop Onset Point (OLOP) Criterion78 was developed by the Deutsches Zentrum für
Luft- und Raumfahrt e.V. (DLR) in Braunschweig, Germany. OLOP was developed from

105
describing-function analysis of rate-limited flight control systems. It is based on the observation
that the location of the rate limiter onset point in the frequency response is highly correlated with
the corresponding phase jump in the quasi-linear frequency response of the closed-loop system.
OLOP uses a linearized model of the rate-limited vehicle and assumes a high-gain pilot with full-
amplitude control inputs.

b. Parameters for the OLOP Criterion

The OLOP is the amplitude and phase of the linear frequency response measured at the
frequency for rate limiting onset. If this amplitude is too high, Category II PIO is predicted to
occur.

c. Evolution of the OLOP Criterion

The criterion has undergone only minor revisions and refinements over the last few years. It is
relatively new – as are all of the criteria in this section – and hence has had little time for
“evolution.”

d. Strengths of the OLOP Criterion

The most apparent strength is that the criterion deals directly with rate limiting, yet it does so in
the context of a linear aircraft model. This makes it relatively simple in application. It
eliminates issues about pilot input amplitude by assuming full stick.

e. Shortcomings of the OLOP Criterion

OLOP can be applied only to Category II PIOs, not to Category I, and hence the connection with
all previous work is lost. It can only deal with one nonlinear element; all other possible sources
of nonlinearity in the airplane (force feel, command gradients, gain scheduling, etc.) must be
linearized.
A significant shortcoming noted in various applications is that, although the criterion is defined
for full pilot stick inputs, it has typically been applied only after actual control usage can be
determined. For example, in its application to the HAVE LIMITS data, Ref. 79 states that the
closed-loop onset frequency is determined for maximum stick force inputs, and further, that,
Unfortunately, no time histories of the original experiment were available….
These could have been used to determine the pilot gain during the flights. A
maximum stick deflection … of 1.5 inches was assumed.
In other words, the criterion was applied only after determining pilot control inputs! The
apparently good correlation with the HAVE LIMITS data (discussed below) is mitigated by the
need to apply this intermediate step. It is not known how the correlation would be if full control
input had been used, as intended. Other applications of the criterion show the same reliance on a
priori knowledge of input amplitude.80

2. Pilot-Vehicle Dynamics NonLinear (PVDNL)

a. Theory Behind the PVDNL Criteria
PVDNL is based on the Structural Model of the human pilot, developed by Hess.24 It assumes
compensatory closed-loop tracking with vestibular, proprioceptive, and visual feedback loops. A

106
target pilot-vehicle crossover frequency of 2.0 rad/sec is assumed and parameters of the pilot
model are either fixed at nominal values or adjusted to obtain the desired crossover frequency.

b. Parameters for the PVDNL Criteria

A “Handling qualities Sensitivity Function,” HQSF, and the normalized power spectral density
(PSD) of the proprioceptive feedback signal, Φ u mu m (ω) , are computed from the pilot-vehicle
system. Both are functions of frequency; HSQF is a measure of the closed-loop performance in
compensatory tracking, and Φ u mu m (ω) is a measure of input power.

c. Evolution of the PVDNL Criteria

The Structural Model has been used to develop linear handling qualities criteria based on HQSF
as a function of frequency. This was extended to Category I PIO, and most recently to Category
II PIO, through the computation of Φ u mu m (ω) .

d. Strengths of the PVDNL Criteria

The criteria provide a direct replication of piloting activity for such high-gain tasks as attitude
tracking (as was used in the HAVE LIMITS flight research program). They can be applied for
both linear and nonlinear aircraft models, thus preserving the connection with past data (e.g.,
Section V).
The developers of the criteria have produced Matlab software for performing the analysis, thus
making them among the simplest to apply. 81

e. Shortcomings of the PVDNL Criteria

The most serious shortcoming is that the parameters are functions of frequency, and thus must be
judged across a span of frequencies. This opens the usual questions about occasional, minor
exceedances of a limit: do we consider that to be the same as a more obvious violation of the
limit?
In terms of PIO, the gradations in PIOR raise some questions, especially if the decision-tree, six-
point PIO Tendency classification scale is used. The possible PIOR ranges are 1”3,25” ±
which translates only to PIORs of 1 and 2; 2<PIOR<4, which translates only to a PIOR of 3; and
PIOR• ZKLFK WUDQVODWHV WR 3,25V RI   DQG  7KH PLGGOH UDQJH VHHPV TXLWH OLPLWHG
fortunately, for our purposes, if we accept that the sort of PIO we are concerned about is severe,
we can worry only about the regions for PIOR<4 and PIOR•
As with OLOP, PVDNL works only for one nonlinear element. All other nonlinearities in the
aircraft must be linearized.
An assumption of a crossover frequency of 2.0 rad/sec may be questionable for some aircraft and
flight conditions. Certainly, in a PIO the relevant frequency is not crossover but the PIO
frequency, which is usually higher than 2.0 rad/sec. On the other hand, this is at least a
reasonable value based on analysis of tracking data,10,67 and there is certainly some attractiveness
to the concept of holding as many parameters constant as possible, to reduce the degrees of
freedom in the closed-loop analysis.

107
3. Time-Domain Neal-Smith (TDNS)
a. Theory Behind the TDNS Criterion
The TDNS criterion18 was described in Section V. It was originally developed as a criterion for
Category II PIO, but when the initial work was performed there were no research data for
longitudinal Category II PIOs to apply to it.

b. Parameters for the TDNS Criterion

The single parameter, the 2nd derivative of rms tracking error, d 2 (rmsθ e | t > D ) dD 2 , for a pilot
model following a step commanded change in attitude, was described in Section V.

c. Evolution of the TDNS Criterion

This is a relatively new criterion. A lateral-axis version has been proposed,82 but only for
handling qualities, and the only extension to Category II PIO has been in a review by the Air
Force.83

d. Strengths of the TDNS Criterion

See Section V.

e. Shortcomings of the TDNS Criterion

See Section V.

4. Airplane Bandwidth
a. Theory Behind the Bandwidth Criteria
The basic theory behind the criteria is discussed in Section V. Because this is the first
publication of the work, most of the remainder of this section of the report is devoted to
extending the criteria to Category II PIOs.

b. Parameters for the Bandwidth Criteria

See Section V.

c. Evolution of the Bandwidth Criteria

See Section V.

d. Strengths of the Bandwidth Criteria

These are the only criteria for Category II PIO that can be applied through all stages of aircraft
development, and they are especially amenable to flight testing. There is no need for
linearization or simplification.

e. Shortcomings of the Bandwidth Criteria

The most glaring shortcoming is that the basic data for application of the criteria can be difficult
to obtain. For example, later in this section it is indicated that frequency-response data are
required at elevated load factors – but getting such data in flight is not easy. In addition, the
particulars of pilot input size, frequency range for sweeps, etc., must be carefully defined and
controlled.

108
The Bandwidth criteria are not necessarily more or less difficult to apply than the other three.
The source of the difficulty has simply been shifted, from assumptions about pilots and aircraft
models, to the rigors of generating the needed data. By definition, the problem of predicting
susceptibility to Category II PIOs is extremely challenging, and the criteria are correspondingly
complicated in one way or another. Bandwidth is no exception.

5. Application of the Criteria to HAVE LIMITS

Since all of the criteria described above have been applied to at least some portion of the HAVE
LIMITS data, it makes sense to use those data to determine their effectiveness at correlating with
what happened in flight. It is not possible, with such a limited data base, to conclude whether
one set of criteria is better overall, but only relatively.

a. Overall Effectiveness
Here we will look only at the effectiveness of the four criteria for predicting PIO susceptibility,
and for PVDNL and Bandwidth we can also look at the handling qualities predictions. A
summary is given in Table 14, for which considerable discussion is needed.

Table 14. Effectiveness of the Criteria for HAVE LIMITS Data (Discrete Task)
Criteria Predictions
Config-
RL PIORs HQRs OLOP TDNS PVDNL Bandwidth
uration
PIO PIO PIO HQ Level PIO HQ Level
10 3R3,2R3 5R5,4R6 Yes Yes Yes 3 No 2-3

20 2 R2 R2 2 R3 R2 No No No 1 No 2

30 2 R1 R1 2 R2 R1 No No (No) (1) (No) (2)

2D
40 2R3,1,2R1,2 3R5,2,4R1,2 No No No 1 (No) (2)

50 2 R1 R2 4 R2 R4 No No (No) (2) (No) (2)

157 2R-R2 2R-R4 No No No 2 Bobble 2

10 3 R5 R3 6 R6 R7 Yes Yes

20 4R3,4,3,4R4,1 5R4,6,4,8R7,3 No Yes Yes 3

30 3R-R3 4R-R6 No No (No) (2)

2P
40 2R4R3,3 3R6R5,5 No No No 2

50 2R3,3R2 4R5,4R3 No No (No) 2

157 2R-R2 4R-R4 No No No 2 No 2

20 6 R6 R5 10R10R10 Yes Yes (Yes) (3) Severe 3

30 5R-R5 9R-R10 Yes Yes (Yes) (3) (Severe) (3)

40 5 R6 R5 10R10R10 Yes Yes (Yes) (3) (Severe) (3)

2DU
50 4,4R5R5 5,9R10R10 Yes Yes (Yes) (3) (Severe) (3)

60 4R3,2R5,5 8R4,3R10,10 Yes Yes Yes 3 Severe 3

157 -R2R3,2 -R2R5,4 Yes No No 2 Bobble 2

109
The ratings from HAVE LIMITS are for the discrete (pitch-and-roll) HUD tracking task only.
This was the most effective task at exposing PIO, according to the test pilots, and it was the
primary basis for all of the criteria applications considered here. The pilot ratings are Pilot
1RPilot 2RPilot 3, with repeats separated by commas. “RL” is the rate limit in deg/sec. PIO
predictions are listed for all four criteria, and predicted handling qualities (HQ) Levels for
PVDNL and Bandwidth.

For PVDNL, only selected configurations were applied;84 based on these configurations, it is
possible to interpolate the likely predictions for most of the others. Even fewer configurations
were assessed using Bandwidth, and interpolations were again made where possible.
Interpolations are listed in parentheses. Blank entries indicate cases for which no interpolations
could be made.
For application of TDNS, the original criterion limit proposed by Bailey and Bidlack18 was not
used for reasons discussed below; instead, limits proposed by Foringer and Leggett,83 also
discussed below, were applied.
“Hits” – where the criteria correlate with flight – are indicated by bold print.
The results of Table 14 show that all of the criteria are reasonably effective at correlating with
the PIO results from flight. OLOP is perhaps the most conservative, as it predicts PIO for
Configuration 2D with 10 deg/sec rate limit and for 2DU with 157 deg/sec limit, and neither
exhibited PIO in flight. It could be argued that a prediction of PIO for 2DU is reasonable, since
this configuration probably could be made to PIO if larger control inputs were applied. But
remember that the input size has already been scaled down from full stick (as the criterion
dictates) to 1.5 inches, so correlation should be better for this case.
TDNS is very effective when applied to the Air Force-modified limits. The only failures are two
cases (2D/10 and 2P/40) for which other criteria also fail.
In applying PVDNL, an assumption had to be made about the impact of even slight exceedances
of the PIO or handling qualities limits. A conservative approach was taken, assuming even the
slight exceedance meant the criteria predicted the worst case. These criteria still work very well,
for both PIO and handling qualities. In the case of the latter, it is interesting to note that PVDNL
predicts an improvement in handling qualities as the rate limit is reduced (correctly predicting
Level 1 for 40 and 20 deg/sec rate limits) before degrading again at the lowest limit.
The Bandwidth predictions are based on the ±2-g frequency sweeps, discussed later in this
section.

b. Graphical Trends
As with the criteria and Category I PIOs, it is insightful to examine the results of the data
correlations graphically. This is not so easy with the PVDNL criteria, since each configuration
requires two plots – one each for HQSF and Φ u mu m (ω) as functions of frequency. An example,
to illustrate the criteria, is shown in Figure 45. The relevant limits on handling qualities Level
(left plot) and PIO (right plot) are shown. The actual results are the bold lines for the no-rate-
limit case (labeled “analysis”) and the jagged lines for the actual rate-limited case (labeled
“simulation”).

110
 Figure 45. Example of PVDNL for Configuration 2D, 10 deg/sec Rate Limit84

The relevant results in Figure 45 – the jagged lines – illustrate the dilemma with criteria such as
this: both lines exceed the worst boundaries around 1.5 rad/sec, but neither by very much. So is
this configuration predicted to be Level 3 with PIO, or Level 1 with no PIO? For the listing in
Table 14 the former was assumed – i.e., any violation constitutes the prediction. Comparison of
the linear (analysis) and nonlinear (simulation) lines also shows the apparently beneficial effect
of the rate limiter in suppressing a peak at around 4 rad/sec. This is an interesting phenomenon
not addressed by any of the other criteria.
Summary plots of the HAVE LIMITS criteria for OLOP are shown in Figure 46 (the non-PIO
region is below the curved line). OLOP confirms the extreme PIO susceptibility of the 2DU
configurations, but includes the 157-deg/sec rate limit. The plot on the right in Figure 46
includes some information about PIO ratings. Shadings indicate the range of change in PIOR
(DPIOR) from the highest-rate-limit case.

Figure 46. Application of OLOP to HAVE LIMITS Configurations79

111
In its original formulation, TDNS was a single criterion: if the 2nd derivative of rms tracking
error, d 2 (rmsθ e | t > D ) dD 2 (discussed further in Section V), was greater than 100, PIO was
predicted. But this limit was based on data for Category I PIOs only. Foringer and Leggett83
found that any HAVE LIMITS configuration could be made to exceed this limit simply by
tightening up on the task demands (reducing acquisition time, D, in seconds). They then pursued
several alternative approaches, all showing some promise but all requiring very specific
constraints on the methods (step size, pilot model, etc.). One promising criterion involves a
combination of d 2 (rmsθ e | t > D ) dD 2 and the value of D where d 2 (rmsθ e | t > D ) dD 2 first exceeds
100. The resulting possible criterion is as shown in Figure 47.

Figure 47. Modification to TDNS and HAVE LIMITS Data83

C. EXTENSION OF BANDWIDTH CRITERIA IN THE PRESENCE OF

NONLINEARITIES
Measurement of the parameters needed to apply the Bandwidth criteria obviously requires an
accurate frequency-response representation of the airplane. For an analytical model, and for
simple airplanes with no significant nonlinearities, this response is not difficult to obtain.
If the frequency response is obtained from flight data, the usual input is a time-varying frequency
sweep of the longitudinal cockpit control. The time responses of pitch attitude (or, more
commonly, pitch rate since rate gyros are usually more available and angular rate will have more
output signal power) and control force are Fast-Fourier Transformed (FFT’d) to produce the
required frequency response.
If there are no nonlinear elements, the amplitude of the frequency sweep must be only as large as
required to perturb the airplane, but not so large that changes from the trim flight condition
result. Coherence between output and input will be affected only by the power content of the
two signals, and presence of any extraneous factors, such as gyro noise, turbulence, and
uncorrelated control inputs from other controllers.
In the presence of nonlinear elements, however, the task can be much more challenging.
Nonlinearities in aircraft response will contaminate the output signal and can reduce overall

112
coherence by either introducing uncorrelated response or reducing the signal of the correlated
response.
If handling-qualities and PIO susceptibility criteria are to be based on the frequency responses
for a real airplane, the severity of common nonlinearities on measurement of the parameters used
by those criteria must be understood.

1. Frequency Response Measures

The output time history data from a frequency sweep are analyzed with some form of fast
Fourier transform to produce the desired frequency response. A typical set of such data, for a
piloted sweep of an airplane with actuator rate limiting, is illustrated by Figure 48. Different
specific techniques produce different-looking frequency responses; the example in Figure 48,
and all results presented here, were generated by FREquency Domain Analysis (FREDA)
software, proprietary to Systems Technology, Inc. FREDA assembles “bins” of adjacent
frequencies to improve the FFT at any one frequency, producing a series of discrete data points.
Other software, such as CIFER,85 may use different methods.
For the Figure 48 example, the frequency response appears to be of good quality (in terms of
lack of point-to-point scatter) for frequencies below about 10 rad/sec.
The middle plot is the coherence as a function of frequency. Coherence is a measure of the
amount to which the output is linearly related to the input, and is defined as
2
G xy
ρ 2xy =
G xx G yy

where G yy is the output autospectral density. For good data, coherence should be above about
0.6-0.8. The example in Figure 48 shows excellent coherence up to about 5 rad/sec, then falls
off rapidly. (Note that the Bode plot continues to look reasonable beyond this frequency.)
The third plot on Figure 48 shows the autospectra for the input (G xx ) ( )
and output G yy ,
expressed in terms of units of power decibels, 10log10. Also on this plot is the power for the
remnant cross-spectrum, G yynn ; it is calculated by FREDA as a measure of the uncorrelated
response. For the example in Figure 48, input power is high up to about 10 rad/sec, then drops
off. Output power, however, drops at a much lower frequency. In addition, the remnant is lower
than output until around 8 rad/sec, where the two sets of data merge, indicating that the content
of the output at higher frequencies consists entirely of remnant.
As we will see, the example in Figure 48 is typical of frequency-sweep responses in the presence
of severe rate limiting: the limiting inhibits aircraft response (output power), results in nonlinear
response that reduces coherence, but does not prevent the acquisition of trustworthy frequency-
response data.

113
 20 0
ωB W = 2 .0
Φ
0 -1 8 0
ω1 8 0 = 2 .6
M ag (dB) θ P h. (deg)
-2 0 ω -3 6 0
B W θ = ωB W G = 0 .3 7
τp = 0 .2 8 s e c
-40 θ -540
M agnitude P ha se

-6 0 -7 2 0
1
0 .8
0 .6
ρ2
0 .4
0 .2
0
200.1 1 10

P o w e r -2 0
(d B ) Input
-4 0 Output
Rem n ant
-6 0
0.1 1 10
F req u ency (rad /se c)

Figure 48. Example Output Frequency Response for a Rate-Limited Airplane

2. Airplane Model
The model used for this analysis was based on the variable-stability NT-33A as used in the
HAVE LIMITS flight research study by students of the US Air Force Test Pilot School.21,86 The
HAVE LIMITS Test Management Project investigated the effects of elevator rate limiting on
PIO for three longitudinal configurations. The baseline, configuration 2D, was designed to have
Level 1 handling qualities for up-and-away pitch tracking. In flight, this configuration exhibited
pitch rate overshoot that was marginally acceptable, resulting in some Level 2 Cooper-Harper
Handling Qualities Ratings.28 The model was implemented with no feedbacks around the limiter,
so the only consequence of reaching the limiter was the resulting loss of phase and amplitude
attenuation.
A second model, Configuration 2DU, had approximately the same dynamics as 2D but by
augmentation. The unaugmented airplane (as seen whenever the rate limiter was saturated) was
very unstable and almost unflyable.
For this study, the models of the configurations were based on analytical models from Calspan
Corp. The short-term characteristics of the configurations actually evaluated in flight were

114
identified from frequency sweeps44 and differ slightly from the Calspan models. The most
significant difference was the presence of an additional equivalent time delay in the flight-
derived models. A pure delay term was added to the analytical models to account for this
equivalent time delay. Elevator control effectiveness was determined from the flight data to be
approximately 70% of that in the analytical model, and an adjustment was made to account for
the difference. Because the short-period dynamics (specifically, short-period damping and
frequency) were similar, no further attempt was made to adjust the analytical model to match the
flight data.
Transfer functions for the pitch attitude response to longitudinal stick force for the analytical
models are listed in Table 15. The dynamics for 2D and 2DU, in the absence of rate limiting,
were identical. Table 15 also lists the transfer function for 2DU without augmentation. The
models include the dynamics of the elevator actuator and the control stick feel system. They are
written in “shorthand notation,” so that (a) ≡ (s + a) and [ζ, ω n ] ≡ [s 2 + 2ζω n s + ω 2n ].
Table 15. Transfer Functions of Analytical Models
Conf. θ/Fes

2D
10(0.029)(1.3)e -0.06s
and
[0.333,0.036][0.7,4.73][1.0,31][0.7,72.5]
2DU

2DU 10(0.029)(1.3)e-0.06s
(unaugmented) (-1.26)(2.23)[0.255,0.180][1.0,31][0.7,76.4]

3. Sources of Nonlinearity
The modern fly-by-wire (FBW) flight control system (FCS) has many possible sources of
nonlinearity. A sketch of a typical FBW FCS shows some of these sources (Figure 49),
including command shaping from the pilot’s controller, and rate limiting, in the form of either
hardware (a hydraulic actuator) or software (incorporated to prevent hardware rate limiting).
These two sources of nonlinearity were the focus of this study.

B re ak ou t & C om m a nd S oftw are G ain R a te and pos itio n A erod yna m ic

frictio n ra te lim it rate lim it s c he duling lim its no nlin earities
. .
δ δ o ut e o ut y
F es δes + δe c δe
. . A ctu a tor A ircra ft
F δ in - e in
x (U ,h,M )

H1 1

+
H2 2

Figure 49. Sources of Nonlinearity in Flight Control Systems

115
a. Controller Command Shaping
Some form of command shaping is used on the cockpit controller of every airplane in existence.
Because the Bandwidth criteria are written in terms of control force inputs, controller breakout
and hysteresis will introduce nonlinearities. In addition, command shaping is usually designed in
to provide proper sensitivity for small inputs and high control authority for large inputs.
Command shaping can usually be approximated for linear analysis by an equivalent single gain
element;87 the issue for flight testing is the effect of the shaping on overall coherence and on
measurement of the parameters for the Bandwidth criteria.
The importance of command shaping on flying qualities was demonstrated in a flight research
program conducted by Calspan on the NT-33A.88 For roll maneuvering, configurations with
high roll damping and essentially no command shaping were susceptible to high-frequency roll
oscillations identified as “roll ratchet;” addition of only a slight amount of shaping, around zero
control input, improved the flying qualities drastically. 89
Three command shaping curves were investigated in this study, as sketched in Figure 50. (Only
positive commands are represented in the figure; for this analysis, the commands were assumed
to be symmetric.) The curve labeled “Calspan” is the shaping used on the HAVE LIMITS flight
experiment21 and includes a 0.02-in. deadband around zero. The linear command was selected to
give approximately the same overall response for full control input (3.6 in. of stick deflection).
The power curve is simply a cubic gradient of the form y = x3 and is meant to represent an
extreme form of nonlinearity. The power gradient would probably not be acceptable in flight,
since it has very low control command at low deflections, and extremely high command at
higher deflections.
There was no mechanical breakout force on the control stick in HAVE LIMITS, and breakout
was not included in this analysis.

50
Calspan
40 Linear
30 Power
Output
(deg)
20

10

0
0 1 2 3 4
Input (in.)

Figure 50. Command Shaping

116
b. Rate Limiting
Rate limiting is a serious problem for modern, highly-augmented aircraft. If feedback
stabilization is required for an unstable airplane, the consequences of rate-limiting can be
especially dire, since the airplane is effectively open-loop with no control authority for the pilot
to even attempt to control the instability.
There is evidence that rate limiting has been a factor in most recent Category II PIOs.1 The most
likely source of rate limiting is the control surface actuator; in reality, however, for many of the
PIOs in which rate limiting occurred, the actual rate-limiting element was a software limiter,
included in the flight control system for other reasons.47
As with command shaping, describing functions for software rate limiting are available,87 as are
models for a first-order actuator with rate limiting.15
In the HAVE LIMITS flight program,21 ,86 rate limiting was varied from none (effectively the
rate limit of the NT-33A actuator itself, at approximately 157 deg/sec) down to 10 deg/sec.
Specific values investigated were 60, 50, 40, 30, 20, and 10 deg/sec. The highest value was not
flown on configuration 2D and the lowest was not flown on 2DU. For this analytical study, the
rate limits selected were 157, 60, 20, and 10 deg/sec.

c. Effects of Multiple Nonlinearities

There have been no systematic studies of the effects of multiple nonlinearities on either the
quality of frequency-response data, or the accuracy of the measurement of Bandwidth
parameters. Application of another proposed PIO criterion, Time-Domain Neal-Smith, has
shown some shortcomings when there is more than one nonlinearity present in the aircraft
model.83 Since it is almost impossible to have a true linear airplane, the impact of multiple
nonlinearities must be understood.

4. Application of Frequency Sweep Technique to HAVE LIMITS Configurations

a. Summary of the Program
In general, the results from the in-flight piloted assessments21,86 of configurations 2D and 2DU,
for a discrete pitch and roll HUD tracking task, may be summarized as follows.
None of the three evaluation pilots reported PIOs for configuration 2D for any rate limit, down to
the lowest tested value of 10 deg/sec. Level 2 Cooper-Harper Handling Qualities were assigned
for this value, however, indicating a degradation in handling qualities. For all higher values of
rate limiting the airplane was considered to be Level 1.
By contrast, for any value of rate limiting, configuration 2DU exhibited a rapid and severe
divergent PIO. One of the three pilots did not see the PIO for the 60-deg/sec limit, but all pilots
experienced PIO for all values down to the lowest tested, 20 deg/sec.

b. Analytical Approach
Analysis was performed on a Matlab™ model of the airplane with no nonlinearities and in the
presence of rate limiting and stick shaping, separately and in combination. The input was a
computer-generated frequency sweep (chirp) command of uniform amplitude. Sweeps of
different amplitudes were applied to achieve different levels of peak normal load factor during
the sweeps. “Peak” load factor was measured as the largest change from trimmed flight

117
throughout the sweep, and it usually occurred very near the neutral-stability frequency. The
input frequencies varied from 0.1 to 20 rad/sec over 100 seconds. Resulting time-history data
were FFT’d using STI’s FREDA software.

c. Rate Limiting
Analysis of the effects of rate limiting was performed for the two configurations with the
nominal Calspan command shaping gradient. This represents the closest approximation to the
actual in-flight configurations.
Configuration 2D – Identified frequency responses for this configuration, for rate limits of 60
and 20 deg/sec, are shown in Figure 51a. For the higher rate limit (circle symbols), coherence is
approximately unity for all frequencies except those at the very low and high ends of the data
range. With a rate limit of 20 deg/sec, the coherence is similar at frequencies below about 5
rad/sec, but falls off more rapidly as frequency increases. This reflects the fact that the rate
limiter was reached when the frequency of the sweep reached approximately 5 rad/sec. The rate
limiter at 20 deg/sec produces attenuation in magnitude and greater rolloff in phase at the higher
frequencies compared to 60 deg/sec.
20 20
6 0 de g /se c 2 0 d e g /se c 60 d e g /se c 2 0 d e g /s ec
0 0
M a g n itu d e M a g n itu d e
-2 0 -2 0
(d B ) (d B )
-4 0 -4 0

-6 0 -6 0
0 0

-9 0 -9 0

P h a se -1 80 P h a se -1 8 0
(d e g ) -2 70 (d e g ) -2 7 0

-3 60 -3 6 0

-4 50 -4 5 0
1 1

0 .8 0 .8

0 .6 0 .6
C o h e re n ce C o he re n ce
0 .4 0 .4

0 .2 0 .2

0 0
20 20
Input Input

0 0
Power Power
O utput (d B ) Output
(d B )
-2 0 -2 0

-4 0 -4 0
0 .1 1 10 0 .1 1 10
F re q u e n c y (ra d /se c) Fre q u e n cy (rad /se c)

a ) C o n fig u ra tio n 2 D b ) C o n fig u ra tio n 2 D U

Figure 51. Effect of Rate Limiting on Pitch-Attitude-to-Stick-Force Frequency Responses

of Configurations 2D and 2DU (Sweep Amplitude Achieved ±1g Change in Load Factor)

118
For a linear physical system, a loss of coherence such as that exhibited at high frequencies by the
20-deg/sec case in Figure 51 would indicate some reason to doubt the results of the frequency
response. In this case, however, it is simply an indication of the presence of the rate limiter:
coherence is low not because the input and output are uncorrelated, but because they are not
linearly correlated. In this case, the magnitude and phase of the frequency response certainly
appear to be reasonable and typical of the effects of rate limiting.15
Further evidence of the validity of the data in Figure 51 is the fact that output powers (bottom
plot in the figure) for all configurations are quite similar. In all cases output power falls off with
frequency, but reducing the rate limits does not significantly attenuate the output power. In fact,
there is a peak in power for configuration 2DU with the 20 deg/sec rate limit (triangles in Figure
51b) at the frequency where coherence dips. This supports the assumption that there is sufficient
output at this frequency, but that it is not output from a linear system.
Configuration 2DU – Presence of augmentation in this configuration results in a vastly different
trend for the 20-deg/sec case, although the 60-deg/sec data are almost identical to those for
configuration 2D (Figure 51b). There is attenuation in magnitude and loss of phase and
coherence is lower at all frequencies, with a severe drop around 3 rad/sec. Still, except for the
single drop and the rapid loss as frequency increases, coherence for the 20 deg/sec case is still
generally above about 0.7, reflecting some linear correlation.
Effect of Input Amplitude – By its nature, rate limiting is a function of the amplitude of the input.
We should expect that, if the amplitude of the frequency sweep were reduced, the rate limiter
would be encountered at a higher frequency and the effect on the overall frequency response
would be less. There is, in fact, an input amplitude that will result in a response of the 20-
deg/sec model for configuration 2DU that looks like that for the 60-deg/sec case.
Conversely, increasing the amplitude of the sweep will mean reaching the rate limiter at a lower
frequency. As an example of the effect of increasing input size, configuration 2DU, with a rate
limit of 60 deg/sec, was driven with a sweep that produced peak load factors in excess of ±2g.
The resulting frequency response is plotted in Figure 52. For comparison, the 1g sweep data
from Figure 51b are also shown in Figure 52.
The larger input size has the expected effects: magnitude is higher at low frequencies (more of
the sweep is on the second slope of the Calspan command gradient), but lower at high
frequencies (attenuated by the rate limiter); phase rolls off more rapidly; and coherence is lower
overall. In fact, the response for the 2g sweep, with a 60-deg/sec rate limit, looks very similar to
that for the 1g, 20-deg/sec case in Figure 51b.
Confirmation of Low-Coherence Data – The drops in coherence that accompany rate limiting are
cause for concern about the quality of the data. If we had only the low-rate-limit sweep for
Configuration 2DU in Figure 51b, for example, we would not know whether to trust the results
based simply on the measurement of coherence. The severe loss of coherence at 3 rad/sec
corresponds to magnitude attenuation and phase loss, indicating that this is the frequency at
which the rate limiting is reached. It is not surprising, therefore, that coherence drops: as the
sweep transitions though the limiter, the effective airplane is a mix of the augmented and
unaugmented models, with the limiter thrown in to further muddle the answer, and linear
output/input correlation breaks down.

119
 20
1 g in p u ts 2 g inp u ts
0
M a g nitu d e
-2 0
(d B )
-4 0

-6 0
0

-9 0

-18 0
Phase
(de g ) -27 0

-36 0

-45 0
1

0 .8

C oh e re n ce 0 .6
0 .4

0 .2

0
0.1 1 10
F re q ue n c y (rad /se c )

Figure 52. Effect of Input Amplitude on Frequency Response (Configuration 2DU, 60

deg/sec Rate Limit)

The question posed by the results of both Figure 51 and Figure 52, especially for configuration
2DU, is: how do we know if the identified frequency response is believable? To address this,
the model for configuration 2DU, 20 deg/sec rate limit, was driven by a sweep that produced
peak load factors of ±2g, and then by single sinewaves at specific input frequencies
corresponding to the frequencies of selected low-coherence data points. Each sinewave time
history was FFT’d with FREDA, resulting in a magnitude, phase angle, and coherence value at
the input frequency of the sinewave. The high input size and low rate limit make a relatively
small change in the response, as Figure 53 shows (compare the open triangles with the responses
in both Figure 51b and Figure 52). The loss of coherence is greater, however, as there are now
two distinct regions where coherence drops off precipitously.
Single sinewaves were applied around the two regions where coherence is lost (solid triangles on
Figure 53). Two points around 1-1.5 rad/sec, and three more between 3 and 5 rad/sec, confirm
that the results of the sweep were accurate, that is, the data from the two input types effectively
overlay. This confirms that, as hypothesized, a loss of coherence does not necessarily indicate a
loss of quality data. It also suggests that in-flight verification could be made by a similar
approach, using a sweep to identify the overall effective airplane and single sinewaves to check
questionable regions.

120
 20
S w eep S in g le s in ew a v e s
0
M a g n itu d e
-2 0
(d B )
-4 0

-6 0
0

-9 0

P h a se -1 8 0
(d e g ) -2 7 0

-3 6 0

-4 5 0
1

0 .8

C o h e re n ce 0 .6
0 .4
0 .2

0
0 .1 1 10
F re q u e n cy (ra d /se c)

Figure 53. Single-Sinewave Verification of Frequency Responses (Configuration 2DU, 20-

deg/sec Rate Limit, ±2g Load Factor Change)

d. Stick Command Shaping

The effect of command shaping was investigated using configuration 2DU with a 20 deg/sec rate
limit, and input sizes selected to achieve peak load factors of ±1g from trim. As the response in
Figure 51b (triangle symbols) shows, this configuration exhibits low coherence around 3 rad/sec,
and again above about 10 rad/sec. Any additional complication from the change in input shaping
should be apparent.
Frequency responses of pitch attitude to stick force for the three command gradients of Figure 50
are shown in Figure 54. The response labeled “Calspan” is identical to that in Figure 51b.
Changing the command shaping to purely linear (inverted open triangles in Figure 54) results in
slightly higher magnitude overall. This translates to a slightly greater loss of phase at high
frequencies, since the increased magnitude meant reaching the rate limiter at a slightly lower
frequency. There is a slight improvement in coherence at almost all frequencies, especially
above 10 rad/sec, possibly because of the more linear nature of the command in combination
with the nonlinearity of the rate limiting. In general, however, the frequency response is not
significantly changed from the Calspan shaping case.

121
 20
Lin ear C a lspa n P ow e r
0
M ag nitu de
-20
(dB )
-40

-60
0

-90

P h ase -1 80
(de g) -2 70

-3 60

-4 50
1

0.8

0.6
C oh erence
0.4

0.2

0
0.1 1 10
Frequ ency (rad/se c)

Figure 54. Effect of Command Shaping on Frequency Response (Configuration 2DU, 20

deg/sec Rate Limit, ±1g Load Factor Change)

Trends for the power curve are almost the opposite of those for linear in Figure 54: magnitude is
slightly lower and phase loss is slightly less. As with the linear case, coherence is generally
better at all frequencies, but in the ballpark of both other cases. (The increase in coherence
toward unity for frequencies above 10 rad/sec is not fully understood.)

5. Impact on Bandwidth Parameters

a. Rate Limiting
Pitch attitude Bandwidth parameters (Bandwidth frequency, ω BWθ , and Phase Delay, τ p θ ) were
measured from selected identified frequency responses for configurations 2D and 2DU, Calspan
command shaping, and rate limits of 157, 60, 20, and 10 deg/sec. For most runs, the input
amplitude was adjusted to give peak normal load factors of ±1g from trim. Selected cases were
also run with inputs to give ±2g.
Results for the 1g inputs are plotted in Figure 55 with the relevant PIO boundaries. In all cases,
the only parameters of interest are pitch attitude Bandwidth frequency and Phase Delay.

122
 .
C on f.
2D 2 D U
N o P IO in flig ht
0.3
P IO in flight

0.25 10
P hase 20
P IO
D elay,
τp θ 0.2 10
P IO P os sib le
(s ec)
(if fligh t p ath B an dw id th is lo w ) 20
0.15
1 57 , 6 0
0.1 N o P IO 60
15 7
P IO P o ss ible
0.05 (if pitc h rate A ctu ato r rate lim it
ov ersh o ot (deg /se c)
is ex ce s sive )
0
0 1 2 3
P itc h A ttitude B andwidth, ωB W (rad/s ec)
θ
Figure 55. Effects of Rate Limit on Bandwidth Parameters (±1g Load Factor Change)

Configuration 2D – For the two highest rate limits of 157 and 60 deg/sec, the effects on
magnitude and phase angle for this configuration are sufficiently high in frequency that the
parameters are unchanged. (Remember that this configuration was not evaluated in-flight21 with
a 60 deg/sec limit; this limit was used here as a comparison for configuration 2DU.) As
indicated by Figure 55, no PIO would be predicted. There is a reduction in Bandwidth frequency
and increase in Phase Delay when rate limit is reduced to 20 deg/sec, but still no prediction of
PIO. Finally, at 10 deg/sec rate limit, Bandwidth is unchanged, and Phase Delay is just below
the PIO-susceptible line.
No PIO is predicted for configuration 2D with any rate limit. For the discrete HUD tracking
task, no pilot reported a PIO in flight at any rate limit. All three pilots noted small-amplitude
pitch oscillations about the target with a rate limit of 10 deg/sec, though none considered these
oscillations to be PIO.
Configuration 2DU – With a 1g input sweep, the parameters for configuration 2DU with rate
limits of 157 and 60 deg/sec overlay those for 2D on Figure 55, predicting no PIO susceptibility.
Reduction in rate limit to 20 and 10 deg/sec results in a substantial reduction in Bandwidth and
increase in Phase Delay, suggesting PIO. In flight, this configuration exhibited divergent PIOs
for all rate limits at and below 60 deg/sec. This is clearly not predicted by Figure 55 for the 1g
input sweep.

123
Effect of increasing input size – A review of the data from the HAVE LIMITS flight program
indicates that it was common for the pilots to attain load factors above 2g during the course of
the HUD tracking maneuver. Typical maneuver load factors were more in the range between 0.5
and 2.5g. In order to bracket this input range, a frequency sweep was generated for both
configurations that provided approximately ±2g variations in load factor (-1 to +3g). When this
input is applied to Configuration 2DU, the Bandwidth parameters are very near those for the
lower input size with reduced rate limits (Figure 56), even for the 60-deg/sec rate limit. Thus
there is a severe change in effective vehicle dynamics with load factor for 2DU.
The change for 2D with a 60 deg/sec limit is much less dramatic: Phase Delay increases
somewhat, but even for the lowest rate limit of 10 deg/sec it is still not above the PIO limit in
Figure 56. Bandwidth frequency steadily decreases as the rate limit is reduced. Rate limiting
has a smaller overall effect on Configuration 2D because the only consequence of reaching the
limit is the added phase lag from the limiter, while for 2DU the basic characteristics of the
airplane are modified.
Comparing the ±2g-input data in Figure 56 with the ±1g-input data of Figure 55, it is obvious
that Configuration 2DU should be susceptible to PIO for small inputs when the rate limit is 20
deg/sec or less, and for larger inputs when it is 60 deg/sec. In addition, the large change in
dynamics (as evidenced by the reduction in Bandwidth) would suggest that the response of the
airplane changes drastically when the rate limit is reached. These observations are consistent
with the flight results.
.
C onf.
2D 2D U
N o P IO in flig ht
0 .3
P IO in flig ht
20
0 .2 5 60
P ha se P IO
D e la y,
τp θ 0 .2
P IO P ossible 10 20
(s e c ) (if flight path B andw idth is low ) 60
0 .1 5

157
0 .1 P IO P ossible N o P IO
(if pitch rate 157
overshoot A ctuator rate lim it
0 .0 5
is excessive) (deg/sec)

0
0 1 2 3
P itc h A ttitud e B a nd w idth, ωB W (ra d/se c )
θ
Figure 56. Effects of Rate Limit on Bandwidth Parameters (±2g Load Factor Change)

124
By contrast, Configuration 2D is never predicted to be susceptible to PIO, and the change in both
Bandwidth frequency and Phase Delay is gradual. This, too, is consistent with the flight results.

b. Command Shaping
The data of Figure 54 suggest that nonlinear command shaping will not adversely impact the
measurement of Bandwidth parameters, as it has with the Time-Domain Neal-Smith criterion.83
The effect on prediction of PIO susceptibility would also not change appreciably, as the PIO
parameters in Table 16 indicate.

Table 16. Bandwidth Parameters for Configuration 2DU with Different Command
Shaping Curves
Command ω BWθ τ pθ Predict
Shaping (rad/sec) (sec) PIO?
Linear 2.1 0.23 Yes
Calspan 1.0 0.21 Yes
Power 0.7 0.20 Yes

6. Application of Bandwidth Criteria

This analysis of the effects of nonlinearities on Bandwidth parameters has demonstrated that:
1) It is possible to obtain good frequency-response data even with the presence of extreme
surface rate limiting or nonlinear command shaping.
2) A drop in coherence for a small range of frequencies indicates only that the input and output
are not linearly correlated, not whether the resulting frequency response data are valid.
3) If the drop in coherence is accompanied by high input and output power, it is likely that the
frequency-response data are still usable.
4) Single-sinewave forcing functions may be used to verify the responses at frequencies where
coherence is low.
5) For in-flight application, and especially whenever substantial nonlinearity is anticipated,
several frequency sweeps should be applied. Each sweep should have roughly constant input
amplitude, starting with small inputs. Input amplitude should then be increased on
subsequent sweeps, to cover the load factor range expected for the airplane.
6) The combined effect of rate limiting and stick shaping, at least for the range of both
parameters investigated here, is not significant on the frequency response, or on the
prediction of susceptibility to PIO.
7) Bandwidth-based PIO criteria are effective at explaining flight results from the HAVE
LIMITS research program when the effects of rate limiting are accounted for in the
measurement of the required parameters.

125
 VII. PREVENTION OF PIO BY DESIGN (CONTINUED): LATERAL
CRITERIA FOR CATEGORY I AND II PIOS

A. BACKGROUND
Despite the fact that roughly half of all documented PIOs are lateral,15 development of criteria
for the prediction of lateral-directional PIOs lags far behind that for longitudinal. This is due in
part to two major factors: 1) the number of possible contributors to PIOs is smaller laterally than
longitudinally; and 2) there is a lack of a quantitative research data base where PIOs were
investigated and where PIO tendency ratings were obtained.
The lateral data base for Category I (linear) PIOs consists of two reports from Calspan.88,9 In
both experiments the only pilot ratings obtained were HQRs, so evidence of PIOs or their
tendencies must be gleaned from documented pilot comments.
For Category II PIOs, there have been several recent studies of the effects of rate limiting,
focusing on a search for possible solutions.90,91,49 Detailed analysis of the data generated by these
studies, in terms of PIO prediction criteria, requires extensive computation from time-history
data along with models for Matlab/Simulink. Given the primary interest in this research program
on longitudinal PIOs, such an effort was considered beyond the scope of the program and was
not performed. There are no specific recommendations for criteria for the prediction of lateral
Category II PIOs, except to use the Category I Bandwidth/Phase Delay criteria described below
with the test techniques described for longitudinal Category II applications in the previous
section of this report.
The only lateral PIO criteria are Bandwidth/Phase Delay45 and Smith-Geddes criteria.68 The
discussions that follow are shorter than those for the longitudinal criteria as a result. The flight
research data base is applied as a part of the descriptions of the criteria. Much of what follows is
taken from reference 45.

B. EFFECTS OF COMMAND/RESPONSE SENSITIVITY

It was shown in Section V that pitch command sensitivity had an influence on the occurrence of
PIOs. A large portion of the pitch data base was generated using pilot-selected sensitivities,
either by each pilot before evaluating or by one pilot before evaluations by the remaining pilots.
In some experiments, a fixed value of pitch command sensitivity was used.
Roll sensitivity is the initial acceleration per pound, written as either the lateral sensitivity
derivative LFAS, or simply as p /FAS , both in units of deg/sec2/lb. The two roll experiments
reviewed here included variations in sensitivity in their test matrices. The advantage of varying
roll sensitivity is that it allows us to define the best values; the disadvantages are that it is
possible that the “best” value was not flown, and now we have an additional variable to account
for.
In both experiments, lateral control/response variations were defined by control power, defined
as the achievable steady-state roll rate for lateral force inputs, pss/Fas, in deg/sec/lb. In the
remainder of this section, the configurations are defined by control power; for consistency with
the pitch discussion in Section V, here we will use the initial acceleration sensitivity as our
correlating parameter.

126
Figure 57 shows a plot of roll command sensitivity versus added time delay for a configuration
from the LATHOS program with roll damping TR = 0.3 sec. Unfortunately, because PIO
Tendency ratings were not gathered in this experiment, we must rely on pilot comments and
Handling Qualities Ratings to determine where PIO occurred. The ratings in Figure 57 are for
the three pilots, separated by slashes; commas separate repeat evaluations by the same pilot. In
several cases, at least one of the pilots reported a tendency for roll ratchet, rather than PIO, and
this is noted by an “R” beside that symbol.
The data in Figure 57 clearly show a trend for reduced PIO tendency as roll sensitivity is
reduced. Generally, pilot ratings degrade to Level 3 at about the same value of added time delay
for all three values of sensitivity. There is a clear need for some measure of the effect of roll
sensitivity on PIO susceptibility, either through an adjustment on existing criteria or as a new
criterion.

C. BANDWIDTH/PHASE DELAY CRITERIA FOR CATEGORY I PIOS

In the development of longitudinal requirements based on Bandwidth, it was found that pitch
attitude Phase Delay was a prime measure of the potential for PIOs. Since the two roll
experiments focused primarily on roll damping, control sensitivity, and delays and lags, it seems
logical that this would be the case here as well. Roll Bandwidth does not appear to be a strong
factor in the ratings from these experiments,45 so only control sensitivity and Phase Delay need
to be considered.

60
R R R

3,2/3/5,3 5/-/- 6,7/6/- -/9/-

Open -- No PIO reported
50
Solid -- PIO reported
R indicates ratchet reported
HQRs are Pilots B/G/P
R (repeats separated by comma)
40
2,4/3/2,3 3,4,4/-/6 5/-/6,4.5 9/8/-
Roll command
sensitivity,
LFAS
2 30
(deg/sec /lb)

20 3,2,4/5,3/3 2/2/- 6,5/-/- -/7,7/- 8,9/-/8

PIO
No PIO
10

0
0 0.05 0.1 0.15 0.2 0.25
Added time delay (sec)

Figure 57. Effects of Roll Command Sensitivity on PIO Tendency as a Function of Added
Time Delay (Configuration with Roll Damping TR = 0.3 sec)

127
The lateral experimental data must be carefully screened before using any of it. Because control
sensitivity (described in both references in terms of roll rate per lb, deg/sec/lb) was a part of the
experimental matrices, it is sometime difficult to determine if a poor HQR is due to sensitivity,
roll damping, or time delay. In addition, the number of different values of control sensitivity
tested was small, so there is no assurance that the "best" value was ever tested with a given set of
dynamics. This complication does not invalidate the data, but it does make their interpretation
much more difficult. Only a subset of the data from either experiment will be used here.
Figure 58 is a plot of HQR versus Phase Delay for the relevant data. There is considerable
scatter in the data, especially for Category A (Figure 58a). Because PIO ratings were not
obtained, it is difficult to determine if PIO tendency was a strong factor in the pilots' ratings.
Some measure of the occurrence of PIOs can be made by reviewing published pilot comments
and searching for specific mention of PIOs. In both experiments the pilots were asked
specifically to comment on any PIO tendency. Unfortunately, positive comments here usually
reflect the existence of a PIO, as opposed to a tendency for PIO. It is always possible that some
of the configurations for which there are no comments on PIO tendency would still have
received poor PIO ratings if such ratings had been taken. Still, it is reasonable to expect that any
PIO requirement will successfully capture most of the cases for which PIOs were reported. The
majority of cases where PIO was reported (flagged symbols) are to the right in both plots. The
following observations can be made from Figure 58:
• There is a relatively strong correlation between HQR and Phase Delay;
• Control sensitivity appears to be a factor in the pilot ratings: the lowest values (indicated
in both plots by solid symbols) may have been too low for small values of Phase Delay
(the ratings are generally worse), but preferred for higher values of Phase Delay (where
the ratings are generally better);
• All but one of the PIO cases are above the Level 1 limit of 0.14 sec, and:
Between Phase Delays of about 0.14 and 0.17 sec, the only PIO cases were those
with the higher roll gain, and in several of these cases the pilots specifically
mentioned the excessive control sensitivity as a factor in their ratings. Hence
these cases may be PIO-prone more because of control sensitivity than Phase
Delay;
Above a Phase Delay of about 0.17 sec, only three of the 12 cases did not involve
PIOs. Two of the three non-PIO cases had low sensitivity, again suggesting that
decreased responsiveness may have helped alleviate the potential for PIO. But of
the remaining nine PIO cases, both low and high sensitivity cases are involved,
further suggesting that at this point the potential for PIO is high no matter what
the control gain.
On the basis of these observations it was concluded that: 1) PIO is unlikely if Phase Delay is
less than 0.17 sec, as long as sensitivity is separately optimized; and 2) PIO is likely always if
Phase Delay is above about 0.17 sec.

128
Figure 58. Comparison of Handling Qualities Ratings from Two Roll Experiments88,9 with
Roll Attitude Phase Delay (Reproduced from Ref. 45)

129
D. SMITH-GEDDES CRITERIA FOR CATEGORY I PIOS
It has been stated by Smith54 that the Smith-Geddes criteria apply equally in roll and pitch. The
only exception is that there is no roll counterpart to the normal acceleration phase parameter.
Hence the only parameter with which we must be concerned is the roll attitude phase angle at
the criterion frequency, ∠ϕ/Fas(jωc). Figure 59 shows the relevant lateral pilot rating data plotted
against this phase angle. As with Bandwidth, there is considerable scatter in the data. In fact
there is only a very slight trend seen in these plots. Based on this figure, it is clear that
• There is a large variation in ratings at any one phase angle, especially for the Category A
ratings of Figure 59a, although some of this variation may be attributed to the effects of
other experimental factors, such as control sensitivity;
• A limit on phase angle at the criterion frequency of -180 degrees results in many good
ratings on the bad side.
In Figure 59 those cases for which the pilots explicitly reported a tendency for, or an actual, PIO
are indicated by flagged symbols. (Remember that some of these PIOs may be due more to
control sensitivity, which is not addressed by the Smith-Geddes criteria.) Most comments about
PIO occurred during the landing tests. In Figure 59b, almost all PIO cases are to the right of the
-180 degree line. Unfortunately, so are about half of the non-PIO cases. Thus these criteria are
not very discerning of the potential for PIOs.

130
Figure 59. Comparison of Handling Qualities Ratings from Two Roll Experiments88,9 with
Roll Attitude Phase Angle from Smith-Geddes Criteria (Reproduced from Ref. 45)

131
 VIII. GROUND SIMULATION AND FLIGHT TESTING FOR PIO

A. INTRODUCTION
It should be obvious that the focus of this report is on the prevention of PIO by design, with a
concerted effort to apply analytical criteria throughout the full-scale development of any
airplane. But there will always be a need to perform pilot-in-the-loop testing of the airplane, first
in ground-based simulation, and then in flight, to verify that the design is resistant to PIO.
This section investigates some of the issues involved with testing for PIO, focusing on the
expectations in testing and on the pilot’s role in the process.

B. PIOS IN THE DEVELOPMENT PROCESS

There is a clear need for a fundamental shift in philosophy about the occurrence of PIOs. Rather
than react with surprise if a PIO is encountered late in full-scale development, we should be
prepared for the eventuality and plan for it from the initial design stages. This includes
guidelines used by the contractor in the initial system definition, through preliminary design, and
into flight testing. Flying qualities specifications contain criteria that will help reduce the
propensity for PIO, even if they would not prevent it altogether.
Flight test organizations, such as the Air Force Flight Test Center at Edwards AFB and the Naval
Air Warfare Center at NAS Patuxent River, have come to expect that every new prototype will
exhibit some tendency to PIO. Conversely, success-oriented development programs are
inherently structured to produce PIO-free, mission-effective airplanes starting with the first test
article. Clearly, there is a disconnect.

1. Flight Test and PIOs

Historically – especially at the Air Force Flight Test Center at Edwards AFB – flight testing of a
prototype has consisted primarily of “airworthiness” evaluations. In-flight demonstration of
flying qualities has involved open-loop maneuvers for data gathering: pulses and steps to check
short period and Dutch roll, speed changes for speed stability, trim changes for trimmability, etc.
The classical image of the test pilot in this scenario has been as a servo generating information
for the engineers. Closed-loop (pilot-in-the-loop) testing was not required by the RFP, provided
for in the contract, or anticipated by either the contractor or the procuring activity. As a result,
any pilot-in-the-loop testing that occurred was either accidental or as a consequence of testing for
mission suitability, as opposed to pilot acceptability.
More recently, this philosophy has changed at the flight test facility, and it is starting to change
from the ground up. Tracking tasks that are completely unrelated to mission suitability, but
valuable for handling qualities and PIO assessment, have been adopted by flight test
organizations. A prime example is the Handling Qualities During Tracking (HQDT) task
favored at Edwards AFB. Unfortunately, for some recent aircraft programs, a need to perform
tracking tasks such as HQDT has come as a surprise to the developer, discovered only after the
first prototype has been delivered.

2. A Change in Philosophy
Fortunately, the value of extensive pilot-in-the-loop testing has been recognized by all facets of
the aircraft development process, and there is a shift in philosophy. In its review of the complex

132
problem of adverse aircraft-pilot coupling (APC), the National Research Council’s Committee
on the Effects of Aircraft-Pilot Coupling on Aircraft Safety recommended that extensive testing
for PIOs be applied from beginning to end.48 As a result of the YF-22’s mishap, Lockheed
Martin instituted a system for piloted evaluations of PIO potential in fixed-base simulation, and
continuing into flight test.92 Boeing Commercial Airplane Group has begun to apply such
“fighter-type” tasks as close formation and simulated aerial refueling as a consequence of several
PIOs experienced early in the development of the 777 transport. Even the FAA has proposed
certain tasks for PIO assessment,77 and work is continuing on an explicit set of tasks.

3. Selection of Pilot-in-the-Loop Tasks

There is, understandably, great reluctance to initiating a full-scale flight test program to look for
PIO. First, such a program is expensive and time-consuming. Second, it presumes that PIO is a
possibility. Third, it is extremely risky, especially for a one-of-a-kind airplane. And fourth,
there is an assumption that the specific maneuvers and tasks necessary to expose PIO potential
are known.
The first two issues must be dealt with from the beginning. For the third issue, risks may be
minimized through common-sense buildups to the most extreme conditions for PIO. The fourth
issue, however, requires assembly of a set of potential maneuvers for all airplane types, flight
conditions, and test conditions.
There are indications that some of the most effective maneuvers in flight may not consistently
expose PIO susceptibility in a simulator. Likewise, there are possible contrived tasks that can
only be performed in a simulator that may be needed to expose PIO potential before flight test.
Many maneuvers developed for handling-qualities assessment will also serve to look for PIO,29
but some are better than others, especially on a ground-based simulator.

C. PILOT EVALUATION OF PIOS

If it is assumed that PIOs will be encountered at some time during the developmental flight test
of any new airplane, there must be a structure for assessing the severity of the event and
determining the risks involved in continuing with the test program once the event has been
uncovered.

1. Pilot Evaluation: The PIO Tendency Rating Scale

In flight research and full-scale development testing, there is an obvious need to find a way to
quantify the pilot’s opinions about the susceptibility to PIO for a particular airplane/flight
condition/task combination. This is especially critical when PIO does not occur; once a PIO has
happened, the interest shifts from susceptibility to severity, type, etc.
Since the 1950’s at least, pilot opinion about susceptibility to PIO has been obtained through
some form of numeric PIO rating. PIO tendencies have generally been measured on a six-point
scale, sometimes referred to as a “classification” but more generally as simply a pilot rating. As
an example of an early scale, Figure 60 was used for pilot assessment of PIOs in a flight
evaluation of the low-altitude high-speed characteristics of the F-4 airplane. 93 By the “Adjective
Rating” column in this scale, a numerical rating of 2 indicated a possibility for PIO. The
comments change dramatically from 3 to 4: “Undesirable motion” for a 3, with no mention of
PIO or any oscillations specifically, to “Oscillations tend to diverge” for a 4.

133
PIO rating scales have evolved since the 1960’s, and this evolution has not been entirely for the
better. There is compelling evidence of a “PIO rating inflation” in the past twenty years or so,
and much of this inflation can be traced to the evolution of the rating scale itself. Early versions
of the scale defined the ratings with adjectival descriptions (Figure 61). This scale was used, for
example, in the assignment of PIO ratings in the ubiquitous Neal-Smith experiment51 reported in
1970 and the LAHOS experiment of 1978.43
The decision-tree scale in Figure 10 was first introduced by Calspan Corporation in 1980. It
provides a structure similar to the Cooper-Harper28 scale (Figure 9) but retains the six-point
division of the older scales such as Figure 61. For some flight research projects36 conducted
since 1980, both the adjectival scale of Figure 61 and the decision-tree scale of Figure 10 have
been given to evaluation pilots, usually because the pilots were more familiar with the first scale.
The decision-tree scale in Figure 10 has become the standard for flight research and testing, and
is introduced to students at both the Air Force and Navy Test Pilot Schools. It has proven its
value over the years, yet there are several significant shortcomings, especially as the scale is
applied in developmental flight test as opposed to flight research. These shortcomings may be
summarized as follows:

F-4A Airplane
BuNo 146817

PILOT OPINION RATING SYSTEM

FOR
PILOT INDUCED OSCILLATION TENDENCY

NUMERICAL ADJECTIVE
RATING__ RATING_ COMMENTS
1 None. No tendency for pilot to induce
undesirable oscillation - no
tendency to get out of phase with,
or to lag behind aircraft motion.
2 None to Undesirable motion may be induced
possible but can be damped by pilot effort.
3 Possible Undesirable motion easily induced
but can be damped by pilot effort.
4 Possible to Oscillations tend to diverge.
probable Pilot may be required to fix stick.
5 Probable Oscillations tend to diverge.
Pilot must fix stick to stop motion.
6 Highly Disturbance or normal pilot control
probable may cause divergent oscillation.
Pilot must fix stick to stop motion.

Figure 60. An Early PIO Tendency Rating Scale from Naval Air Test Center, 196393

134
 N U M ER IC AL
D ES C R IPTIO N R AT IN G

N O TE N D EN C Y FO R PILO T TO IN D U C E U N D ESIR AB LE 1
M O T IO N S

U N D ESIR AB LE M O TIO N S T E N D TO O C C U R W H E N 2
P ILO T IN ITIAT ES A BR U P T M AN EU VE R S O R ATT EM PTS
T IG H T C O N TR O L. T H ESE M O T IO N S C AN B E PR EVEN T ED
O R ELIM IN AT ED BY PILO T T EC H N IQ U E.

U N D ESIR AB LE M O TIO N S E ASILY IN D U C ED W H EN P ILO T 3

IN IT IAT ES ABR U P T M AN EU V ER S O R ATT E M PT S T IG H T
C O N T R O L. TH ESE M O TIO N S C A N B E P R EVEN TE D O R
E LIM IN ATED B U T O N LY AT SA C R IFIC E TO TA SK PER -
FO R M A N C E O R TH R O U G H C O N S ID E R A B LE PILO T
ATTEN TIO N A N D E FFO R T.

O SC ILLAT IO N S T EN D TO D EVELO P W H E N PIL O T IN IT IAT ES 4

A BR U PT M A N EU V ER S O R AT T EM PT S TIG H T C O N T R O L.
P ILO T M U ST R ED U C E G A IN O R A B A N D O N TA SK TO
R EC O V ER .

D IVE R G EN T O SC ILLAT IO N S T EN D TO D E VELO P W H EN 5

P ILO T IN ITIAT ES A BR U P T M AN EU VE R S O R ATT EM PTS
T IG H T C O N TR O L. PILO T M U ST O PEN LO O P B Y R ELEA SIN G
O R FR EEZIN G TH E S TIC K .

D IST U R BAN C E O R N O R M AL PILO T C O N T R O L M AY 6

C AU SE D IVER G EN T O SC ILLAT IO N . PILO T M U S T O PEN
C O N TR O L LO O P B Y R ELEA S IN G O R FR EEZIN G TH E
S TIC K .

Figure 61. Scale for Classification of PIO Tendency, from Cooper and Harper28 (Phrases
in Bold Italics are Missing from Decision-Tree Scale of Figure 10)

• Only PIORs of 4, 5, and 6 specifically refer to PIO. PIORs of 2 and 3 address

“undesirable motions” and are not directly relevant to PIO. Any such motions may or
may not be a precursor to PIO, and if they are not, these ratings say nothing about
PIO tendency. Some pilots, however, may interpret the reference to “undesirable
motions” as a specific indicator of PIO tendency. Certainly the Navy scale in Figure
60 makes the connection between “undesirable motions” and PIO, since a numerical
rating of 2 indicates a possibility for PIO with undesirable motion that “may be
induced but can be damped by pilot effort.” This is much more detailed than the
block in Figure 10 that asks, “do undesirable motions tend to occur?” with no other
amplification. One evaluation pilot in the simulation documented in Appendix A
commented on his interpretation: “The terms undesirable motion and poor response
get a little bit clouded, a little gray. Undesirable motion is something that's occurring
without input. And if I'm making an input and I just get poor response, it's an
undesirable response…. It first of all has to center on … the definition of the
undesirable motion. Is it a motion that occurs without input or is it part of the
response to the pilot's input?” The FAA makes an interesting clarification for
“undesirable motions” in published criteria for rating aircraft-pilot coupling (A-PC,
Section II), as shown in Figure 62. For a rating of “SAT” the motions are specified to
be overshoots, and for “ADQ” they are “unpredictability or over control.”

135
 FAA HQ
RATING A-PC CHARACTERISTICS DESCRIPTION

SAT NO TENDENCY FOR PILOT TO INDUCE UNDESIRABLE MOTION.

UNDESIRABLE MOTIONS (OVERSHOOTS) TEND TO OCCUR WHEN PILOT

INITIATES ABRUPT MANEUVERS OR ATTEMPTS TIGHT CONTROL. THESE
MOTIONS CAN BE PREVENTED OR ELIMINATED BY PILOT TECHNIQUE. (NO
MORE THAN MINIMAL PILOT COMPENSATION REQUIRED)

ADQ UNDESIRABLE MOTIONS (UNPREDICTABILITY OR OVER CONTROL) EASILY

INDUCED WHEN PILOT INITIATES ABRUPT MANEUVERS OR ATTEMPTS
TIGHT CONTROL.

THESE MOTIONS CAN BE PREVENTED OR ELIMINATED BUT ONLY AT

SACRIFICE TO TASK PERFORMANCE OR THROUGH CONSIDERABLE PILOT
ATTENTION AND EFFORT. (NO MORE THAN EXTENSIVE PILOT
COMPENSATION REQUIRED)

CON OSCILLATIONS TEND TO DEVELOP WHEN PILOT INITIATES ABRUPT

MANEUVERS OR ATTEMPTS TIGHT CONTROL. ADEQUATE PERFORMANCE
IS NOT ATTAINABLE AND PILOT MUST REDUCE GAIN TO RECOVER. (PILOT
CAN RECOVER BY MERELY REDUCING GAIN)

UNSAT DIVERGENT OSCILLATIONS TEND TO DEVELOP WHEN PILOT INITIATES

ABRUPT MANEUVERS OR ATTEMPTS TIGHT CONTROL. PILOT MUST OPEN
LOOP BY RELEASING OR FREEZING THE CONTROLLER.

DISTURBANCE OR NORMAL PILOT CONTROLMAY CAUSE DIVERGENT

OSCILLATION. PILOT MUST OPEN CONTROL LOOP BY RELEASING OR
FREEZING THE CONTROLLER.

SAT = Satisfactory
ADQ = Adequate
CON = Controllable
UNSAT Unsatisfactory or Failed

Figure 62. FAA A-PC Rating Criteria77

• An ambiguity sometimes arises with the PIOR of 4 (reflecting non-divergent

oscillations). Mild pitch bobble is usually manifested as a high-frequency, low-
amplitude oscillation; some pilots, following Figure 10, will ultimately assign a PIOR
of 4 for the resulting motions. Other pilots, however, consider these bobbles to be
more of a nuisance motion than a true oscillation (or that the Figure 10 scale has to do
with PIOs and the pitch bobble does not act like their concept of a PIO), and hence a
PIOR of 3 will be given. These pilots are usually more familiar with the old Figure
61 scale, where a PIOR of 4 also implies that "[the] pilot must reduce gain or abandon
task to recover" – not at all consistent with an annoying pitch bobble. It is possible
also to encounter a rather severe PIO, that shows no signs of being divergent, and
this, too, can (correctly) receive a PIOR of 4. Yet mild pitch bobble and a severe PIO
are hardly comparable. Interpretation of a PIOR of 4 requires additional information,

136
 such as a Cooper-Harper Handling Qualities Rating (Figure 9), and even then there
may be other factors affecting the rating. On more than one occasion, pilots have
objected to assigning a PIOR of 4 to an extremely horrifying – but non-divergent –
PIO. Some pilots insist on assigning a 5 just to convey the severity, even though the
wording for a PIOR of 5 is inconsistent with their experiences.
• PIORs of 5 and 6 are applicable to “divergent” PIOs, and it is common for the pilot to
have difficulty judging divergence, especially if the pilot is able to arrest the
oscillation quickly. In flight research experiments, it has not been uncommon for
evaluation pilots to spend an inordinate amount of time trying to decide between
PIORs of 4 and 5 – when the most important information has been conveyed: there
was an oscillation.
• The scale has generally been recognized as a PIO tendency scale, yet the word
tendency does not occur anywhere. This is a crucial distinction: pilots who are not
properly briefed on the scale will assign a PIOR of 4, 5, or 6 only if a fully-developed
PIO is experienced. Other pilots will assign a PIOR of 4 if they feel there was any
tendency to PIO, even if a real PIO was not experienced. This degree of
subjectiveness and interpretation is a serious drawback to the scale. The Federal
Aviation Administration is attempting to correct this deficiency by creating a
modified scale that includes the word tendency,77 as shown in Figure 62 (there is a
danger here, too, of course: flawed or not, there is an extensive knowledge base for
the Figure 10 scale, and test pilots have been trained in its use.) The FAA’s A-PC
scale has words similar to the adjectival PIO Rating scale of Figure 61.
Fundamentally, the current Figure 10 scale attempts to account for both the tendency
to PIO and the actual event of a PIO.
• It is reasonable to question whether “tendency to PIO” should occur on a rating scale
at all. A “tendency” is not a PIO, so there is no compelling argument to fix the
airplane. But it indicates the potential for PIO, so perhaps the airplane should be
fixed after all. In real applications, if an experienced test pilot comments on a
“tendency” to PIO, this should be taken as a challenge to that pilot: redo whatever
you were just doing, force yourself to attempt tighter performance standards and
achieve better performance, and see if the “tendency” develops into a PIO.
• There are no specific directions to the pilot to indicate required actions or demands.
One option to fixing these shortcomings is to change the decision-tree PIOR scale. A
clarifying statement could be added to the PIOR scale to tell the evaluation pilot that
mild bobble should not be considered a PIO. A more appropriate change would be to
add some alternative path that allows the pilot to assign a different number to bobble.
Perhaps a PIO severity rating could be obtained along with the PIO Tendency rating.
Certainly the statements in the old scale that describe pilot actions (phrases shown in
bold italics in Figure 61) should somehow be incorporated into the newer one, similar
to the "Demands on the Pilot" column on the Cooper-Harper scale (Figure 9). Older
versions of the PIOR scale had some information that was not incorporated into the
Figure 10 version.

137
2. PIO Rating Inflation
By “PIO Rating Inflation” we refer to the propensity for PIO tendency ratings to worsen when
compared with their HQR counterparts. This rating inflation will be obvious when we review
the ratings trends from selected experiments spanning 1970 through 1998. It will be suggested
that the inflation has arisen as a result of the greater use of the decision-tree PIO Tendency rating
scale of Figure 10 in place of scales such as that in Figure 61. Graphical evidence is provided in
Figure 63.
For many flight research programs conducted in the 1960’s and 1970’s, there is a strong
correlation between HQR and PIOR; as an example, for the Neal-Smith experiment51 the PIORs
are almost always one-half the corresponding HQRs. Consider the plot in Figure 63a.
Figure 63a is a crossplot of the HQRs and PIORs assigned by the two evaluation pilots in the
Neal-Smith experiment. Indeed, there is a strong trend between the ratings. This strong trend
helps to explain why so many criteria for both handling qualities and PIO may be developed
from the data generated by Neal and Smith: any criterion that explains one rating will probably
explain the other as well. Besides an apparent upper limit on PIOR as a function of HQR (i.e.,
there are no PIORs of 4 with an HQR better than 7), there is a lower limit as well (no
configuration with an HQR of 7 or worse received a PIOR better than 3). Notice, too, a large
number of half-ratings, for both PIOR and HQR. The two pilots clearly felt that assigning such
half-ratings, using the adjectival scale of Figure 61, was perfectly acceptable and best described
their impressions.
Interestingly, the evaluation pilots for the Neal-Smith experiment did not feel compelled to
assign PIORs of 4 or worse to PIO tendency. There is even evidence among pilot comments in
the published report51 of configurations that exhibited PIO tendency, but that were assigned
PIORs of 3.5, 3, and occasionally even 2.5 and 2. In 11 individual cases, the PIOR was 3.5 or
better and the HQR was 6 or better, as listed in Table 17. Some of these combinations of ratings
seem especially incongruous with the comments, such as Configuration 2J, where “some PIO’s”
occurred but the PIOR was only a 2. Such ratings and comments reflect what must certainly
have been, in most cases, a tendency for only a “mild” PIO (or pitch bobble) at worst. Based on
the examples listed in Table 17, it may not be assumed that PIO tendencies correspond only to
PIORs of 4 or worse – or to HQRs of 7 or worse – for the Neal-Smith experiment.
Even in the landing approach study LAHOS43 the PIORs are generally half the corresponding
HQRs – or less in some cases (Figure 63b): the two pilots in LAHOS assigned a few HQRs of 7
or worse with PIORs of 2 or better. The worst PIORs in the program were 4s, but PIORs of
worse than 3 were assigned only when HQRs were 7 or worse.
In the Test Pilot School’s HAVE PIO experiment,36 it still took a Level 3 airplane (HQR of 7 or
worse) to elicit a PIOR of 4 or worse (documentation for this experiment shows both the old
Figure 61 scale and the newer Figure 10 scale, suggesting both were used for assigning PIORs).
The author of Reference 36 considered a PIOR of 2 or worse to indicate a PIO tendency. This is
far lower than most researchers (and most research pilots) would consider a sign of PIO; first, to
assign a PIOR of 3 or better on the Figure 10 scale, the pilot must answer “no” to the question
about “causes oscillations,” and then the issue becomes one of interpreting “undesirable
motions.” Many pilots consider such motion to not be signs of an incipient PIO, yet if we
defined “undesirable motions” to mean “precursors to PIO,” the decision-tree scale of Figure 10
would be more useful, and more effective at classifying both tendencies to PIO, and real PIO.

138
 6 6

5 Pilot M 5 Pilot A
Pilot W Pilot B
4 4
PIOR

PIOR
3 3

2 2

1 1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
HQR HQR

D 1 HDO6P LWK E /$ + 2 6

6 6
1
A 2
5 5
B 3
C 4
4 4
PIOR

PIOR

3 3

2 2

1 1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
HQR HQR

F + $ 9( 3,2 IOLJK W G  + $ 9( 3,2 $ LU )RUFH 6LP 

6 6
1 A B
2 C D
5 5
3 E F
5 G H
4 4 I
PIOR

PIOR

3 3

2 2

1 1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
HQR HQR

H + $ 9( 3,2 1 $ 6$ 6LP  I + $ , 3,2 6LP X ODWLRQ

Figure 63. Comparison of PIO Ratings and Handling Qualities Ratings from Selected
Research Experiments

139
 Table 17. Configurations from Neal-Smith Experiment51 with HQRs 6 or Better, and
PIORs 3.5 or Better, and Pilot Comments of PIO Tendencies
Config. Pilot PIOR HQR Comments Relevant to PIO
1C M 2.5 5 The tracking tasks show up some tendency to PIO
1E M 3.5 6 PIO tendency in tracking or attitude control
A PIO tendency…. Not full blown PIO, but lots of oscillations and weird
2B M 2.5 6
motions
[Tracking random disturbances] showed up a PIO tendency that was not
2B M 3 6
apparent before
2H M 2.5 5 Tracking tasks showed up more PIO tendencies
2H M 2.5 6 PIO tendency on target
2J M 2 6 Tracking task often caused some PIO’s
5A W 3 6 Small PIO’s occurred…. PIO tendencies IFR
6D M 2.5 5.5 Has a tendency to PIO – large amplitude…. Could get into a PIO
PIO tendencies if your gain is high…. Close to a PIO on the random
7D M 3 5.5
noise tracking task
14 M 3 6 [Tracking random disturbances] … really brings out PIO tendencies

Researchers at the Air Force Research Lab, Wright-Patterson AFB, OH, and at NASA Ames
Research Center, attempted to replicate the HAVE PIO experiment on ground-based simulators.
Their ratings comparisons show evidence of the “PIO rating inflation.”
First, the Air Force conducted their research on two facilities, the fixed-based Manned
Simulator-1 (MS-1) and the moving-base Large Amplitude Multimode Aerospace Research
Simulator (LAMARS), in 1995-1996.94 For these experiments only the decision-tree PIOR scale
of Figure 10 was used. Ratings results from the MS-1 phase are shown in Figure 63d. Besides
some obvious overall differences from the flight ratings (compare trends in Figure 63c), the most
glaring is the shift in PIOR-HQR correspondence. PIORs of 4 and even 5 have been assigned to
airplanes that received HQRs of 5 and 6. Remember that a PIOR of 4 implies a definite PIO, and
a 5 is a divergent PIO (Figure 10), so it must be inferred that the pilots who assigned these
ratings considered the PIOs to be deficiencies that only warrant but do not require improvement
(see left side of HQR scale in Figure 9).
The second set of HAVE PIO replication data comes from NASA Ames Research Center’s
Vertical Motion Simulator (VMS). The pilots in this experiment evaluated the configurations
with full VMS motion, no motion at all, and a reduced motion range corresponding to a hexapod
facility. The ratings in Figure 63e, for the full-motion data, shows a further inflation, as
numerous cases of PIOR 4 or worse correspond to HQRs of 6 or better – including several PIOR
= 4 with HQR = 4, and PIOR = 6 with HQR = 6! An HQR of 4 implies only minor deficiencies,
and desired performance is possible (Figure 9), while a PIOR of 4 implies PIO occurred. Even
an HQR of 6 indicated an airplane that has deficiencies that warrant, but do not require,
improvement, but a PIOR of 6 indicates divergent PIO just from attempting to control the
airplane. These apparently incongruous combinations may be possible if the pilot is able to
somehow find a way to compensate for the PIO once it has been exposed – but they are directly
counter to past experience (Figure 63a and b).

140
The authors of this report are not immune from PIO rating inflation, as reflected in Figure 63f
(data from Appendix A). There was a conscious effort to avoid some of the shortcomings with
the current scale discussed above. For example, a hybrid PIO rating scale (described below) was
used, and every pilot was instructed to differentiate bobble from PIO, assigning a PIOR of 4 or
worse only if the oscillations required reducing gain or abandoning the task entirely. Despite our
best efforts, there are still some PIORs of 4 and 5 with HQRs of 5 and 6.
What the data in Figure 63 suggest is that the test pilot’s interpretation of what constitutes “PIO”
has changed since the 1970’s. Certainly the pilot’s judgment about what constitutes PIO that
needs fixing (interpreted as HQR of 7 or worse) has changed, since PIO (PIOR of 4) can occur
on configurations that are almost Level 1 (HQR of 4).
The foregoing is not in any way intended to point blame at any of the researchers or pilots
involved in this rating inflation. Rather, it is meant to show that we need to look carefully at
what “PIO” really is, especially if it is reported in a recent experiment, and we need a new way
of conveying the existence of PIOs, and their severity, from the test pilot to the engineer.

D. THE NEED FOR A NEW METHOD FOR EVALUATING PIO

The “PIO rating inflation” documented above is not surprise to some. Evidence of it has been
seen by other researchers, and especially the shortcomings of the decision-tree scale of Figure 10
when compared to the adjectival scale of Figure 61. Some researchers21 have found it useful to
attach the words from the older versions of the scale onto the new version (Figure 64). This is
certainly a step in the right direction, and, in addition to a thorough briefing about the proper use
of the PIO Tendency scale, should help reduce rating inflation. But it is more a compromise than
a solution.
Besides the shortcomings discussed above, a fundamental problem with the decision-tree scale in
Figure 10 is its basic philosophy: the scale is structured so that it parallels the general format of
the Cooper-Harper HQR scale, suggesting PIOs are simply another aspect of flying qualities for
the pilot to judge. But this is not always true, and this is where a change in our approach to PIO
must be made.
“Flying qualities” are metathetic;95,96 that is, they are quantified by the pilot in relative terms
(“this airplane has better flying qualities than that one”), based on the pilot’s experience.*
(Weight is an example of a quantity that is prothetic: there is a universal measure for weight and
we all know what it is.) Flying qualities, one could argue, are always present as long as the pilot
is performing some degree of closed-loop control of the aircraft. Then the use of rating scales
such as the HQR scale in Figure 9 makes perfect sense.
The conditions for pilot-in-the-loop operations in which PIO is encountered can be divided into
three regions: the operation of the airplane before the event, the PIO, and the operation of the
airplane following the event. Then a discussion of “flying qualities” is appropriate for the pre-
and post-PIO segments, up to and including the onset of the PIO itself. In fact, the dynamics of
the airplane at the point of PIO onset should be viewed from the standpoint of flying qualities.

*
This is the main reason we prefer to use test pilots for PIO evaluation: they have been provided
the opportunity to experience the handling qualities of many diverse aircraft and can use their
experience to provide perspective.

141
 no tendency to induce
1 undesirable motions

undesirable motions can be

no 2 prevented or eliminated by
no pilot technique
Is task
Do undesirable yes performance
motions tend to
compromised? undesirable motions can be
occur?
yes prevented or eliminated through
3
considerable pilot attention and effort

no pilot must reduce gain

4 or abandon task to recover
no
Causes yes
oscillations? Divergent?

yes pilot must open loop by

5 releasing or freezing stick
Pilot initiates abrupt
maneuvers or tight control

no
Causes
divergent yes pilot must open control loop by
6 releasing or freezing stick
oscillations?

Pilot attempts to
enter control loop

Figure 64. Hybrid PIO Tendency Rating Scale21

In reality, of course, it has proven difficult to determine precisely when “PIO onset” ends and the
fully-developed oscillation begins. In the case of the YF-22, Figure 7b, the PIO started shortly
after the gear-up command at t = 35 sec, and the pilot recognized a problem shortly before
cutting power about three seconds later. The PIO ended with impact on the runway.
But the PIOs themselves are not flying qualities,* nor are they always present (one hopes!). PIOs
are sporadic events, and when they occur in flight they are almost always a surprise to the pilot.
Tom Morgenfeld, the YF-22 pilot in the incident that ultimately led to loss of the airplane, is
quoted as saying that “I thought something had broken and I didn’t see any [warning] lights… I
thought, boy, that’s it, something’s drastically wrong here…”2 In fact, “it felt like it broke” is a
common comment from pilots who encounter PIOs when they are not expecting them.
If PIOs are not flying qualities, it does not seem sensible to attempt to use a flying-qualities-
oriented rating scale, such as that in Figure 10, for piloted evaluations. While the decision-tree
scale has value for flight research, and perhaps conveys some useful information in
developmental and operational flight, a more meaningful method for assessment must be used.

*
PIO is one manifestation of poor flying qualities, and, in fact, a very common one, especially
when the poor flying qualities result from excessive filtering, lags, time delays, etc.

142
E. SOME POSSIBLE ALTERNATIVES TO THE DECISION-TREE SCALE
During this research program, several possible alternative PIO rating scales were developed and
discussed with other researchers and pilots. The scales were not intended to be the final answer,
but instead to provoke discourse on the need for a new scale. The interim approaches are
documented here for the same reason, and in the hope that there will be a dedicated effort to find
a totally new method for evaluating PIOs and their tendencies.

1. A True “PIO Classification” Method

It is natural for pilots – and, possibly, most common for highly experienced test pilots – to
change their flying technique if the slightest hint of PIO is encountered. As a result, for some
pilots a serious PIO may go completely undiscovered. In designing tasks that focus on handling
qualities, engineers encourage this adaptation: we set limits on “desired performance” that we
want the pilot to achieve, but we also have “desired performance” limits that we will allow the
pilot to achieve. In truth, for PIO it is desirable to always strive for some high level of
performance. One way is to demand that the pilot always try to meet the objectives of desired
performance and not “back out” to adequate if things are not going well. (Another way, of
course, is to demand maximum performance with no error tolerance whatsoever – the premise of
tasks like “Handling Qualities During Tracking” or HQDT.)
Experience has shown that when pilots are challenged to “fly tighter,” one of three scenarios
occurs. The pilot will 1) comply to the extent possible; 2) actually back further out of the task
instead of trying to achieve what is seen as an unachievable “zero-error” goal; or 3) interpret the
request as a need for more high-frequency control inputs, even if they are not correlated with the
task.
If we demand and receive a desired level of performance, then we may have some confidence
that the pilot has given maximum effort in a task that still has some operational relevance. Then
we may ask if any PIO tendencies, or any real PIOs, showed up during the performance of the
task. A true “PIO classification” scale, with no “rating” numbers, would then look something
like the scale in Figure 65.
The classification flowchart in Figure 65 (the entry point is the lower left box) is applicable only
if the pilot achieved, or employed maximum effort in an attempt to achieve, desired task
performance. The questions lead the pilot to classifications of “ratchet” – intended to cover pitch
bobbles and roll ratchet – or of handling-qualities problems if the oscillations did not detract
from the primary task. If the pilot had to pay attention to stopping the oscillations, the airplane is
prone to PIO, and the only issue is if it is severely PIO prone.
This scale is not in any final format, and would need some clarification if applied (e.g., what are
the implications of merely being “PIO prone” compared to “severely PIO prone”?), but it has the
fundamental advantages of forcing the pilot to use maximum effort in performing the task and
avoiding numerical scales. It also provides more methods for identifying the form of the
oscillation and not just determining whether the oscillation is divergent.
It has some drawbacks, as noted by several reviewers, in that it requires achieving desired
performance or applying “maximum effort” to do so. Many PIO-resistant airplanes may not be
able to achieve desired performance, and handling-qualities problems besides PIO could arise if
the pilot attempted to do so. There are some complaints that the scale should demand zero error
instead of only desired performance, to force high-gain activity on the part of the pilot.

143
 $LUFUDIW LV VHYHUHO\ 3,2 SURQH
6HYHUHO\ 3,2
SURQH
1R
1R
'LG WKH UHVXOWLQJ FRQWURO VWUDWHJ\ DOORZ
<HV &RXOG WDVN EH
\RX WR UHGXFH WKH XQGHVLUDEOH
RVFLOODWLRQ VR WKDW DWWHQWLRQ FRXOG EH DFFRPSOLVKHG"
UHVWRUHG WR WKH SULPDU\ WDVN"
<HV

3,2 SURQH

'HVFULEH UHDVRQ
IRU RVFLOODWLRQ

1R

:DV RVFLOODWLRQ D +DQGOLQJ TXDOLW\

<HV
FRQVHTXHQFH RI SUREOHPV QRW
WU\LQJ WR QXOO HUURUV" FODVVLILHG DV 3,2

<HV

1R
:DV D QRWLFHDEOH SDUW RI \RXU FRQWURO
DFWLYLW\ GLUHFWHG DZD\ IURP WKH SULPDU\ 1R :DV RVFLOODWLRQ WRR

WDVN LQ RUGHU WR GHYRWH DWWHQWLRQ WR IDVW WR WUDFN"

VWRSSLQJ DQ XQGHVLUDEOH RVFLOODWLRQ"

<HV
5DWFKHW

1R 3,2 WHQGHQF\ IRU WKLV WDVN

<HV

'LG DQ RVFLOODWLRQ RFFXU WKDW DIIHFWHG 1R

1R 3,2 WHQGHQF\ IRU WKLV WDVN
\RXU DELOLW\ WR DFKLHYH GHVLUHG
SHUIRUPDQFH"

<HV

:DV GHVLUHG SHUIRUPDQFH DFKLHYHG "

1R
7KLV VFDOH GRHV QRW DSSO\
,I QRW GLG \RX HPSOR\ PD[LPXP HIIRUW LQ
DQ HIIRUW WR DFKLHYH GHVLUHG SHUIRUPDQFH"

Figure 65. A PIO Classification Scale Based on Demanding Desired Task Performance

144
There is no question that the classification scale of Figure 65 is a huge departure from the
decision-tree scale of Figure 10, but it is one possible approach to really providing a
classification method for PIOs.

2. A Modified PIO Tendency Rating Scale

The hybrid scale that was presented in Figure 64 and discussed above is a much less
revolutionary – and hence much less controversial – attempt at improving the PIO tendency scale
than that shown in Figure 65.
A slightly more radical modification was developed under this contract, in a similar attempt to
merge the adjectival scale of Figure 61 with the decision-tree scale of Figure 10. The result is
shown in Figure 66. Text in bold italics comes from the adjectival scale of Figure 61; the
remaining text is taken, with slight modification, from the decision-tree scale of Figure 10. The
decision-tree structure is maintained, as are the first boxes (compare the scale in Figure 10), but
there is a departure for PIO ratings of 5 or better.

N o te nde nc y to induc e 1
oscillations

No
2

Is
Ta sk
O scillations can be prevented
P e rform an ce
or e lim inated by pilot technique
C o m pro m is ed?

No Yes
3

Yes
C aus es P ilot m ust reduce ga in or 4
O s c illa tio ns a bandon task to recove r

P ilot m ust open control loop

5
by relea sing or freezing
the stick

P ilot Initiate d
A b ru pt M ane uv ers
or
Tight C ontro l

No

C au ses Yes P ilot m us t open c ontrol loop

D iv erge nt 6
by relea sing or freezing
O sc illation the stick

P ilot A tte m pts

to E nter C ontrol
Lo op

Figure 66. A Possible PIO Rating Scale Combining Decision-Tree and Adjectival Scales

145
The advantages of the Figure 66 scale are as follows:
1) Only one rating corresponds to “no PIO tendency” – a rating of 1; any other rating
suggests PIO or the tendency to PIO. This avoids the controversy over whether
“undesirable motions” are a PIO tendency or not.
2) The “demands on the pilot” verbiage from the adjectival scale is retained; PIORs of 2
through 5 are dominated by these demands.
3) There is no requirement for the pilot to decide if the oscillations were divergent (the only
discriminator between PIORs of 4 and 5 on Figure 10).
4) Separation between PIORs of 4 and 5 are based on whether the oscillations required only
a reduction in gain/abandoning the task (4) or a requirement to release or freeze the stick
entirely (5).
5) PIORs of 2 and 3 are consistent with such “PIOs” as pitch bobbles and roll ratchet, but
they also allow for determination of “moderate” PIOs or tendencies; such oscillations are
prevented or eliminated by pilot technique, and the only issue is whether they interfere
with task performance. This distinction is felt to be more in keeping with the intent of the
original PIO scale of Cooper and Harper (Figure 61), and with the results of the Neal-
Smith experiment,51 where occasional PIO Ratings of 3.5 and better were assigned to
configurations that showed some tendency to PIO (Table 17, discussed earlier in this
section).
Unfortunately, the modified scale of Figure 66 also encountered some resistance, perhaps in part
because it is a change from tradition, and in part because it relies so heavily on oscillations
actually occurring. That is, there is no mention of “tendency” in any of the decision-tree boxes,
and some pilots feel that they need to be able to give a degraded rating based on a sense of
susceptibility to PIO without actually encountering a PIO. This criticism, of course, is central to
the current decision-tree scale of Figure 10 in the first place, and is perhaps the single greatest
source of confusion over use of that scale.
As with the radically different classification method of Figure 65, the modified scale in Figure 66
has not been applied in any formal evaluation. It is presented here to evoke discussion and
possibly to lead toward the definition of a truly comprehensive classification system.

F. A PROPOSED COMPREHENSIVE PIO CLASSIFICATION METHOD

Based on the overall reactions received for the modified PIO scales presented above, it was
determined that there is a dire need for a method for classifying PIOs and their tendencies. Such
a method may take the place of the current scale, but it may also be used in conjunction with a
PIO Tendency rating scale (whether it is the current scale or a new one). It is viewed as a cross
between a systematic pilot comment sheet and a simple rating scale.
It is possible to derive a more comprehensive method for classifying PIO if we recognize that
PIO is less like flying qualities and more like a flight regime – that is, it is a loss-of-control
phenomenon akin to departure or spin, or perhaps, to an encounter with severe turbulence or
windshear. All of these phenomena have been studied for years, and for all of them some form
of pilot rating scale has been developed. As an example, a departure/spin scale (shown in Figure
67) was generated jointly by test pilots and engineers for a simulation research study, 97 and a

146
slightly modified version of that scale* is published in the military flying qualities standard MIL-
STD-1797A.4 The following scale for PIO was modeled after that scale.
In order to determine the severity of the PIO, assess the risk to continued flying, and direct
efforts for fixing the problem, we need to know much more than just a PIO Tendency Rating
(even if it is augmented with an HQR and pilot comments). One possible scale is presented in
Figure 68.
The four-part scale in Figure 68 conveys much more information about PIO than the PIOR scale
in Figure 10. It also attempts to relate more directly to the pilot’s experiences with the PIO. In
fact, none of the four topics in Figure 68 corresponds directly to any part of the Figure 10 scale,
but all provide much-needed information.
Severity — Tells us how bad the event was. This assessment by the pilot must always come in
the context of the task and flight condition: perception of severity will be proportional to
proximity to the ground or other aircraft.
Duration — Indicates the length of time the event was experienced. A short-duration episode
may be perceived to be more tolerable than one that never stops.

Figure 67. Departure Rating Scale97

*
The only change between the scale in Figure 67 and that published in MIL-STD-1797A is to
the end points on the final scale: “How would you like to fly this configuration in the high α
condition?” In MIL-STD-1797A, in place of the pilot-centered responses, “Sierra Hotel” for
best and “Delta Sierra” for worst, are the more politically correct but less pilot-oriented “Lovely”
and “Atrocious.”

147
 S e ve rity D ura tio n D em a n d s O vera ll
o n p ilo t a sse ssm e nt
D an ge rou s N e ver stop pe d C o uldn 't p reve nt it W h at airplan e?
(ab an do n airplan e) (a ban do n airplan e) (D elta Sierra )

Se ve re C o uldn 't p reve nt it In to le rab le fo r

(a ba nd on tas k) M os t o f the tim e (ab and on ta sk) th e ta sk
(fix the a irp la ne )

A lleviate d
M od era te by tech niqu e O bjection able
(ca n't ig no re it) O ccas io na l (task p erform a nc e (w a rrants
com prom ise d) im p rove m en t)

Pre ven ted o r e lim inate d To le rable

M ild O n ly a v ery sho rt tim e by tech niqu e (satisfacto ry w ith ou t
(ca n ig no re it) (task p erform a nc e im p rove m en t)
no t com prom ise d)

N one N ev er saw on e N o ten den cy to W h at PIO ?

ind uce osc illa tion s (S ie rra H o tel)

Figure 68. A Proposed Comprehensive PIO Classification Method

Demands on pilot — One of the key ingredients missing from the decision-tree scale, but taken,
with alterations, from older versions of the PIO Tendency Rating Scale. This rating incorporates
impact on task performance as well.
Overall assessment — The chance for the pilot to tell us whether the airplane should be
considered safe or not. It is a final assessment of the airplane in context of the mission task and
the resulting PIO.
The proposed classification method in Figure 68 intentionally avoids the use of any numbers;
human factors experts maintain that the pilot is better able to judge a quality when it is associated
with words and not numbers. (In actual practice – being the numbers-oriented people we are –
engineers are likely to just assign numbers to the scales after the pilots have made their ratings
anyway.)
In addition, Figure 68 is designed to look like an interval scale (continuous gradations) as
opposed to an ordinal scale (ordering of numbers from good to bad). It is intended that the pilot
place a tick mark on each scale to indicate a response; unlike the PIOR and HQR scales, where
fractional ratings are not only impossible, they are also meaningless, a fractional marking here is
acceptable. As an example, consider the PIOR scale in Figure 10: what would a PIOR of 3.5
mean? That the airplane half-PIOs and half-doesn’t? But on the Figure 68 scale, if the pilot
can’t decide whether to call an event “moderate” or “severe,” a tick mark between the two tells
us just that.
Ordinal scales such as PIOR and HQR frequently pose a challenge to the pilot. There are times
with HQR, for example, when the wording that best describes the airplane characteristics does
not fit with the wording for demands on the pilot (Figure 9). This is a result of a combination of
the attempted melding of different topics into a single rating, and the nonuniform character of
many of those words. McDonnell98 found, in a survey of flying qualities specialists, that the

148
interpretation of many of the phrases on the HQR scale produces a nonlinear trend when the
ordinal scale is converted to an equivalent interval scale.
The solution is to break the rating process into pieces, as is proposed in Figure 68. One can see
by reviewing this new classification method that it is possible to convey information about many
different types of PIO. For example, a severe but non-divergent event that occurred for only a
few seconds will look very different in terms of pilot responses on the Figure 68 scale than a
nuisance event that occurs continuously. Yet both would likely be assigned PIORs of 4 from
Figure 10, and we would be left with HQRs and pilot comments to try to discern between the
two.

G. PIO AND RISK ASSESSMENT

If we are to accept that PIOs can and will occur in developmental flight testing, then we must
decide: is the event just experienced sufficiently dangerous to justify shutting down the testing
until the airplane is fixed?
It is also possible – but potentially dangerous – to address the question of whether the problem
even needs to be fixed in the first place. The procuring activity might decide, for example, that a
PIO that occurs only infrequently and is not severe is not worth fixing. It is even possible to
determine that a relatively severe event is so rare that it, too, does not deserve to be fixed. (This
is the potentially dangerous side, of course, especially since just one such event with a passenger
airplane full of people is unacceptable.)
Still, the rating scale in Figure 68 is directly amenable to risk assessment in developmental
testing. As an example, a tradeoff between severity and duration could be made, as illustrated in
Figure 69. With this form of risk assessment, it is possible to divide the potential risks into Low,
Moderate, and High categories.

Duration
Never Most of the Occasional Only a very Never saw
stopped time short time a PIO
Dangerous
(bail out) High High High High
Severe
(abandon High High Moderate Moderate
task)
Severity

Moderate
(can’t ignore High Moderate Moderate Low
it)
Mild
(can ignore it) Moderate Moderate Low Low
None

Figure 69. Risk Assessment for PIO Severity versus Duration

149
This is similar to the assessments made at Edwards AFB by safety review boards for every flight
test effort conducted there.99 Even more categories of risk could be defined, of course, and it
remains to determine what each level of risk means in terms of fixing or flying.
This process for risk assessment is simply not possible with the current PIOR and HQR scales.
We can never be sure that a PIOR of 4 – in combination with a very poor HQR – really results
from a severe event, or from a mild event and some other flying qualities problem, unrelated to
the PIO.

H. MANEUVERS TO EXPOSE PIO

Because PIOs are sporadic by nature, the maneuvers that will identify PIO must be robust; that
is, they must be able to expose real PIO problems on a regular basis, so that the answer can be
believed, without giving false indications of a tendency to PIO where none actually exists.
There is an ongoing discussion among the user community over the best form of these evaluation
maneuvers. For example, the Handling Qualities During Tracking (HQDT) task is favored at
Edwards Air Force Base; but the most common flight condition for PIOs is landing.52 Precision
offset landing is a regular handling-qualities task, but some feel that it may not be the best task
for revealing PIO tendencies. On the other hand, HQDT is frequently not flown as it was
designed (HQDT, when properly performed, involves tracking of a target while allowing no
error at all).
When flown as intended, HQDT is not a real “handling qualities” task since limits on desired and
adequate performance are not specified. But addition of such limits, thereby allowing for
application of the Cooper-Harper scale, removes the fundamental precept of the task. Since
“HQDT” is a misnomer, the task should be renamed as either “PIO During Tracking” (PIODT)
or “High-Gain Tracking” (HGT).
Both the offset landing and a high-gain tracking task such as simulated aerial refueling at least
relate to real mission elements. Both also appear to have their place in the maneuver list for
flying qualities and PIO.92 The landing task forces control of both attitude and flight path near
the ground, thereby inducing high-gain control activity, at landing speeds. The very requirement
to correct for a lateral offset from centerline will help expose any tendency for lateral actuator
rate limiting.
By contrast, simulated aerial refueling does not guarantee large control amplitudes, but it does
lead to high pilot gain, and safely away from the ground.
Use of fly-by-wire control systems and full-envelope tailoring dictates special attention to
maneuvers specific to the design under evaluation. For example, large-amplitude maneuvering
at high speeds may be sufficient to evaluate PIO tendencies if the control system contains logic
switches and gain changers in this speed regime. The bottom line is, there will always be a need
to devise new maneuvers, or tailor existing ones, on an ad hoc basis. Such maneuvers should
never take the place of fundamental PIO evaluations, but should be added to them.

I. RECENT SIMULATION EXPERIENCE WITH REPLICATING PIOS

Even with the continuing development of analytical criteria to predict and prevent both linear
and nonlinear PIOs, ground simulation will always be considered an indispensable tool in the
development of any new aircraft design. As such – and with the trend toward shrinking money

150
for flight testing, for the foreseeable future – simulation will be used increasingly to investigate
PIO.
Capabilities for visual and motion systems in simulators have evolved tremendously over the last
few years. Unfortunately, no matter how far this evolution goes, it will never completely
replicate the real world. So some level of doubt will exist about the ability of ground simulation
to supplant actual flying, especially when investigating a random, sporadic event like PIO.

1. The HAVE PIO Truth Model

How effective is ground simulation as a PIO assessment tool today? This question has been
investigated by several researchers recently, using the results of the US Air Force Test Pilot
School student project HAVE PIO36 as a “truth model.” The ratings and quantitative data from
this experiment have been used throughout this report, including Figure 63 in this section.
HAVE PIO was performed by three test pilot students in the USAF variable-stability NT-33A.
Eighteen pitch configurations were evaluated in the offset approach/precision landing task.
Variations in short-period frequency and added lag filters produced airplanes that ranged in
flying qualities from excellent to highly PIO-prone. The NT-33A was a two-place airplane with
a safety pilot capable of taking over manual control at any time. On several approaches in
HAVE PIO, the safety pilots felt it necessary to do so.

2. Simulator Replications of HAVE PIO

The objectives of the simulations were to replicate the conditions of HAVE PIO, including
dynamic models of the airplane and the visual runway environment, on ground-based simulators
and then measure their effectiveness at reproducing the pilot ratings of the “truth model.”
The first of these studies94 was conducted by engineers of the Air Force Research Labs, Wright-
Patterson AFB, Ohio. The facilities were a fixed-base dome, the Manned Simulator 1 (MS-1),
and a moving base simulator, the Large-Amplitude Multimode Aerospace Research Simulator
(LAMARS). Since publication of Reference 94, simulation engineers at Wright-Patterson AFB
have reported that the motion setup for the LAMARS portion was not correct, so these data will
not be discussed here. While the dynamics of the HAVE PIO configurations were modeled, the
pilots operated as if they were in a single-pilot airplane with no safety pilot, and with the heads-
up display (HUD) setup of the basic simulator. Five pilots participated.
The more recent study was conducted at NASA Ames Research Center using the large-amplitude
Vertical Motion Simulator (VMS).23 In this case, the simulator was operated with three motion
gain sets: fixed-base, a reduced gain set that emulated a small-displacement hexapod-type
motion system, and the full motion capability of the VMS. In this study, a concerted attempt was
made to fully reproduce all possible elements of the HAVE PIO flight program. This included a
computer-controlled “safety pilot” that took control away from the test pilot if nose-gear sink
rate was too high near the ground, and more accurate visual scenes, both in the cockpit and out
the window. As with the Air Force study, five pilots participated in this simulation, flying all
configurations with all levels of motion.

151
3. Qualitative Comparisons
In both simulation experiments, the primary metrics for effectiveness were the subjective pilot
ratings, HQR and PIOR. The closer the ratings to the “truth model,” the more effective the
simulators would be judged.
The HQR/PIOR results were shown for both experiments in Figure 63. A crossplot of HQRs
from HAVE PIO and from the simulations is shown in Figure 70. For simplicity, only the mean
ratings are plotted. A similar plot, for PIO Tendency Ratings, is shown in Figure 71. In both
plots, the diagonal line indicates perfect agreement, and dashed lines represent a rating difference
from perfect agreement of one rating point.
Both figures show the same basic trend: airplanes that were considered to be very bad in flight
were not as bad in any simulation. For the VMS experiment, especially, the spread in HQRs was
much less from the simulator. Configurations that were good in flight (HQR of 3 or better) were
not as good in the simulation, and those that were very bad in flight (HQRs of 7 or worse) were
not as bad in the simulation. This suggests that simulation can mask the positive features of a
particular airplane, but it can also mask the negatives.
Three of the eighteen configurations were assigned HQRs of 3 or better by all three pilots in the
flight program. In the VMS, these configurations were at least one rating point worse, and in the
MS-1, two of the three were worse. At the bad end, three configurations in flight received HQRs
of 10 from at least two of the three pilots. No rating of 10 was assigned at any time in any of the
ground-based simulations.
10

8
Mean HQR Flight

4
VMS Large motion
3 VMS Hexapod motion
VMS Fixed base
2
MS-1 Fixed base
1
1 2 3 4 5 6 7 8 9 10
Mean HQR Simulation
Figure 70. Comparison of Mean Handling Qualities Ratings from Simulations and Flight

152
 6

5
Mean PIOR Flight

VMS Large motion

2 VMS Hexapod motion
VMS Fixed base
MS-1 Fixed base
1
1 2 3 4 5 6
Mean PIOR Simulation
Figure 71. Comparison of Mean PIO Tendency Ratings from Simulations and Flight

The trend for the PIO Tendency Ratings in Figure 71 is for better (lower numerically) ratings in
the simulator: most of the mean ratings lie to the left of the line of perfect agreement. Even
more dramatic is the spread in ratings that is not shown by the mean ratings in Figure 71: in the
HAVE PIO flight program, nine of the 18 configurations were not considered to exhibit PIO.
Taking all simulation results as a group, only one of the 18 configurations did not receive a PIOR
of 4 or worse (indicating PIO, see Figure 10) from at least one pilot. If we are to accept the
hypothesis that PIOs are sporadic events, the fact that at least one pilot saw PIO for a particular
configuration means that we should consider it to be susceptible to PIO. Hence 17 of the 18
cases were susceptible, if we use the simulation results.
Several factors, however, influenced the PIO rating process. As discussed earlier, the PIOR
scale has been found to be subject to pilot interpretation, and this was true for at least the Air
Force’s MS-1 study. Specifically, several of the test pilots interpreted pitch bobbles to be PIOs
and assigned PIORs accordingly. The result is that many PIO Tendency Ratings of 4 were for
this “bobble” and not for more serious PIO. (This was commonly evidenced by the pilot’s
assignment of an HQR of 4 or 5 to go with the PIOR of 4 – indicating an airplane that is less than
perfect, but not too bad for the task.)
The authors of Reference 23 conclude from their VMS experiment that the large motion of the
VMS produced ratings that were closest to the “truth model.” The data in Figure 71 certainly
support this conclusion, since the solid circles tend to lie more to the right than the other
symbols, but the actual correlation with flight is still not very good.

153
As a tool for predicting the occurrence and severity of PIO, simulation does not appear to be very
effective. In HAVE PIO, seven configurations were judged by at least two of the three pilots to
both be PIO-prone (PIOR 4 or worse) and have Level 3 handling qualities (HQR 7 or worse). In
the MS-1 simulation, only one was judged to be so, and one was almost Level 1 (average HQR
of 3.6) with no PIOs (five PIORs of 2). Likewise, in the VMS study, with large motion only one
case received a Level 3 mean HQR and a mean PIOR worse than 4.

4. Quantitative Results
The results of Figure 71, especially, raise the question of the existence of PIO versus the pilot’s
perception of the PIO. We cannot, based purely on pilot ratings, determine if it is the pilots, the
dynamic model, or the simulator itself that precludes a match with the “truth model.”
A brief look at a representative set of time histories suggests the answer. As an example, Figure
72 shows the time responses of pitch attitude and pitch rate for one configuration.
20
MS-1 (Pilot 2) Runs 16-18
MS-1 (Pilot 2) Runs 22-24
15 Flight (Pilot A) Flt 3 Runs 6-7

10
pitch
attitude
(deg)
5

0
0 10 20 30 40
time (sec)
-5
20

15

10

pitch 5
rate
(deg/sec) 0
0 10 20 30 40
-5 time (sec)

-10

-15

Figure 72. Comparison of Responses from HAVE PIO Flight and Simulation

154
The configuration for Figure 72 is 5-10, the worst from the flight experiment. In flight, this case
received HQRs of 10 and 10, and PIORs of 5 and 5, from the two pilots who evaluated it.
The two thick lines in Figure 72 are the traces for one pilot from the flight study. The thin
curves are for one pilot from the MS-1 study,94 with six separate runs. The simulation traces end
at main gear touchdown, at which point the pitch rate value is “frozen” in the time histories. The
simulation pilot assigned an HQR of 6 and a PIOR of 4 for this case, indicating PIO, but
apparently not an especially severe one.
The differences in peak values on the time traces are dramatic: in the simulator, peak pitch rate
exceeded 15 deg/sec on several occasions, while in flight, peak pitch rate is about 6 deg/sec or
less. These data are typical for those from the two experiments.
The question, then, is: why did the pilot in the simulator consider this configuration to be only
Level 2 (HQR 6), while the pilot in flight thought it was uncontrollable (HQR 10)? It is not
possible, with the data at hand, to come up with a single reason. Certainly the lack of motion
cues in the fixed-base simulator must have contributed, since the pilot would not have been as
aware of the large motions in the PIO. Lack of a compelling outside visual scene probably made
a difference as well, for the same reason.

5. Observations About Simulation

Despite the apparent failings of simulation to replicate the flight results of HAVE PIO, there is
still some promise in the data. The following observations can be made:
• If a particular configuration is bad (mean HQR worse than 6.5 and PIOR of 4 or worse)
in the simulator it will be even worse in flight. If it is moderately good (mean HQR 4 or
better and PIOR better than 3) in the simulator it may be even better in flight. No real
conclusions can be made about configurations between these extremes, as they may be
either better, worse, or the same in flight.
• Insight can be gained by looking at quantitative data (time histories) from the simulator
and comparing these data with known in-flight PIOs. Generally, it is expected that the
simulation events will be much more severe, but that the pilots may not be aware of this.
Some other general observations, not specifically demonstrated in this section, can also be made:
• A large pilot population must be used, and it appears to be more important in the
simulator than in flight. Both simulations94,23 used five pilots, and in both cases at least
one of the pilots failed to see PIO for most of the PIO-prone configurations. Five would
seem to be a minimum pilot population; if it were possible to select the specific subjects,
a smaller number could be used, as long as more than one of the pilots has proven
capable of identifying PIOs in a simulator.
• It is imperative that all pilots receive a thorough briefing on the use of the PIO Tendency
Rating Scale, and that all understand what constitutes full-blown PIO as opposed to
nuisance “bobbles” with respect to the assignment of PIORs.
• Ratings from the current PIOR scale (Figure 10) must always be interpreted in
conjunction with HQRs (Figure 9) and are not to be considered stand-alone ratings.

155
 IX. PREVENTION OF PIO IN FLIGHT: A METHOD FOR
REAL-TIME DETECTION

A. REASONS FOR REAL-TIME DETECTION

It has been stated repeatedly in this report that the goal should always be to prevent PIOs by
proper design of the aircraft and all of its components. Still, there are many reasons to consider a
method for real-time detection of PIO after the airplane has been built. The uses for such a
device include:
• Prototype flight test. Many PIOs occur during initial flight testing, on prototype and
experimental aircraft before full-scale production has begun. In fact, it is likely that just
about every airplane ever developed has experienced some form of PIO during its
development, but the vast majority of these events are relatively minor, easy to identify
and fix, and do not adversely impact the development process. Precursors to the few
catastrophic events that have occurred could well have been spotted with identification
methods and repairs made, or at least precautions implemented.
• Envelope expansion. Possibly as a part of initial flight test, or even years later when the
missions and capabilities of a particular airplane are increased, envelope expansion
testing could include real-time monitoring for PIOs. Otherwise PIO-resistant production
airplanes with new weapons loadings can become susceptible to PIO. This was the case
with the F-16, when added fuel tanks and weapons placed the airplane outside the normal
envelope for U.S. operation, and PIOs occurred. Data for such events were provided for
this project by Lockheed Martin Tactical Aircraft Systems (LMTAS) under a subcontract.
• Operational uses. As flight control systems become more advanced, airplanes are
operated with more relaxed stability, and control surfaces become smaller or move from
aerodynamic to thrust vectoring effectors, the possibility for severe PIO on a production
airplane, years after its acceptance for operational use, will only increase. A method for
detecting PIO – in conjunction with a training program for exposing operational pilots to
the phenomenon to help them identify it – can provide a safety element, similar to stall
warning or other current devices. This has been criticized by some as nothing more than
a “crutch” for imperfect design, but we re-emphasize that design is only the first layer of
prevention. Sometimes, it may be necessary, because of funding limitations, to decide
that an airplane will enter full-scale production even with a suspected susceptibility to
PIO in some combination of loadings and flight conditions. This compromise is in effect
made today, when airplanes with degraded handling qualities in certain conditions are
still sent into fleet use. An on-board detector can provide a level of protection from
unexpected catastrophic PIO.
• Post-flight analysis. While this is hardly “real-time,” the methods for real-time detection
can be applied to any set of flight data to determine if PIOs occurred. The method
developed under this contract has been applied this way hundreds of times, even though it
has not yet been tested in a real-time environment. If a particular airplane has been
suspected of exhibiting PIO, but is not equipped with a real-time detector, the time
history data can be analyzed using the same methods.

156
• Failures. Impaired vehicles are likely to be even more susceptible to PIO. The more
advanced the aircraft and its flight control system, the more this is going to be true.
While such exotic schemes as reconfiguration and active envelope limiting may help to
minimize the risk, there is still a risk. Real-time monitoring would help reduce the risk of
PIO while operating in a failed state, though it remains to be determined how much such
a monitor might help during the transition to the failed state.

B. THE REALITIES OF REAL-TIME DETECTION

While the goal of this report is to provide methods for preventing PIO, there are some realities
that must be considered if the final protection against catastrophic PIO is going to be real-time
detection. These realities may be summarized as follows.
• It may be impossible to “detect” PIOs before they happen. In conducting this study, we
have found that the possible causes for PIO – the combination of airplane, pilot, task, and
trigger – are widely varied. It is not possible to coalesce all of the known PIOs, ranging
from minor to catastrophic, into a small list of common circumstances. The fact that
researchers have found it necessary to create three (or more) Categories of PIO illustrates
the broad range of events and causes that can be identified. There simply is not a
consistent, identifiable “pre-PIO” condition that can be used to reliably indicate that a
PIO is about to happen. It will be difficult, if not impossible, to determine an identifiable
precursor to PIO that we can watch for in real time. Rather than monitor for the signs of
PIO onset, we may have no choice but to wait for a real PIO to develop before we do
anything about it. The issue, then, is how soon we can identify it as a PIO.
• Some pilots use oscillatory inputs as part of normal control strategy. There is evidence
of oscillatory behavior by pilots during normal flying in what are considered to be PIO-
resistant airplanes. PIO detection must be capable of separating these intentional inputs
from real PIO.
• Some or all of the typical PIO “signatures” may be observed during normal operations.
Pilots do not fly airplanes at one operating frequency. As the task demands it, they will
apply inputs at frequencies well above the normal range for closed-loop control, and
momentary out-of-phase conditions may be observed. In fact, all of the telltale signs of
PIO – apparent oscillatory input-output behavior with the airplane out of phase with the
pilot – may occur from time to time, even for the best of airplanes. This is not PIO, and
if the pilot needs this response to get the job done, it must not be falsely identified as PIO.
• Common precursors to PIO, such as actuator rate limiting, can occur without PIO. It is
appealing to think that we could monitor actuator rates and use rate limiting as a sign of
impending PIO. Unfortunately, there can be times when rate limiting is common and
there is no accompanying PIO. In other cases, of course, the first nibble at rate limits
may lead to a rapidly divergent PIO. It is not possible to reliably use rate limiting as a
PIO test in general. It may be applicable for specific aircraft, especially for airplanes that
are highly unstable and rate saturation means an immediate departure. In such cases,
there is usually a software rate limiter somewhere in the flight control system, and this
software limiter can be monitored. Hydraulic actuators do not have a single, repeatable
rate limit – the limit depends upon loads on the surface and demands on hydraulic flow
rates throughout the system – so a monitor on “rate limiting” would be impossible.

157
• What constitutes a PIO, as opposed to overcontrol or just normal operation, varies with
flight condition. It is hopeless to try to define a single set of parameters that separates
“non-PIO” conditions from true PIO. For example, a cycle of oscillation in load factor
of, say, 0.3g from trim is perfectly normal for a fighter airplane maneuvering at high
speeds, but a sign of a possible PIO for any class of airplane at landing speeds.
• There may be times when it will be better to allow the PIO than to prevent it. This is a
fundamental concern for any practical use of real-time detection: should the flight
control system take active measures to stop the PIO, or should this be left entirely up to
the pilot? Until a reliable method for detection can be developed, of course, this issue is
moot: the ideal application of active intervention will be only when the life of the pilot is
at stake. In the meantime, there will be situations where the life of the pilot could be
compromised more by active intervention than by simply allowing the PIO to continue –
for example, if the PIO occurs during an emergency collision avoidance maneuver. Then
it must be left to the pilot to determine if backing out of the control loop is possible.

C. PHILOSOPHIES FOR PIO DETECTION

Several organizations have been active in the development of methods to detect PIOs. All
known methods require that a PIO occur before it is detected, but each has its own unique
approach to detection. And all known methods have been funded by, or developed by, the US
Air Force Research Lab (AFRL) at Wright-Patterson AFB, Ohio.
A significant in-house effort at AFRL, supported by Dr. Dominick Andrisani from Purdue
University, has led to a set of algorithms called FIDO, for Frequency Investigator and Detector
of Oscillations. Initial work was directed at adopting a moving-window fast-Fourier-transform
(FFT) of pitch rate and stick force: the time responses were processed to identify amplitude and
phase characteristics. This method, however, proved too complicated, as the window size had to
be quite large to get sufficient data for a good FFT.
Modifications to FIDO were made to instead perform a time-domain analysis. 100 FIDO searches
time-history data, identifies peaks in pitch rate, forms the data around the peak into a window,
and fits the data to a sinusoid. A second sinusoid of the same frequency is fitted to the pilot’s
control input. If the data match the sinusoidal form sufficiently well, the frequency, amplitude,
and phase of the two sinusoids are compared to determine if the response is a PIO.
FIDO has proven to be an extremely useful tool at AFRL for post-run analysis of time-history
data. Unfortunately, because of its basic methods, FIDO is not amenable to real-time use.
A second development effort, sponsored by AFRL, has been performed by Accurate Automation
Corporation of Chattanooga, TN. The AAC approach has been to use neural-network methods to
identify PIO. The neural net has been “trained” with time-history data from actual PIOs, and
AAC has conducted a simulation at Wright-Patterson AFB to evaluate the method. AAC
extended the neural-net detection to include methods for alerting the pilot of the development of
the PIO. The AAC neural-net detector was very effective in its Air Force evaluations on the
piloted simulator.
The third known development effort was undertaken by HAI and is the subject of the remainder
of this section of the report.

158
D. DETECTION METHOD DEVELOPED FOR THIS STUDY
Work on a method for real-time detection in this research study was third in priority to
development of design criteria and testing methods. Still, it was recognized that there is value to
such a method, so a set of algorithms was devised using our definitions of PIO documented in
this report. While our definitions are relatively simple, the real challenge in developing PIO
detection was in determining the best ways to filter the data so that oscillations could be
identified and separated from normal control inputs and airplane responses.
Much of the development of the algorithms was performed in conjunction with Mr. Robert K.
Heffley of Robert Heffley Engineering. They were implemented in Matlab as a series of m-files.
The remainder of this section describes the theory behind our method and evaluates its overall
effectiveness for a selected set of flight and simulation data – again, applied not in real time, but
after the event, sometimes many years after the event. The intent is to demonstrate that such a
method can be an effective detection device for the future occurrence of PIOs.
The fundamental philosophy for PIO detection was that the pilot will control an airplane in a
more or less sinusoidal fashion even when there is no tendency to PIO. The objective, then, is to
monitor this oscillatory activity to determine if it appears to be normal or if there is evidence of a
developing PIO.
In truth, pilot control inputs are not precisely sinusoidal, and can resemble pulses, ramps, and
even steps, with only occasional obvious sinusoidal character. As an example, Figure 73 shows
the longitudinal stick position traces for a selected 20-second segment from the LAMARS
simulation documented in Appendix A. These time traces are for the highest-crossover pilot
(Pilot A) and the lowest-crossover pilot (Pilot D) in the simulation, as identified from tracking
data (Section IV), flying the baseline pitch configuration, 2DR30. They are the inputs for the
identical pitch HUD commands. The traces for Pilot A show considerably more oscillatory-
looking inputs than those for Pilot D, but for either pilot there is a significant amount of high-
frequency, random-looking control activity that precludes direct observation of any oscillatory
input tendencies.
By contrast, Figure 74 shows similar traces for the same two pilots, over the same time sequence,
while flying one of the most PIO-prone pitch configurations in the Appendix A simulation
(1DL3). Both pilots show some evidence of oscillatory inputs near the PIO frequency – roughly
4.5 rad/sec (1.4 seconds per cycle) for pilot A and 3.6 rad/sec (1.7 seconds per cycle) for Pilot D.
But even in this case, where the pilots clearly saw PIOs (Pilot A felt they were divergent at
times, as indicated by his PIOR of 5), there is not a clear oscillatory characteristic to be
identified.
Extraction of sinusoidal behavior from time traces such as those in Figure 74 requires extensive
filtering. First, a high-frequency filter will eliminate small extraneous inputs that are not related
to the PIO itself. Second, a washout filter removes the dc component, leaving only data at
frequencies around the PIO.
The same filtering must be applied to the response state variables of interest, or the dynamics of
the filters themselves will introduce phase errors that can hide critical output-input relationships,
or introduce artificial relationships that aren’t really there.

159
 4
Pilot A (Run 66: HQR 4, PIOR 2)
Pilot D (Run 304: HQR 2, PIOR 1)
3

long. 2
stick
pos.
(in.) 1

0
70 72 74 76 78 80 82 84 86 88 90

-1
time (sec)

Figure 73. Time Histories of Longitudinal Stick Activity for Two Pilots from LAMARS
Simulation, PIO-Resistant Configuration (2DR30)
4

Pilot A (Run 61: HQR 8, PIOR 5)

3 Pilot D (Run 268: HQR 7, PIOR 4)

long. 2
stick
pos.
(in.) 1

0
70 72 74 76 78 80 82 84 86 88 90

-1
time (sec)

Figure 74. Time Histories of Longitudinal Stick Activity for Two Pilots from LAMARS
Simulation, PIO-Prone Configuration (1DL3)

A key precept for our approach to PIO detection is that there will always be some state of the
airplane that is out of phase with the pilot’s inputs. Most pitch PIOs analyzed in this project
show that pitch attitude is usually close to 180 degrees out of phase with stick input at the
beginning of the PIO (though the phase difference typically becomes greater as the PIO
progresses, as shown be the frequency-response data in Section IV). Other states, such as normal
acceleration or flight path angle, may be effective as well, and in a few rare cases may work
where pitch attitude doesn’t appear to work. Still, for the vast majority of PIOs that were
examined in the course of this study, pitch attitude is the appropriate parameter, and no other

160
states were needed to obtain good information. It is not known at this time how other states
might be blended, monitored, or selected to augment the pitch attitude information.
While the principal out-of-phase state is pitch attitude, it did not turn out to be the best parameter
for detection. As with FIDO,100 pitch rate was selected instead. This is reasonable since pitch
rate typically has more output power than attitude and is only one integration away from attitude.
In addition, pitch rate is a body-referenced signal, while attitude is Earth-referenced, so it is more
appropriate during turning flight. If attitude were to be used, besides the basic problems with
input power (and signal quality in general, since vertical gyros tend to be noisier and less reliable
than rate gyros), some correction for bank angle would be required.
With pitch rate as the relevant output state, the out-of-phase condition becomes 90 degrees
instead of 180 degrees.

1. The Realtime Oscillation VERifier (ROVER)

The method developed under this contract searches for oscillatory signals in stick input and
angular rate output states, and compares the characteristics of the two sets of signals. Because it
“verifies” the oscillation, and because it is by design simple enough for real-time
implementation, the method has been named ROVER, for Realtime Oscillation VERifier.
Four parameters are monitored, as follows:
Pitch rate, q: After processing through a bandpass filter, as described above, positive and
negative peak values of pitch rate are measured. Based on the initial assumption that these
values constitute one-half-cycle of an oscillation, the effective frequency and amplitude of this
oscillation are computed. The amplitude is compared with predefined thresholds for PIO.
Stick input, δes or Fes: After preprocessing with the same filter as pitch rate, peak values of stick
input are also identified and treated as if they were sinusoidal. The magnitude of the stick input
oscillations is compared with threshold values specified for PIO.
Oscillation frequency, ωq: That an identified “oscillation” in filtered pitch rate is within a
certain frequency range is not necessarily a sign of PIO. In fact, the pilot’s normal operating
frequency for closed-loop control will typically be well within the range for PIO, so an effective
oscillation in this frequency range may be good, not bad.
Phase between stick and pitch rate, ΦqF: Besides the fundamental assumption of oscillatory
behavior in both input and output, it is further assumed that the identified half-cycle of pitch rate
is directly correlated to the immediately previous half-cycle of stick input. Thus there is always
a phase lag, and if this phase lag is in the proper range, PIO is possible.
The last two parameters may, in fact, occur during normal closed-loop control and are not, alone,
indications of the presence of PIO. Even an occasional exceedance of the threshold for either
pitch rate or stick input can occur without any hint of a PIO. But experience has found that if the
phase angle and frequency are within the proper ranges, and if the magnitude of either input or
output is also in the proper range, for at least one full cycle, a PIO is developing. Such an event
is usually considered “mild” or “moderate” by the pilot who experienced it – PIO ratings may be
4 or 5 (or sometimes even 3), while Cooper-Harper Handling Qualities Ratings will typically be
only Level 2 (5 or 6), with an occasional rating of 7. The primary task can still be performed
even through a fully-developed PIO.

161
If all four parameters reached their thresholds, however, PIO is definitely occurring, and almost
certainly it is severe.

2. PIO Severity
Given the importance of the four parameters described above, ROVER computes what is
euphemistically referred to as “PIO severity” – a relatively simple summation of flags based on
whether the parameters reached their thresholds. As each parameter reaches its threshold value,
a flag is set to one, and the values of the flags are added together. A value of 1 is desirable if that
1 comes from frequency alone: the pilot is providing closed-loop control at a reasonable
frequency, but the airplane is not out of phase and input and output magnitudes are reasonable.
A single occurrence of a 3 (meaning three of the four parameters indicate possibility of PIO)
does not register as a PIO, but a second occurrence on the very next half-cycle is considered to
be a “moderate” PIO and an intermediate value of 3.5 is assigned – not as bad as a 4 but worse
than a 3.

E. APPLICATION OF ROVER FOR IDENTIFICATION OF PIOS

1. Application of ROVER for this Report

ROVER was applied in Section IV of this report, for the measurement of human pilot behavior
in PIO. It was shown there that ROVER is a very powerful tool for identification of control
frequencies and PIO frequencies. It was applied with equal success to the flight data from the
HAVE PIO36 experiment and to the simulation data in Appendix A. Any reference to PIO
frequency in this report for data from these two sources is taken directly from ROVER. All of
the data analysis that follows is also based on the output from ROVER.

2. Application of ROVER to the YF-22A PIO

The best illustration of the utility of ROVER is its application to a now-familiar event: the low-
altitude PIO of the YF-22A that ended in impact with the runway at Edwards AFB. This PIO
was discussed in Section II, where a time history of the event is provided (Figure 7b). Lockheed
Martin Tactical Aircraft Systems provided time-history data for both the event itself, and for the
previous low approach, which had similar conditions but did not result in a PIO (Figure 7a). In
the first pass, the pilot used only partial power, while in the PIO run full power was applied.
Graphical output from ROVER is in the form of several time-history plots, as shown in Figure
75 for the two YF-22A runs. The top two plots show the unsmoothed and smoothed (bandpass
filtered) time histories of pitch rate and longitudinal stick force. Stick force is distinguishable in
the “unsmoothed I/O” traces for each run because the pilot trimmed nose-up, requiring a push
force to maintain level altitude, at 10-15 seconds into each run. A third trace that can be seen in
the “unsmoothed I/O” plots is normal acceleration, plotted but not used in the ROVER
computations.
All of the other plots in Figure 75 are computed from the smoothed signals and every data point
corresponds to one half-cycle of oscillation identified from the smoothed signals.
The two plots below the smoothed I/O are the identified frequency of pitch rate oscillations and
the computed phase lag between stick force input and pitch rate response.

162
 20 20

10 10

S m o othed 0 S m oo th ed 0
I /O I/O
-10 -10

-20 -20
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
T im e (s ec ) T im e (sec )
Ru n: y f2 2 1 Ru n: y f22 pio
20 20

10 10
Uns m oo th ed 0 Uns m o oth ed 0
I /O I/O
-10 -1 0

-20 -2 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
T im e (s ec ) T im e (s ec )
8 8

6 6

F requ enc y 4 F requ enc y 4

(rad/s ec ) ( rad/s ec )
2 2

0 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
T im e (s ec ) T im e (s ec )
2 00 20 0

1 50 15 0

P has e 1 00 P h ase 10 0
(deg ) (deg)
50 50

0 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
T im e (s ec ) T im e (s ec )
60 60

40 40
P eak P eak
v alu es values
20 20

0 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
T im e (s ec ) T im e (s ec )
4 4

3 3

P IO 2 PIO 2
s ever ity s everity
1 1

0 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
T im e (s ec ) T im e (s ec )
s tic k s tic k

q q
P IO PI O
F lag s ph as e F lag s ph ase

om ega o m ega

0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
T im e (s ec ) T im e (s ec )

a) F irst p ass (n o P IO ) b ) P IO startin g at t = 3 7 sec

Figure 75. Outputs from ROVER Application to YF-22 Low-Approach Data

163
The “peak values” plot shows the peak-to-peak magnitudes for stick force (circles) and pitch rate
(plus signs) for each identified half-cycle.
The bottom plot indicates “hits” for each of the PIO flags: presence of a symbol indicates that
the PIO threshold for that parameter was met for that half-cycle of oscillation.
The next-to-last plot is the measure of PIO severity, effectively a summation of the four PIO
flags. A “PIO severity” setting of 1 is desirable, indicating closed-loop control at the proper
frequency. A value of 2 usually indicates an out-of-phase condition but with only small input
and output amplitudes, unnoticed by the pilot. The severity level can be a 3 only if two of the
flags are frequency and phase, and the third flag is magnitude of either stick force or pitch rate.
An occasional occurrence of a 3 is not entirely unusual, but if there are consecutive 3’s an
intermediate value of 3.5 is defined – an indication of a developing PIO, and the point at which
advisory should be given to the pilot. A PIO severity of 4 is a fully developed, severe PIO,
indicating the need for direct, immediate corrective action.

a. Results for First Pass

During the first pass, Figure 75a, the PIO severity reaches 3.5 on two occasions, around 10 sec
and again around 21 sec in the time history. In both occurrences the suggestion is that there is a
mild PIO developing. There is no way of knowing, in hindsight, if this were perhaps true, and
that this might have been a warning of the possibility for PIO. It also entirely possible that the
thresholds one or more of the test parameters is slightly too low, allowing an otherwise normal –
albeit high-gain – operation to be detected as a PIO. In both cases the oscillation frequencies
were between 5 and 7 rad/sec and the airplane was indeed out of phase with the pilot, but
resulting pitch rate output was rather small. If an onboard, real-time device had been used, these
exceedances would have triggered a warning that the pilot was nibbling at PIO, and it is possible
that the next low approach would not have been made.
In general, for the first pass there is no clear, consistent signature in any of the traces in Figure
75a that indicates a PIO.

b. Results for PIO Run

During the next low approach, Figure 75b, the pilot selected full power at about 32 sec and gear
up at 35 sec. It is a subject of some debate as to when the PIO really started, but close inspection
of the time histories (see Section I) suggests that the real event began somewhere just before the
37-second mark in Figure 75b.
For the first 35 seconds, the data in Figure 75b looks quite similar to that in Figure 75a, with a
PIO severity of 3.5 first occurring at 31 seconds, or 6 seconds before the full-blown PIO. To
examine the events of the PIO itself in greater detail, selected ROVER output plots for the 15-
second segment between 30 and 45 seconds are shown in Figure 76. PIO flags and estimated
PIO severity for the same time scale are given in Figure 77.
The calculated signals in Figure 76 show an unambiguous signature of severe PIO: from 35 sec
to about 39 sec, frequency drops for each half-cycle as phase angle increases and both input and
output amplitudes increase. At about the 38-sec point the pilot reduced power and changed
control strategy for a moment – reflected in the Figure 76 data by a one-time reduction in phase
and frequency.

164
 40 Run: y f22pio

20

Uns m o oth ed 0
I/O
-20

-40
30 35 40 45
T im e (s ec )
8

F req uen c y 4
(rad /s ec )
2

0
30 35 40 45
T im e (s ec )
20 0

15 0
P has e
(d eg ) 10 0

50

0
30 35 40 45
T im e (s ec )
60

40
P eak
v alues
20

0
30 35 40 45
T im e (s ec )
Figure 76. Expanded Time-Scale Plots for YF-22 PIO

165
 4

P IO
2
severity
1

0
30 35 40 45
T im e ( s ec )
s tic k

q
PIO
F lags phas e

om ega

30 35 40 45
T im e ( s ec )

Figure 77. PIO Flags and Estimated PIO Severity for YF-22 on Expanded Time Scale

After the 40-second point, the pilot re-enters the PIO and the frequency and phase parameters
look as they did for the first segment.
Shortly after the gear-up command at 35 seconds, PIO severity is 3, 3.5, and 4 on consecutive
half-cycles (Figure 77). The first 4 is computed at 36.4 sec, so in a real-time implementation
some active measure to get out of the PIO would have begun at that point – seven seconds before
runway impact at roughly 43 seconds.

c. A Word of Caution About the Results

There is, of course, simply no way to verify that a device such as ROVER would have made any
difference whatsoever in the occurrence of the YF-22A PIO. After all, we have the benefit of
hindsight and the availability of the data of Figure 75 to help determine the thresholds for PIO.
Nevertheless, there is reason to be encouraged from the results of Figure 75. For one thing, the
numbers used in ROVER for the various test thresholds for PIO were set not from the YF-22A
data, but from application of ROVER to the time-history runs from HAVE PIO36 and then
applied to the YF-22A. Perhaps an occasional warning on the first run would have led to a
decision to abandon the second run, or perhaps the thresholds would have been raised long
before because such intermittent warnings were considered annoying by the pilots.

166
3. Application to HAVE PIO Project
ROVER was applied to the landing approach time-history data from the HAVE PIO36 project.
The parameters for filtering, detection, and flagging were identical to those for the YF-22A
described above (they were, in fact, determined from the HAVE PIO data before application to
the YF-22). In most cases, application was straightforward, since the severe-PIO cases were
easily identifiable as such, and the very good cases had no sign of PIO.
There were some exceptions, however: in a number of the landings, ROVER indicated one-half
cycle of PIO occurring at what appeared to be the moment of main gear contact, even for
configurations that were considered to be resistant to PIO.
Since the NT-33A was a variable-stability airplane with a safety pilot, it is possible that the
apparent out-of-phase condition at the moment of contact came from an input by the safety pilot
to take over control of the airplane; the safety pilot’s control information is not available to
verify this. It is also possible that the pitching moment generated by main-gear impact produced
a pitch rate that was, indeed, out of phase with the pilot’s input. In this case, it would indicate a
need to build logic into ROVER to ignore the instant of landing touchdown. For the assessment
that follows, such one-time indications of possible PIO have been removed if they appear to
correlate with touchdown.
For best illustration of the effectiveness of ROVER, it is desirable to plot the PIO severity output
for every single identified half-cycle of oscillation in every run by every pilot, as was done for
the YF-22A in Figure 75. This amounts to literally thousands of data points and makes for an
almost unnecessary overpresentation of data. For the following, because the number of pilots,
configurations, and data runs is small, it was decided instead to tabulate the results of ROVER in
terms of the maximum value of the “PIO severity” metric for each configuration, and compare
these results with the pilot ratings and reported PIOs.

Table 18. Application of ROVER to HAVE PIO36 (Severe PIO Cases are in Italic Font)
Handling Qualities Ratings PIO Ratings Maximum
PIO in ROVER
Conf. A B C A B C PIO
flight? Detect PIO?
Severity
2-B 7 3 3 3 3 2 2 1 No 3 No
2-1 2 2 3 1 1 1 No 2.5 No
2-5 10 7 10 4 4 5 Yes 4 Yes
2-7 7 4 4 4 3 2 Yes 4 Yes
2-8 8 10 8 4 4 4 Yes 4 Yes
3-D 2 2 1 1 No 3 No
3-1 5 3 4 3 2 2 No 3 No
3-3 7 2 3 3 1 1 No 4 Yes (Pilot A)
3-6 5 4 2 2 No 3.5 Yes (Pilot A)
3-8 8 5 8 4 3 4 Yes 4 Yes
3-12 7 9 4 5 Yes 4 Yes
3-13 10 10 4 5 Yes 4 Yes
4-1 3 2 3 1 1 1 No 2.5 No
4-2 3 3 4 1 1 2 No 3 No
5-1 2 5 1 1 No 2.5 No
5-9 7 8 7 4 5 4 Yes 4 Yes
5-10 10 10 5 5 Yes 4 Yes
5-11 7 7 5 2 4 3 Yes 4 Yes

167
Table 18 repeats the pilot ratings for HAVE PIO from Table 2; an added column lists the
maximum value of “PIO severity” from ROVER, for any landing, by any pilot, and a column
indicating whether ROVER detected PIO.
Correlation is excellent, as only two configurations met the ROVER requirements for PIO with
no PIORs of four or higher. One of these configurations, 3-6, exhibited pitch bobble in flight, so
it is possible that the bobble was sufficient to trigger the PIO flags. And for this configuration
the worst PIO severity reading was one full cycle for which three of the four flags were
triggered, resulting in a PIO severity of 3.5. The other apparent failure, Configuration 3-3, was a
PIO severity of 4 for Pilot A, who assigned an HQR of 7 and a PIOR of 3 – indications of a
definite handling-qualities deficiency, with at least undesirable motions, if not PIO.
Conservatively, ROVER was correct on 16 of the 18 HAVE PIO configurations, but in reality, it
was effective on all 18.

4. Application to LAMARS Simulation Data

ROVER was applied to selected configurations from the LAMARS simulation documented in
Appendix A. Those configurations not analyzed were eliminated because they were flown by
few pilots, or because they were intended to look at secondary factors on PIO. Time-history data
from nineteen pitch configurations were processed through the ROVER algorithms as they were
implemented on Matlab. This processing was done post-simulation, not real-time.
Based on the “PIO severity” metric described above, we would expect that, for a typical tracking
run of over two minutes, an airplane that does not exhibit any tendency to PIO should have a
maximum output from ROVER of 3. An output of 3.5 should correspond to a tendency to PIO,
and the expected PIO tendency rating would be a 4, though an occasional rating of 3 or even 5
would not be completely unreasonable. An output from ROVER of 4 should only happen if
there was a real PIO, and then the PIO rating should be a 4, 5, or 6.
ROVER is not perfect, by any means, and there were minor adjustments to improve its
effectiveness for the simulation data. Sometimes, a single PIO severity of 3.5 may occur only
once during a run for an airplane that is generally considered to be resistant to PIO. Sometimes
this occurrence may signal a lurking PIO, but more likely it is a result of the less-than-optimum
operation of the signal filtering and PIO testing methods.
There are literally hundreds of data runs from the LAMARS simulation, making it almost
impossible to find a clear, yet concise, method for presenting the results from ROVER. Simply
tabulating maximum values, as was done for HAVE PIO in Table 18, makes for a compact
presentation, but it does not allow for insight into the circumstances of a particular event. Thus,
a generally good airplane that fails the tests only once may look as bad as a severely PIO-prone
airplane in terms of maximum PIO severity.
Still, even knowledge of the number of times a certain PIO severity value is achieved is not the
whole story, either. For example, the most PIO-prone configurations for Category I PIO in the
simulation may show prolonged periods where the output is a 4, while the severely divergent
Category II configuration might have only a single occurrence of 4 because control is lost in the
first half-cycle.
For this discussion, the data will be shown graphically in a series of crossplots of PIO Tendency
Rating against maximum output for PIO severity. In this way, we can see the effectiveness of

168
ROVER both as a general tool for detecting PIO and specifically as a method for identifying any
interpilot differences.
The data on the crossplots can be interpreted as indicated in the sketch shown in Figure 78, and
in words as follows.
• A PIO-resistant airplane is indicated by a combination of PIOR of 3 or lower and
maximum output of 3 or lower (lower left corner in Figure 78).
• A PIO is indicated by a combination of PIOR 4 or higher and PIO severity of 3.5 or 4
(upper right corner).
• The specific combination of PIOR = 3 and PIO severity = 3.5 may indicate a mild PIO
tendency that is seen by the pilot as nothing more than an undesirable motion.
• Bad PIORs – 4, 5, or 6 – with PIO severity output of 3 or lower indicates a lack of
detectable PIO in the time history – at least, by the definitions programmed in ROVER
(upper left corner of Figure 78). It is possible that the pilot did not actually see PIO
either, but is rating only the tendency. In the overall evaluation of effectiveness, this was
considered a failure of ROVER to detect PIO.
• Bad PIO severity values – 3.5 or 4 – with PIORs of 3 or lower indicates the presence of
the conditions for PIO in the time history – at least, as programmed in ROVER (lower
right corner of Figure 78). It is possible that there was no tendency to PIO, and ROVER
has falsely detected one, or that the pilot was unaware of the PIO for some reason. In any
case, in the evaluation of its effectiveness, this is considered to be a sign of conservatism
and is a failure of ROVER to detect non-PIO.

6
P IO re po rted ,
5
n o t d e te c te d P IO
(R O V E R failu re)
4
P IO R
3 P IO d e te c te d ,
N o P IO no t re p orte d
2 (R O V E R
fa ilure )
1
1 2 3 4
M a x O u tpu t

Figure 78. Key for Interpreting ROVER Summary Data Plots

169
A summary plot for the 19 selected configurations is shown in Figure 79. The top two
configurations in Figure 79, 17 and 2DR30, were intended to be baseline PIO-resistant pitch
configurations. Twice for Configuration 17, and once for 2DR30, the conditions for mild
PIO (PIO severity = 3.5) were found. Also, one pilot assigned a PIOR of 4 for one
evaluation of 2DR30, and this case indeed showed the signs of a PIO, with a PIO severity
output of 4.
The remaining configurations in the first column of Figure 79 all show the combined effect
of actuator rate limiting and basic vehicle stability. Configuration 2DR10 is the baseline
with a more restricted rate limit of 10 deg/sec, and while no pilot assigned a PIOR of 4
several runs were detected as exhibiting PIOs. This case was expected to have a mild PIO
tendency, but the extremely low rate limits made it difficult to maneuver this airplane in the
first place, so PIO was not likely to be seen by the pilot even if it occurred.
Configuration 2DUR20 is very similar to 2DU from HAVE LIMITS21 with a 20 deg/sec rate
limit – a highly PIO-susceptible airplane with severe divergence tendencies. Only the low-
crossover pilot, Pilot D, managed to fly this configuration through the entire task without loss
of control, and his PIO severity output was a 2.5. In several instances the departure occurred
so quickly that it was not possible to identify a fully-developed PIO, and the maximum
output from the PIO severity metric was a 3.
For the 2DV cases in the first column of Figure 79, as the rate limit is reduced from 20
(2DVR20) to 15 (2DVR15) and then 10 deg/sec (2DVR10), more pilots experience divergent
PIOs and the maximum output values all migrate towards a 4. Occasionally the conditions for
the PIO were encountered but the pilots assigned PIORs of 2 (2DVR20, Pilot G) or 3 (2DVR15,
Pilot C).
The cases in the middle column of Figure 79 are the baseline with added lags, either first-order
(2F, 2P, 2H) or second-order (2DL1 through 2DL4). For the former cases, several runs showed
occasions of mild-to-moderate PIO (PIO severity = 3.5) but no pilot considered any of them to
actually have PIO tendencies. Configurations 2F and 2H were based on similar cases in the
Neal-Smith in-flight experiment,51 where 2H showed a tendency to PIO. For the cases with
added second-order lags, the trends are quite consistent with no outliers at all.
The third column of data plots in Figure 79 shows the output from ROVER as short-period
damping is reduced from the baseline value of 0.7 to roughly 0.4 (Configuration 15) to 0.18 (5A)
to 0.009 (16). Configuration 1D has lower short-period frequency, 2.2 rad/sec in place of the
baseline frequency of 4.6 rad/sec, and 1DL2 and 1DL3 are 1D with added second-order lags.
As short-period damping is reduced, reports of PIO and detection of PIO both increase. For the
most part, the low-frequency cases 1D through 1DL3 have results similar to 2DL1 through 2DL4
in the second column.

5. Effectiveness for the LAMARS Data

While ROVER, as it was applied to the LAMARS simulation data, was not considered to be a
mature design by any means, it is still reasonable to ask: how effective was ROVER in correctly
identifying PIO?

170
 6
17 Legend for all charts:
5
A B C D
4
PIOR

3 E F G H
2
1
1 2 3 4
6 6 6
2DR30 2F 15
5 5 5
4 4 4
PIOR

3 3 3
2 2 2
1 1 1
1 2 3 4 1 2 3 4 1 2 3 4
6 6 6
2DR10 2P 5A
5 5 5
4 4 4
PIOR

3 3 3
2 2 2
1 1 1
1 2 3 4 1 2 3 4 1 2 3 4
6 6 6
2DUR20 2H 16
5 5 5
4 4 4
PIOR

3 3 3
2 2 2
1 1 1
1 2 3 4 1 2 3 4 1 2 3 4
6 6 6
2DVR20 2DL2 1D
5 5 5
4 4 4
PIOR

3 3 3
2 2 2
1 1 1
1 2 3 4 1 2 3 4 1 2 3 4
6 6 6
2DVR15 2DL3 1DL2
5 5 5
4 4 4
PIOR

3 3 3
2 2 2
1 1 1
1 2 3 4 1 2 3 4 1 2 3 4
6 6 6
2DVR10 2DL4 1DL3
5 5 5
4 4 4
PIOR

3 3 3
2 2 2
1 1 1
1 2 3 4 1 2 3 4 1 2 3 4
Max Output Max Output Max Output

Figure 79. Maximum “PIO Severity” Output from ROVER for Selected Configurations
from LAMARS Simulation (Configuration Number in Boxes on Charts)

171
For this, we will look entirely at PIO tendency rating as a measure of PIO. This is not always a
reasonable assumption, because, as discussed in Section VIII, there are times where a PIOR of 4
may not indicate a real PIO, and where a PIOR of 3 may indicate a tendency to PIO. Still, for
the LAMARS simulation the pilots were briefed carefully about what constitutes PIO as opposed
to bobble, so we may with some confidence take the ratings at face value. In this context, a
PIOR of 4, 5, or 6 indicates a PIO, a 1 or 2 indicates no tendency to PIO, and a 3 indicates no
PIO but a possible tendency. For the following, we have further taken a PIOR of 3 to suggest no
tendency to PIO.
There are 160 evaluations in the data plots of Figure 79. Figure 78 defined four possible
combinations of PIO severity flag and PIO tendency rating, two desirable and two undesirable.
No PIO reported or detected (good): There were 72 of these.
PIO both reported and detected (good): There were 49.
PIO detected but not reported (bad): There were 28 (18 identified as “moderate”
and 10 as “severe”), but for 15 of these the PIOR was a 3, suggesting there were
definitely undesirable motions.
PIO reported but not detected (bad): There were 11 in this category. Five of the
11 were for the low-damped cases (15, 5A, and 16 on Figure 79), where very
small pilot inputs could result in large oscillations, and four were for 2DUR20,
which was so highly divergent that pilots sometimes did not have the time to
apply large inputs before the simulation was halted. All 11 instances were for
configurations considered to be susceptible to PIO by other pilots as well.
The overall effectiveness, counting the successful applications, is (72+49)/160 = 76%. This
is certainly a respectable rate, given the relatively simple approach. ROVER is slightly
conservative, over-reporting cases not given poor PIO ratings (28) more than it under-
reported PIO cases (11). ROVER was effective at detecting PIO when it occurred, however:
there were 60 PIO runs, and ROVER found PIO in 49 of them, for a success rate of 82%.
When considered on a case-by-case basis, ROVER was even more effective (as shown by
Figure 79). On those configurations where PIO was detected but not reported, it was
common for the airplane to actually be very susceptible to PIO. For example, Configurations
2DVr20 and 2DVR15 (first column in Figure 79) had a total of four apparent outliers where
PIORs were 3 or better and maximum PIO severity was 3.5 or 4. Both, and 2DVR15
especially, were highly susceptible to divergent PIOs. For 2DVR15, Pilot commented in one
run that he was “avoiding a PIO” and assigned a PIOR of 3. This case was repeated later,
and Pilot C gave a PIOR of 4 the second time, so the 3 should be considered a hit, rather than
a miss. There are certainly other similar isolated cases among the entire simulation data
base.

F. THE NEXT STEPS

If ROVER is to become a “real-time” tool, several steps must be undertaken.
• Allowance for other states: PIO can occur, and has occurred, where a state other than
pitch rate is the primary out-of-phase parameter. Load factor is an obvious example. The
heavy reliance on pitch rate, and its success in the limited application of ROVER in this
report, is a result of the characteristics of our data base, which comes entirely from

172
 relatively low-speed operations. Low-altitude, high-speed flight can also cause PIO, and
the likely state in that case is load factor. Data and PIO thresholds are needed to
incorporate load factor into ROVER.
• Adjustment of parameters with flight condition: ROVER has been used successfully on
data for both landings and up-and-away flight, but with differing thresholds for PIO. A
comprehensive method must include provision for changing the thresholds with flight
condition, and possibly even task or configuration. For example, tighter constraints
might be desired for aerial refueling than for air combat, even if the operating conditions
are sometimes similar.
• Extension to other axes: The focus on pitch in both flight research and criteria
development has left our knowledge base for other axes somewhat lacking. Thresholds
for roll, in particular, need to be developed, but the data necessary to develop them may
not exist at this time.
• Protection for data spikes, dropout, and noise: The time-history data applied to
ROVER have been either fairly clean or perfectly clean, with no evidence of data
problems that are common for real-world operations. The filtering necessary to account
for these types of contamination should be incorporated before ROVER is evaluated in a
real-time environment.
• A real-time test: ROVER has yet to be applied real-time. The current method seems to
be quite effective but it needs to be tried in a real environment.

173
 X. RECOMMENDATIONS FOR FUTURE WORK

A. PIO PREVENTION CRITERIA

1) Criteria for Category I PIOs in axes other than pitch: Criteria based on pitch attitude and
flight path Bandwidth have proven to be very effective at predicting the occurrence of pitch-
axis Category I PIOs. There has been relatively little work to define criteria for the
prevention of roll, yaw, or even axial (thrust) axis PIOs. This work will require basic flight
research to gather the necessary data base.
2) Practical methods for extending Bandwidth criteria to elevated load factors: Work remains
to be done to verify methods for obtaining the necessary data, from either simulation or
flight, to test for possible nonlinear contributors to PIO. For example, is a control-input
frequency sweep still the best forcing function form when the criteria are to be tested at high
load factors? What are the challenges to obtaining such data, and how can these challenges
be overcome?
3) Criteria for characteristics of the pilot’s control interface: Essentially all of the existing
flight research data base has been generated assuming ideal input/response characteristics.
This includes the control/response sensitivity (measured in terms of pitch acceleration or load
factor per pound of force in pitch); linearity of response; inceptor breakout, centering,
gradients, and displacements. All of these characteristics have contributed to several PIOs –
most of them mild or moderate in severity – but none has been systematically investigated.
Their interaction with basic aircraft dynamics, especially, could be critical and should be
studied.

B. TESTING FOR PIO

1) Maneuvers to expose susceptibility to PIO: Most PIOs in flight research have been
discovered during the performance of handling-qualities tasks. While this is certainly
important, since PIOs are one manifestation of poor handling qualities, it may not be
sufficient to expose dangerous PIOs that require more than a normal level of pilot activity.
On the other hand, PIO-specific tasks such as Handling Qualities During Tracking (HQDT)
have not been rigorously tested to verify that they are any better. A series of flight research
efforts, using a variable-stability airplane, could determine the most appropriate maneuvers
to expose susceptibility to PIO.
2) Data for Category III PIOs: There is a wealth of information on the causes of Category I
(linear) PIO, and a much smaller data base for Category II (rate-limiting) PIO, but
essentially no data for Category III (such contributors as mode switching, failure transients,
etc.) PIOs. As this report suggests, in reality Category III PIO may always be preceded by
identifiable Category I or II events. Although there are stories of PIOs that have resulted
entirely from inappropriate mode switching, any documented cases show that the basic
airplane was usually susceptible to Category I PIO to begin with, and the fixes have been to
both the switching logic and the basic dynamics of the airplane. A flight research effort
should be conducted to investigate the unique character of Category III PIO.

174
C. REAL-TIME PIO DETECTION AND PREVENTION
1) Devices for real-time detection of PIO in flight: The Realtime Oscillation VERifier
(ROVER) has never been tested in a real-time environment. It shows considerable promise,
given its simplicity, but the final proof is its application. In addition to its value as a PIO
detector, ROVER can assist in flight testing by providing a quick-look indication of data
quality from open-loop test inputs. To be fair, ROVER should be compared against other
similar real-time detection methods to determine which is most effective.
2) Devices for prevention of PIO in flight: Ultimately, ROVER should be incorporated into a
package that includes passive warning to the pilot of a developing moderate PIO, and active
intervention if the PIO is severe. But this is true of any PIO detection method, and the
specifics of the warning and intervention need to be defined. For example, an automatic fly-
up command that locks out the pilot’s commands may be perfectly appropriate for a PIO on
an inherently stable airplane during landing. But it could be worse than the PIO itself if the
unaugmented airplane is unstable and on rate limits, or if it is under a tanker receiving fuel
when the PIO occurs.
3) Expansion of ROVER to incorporate other aircraft states: The implementation of ROVER
as described in this report consists entirely of tests on pitch rate. There are some known
shortcomings with using rate information alone, and other response states, such as normal
acceleration, should be merged into the algorithms. Proper methods for doing so, how to
weight different states, etc., must be developed.
4) Effects of noise, excessive signal filtering, signal sample rates, etc., on effectiveness of PIO
detection devices: In reality, gyros and accelerometers on modern airplanes are not always
of highest quality. If the sensors are flight-critical, i.e., if the airplane uses full-time
augmentation to obtain certification or acceptance, the quality must be better, with multiply-
redundant sensor packages. Still, in all cases there can be output noise that usually requires
conditioning of the sensor signal before it is used by the flight control computers. Such
filtering itself can adversely impact performance of PIO detection devices. If the sensors are
to be unfiltered or minimally filtered, the likelihood for occasional data spikes and noise
increases. A sensitivity analysis must be performed on any potential detection device before
it can be considered mature enough for practical implementation.
5) Comparative Evaluation of PIO Suppression Methods: Engineers at McDonnell Douglas
(now Boeing) investigated improvements to the PIO suppression filter developed for the
Space Shuttle. Their intent was to look at the use of the filter on the C-17A. SAAB
engineers have implemented software on the Gripen that protects against hitting rate limits.
Both of these forms of PIO suppression may be effective in their own right, or perhaps as a
part of an integrated PIO prevention device. In any case, the improved PIO suppression
filter has not been evaluated real-time, and it should be compared with the SAAB approach
to determine the strengths and weaknesses of each.

175
 REFERENCES

1
Aviation Safety and Pilot Control: Understanding and Preventing Unfavorable Pilot-Vehicle
Interactions, National Academy Press, Washington D.C., 1997.
2
Dornheim, Michael A., “Report Pinpoints Factors Leading to YF-22 Crash,” Aviation Week &
Space Technology, 9 Nov. 1992, pp. 53-54.
3
Harris, Jeffrey Jay, and G. Thomas Black, “F-22 Control Law Development and Flying
Qualities,” AIAA-96-3379-CP, AIAA Atmospheric Flight Mechanics Conference
Proceedings, July 1996, pp. 155-168.
4
Department of Defense Interface Standard, Flying Qualities of Piloted Aircraft, MIL-STD-
1797A, Jan. 1990; Notice of Change, 28 June 1995.
5
Hoh, Roger H., David G. Mitchell, and John Hodgkinson, “Bandwidth – A Criterion for
Highly Augmented Airplanes,” Criteria for Handling Qualities of Military Aircraft,
AGARD-CP-333, Apr. 1982, pp. 9-1 – 9-11.
6
Hodgkinson, J., W. J. LaManna, and J. L. Heyde, "Handling Qualities of Aircraft with Stability
and Control Augmentations Systems - A Fundamental Approach," Aeronautical Journal,
Feb. 1976, pp. 75-81.
7
Gibson, J C, “The Prevention of PIO by Design,” Active Control Technology: Applications and
Lessons Learned, AGARD-CP-560, Jan. 1995, pp. 2-1 – 2-12.
8
Smith, Rogers E., and Shahan K. Sarrafian, "Effect of Time Delay on Flying Qualities: an
Update," J. Guidance, Control, and Dynamics, Vol. 9, No. 5, Sept.-Oct. 1986, pp. 578-584.
9
Bailey, R. E., and L. H. Knotts, Interactions of Feel System and Flight Control System
Dynamics on Lateral Flying Qualities, NASA CR-179445, Dec. 1990.
10
Mitchell, David G., Bimal L. Aponso, and David H. Klyde, Effects of Cockpit Lateral Stick
Characteristics on Handling Qualities and Pilot Dynamics, NASA CR-4443, June 1992.
11
McRuer, D. T., and E. S. Krendel, Mathematical Models of Human Pilot Behavior, AGARD
AG-188, Jan. 1974.
12
Ashkenas, Irving L., Henry R. Jex, and Duane T. McRuer, Pilot-Induced Oscillations: Their
Cause and Analysis, Norair Report NOR-64-143, July 1964.
13
Smith, R. H., A Theory for Handling Qualities with Applications to MIL-F-8785B, AFFDL-
TR-75-119, Oct. 1975.
14
Smith, Ralph H., A Theory for Longitudinal Short-Period Pilot Induced Oscillations, AFFDL-
TR-77-57, June 1977.
15
Klyde, David H., Duane T. McRuer, and Thomas T. Myers, Unified Pilot-Induced Oscillation
Theory, Volume I: PIO Analysis with Linear and Nonlinear Effective Vehicle
Characteristics, Including Rate Limiting, WL-TR-96-3028, Dec. 1995.
16
Preston, Jeff D., John Hodgkinson, Jim Buckley, Kevin Citurs, David G. Mitchell, and Roger
H. Hoh, Unified Pilot-Induced Oscillation Theory, Volume II: Pilot-Induced Oscillation
Criteria Applied to Several McDonnell Douglas Aircraft, WL-TR-96-3029, Dec. 1995.
17
Anderson, Mark R., and Anthony B. Page, Unified Pilot-Induced Oscillation Theory, Volume
III: PIO Analysis Using Multivariable Methods, WL-TR-96-3030, Dec. 1995.

176
18
Bailey, Randall E., and Timothy J. Bidlack, Unified Pilot-Induced Oscillation Theory Volume
IV: Time-Domain Neal-Smith Criterion, WL-TR-96-3031, Dec. 1995.
19
Investigation of Pilot Induced Oscillation Tendency and Prediction Criteria Development,
WL-TR-96-3109, May 1996.
20
Nguyen, Ba T., Thomas J. Cord, and David B. Leggett, Comparisons of Flight to Ground-
Based Pilot-Induced Oscillation Evaluation Methods, AFRL-VA-WP-TR-1998-3032, May
1998.
21
Kish, Brian A., Capt., USAF, et al., A Limited Flight Test Investigation of Pilot-Induced
Oscillation due to Elevator Rate Limiting, AFFTC-TR-97-12, June 1997.
22
Markofski, Andrew R., and Randall E. Bailey, “Analysis of Pitch Pilot-Induced Oscillation
Tendencies Due to Actuator Rate Limiting,"”AIAA-99-4005, AIAA Atmospheric Flight
Mechanics Conference Proceedings, Portland, OR, Aug. 1999, pp. 20-30.
23
Schroeder, Jeffrey A., et al., “Pilot-Induced Oscillation Prediction with Three Levels of
Simulation Motion Displacement,” AIAA Atmospheric Flight Mechanics Conference
Proceedings, Aug. 1998, pp. 390-400.
24
Hess, R. A., “Unified Theory for Aircraft Handling Qualities and Adverse Aircraft-Pilot
Coupling,” J. Guidance, Control, and Dynamics, Vol. 20, No. 6, Nov.-Dec. 1997, pp. 1141-
1148.
25
Nolan, Michael, Capt., USAF, et al. “Evaluation of Operational Closed Loop Handling Tasks
for Use in MIL-STD-1797,” USAF Test Pilot School, viewgraphs from final briefing, 1993.
26
Fedors, David C., Capt. USAF, et al., Limited Evaluation of Proposed Closed-Loop Handling
Qualities Tasks for MIL-STD-1797B, AFFTC-TLR-93-36, Dec. 1993.
27
Kelly, James M., Capt., USAF, et al., A Limited Evaluation of Closed-Loop Handling
Qualities Tasks for MIL-STD-1797B, AFFTC-TLR-94-30, May 1994.
28
Cooper, George E., and Robert P. Harper, Jr., The Use of Pilot Rating in the Evaluation of
Aircraft Handling Qualities, NASA TN D-5153, Apr. 1969.
29
Klyde, David H., and David G. Mitchell, Handling Quality Demonstration Maneuvers for
Fixed-Wing Aircraft, Volume II: Maneuver Catalog, WL-TR-97-3100, Oct. 1997.
30
Hall, G. Warren, and Robert P. Harper, Jr., "In-Flight Simulation of the Light Weight
Fighters," AIAA Paper No. 75-985, Aug. 1975.
31
Taylor, Lawrence W., Jr., Analysis of a Pilot-Airplane Lateral Instability Experienced with the
X-15 Airplane, NASA TN D-1059, Nov. 1961.
32
Smith, John W., Analysis of a Lateral Pilot-Induced Oscillation Experienced on the First
Flight of the YF-16 Aircraft, NASA TM-72867, Sept. 1979.
33
Kullberg, E., and Per-Olov Elgcrona, “SAAB Experience with PIO,” Flight Integration Panel
Workshop on Pilot Induced Oscillations, AGARD-AR-335, Feb. 1995, pp. 9-1 – 9-9.
34
McRuer, Duane T., Pilot-Induced Oscillations and Human Dynamic Behavior, NASA CR-
4683, July 1995.
35
Newell, Fred, Investigation of Phugoid Effects on GCA Landings, WADC TR 57-650, Dec.
1957.

177
36
Bjorkman, Eileen A., Capt., USAF, Capt. Richard L. Bennett, Capt. David W. Eidsaune, Capt.
Seiichi Miyamoto, and 1LT Oliver C. Wilkinson, NT-33 Pilot Induced Oscillation Prediction
Evaluation, USAFTPS-TR-85B-S4, June 1986.
37
Bjorkman, Eileen A., Capt., USAF, Flight Test Evaluation of Techniques to Predict
Longitudinal Pilot Induced Oscillations, Master's Thesis, AFIT/GAE/AA/86J-1, Dec. 1986.
38
Smith, John W., and Terry Montgomery, Biomechanically Induced and Controller Coupled
Oscillations Experienced on the F-16XL Aircraft During Rolling Maneuvers, NASA TM
4752, July 1996.
39
Hoh, Roger H., David G. Mitchell, Irving L. Ashkenas, Richard H. Klein, Robert K. Heffley,
and John Hodgkinson, Proposed MIL Standard and Handbook -- Flying Qualities of Air
Vehicles, Volume II: Proposed MIL Handbook, AFWAL-TR-82-3081, Nov. 1982.
40
Smith, Ralph H., and Norman D. Geddes, Handling Quality Requirements for Advanced
Aircraft Design: Longitudinal Mode, AFFDL-TR-78-154, Aug. 1979.
41
Mitchell, David G., Roger H. Hoh, Bimal L. Aponso, and David H. Klyde, “The Measurement
and Prediction of Pilot-in-the-Loop Oscillations,” AIAA-94-3670-CP, AIAA Guidance,
Navigation and Control Conference, Scottsdale, AZ, Aug. 1994, pp. 1167-1177.
42
Gibson, J.C., Development of a Methodology for Excellence in Handling Qualities Design for
Fly by Wire Aircraft, Delft University Press, 1999.
43
Smith, Rogers E., Effects of Control System Dynamics on Fighter Approach and Landing
Longitudinal Flying Qualities, AFFDL-TR-78-122, Mar. 1978.
44
Klyde, David H., Bimal L. Aponso, and David G. Mitchell, Handling Quality Demonstration
Maneuvers for Fixed-Wing Aircraft, Volume I: Maneuver Development Process, WL-TR-
97-3099, Oct. 1997.
45
Mitchell, David G., Roger H. Hoh, Bimal L. Aponso, and David H. Klyde, Proposed
Incorporation of Mission-Oriented Requirements into MIL-STD-1797A, WL-TR-94-3162,
Oct. 1994.
46
McRuer, D. T. and H. R. Jex, “A Review of Quasi-Linear Pilot Models,” IEEE Transactions
on Human Factors in Electronics, Vol. HFE 8, Number 3, Sept. 1967.
47
Iloputaife, Obi, “Minimizing Pilot-Induced-Oscillation Susceptibility During C-17
Development,” AIAA Flight Mechanics Conference, New Orleans, LA, Aug. 1997, pp. 155-
169.
48
Nelson, Tim A., and Ross A. Landes, “Boeing 777 Development and APC Assessment,”
presented at the SAE Control & Guidance Systems Committee Meeting, Mar. 1996.
49
Deppe, P.R., C.R. Chalk, and M. Shafer, Flight Evaluation of an Aircraft with Side and
Centerstick Controllers and Rate-Limited Ailerons, Calspan Final Report No. 8091-2, Apr.
1994.
50
Mitchell, David G., and Roger H. Hoh, “Low-Order Approaches to High-Order Systems:
Problems and Promises,” J. Guidance, Control, and Dynamics, Vol. 5, No. 5, Sept.-Oct.
1982, pp. 482-489.
51
Neal, T. Peter, and Rogers E. Smith, An In-Flight Investigation to Develop Control System
Design Criteria for Fighter Airplanes, AFFDL-TR-70-74, Volume I, Dec. 1970.

178
52
Mitchell, David G., and Roger H. Hoh, “Development of a Unified Method to Predict PIO,”
AIAA Atmospheric Flight Mechanics Conference Proceedings, San Diego, CA, July 1996,
pp. 611-622.
53
Efremov, Alexander V., et al., “Analysis of Reasons for Pilot Induced Oscillation Tendency
and Development of Criteria for its Prediction,” Investigation of Pilot Induced Oscillation
Tendency and Prediction Criteria Development, WL-TR-96-3109, May 1996.
54
Smith, Ralph H., “The Smith-Geddes Criteria,” presented at the SAE Aerospace Control and
Guidance Systems Committee meeting, Reno, NV, Mar. 1993.
55
Gibson, John C., “Piloted Handling Qualities Design Criteria for High Order Flight Control
Systems,” Criteria for Handling Qualities of Military Aircraft, AGARD CP-333, Apr. 1982,
pp. 4-1 – 4-15.
56
Kivioja, David A., et al., A Limited Evaluation of Predicting Pilot Opinion of Aircraft
Handling Qualities in the Landing Phase of Flight Using the Control Anticipation Parameter
and Bandwidth Criterion (HAVE CAP), AFFTC-TR-95-78, Jan. 1996.
57
Field, Edmund J., and Ken F. Rossitto, “Approach and Landing Longitudinal Flying Qualities
for Large Transports Based on In-Flight Results,” AIAA-99-4095, AIAA Atmospheric Flight
Mechanics Conference, Portland, OR, Aug. 1999, pp. 219-230.
58
Hoh, Roger H., et al., Development of Handling Quality Criteria for Aircraft with Independent
Control of Six Degrees of Freedom, AFWAL-TR-81-3027, Apr. 1981.
59
Mitchell, David G., and Richard J. DiMarco, Analysis of Industry and Government Comments
on Bandwidth, Systems Technology, Inc., Technical Memorandum TM-1163-1, May 1985.
60
Chalk, Charles R., Dante A. DiFranco, J. Victor Lebacqz, and T. Peter Neal, Revisions to MIL-
F-8785B (ASG) Proposed by Cornell Aeronautical Laboratory Under Contract F33615-71-
C-1254, AFFDL-TR-72-41, April 1973.
61
Chalk, C. R., Calspan Recommendations for SCR Flying Qualities Design Criteria, NASA
CR-159236, Apr. 1980.
62
Radford, R. C., R. E. Smith, and R. E. Bailey, Landing Flying Qualities Evaluation Criteria
for Augmented Aircraft, NASA CR-163097, Aug. 1980.
63
Smith, Rogers, John Hodgkinson, and Richard C. Snyder, Equivalent System Verification and
Evaluation of Augmentation Effects on Fighter Approach and Landing Flying Qualities,
Volume II -- Program Plan, Test Data and Analysis, AFWAL-TR-81-3116, Sept. 1981.
64
Mooij, H. A., Criteria for Low-Speed Longitudinal Handling Qualities of Transport Aircraft
with Closed-Loop Flight Control Systems, NLR TR 83037 U, Sept. 1984.
65
Weingarten, Norman C., and Charles R. Chalk, In-Flight Investigation of Large Airplane
Flying Qualities for Approach and Landing, AFWAL-TR-81-3118, Sept. 1981.
66
Berthe, C. J., C. R. Chalk, and S. Sarrafian, Pitch Rate Flight Control Systems in the Flared
Landing Task and Design Criteria Development, NASA CR-172491, Oct. 1984.
67
Mitchell, David G., Bimal L. Aponso, and Roger H. Hoh, Minimum Flying Qualities, Volume
I: Piloted Simulation Evaluation of Multiple Axis Flying Qualities, WRDC-TR-89-3125,
Volume I, Jan. 1990.
68
Smith, Ralph H., Note on Lateral-Directional Pilot Induced Oscillations, AFWAL-TR-81-
3090, Mar. 1982.

179
69
McRuer, Duane, Dunstan Graham, Ezra Krendel, and William Reisner, Jr., Human Pilot
Dynamics in Compensatory Systems: Theory, Models, and Experiments with Controlled
Element and Forcing Function Variations, AFFDL-TR-65-15, July 1965.
70
Onstott, E.D., and W.H. Faulkner, Prediction, Evaluation, and Specification of Closed Loop
and Multiaxis Flying Qualities, AFFDL-TR-78-3, Feb. 1978.
71
Lindsey, Steven W., Prediction of Longitudinal Pilot Induced Oscillations Using the Optimal
Control Model, AFIT/GAE/ENY/90M-1, Dec. 1989.
72
Twisdale, Thomas R. and Paul W. Kirsten, “Prediction and Occurrence of Pilot-Induced
Oscillations in a Flight Test Aircraft,” J. Guidance, Control & Dynamics, Vol. 7, No. 4, July-
Aug. 1984, pp. 410-415.
73
Givens, Margo L., Evaluation of B-2 Susceptibility to Pilot-Induced Oscillations, Northrop
Grumman white paper, 1 Mar. 1994.
74
Aircraft Accident Report, Inadvertent In-Flight Slat Deployment, China Eastern Airlines
Flight 583, National Transportation Safety Board, NTSB/AAR-93/07, Oct. 1993.
75
Craig, S. J., and I. L. Ashkenas, Background Data and Recommended Revisions for MIL-F-
8785B(ASG), “Military Specification – Flying Qualities of Piloted Aircraft”, Systems
Technology, Inc., TR-189-1, Mar. 1971.
76
A’Harrah, Ralph C., and Raymond F. Siewert, “Pilot-Induced Instability,” Stability and
Control, Part 2, AGARD CP No. 17, Sept. 1966, pp. 703-727.
77
Flight Test Guide for Certification of Transport Category Airplanes, Federal Aviation
Administration Advisory Circular, AC 25-7A, Mar. 1998.
78
Duda, Holger, “Prediction of Pilot-in-the-Loop Oscillations due to Rate Saturation,” J.
Guidance, Control and Dynamics, Vol. 20, No. 3, May-June 1997.
79
Duus, Gunnar, and Holger Duda, “Analysis of the HAVE LIMITS Database using the OLOP
Criterion,” AIAA 99-4007, AIAA Atmospheric Flight Mechanics Conference, Portland, OR,
Aug. 1999.
80
Duda, Holger, and Gunnar Duus, New Handling Qualities Database on PIO due to Rate
Saturation, DLR Forschungsbericht 97-53, 1997.
81
Zeyada, Yasser, and Ronald A. Hess, PVDNL – Pilot/Vehicle Dynamics NonLinear, Dept. of
Mechanical and Aeronautical Engineering, University of California, Davis, undated.
82
Bailey, Randall E., and Paul T. Schifferle, “Lateral Time Domain Neal-Smith Criterion and
PIO Metric Development, AIAA Atmospheric Flight Mechanics Conference and Exhibit,
Boston, MA, Aug. 1998, pp. 438-448.
83
Foringer, Lori A., and David B. Leggett, “An Analysis of the Time-Domain Neal-Smith
Criterion,” AIAA Atmospheric Flight Mechanics Conference and Exhibit, Boston, MA, Aug.
1998, pp. 154-162.
84
Zeyada, Yasser, Ronald A. Hess, and Wichai Siwakosit, “Analysis of Aircraft Handling
Qualities and Pilot-Induced Oscillation Tendencies with Actuator Saturation,” AIAA
Atmospheric Flight Mechanics Conference and Exhibit, Boston, MA, Aug. 1998, pp. 401-
414.
85
Tischler, M. B., and M. G. Cauffman, “Comprehensive Identification from Frequency
Responses (CIFER),” USAATCOM TR-94-A-017, 1994.

180
86
Mitchell, David G., Brian A. Kish, John S. Seo, and William B. Mosle III, “A Flight
Investigation of Pilot-Induced Oscillation Due to Rate Limiting,” IEEE 1998 Aerospace
Conference Proceedings, Paper 270, Snowmass, CO, Mar. 1998.
87
Graham, Dunstan, and Duane McRuer, Analysis of Nonlinear Control Systems, Dover
Publications, New York, 1971.
88
Monagan, Steven J., Rogers E. Smith, and Randall E. Bailey, “Lateral Flying Qualities of
Highly Augmented Fighter Aircraft,” AFWAL-TR-81-3171, June 1982.
89
Mitchell, David G., and Roger H. Hoh, “Influence of Roll Command Augmentation Systems
on Flying Qualities of Fighter Aircraft,” J. Guidance, Control, and Dynamics, Vol. 7, No. 1,
Jan.-Feb. 1984, pp. 99-105.
90
Deppe, Paul R., Flight Evaluation of a Software Rate Limiter Concept, Calspan Final Report
No. 8091-1, Mar. 1993.
91
Ohmit, Eric E., NT-33A In-Flight Investigation into Flight Control System Rate Limiting,
Calspan Final Report No. 7738-24, Feb. 1994.
92
Klyde, David H., Bimal L. Aponso, David G. Mitchell, and Capt. Kelly J. Latimer, USAF,
“Development of Demonstration Maneuvers for Aircraft Handling Qualities Evaluation,”
AIAA Atmospheric Flight Mechanics Conference Proceedings, Aug. 1997, pp. 485-495.
93
McPherson, T. M., and LT E. J. Hogan, Jr., USN, Low Altitude High Speed Flight Evaluation
of the F-4A/B (F4H-1F/-1) Airplane, Report No. 1, Final Report, Naval Air Test Center,
Project No. RAD32-103, Apr. 1963.
94
Kish, Brian A. Capt., USAF, David B. Leggett, Ba T. Nguyen, and Thomas J. Cord, “Concepts
for Detecting Pilot-Induced Oscillation Using Manned Simulation,” AIAA Atmospheric
Flight Mechanics Conference Proceedings, July 1996, pp. 559-568.
95
Stevens, S. S., “The Psychophysics of Sensory Function,” Sensory Communication, Walter A.
Rosenblith, ed., MIT Press, Cambridge, MA, 1961, pp. 1-33.
96
Mitchell, David G., and Bimal L. Aponso, “Reassessment and Extensions of Pilot Ratings
With New Data,” AIAA 90-2823, Aug. 1990.
97
Johnston, Donald E., David G. Mitchell, and Thomas T. Myers, Investigation of High-Angle-
of-Attack Maneuver-Limiting Factors, Part I: Analysis and Simulation, AFWAL-TR-80-
3141, Dec. 1980.
98
McDonnell, John D., “An Application of Measurement Methods to Improve the Quantitative
Nature of Pilot Rating Scales,” IEEE Transactions on Man-Machine Systems, Vol. MMS-10,
No. 3, Sept. 1969, pp. 81-92.
99
Lazerson, David A., and Abigail L. Reuter, “The AFFTC Test Safety Review Process – A
Tailored System Safety Approach,” Society of Experimental Test Pilots Thirty-Ninth
Symposium Proceedings, 1995, pp. 224-243.
100
Nguyen, Ba T., and Thomas J. Cord, “Comparison and Analysis of Pilot-Induced Oscillation
Characteristics from Flight Test and Ground-Based Simulation,” AIAA Atmospheric Flight
Mechanics Conference Proceedings, Aug. 1999, pp. 12-19.

181
182
 APPENDIX A. PILOTED SIMULATION

A. INTRODUCTION
A piloted simulation was conducted on the Air Force’s Large Amplitude Multimode Aerospace
Research Simulator (LAMARS) at Wright-Patterson AFB, Ohio. The simulation was performed
over two periods, with the bulk of the testing in April 1998, and a brief three-day session in
February 1999. The second session allowed a unique opportunity to revisit some of the issues
that were raised in the first simulation, and it provided an opportunity to look at the effects of
design variables on susceptibility to PIO. A total of 478 data runs were made, 424 in the 1998
session and an additional 54 in 1999.

B. OBJECTIVES
There were two primary and four secondary objectives for the simulation. The primary
objectives were to:
1. Investigate the pitch attitude Bandwidth criteria for Category I PIOs. Data gathered for this
objective helped to check the boundaries for PIO and provided a foundation for the next
objective.
2. Extend the Category I Bandwidth criteria to Category II PIOs.
The four secondary objectives were to investigate:
Piloting differences for aircraft that have varying levels of susceptibility to PIO.
Effects of simulator motion on PIO susceptibility.
Effects of control/response sensitivity on PIO susceptibility.
Effects of controller force/deflection gearing on PIO susceptibility.

C. FACILITY
LAMARS provided five degrees of motion freedom (pitch, roll, yaw, heave, and sway) on a 30-
ft arm with displacements of ±10 ft vertical and lateral and ±25 deg pitch, roll, and yaw. The
motion drive was capable of producing instantaneous accelerations of ±3g vertical and ±1.6g
lateral.
A wide-angle visual system projected an out-the-world CGI image onto a 20-ft diameter dome.
The cockpit was a fighter configuration, with variable-feel center stick and rudder pedals. The
pedals and throttle controls were not used for this simulation.
A heads-up display (HUD) based on the HUD in the Air Force’s variable-stability NT-33A
provided basic attitude, altitude, airspeed, and heading information (Figure A-1). A fixed set of
brackets was added to the HUD for this simulation to define the desired tracking area, and the
HUD tracking error was displayed by a moving error bar as sketched in Figure A-1. This bar
was driven by the error between actual and commanded pitch and roll attitudes. The tasks are
described in more detail below.

183
 Error Tracking
Bar 05
Desired Tracking
05
Area
50
15,5
40
15,0
Airspeed
30
14,5
Altitude

Flight Path 05 17 180 19

Marker
05 Heading

Figure A-1. Heads-Up Display

D. TASKS
The majority of the evaluations were made with a pitch and roll HUD tracking task, similar to
that used in the HAVE LIMITS flight and simulation studies.1 The tracking command signals
(Figure A-2) consisted of a series of steps and ramps, lasting slightly over two minutes. The
command signals were developed by Calspan and the task is referred to as the “discrete,” “step-
and-ramp,” or “SAAB” tracking task. The pitch signal is corrected for bank angle, making it
body-referenced attitude command, rather than Earth-referenced (Euler) angle. Details of the
mechanization of the commands and their breakpoints are published elsewhere.2
5 80
Pitch
4 Roll 60
3
40
2
20
1
Pitch Roll
cmd 0 0 cmd
(deg) 0 20 40 60 80 100 120 140 (deg)
-1
-20
-2 time
(sec) -40
-3
-60
-4

-5 -80

Figure A-2. Pitch and Roll Command Signals

184
In most studies using the pitch and roll command signals of Figure A-2, including HAVE
LIMITS,1 the signals were scaled to less than 100% of the values shown (HAVE LIMITS used
90% scale; that is, maximum commanded attitudes were 2.67 deg in pitch and 63 deg in roll).
Scaling was necessary on the NT-33A to provide a margin from the g-limits of the airplane. In
steady level flight, a bank angle of 63 deg corresponds to a load factor of 2.2g, but 70 deg
increases load factor to 2.9g, near the limit of the NT-33A. For this simulation experiment no
such g-limits existed so the full-scale values were used.
A very limited number of evaluations used a pitch-only “sum-of-sines” (SOS) tracking task that
was identical to that used in HAVE LIMITS. It has been found by many researchers that pilots
object to performing the SOS task for several reasons. First, the task is generally frustrating to
the pilot, as the command signal is continuously moving and there seems to be no hope of
obtaining a good tracking solution for any period of time. Second, pilots can find it difficult to
sort out the motions of the cue from the motions of the airplane, resulting in increased
uncertainty over whether observed errors are just part of the task or the precursors to a serious
PIO tendency. For these reasons the task was abandoned early in the simulation program and
was used only sparingly to identify pilot-vehicle dynamics.
The SOS task is ideal for extracting describing-function models of the human operator in
compensatory tracking. It is possible to glean some similar information from the discrete
tracking task as well, but the results are invariantly less consistent.
The pitch and roll error signals – commanded minus actual attitudes, with corrections for non-
zero bank angle – drove an error tracking bar (Figure A-1) and the pilot attempted to null the
errors by flying to the bar. A unique fixed reference (desired tracking area in Figure A-1)
defined task performance for the assignment of Cooper-Harper Handling Qualities Ratings.
Desired performance required keeping the error bar within the brackets of the tracking area for at
least 50% of the run, and adequate required the error bar within or touching the brackets for at
least 50% of the run.
The separation between brackets was established during initial checkout with the assistance of an
experienced guest pilot. RMS tracking errors were computed after each run and provided to the
pilots to help them gauge their level of aggressiveness.

E. PITCH CONFIGURATIONS
The aircraft dynamic model was representative of the USAF variable-stability NT-33A airplane.
A Matlab®/Simulink file, provided to the Air Force by Calspan Corporation, generated the basic
dynamic responses to control inputs. This model was used for the piloted simulation phase of
the HAVE LIMITS project as well.
A single flight condition, 10,000 ft at 250 KIAS, was flown. This corresponded closely to the
primary trim condition for the Neal-Smith experiment3 of the 1970s and the more recent HAVE
LIMITS experiment. Use of a common flight condition provided both continuity with the
previous studies and a known achievable set of aircraft dynamics. It also introduced inherent
limitations in the range of possible closed-loop dynamics. For example, the resulting aircraft had
a specific set of numerator characteristics that could not be varied. In addition, adherence to both
the dynamics of the NT-33A and the structure of that airplane’s flight control system prohibited
certain extreme variations in dynamics. Early in the simulation a set of configurations with high-
frequency simulated short-period dynamics had to be abandoned when it was discovered that

185
even a moderate amount of maneuvering led to rate-limiting of the basic NT-33A’s elevator
actuators.
The longitudinal configurations were based on those evaluated on either or both of the Neal-
Smith and the HAVE LIMITS experiments. A small number of dynamic variations, all
consisting of changes only in short-period damping and frequency, was developed. A large
matrix of variations was created by combining the basic aircraft dynamics with differing filters
and actuator rate limits, and by changing the effective “unaugmented” vehicle dynamics. For the
latter, the NT-33A’s simulated analog variable-stability system (VSS) was used to change the
effective aircraft dynamics from those of the NT-33A to those of the desired form. Then,
augmentation was produced through a simulated digital feedback loop (see Figure A-3).
Thirty-nine different pitch configurations were generated by a combination of basic dynamics,
feedbacks, and rate limits. The resulting short-period characteristics of the configurations, and
the relevant limits and filters, are listed in Table A-1.
Where the short-period dynamics are the same for the “augmented” and “unaugmented” cases,
the listed dynamics were obtained through the VSS feedbacks of Figure A-3 and no
augmentation feedbacks were used. Where the dynamics differ, VSS provided “unaugmented”
characteristics and augmentation the “augmented” characteristics listed. Reaching the rate limits
meant a switch from the augmented to unaugmented dynamics (in addition to the dynamics of
the rate-limited actuator, of course).
The dynamics of the pitch controller feel system were identified from test inputs to be
δ es K fs
= 2 2 ,
Fes s /14 + 2(0.64/14)s + 1
where the control force/deflection gearing K fs was either 1/8 in./lb (for the nominal stick) or 1/22
in./lb (for the stiff stick).
Configuration identifiers in the simulation indicated the vehicle dynamics, including, if
appropriate, the dynamics of the “unaugmented” vehicle, plus any filters and rate limits below
the nominal of 160 deg/sec. For example, “2DR30” denoted the baseline Configuration 2D with
a rate limit of 30 deg/sec. This case received the best pilot ratings and comments in HAVE
LIMITS and was the nominal configuration in this simulation. As another example, an identifier
such as “2D16R10” indicates a configuration with the unaugmented dynamics of Configuration
16, augmented to the dynamics of 2D, through an actuator with a rate limit of 10 deg/sec.

F es F eel δes A dded Rate-lim ited q, θ, α, ...

G earing NT -33A
system filters actuator

V SS
feedbacks

Augm entation
feedbacks

Figure A-3. Schematic of Model and Augmentation System Mechanizations

186
 Table A-1. Longitudinal Configurations

a. Basic configurations and actuator rate limits

Identifier Short Period Actuator Rate Limits (deg/sec)
augmented unaugmented 160 40 30 20 15 10
2D [0.71, 4.57] [0.71, 4.57] > > > >
1D [0.70, 2.20] [0.70, 2.20] >
5A [0.18, 4.56] [0.18, 4.56] >
15 [0.41, 4.56] [0.41, 4.56] >
16 [0.10, 4.56] [0.10, 4.56] >
17 [1.30, 4.57] [1.30, 4.57] >
18 [0.70, 1.51] [0.70,1.51] >
T33 [0.62, 2.49] [0.62, 2.49] >
2DU [0.71, 4.57] (-1.33)(2.18) > > >
2DV [0.71, 4.57] (-0.52)(7.19) > > >
2D16 [0.71, 4.57] [0.10, 4.56] > >
2D18 [0.71, 4.57] [0.70,1.51] >
2DT [0.71, 4.57] [0.62, 2.49] >
17DV [1.30, 4.57] (-0.52)(7.19) > >
1716 [1.30, 4.57] [0.10, 4.56] > >
Note: [ζ sp , ωsp ] = [s 2 + 2ζ sp ωsp s + ω2sp ]; (1/T) = (s + 1/T)

b. Added lead/lag filters

Identifier Basic T1 T2 ζ2 ω2
configuration
2DL1 2D 0.7 18
2DL2 2D 0.7 14
2DL3 2D 0.7 7
2DL4 2D 0.7 4
2A 2D 0.5 0.2
2F 2D 0.2
2H 2D 0.5
2P 2D 0.25
1DL2 1D 0.7 14
1DL3 1D 0.7 7
(T1s + 1)
Format of added filters:
(T2 s + 1)[(s 2 /ω22 ) + (2ζ 2 /ω 2 )s + 1]

Configurations 1D, 2A, 2D, 2F, and 2H were based on similar cases evaluated on the Neal-
Smith3 experiment, and 2DU and 2P (along with 2D) came from the HAVE LIMITS1
experiment. The additional cases were developed to explore different regions of the Bandwidth
PIO criteria, or the effects of rate saturation, or both.

187
F. ROLL CONFIGURATIONS
A much smaller matrix of roll variations, totaling ten configurations, was flown. Roll cases were
used in some evaluations to increase the level of difficulty of the pitch task. The baseline roll
dynamics were representative of an augmented NT-33A with feedbacks to the rudder and
ailerons, and a crossfeed between the two. The augmentation effectively canceled the Dutch roll
mode with the roll numerator and the resulting baseline roll rate response to lateral stick inputs
was approximately
p 100(0)e −0.038s
=
δ a (0.053)(2.38)
A second set of roll dynamics was obtained by zeroing all augmentation gains for the baseline
case. This produced a response based on the unaugmented NT-33A:
p 92(0)[0.19, 1.55]e −0.038s
=
δa (0)(1.28)[0.20, 1.58]
The baseline configuration, identified as LB, was flown with the rate limit of the NT-33A model
(set to 160 deg/sec) and with limits of 80, 60, 40, 20, and 10 deg/sec. The unaugmented case
was Configuration LT33 and it was flown with the limit of the NT-33A model and 20 deg/sec.
A third variant was created by placing the feedbacks of LB through the rate limiter, so that
saturation would open the augmentation loops and result in the dynamics of the NT-33A. This
was Configuration LBT and it was evaluated with rate limits of 60 and 20 deg/sec.

G. RATE LIMITING
A second-order dynamic model of an actuator was used (Figure A-4). Rate limiting was
introduced through the actuator. This is not a typical situation for modern aircraft, where some
form of software rate limiting is typically employed upstream of the actuator,4 but it provided the
most consistent, convenient method for introducing rate limiting. When not rate-limited, the
actuator dynamics were representative of the elevator dynamics of the variable-stability NT-33A
with damping of 0.7, natural frequency of 75 rad/sec, and rate limit of 160 deg/sec.

Figure A-4. Matlab®/Simulink Representation of Rate-Limited Actuator

188
H. CONDUCT OF SIMULATION
Each pilot flew one run with a new configuration, with allowance for repeats if needed, before
assigning a Cooper-Harper Handling Qualities Rating and a PIO Tendency Rating. For the
latter, the modified decision-tree scale developed for HAVE LIMITS1 was used. Extensive pilot
comments were recorded and later transcribed for detailed analysis.
RMS tracking errors in pitch and roll were computed and read to the pilots at the conclusion of
each run. Desired values were established by the first pilot, though these values were considered
more as guides to the other pilots in judging their overall performance as opposed to direct
measures for assigning Cooper-Harper ratings.
Configurations were presented to the pilots initially in “pseudo-random” sequence. An
important break from tradition was employed as a result of experience gained from previous
simulations. Several of these configurations developed into rapidly divergent PIOs that required
operator intervention to avoid reaching motion travel limits. It had been noticed that the pilot
tends to be overly cautious with the next run, thereby contaminating the results. To prevent this,
it was common to insert the baseline configuration (2DR30) or a similar good set of dynamics
immediately after a divergent configuration. The evaluation of the “control” configuration was
counted in the overall analysis. As a result, 2DR30 was evaluated many times (a total of 33)
over the course of the simulation.
Every pilot experienced multiple configurations with out-of-control divergent PIOs. In rare
instances, pilots were able to recover the aircraft before the motions became so large that the
simulation had to be stopped. In 36 of the 478 data runs, the pilots lost control of the aircraft in a
divergent PIO.

I. VERIFICATION OF CONFIGURATIONS
Time delays inherent in the motion and visual systems of LAMARS have been measured. The
effective throughput delay for both pitch motion and the HUD visual driver was estimated to be
57 msec. This time delay is considered to be a part of the dynamics of all of the configurations.
Configurations were verified either from frequency-sweeps or from FFTs of actual data runs.
Only a few configurations were swept, so most required use of the tracking run data. It was
found that very good dynamic models could be obtained from either source. The describing
functions for the high-rate-limit (160 deg/sec) cases were generated and equivalent-system fitting
was performed. The resulting short-period dynamics agree closely (usually within 10%) with the
numbers listed in Table A-1. The values in Table A-1 are the desired analytical values, not the
equivalent-system values.

J. RUN LOG
A run log for the entire simulation, encompassing both the 1998 and 1999 sessions, is given in .
Run numbers were consecutive from Run 1 to 424 for the 1998 session; in 1999, run numbers
started at Run 1 each day. Tasks were either the discrete tracking (disc) or sum-of-sines tracking
(sos). Pilot-generated sweeps that were part of the simulation sessions are labeled as sweeps in
the task column. Configuration ID’s are as defined above. Stick gearing was the value of the
final gain under control of the simulation engineers for pitch and roll command gains
(effectively, degrees/second/inch of stick displacement). RMS errors for pitch and roll were

189
computed only if the task was completed. Comments indicate special conditions, including
fixed-base sessions, and evaluations with the stiff stick force/deflection gradients.
Table A-2. Run Log for Piloted Simulation
Run Conf. ID Stick gearing RMS errors Ratings
No. Date Pilot Task pitch roll KSE KSA pitch roll HQR PIOR Comments
1 4/6/98 A disc 2DR30 LB 0.75 -22 0.21 0.17 3 1
2 4/6/98 A disc 2DL3 LB 0.75 -22 0.38 0.18 6 5
3 4/6/98 A disc 2H LB 0.75 -22 0.37 0.17 6 1
4 4/6/98 A disc 2DR30 LBR80 0.75 -22 0.24 0.18 3 1
5 4/6/98 A disc 2DR30 LT33 0.75 -18 0.32 0.19 Lost control
6 4/6/98 A disc 2DR30 LT33 0.75 -18 0.29 0.18 5 1
7 4/6/98 A disc 2DVR20 LB 0.75 -22 0.24 0.18 3 2
8 4/6/98 A disc 2DR30 LB 0.75 -22 0.2 0.17 3 1
9 4/6/98 A disc 2DVR15 LB 0.75 -22 10 5
10 4/6/98 A disc 18 LB 0.75 -22 0.41 0.2 9 5
11 4/6/98 A disc 2DR30 LBTR60 0.75 -22 6 4 Motion problem - abort
12 4/7/98 A disc 17 LB 0.75 -22 0.29 0.17 Forces/displ. high
13 4/7/98 A disc 17 LB 1.1 -22 0.21 0.17 3 1 Much better
14 4/7/98 A disc 17DVR20 LB 1.1 -22 0.29 0.17 6 5
15 4/7/98 A disc 2DR30 LBR60 0.75 -22 0.73 0.25 8 4
16 4/7/98 A disc2 2DR30 LB 0.75 -22 0.23 0.19 Disc 2 task --
17 4/7/98 A disc2 2DR30 LB 0.75 -22 0.24 0.18 4 1 may be too tough
18 4/7/98 A disc 2P LB 0.75 -22 0.33 0.17 5 1
19 4/7/98 A disc 2DUR40 LB 0.75 -22 0.21 0.17 4 1
20 4/7/98 A disc 2DR20 LB 0.75 -22 0.21 0.16 3 1
21 4/7/98 A disc 2DUR20 LB 0.75 -22 10 5 Lost control
22 4/7/98 A sos 2DR30 LB 0.75 -22 0.45 0.6 Not comfortable rating
23 4/7/98 A sos 2DL3 LB 0.75 -22 0.71 0.6 4
24 4/7/98 A sos 2H LB 0.75 -22 0.54 0.53 3
25 4/7/98 A sos 2DR30 LBR60 0.75 -22 0.51 0.77 4
26 4/7/98 A sos 2DR30 LT33 0.75 -18 0.43 0.6 4
27 4/7/98 A sos 2DVR20 LB 0.75 -22 0.52 0.6 4
28 4/7/98 A sos 2DR30 LB 0.75 -22 0.45 0.58
29 4/7/98 A sos 2DR30 LB 0.75 -22 0.43 0.54 3
30 4/7/98 A disc 15 LB 0.75 -22 0.26 0.17 4 2
31 4/7/98 A disc 2DR10 LB 0.75 -22 0.26 0.17 5 3
32 4/7/98 A disc 2DR30 LBR80 0.75 -22 0.27 0.19 5 4
33 4/7/98 A disc 5A LB 0.75 -22 0.42 0.18 7 4
34 4/7/98 B disc 2DR30 LB 0.75 -22 0.28 0.19 4 2
35 4/7/98 B disc 2DL3 LB 0.75 -22 0.47 0.22 7 4
36 4/7/98 B disc 2H LB 0.75 -22 0.35 0.22
37 4/7/98 B disc 2H LB 0.75 -22 0.37 0.22 5 3
38 4/7/98 B disc 2DR30 LBR40 0.75 -22 0.32 0.25
39 4/7/98 B disc 2DR30 LBR40 0.75 -22 0.28 0.24 5 3
40 4/7/98 B disc 2DR30 LBR20 0.75 -22 0.6 0.36 9 5
41 4/7/98 B disc 5A LB 0.75 -22 0.42 0.23 8 4
42 4/7/98 B disc 2DR30 LB 0.75 -22 0.25 0.2 4 3
43 4/7/98 B disc 2DR30 LB 0.75 -22 0.31 0.2
44 4/7/98 B disc 2DR30 LB 0.75 -22 0.33 0.2 5 3
45 4/7/98 B disc 17 LB 1.1 -22 0.27 0.2 3 2
46 4/7/98 B disc 18 LB 0.75 -22 0.37 0.25 7 4
47 4/7/98 B disc 2D18R20 LB 0.75 -22 0.43 0.21 HUD problems
48 4/7/98 B disc 2D18R20 LB 0.75 -22 0.33 0.21
49 4/7/98 B disc 2D18R20 LB 0.75 -22 0.38 0.21 5 4 Saw oscills in roll!!!
50 4/8/98 A disc 2D LB 0.75 -22 0.23 0.16 3 1
51 4/8/98 A disc 1D LB 0.5 -22 0.28 0.18 6 4
52 4/8/98 A disc 16 LB 0.75 -22 0.51 0.2 8 4
53 4/8/98 A disc 2D16R20 LB 0.75 -22 0.2 0.15 3 1
54 4/8/98 A disc 2DR30 LBR40 0.75 -22 0.29 0.23 8 4
55 4/8/98 A disc 2D16R10 LB 0.75 -22 0.22 0.16 3 1
56 4/8/98 A disc 1716R20 LB 0.75 -22 0.18 0.16 3 1
57 4/8/98 A disc 2DVR20 LB 0.75 -22 0.2 0.16
58 4/8/98 A disc 2DVR20 LB 0.75 -22 0.52 0.16 9 5
59 4/8/98 A disc 2DR30 LBTR60 0.75 -22 0.5 0.2 8 4
60 4/8/98 A disc 2DL2 LB 0.75 -22 0.26 0.16 5 4
61 4/8/98 A disc 1DL3 LB 0.5 -22 0.44 0.18 8 5

190
 Table A-2. Run Log for Piloted Simulation
Run Conf. ID Stick gearing RMS errors Ratings
No. Date Pilot Task pitch roll KSE KSA pitch roll HQR PIOR Comments
62 4/8/98 A disc 2D18R20 LB 0.75 -22 0.21 0.16 4 2
63 4/8/98 A disc 2DL1 LB 0.75 -22 0.22 0.16 4 2
64 4/8/98 A disc 2DL4 LB 0.75 -22 0.39 0.16 6 4
65 4/8/98 A disc 2DR30 LB 0.75 -22 0.26 0.16
66 4/8/98 A disc 2DR30 LB 0.75 -22 0.2 0.16 4 2
67 4/8/98 A disc 2DR30 LT33R20 0.75 -18 0.76 0.41 9 5
68 4/8/98 B disc 2DR10 LB 0.75 -22 0.31 0.19 3 3
69 4/8/98 B disc 15 LB 0.75 -22 0.3 0.21 6 4
70 4/8/98 B disc 2DVR20 LB 0.75 -22 0.25 0.19 4 2
71 4/8/98 B disc 2DR30 LBR60 0.75 -22 0.26 0.2 3 2
72 4/8/98 B disc 16 LB 0.75 -22 0.48 0.23 8 4
73 4/8/98 B disc 2DR30 LT33R20 0.75 -18 0.45 0.33 8 5
74 4/8/98 B disc 2DR30 LB 0.75 -22 0.24 0.19 3 2
75 4/8/98 B disc 2DVR15 LB 0.75 -22 0.28 0.2
76 4/8/98 B disc 2DVR15 LB 0.75 -22 0.26 0.2 5 2
77 4/8/98 B disc 1D LB 0.5 -22 0.34 0.21 7 3
78 4/8/98 B disc 2F LB 0.75 -22 0.33 0.19 2 1
79 4/8/98 B disc 2DL1 LB 0.75 -22 0.3 0.2 5 3
80 4/8/98 B disc 2DVR10 LB 0.75 -22 0.34 0.21
81 4/8/98 B disc 2DVR10 LB 0.75 -22 10 5 Lost control at end
82 4/8/98 B disc 2DR30 LBR80 0.75 -22 0.33 0.2 3 2
83 4/8/98 B disc 2DR30 LBTR60 0.75 -22 0.27 0.19 4 3
84 4/8/98 B disc 1DL3 LB 0.5 -22 0.44 0.23 8 3
85 4/8/98 B disc 2A LB 0.75 -22 0.36 0.23 6 3
86 4/8/98 B disc 2DR30 LT33 0.75 -18 0.43 0.19
87 4/8/98 B disc 2DR30 LT33 0.75 -18 0.3 0.19 4 3
88 4/9/98 A disc 2F LB 0.75 -22 0.27 0.16 5 3 Motion fixed
89 4/9/98 A disc 2A LB 0.75 -22 0.28 0.18 5 3
90 4/9/98 A disc 1DL2 LB 0.5 -22 0.33 0.17 8 4
91 4/9/98 A disc 2DUR40 LB 0.75 -22 0.2 0.17 3 1
92 4/9/98 A disc 2DR30 LBR80 0.75 -22 0.22 0.17 6 3
93 4/9/98 A disc 18 LB 0.5 -22 0.42 0.18 8 4
94 4/9/98 A disc T33 LB 0.5 -22 0.28 0.17
95 4/9/98 A disc T33 LB 0.5 -22 0.21 0.16 4 2
96 4/9/98 A disc 2DG LB 1.5 -22 0.2 0.17 5 2
97 4/9/98 A disc 1716R10 LB 1.1 -22 0.27 0.16 5 2
98 4/9/98 A disc 2DR30 LB 0.75 -22 0.19 0.16 3 2
99 4/9/98 A disc 2DV(o.l.) LB 0.75 -22 0.42 0.17 9 N/A
100 4/9/98 A disc 2DR30 LT33 0.75 -18 0.26 0.17
101 4/9/98 A disc 2DR30 LT33 0.75 -18 0.24 0.17 5 1
102 4/9/98 A disc T33 LB 0.5 -22 0.24 0.17 8 4
103 4/10/98 A disc 1716R10 LB 1.1 -22 0.25 0.16
104 4/10/98 A disc 1716R10 LB 1.1 -22 0.24 0.17 5 2
105 4/10/98 A disc T33 LB 0.5 -22 0.27 0.16 6 4
106 4/10/98 A disc 2DR30 LT33 0.75 -18 0.35 0.18 5 2
107 4/10/98 A disc 2DR30 LBR60 0.75 -22 0.27 0.21 7 4
108 4/10/98 A disc 18 LB 0.5 -22 0.37 0.17 8 N/A
109 4/10/98 A disc 1DL2 LB 0.5 -22 0.38 0.16 9 5
110 4/10/98 A disc 2DUR30 LB 0.75 -22 0.21 0.17 4 2
111 4/10/98 A disc 17 LB 1.1 -22 0.19 0.17 3 2
112 4/10/98 A disc 17DVR20 LB 1.1 -22 0.22 0.16 8 4
113 4/10/98 A sweep 2DVR20 LB Sweep – small ampl.
114 4/10/98 A sweep T33 LB Sweep
115 4/10/98 A sweep 2D LB Sweep
116 4/10/98 A sweep 2DR10 LB Sweep
117 4/10/98 A sweep 16 LB Sweep
118 4/10/98 A sweep 17DVR20 LB Sweep – small ampl.
119 4/10/98 A sweep 17DVR20 LB Sweep – mod. ampl.
120 4/10/98 A sweep 17DVR20 LB Sweep – large ampl.
121 4/10/98 B disc 2D16R20 LB 0.75 -22 0.22 0.18 2 2
122 4/10/98 B disc 2DL4 LB 0.75 -22 0.38 0.22 7 3
123 4/10/98 B disc 2D16R10 LB 0.75 -22 0.25 0.19 3 2
124 4/10/98 B disc 2DR30 LBTR60 0.75 -22 0.23 0.2 4 3
125 4/10/98 B disc 17DVR20 LB 1.1 -22 0.27 0.18 3 2
126 4/10/98 B disc 2DR30 LB 0.75 -22 0.24 0.19 3 2

191
 Table A-2. Run Log for Piloted Simulation
Run Conf. ID Stick gearing RMS errors Ratings
No. Date Pilot Task pitch roll KSE KSA pitch roll HQR PIOR Comments
127 4/10/98 B disc 1716R10 LB 1.1 -22 0.24 0.18
128 4/10/98 B disc 1716R10 LB 1.1 -22 0.29 0.18 5 3
129 4/10/98 B disc 2DG LB 1.5 -22 0.26 0.21 5 3
130 4/10/98 B disc 17DVR10 LB 1.1 -22 0.26 0.18 4 2
131 4/10/98 B disc 17DVR10 LB 1.1 -22 0.35 0.19 6 3 Repeat (blind)
132 4/10/98 B disc 2DR30 LBTR20 0.75 -22 0.33 0.27 8 4
133 4/10/98 B disc 1DL2 LB 0.5 -22 0.34 0.23
134 4/10/98 B disc 1DL2 LB 0.5 -22 0.31 0.21 7 3
135 4/10/98 B disc 2P LB 0.75 -22 0.3 0.2 4 2
136 4/10/98 B disc 2DL2 LB 0.75 -22 0.27 0.2 5 2
137 4/10/98 B disc 2DUR20 LB 0.75 -22 10 5 Lost control
138 4/10/98 B disc 17DVR10 LB 1.1 -22 0.24 0.19 4 2
139 4/10/98 B disc 2DR20 LB 0.75 -22 0.25 0.19 2 2
140 4/16/98 C disc 2DR30 LB 0.75 -22 0.28 0.19 1 1
141 4/16/98 C disc 2DL3 LB 0.75 -22 0.4 0.22 6 3 Saw PIO…
142 4/16/98 C disc 2H LB 0.75 -22 0.39 0.21 7 3
143 4/16/98 C disc 2DR30 LBR80 0.75 -22 0.29 0.2 2 1
144 4/16/98 C disc 2DR30 LT33 0.75 -18 0.42 0.21 4 2
145 4/16/98 C disc 2DVR20 LB 0.75 -22 0.29 0.21 5 3
146 4/16/98 C disc 2DR30 LB 0.75 -22 0.23 0.19 1 1
147 4/16/98 C disc 2DVR15 LB 0.75 -22 0.37 0.21 7 3
148 4/16/98 C disc 18 LB 0.5 -22 0.44 0.23 7 3 "Avoiding a PIO"
149 4/16/98 C disc 18 LB 0.5 -22 0.34 0.19 7 4
150 4/16/98 C disc 2DR30 LBTR60 0.75 -22 0.38 0.2 7 3
151 4/16/98 C disc 17 LB 1.1 -22 0.26 0.18 5 1
152 4/16/98 C disc 17DVR20 LB 1.1 -22 0.28 0.18
153 4/16/98 C disc 17DVR20 LB 1.1 -22 0.26 0.18 5 3
154 4/20/98 D disc 17 LB 1.1 -22 0.27 0.18 2 1
155 4/20/98 D disc 1716R20 LB 1.1 -22 0.27 0.19 2 1
156 4/20/98 D disc 1716R10 LB 1.1 -22 0.32 0.19 7 4
157 4/20/98 D disc 2DL3 LB 0.75 -22 0.42 0.21 7 4
158 4/20/98 D disc 2H LB 0.75 -22 0.35 0.19 3 1
159 4/20/98 D disc 2DR30 LBR80 0.75 -22 0.31 0.22 3 1
160 4/20/98 D disc 2DR30 LT33 0.75 -18 0.39 0.21 4 2
161 4/20/98 D disc 2DR30 LBR40 0.75 -22 0.86 0.26 4 3
162 4/20/98 D disc 2DVR20 LB 0.75 -22 0.39 0.21 3 1
163 4/20/98 D disc 2DVR15 LB 0.75 -22 0.37 0.22
164 4/20/98 D disc 2DVR15 LB 0.75 -22 0.29 0.19 4 2
165 4/20/98 D disc 2DR30 LB 0.75 -22 0.29 0.2 3 1
166 4/20/98 D disc 2DVR10 LB 0.75 -22 10 5 Departed
167 4/20/98 C disc 2DR30 LB 0.75 -22 0.26 0.18 2 2
168 4/20/98 C disc 2DR10 LB 0.75 -22 0.29 0.19 6 3
169 4/20/98 C disc 1716R20 LB 1.1 -22 0.22 0.18 1 1
170 4/20/98 C disc 15 LB 0.75 -22 0.26 0.19 7 4
171 4/20/98 C disc 2DR30 LBR40 0.75 -22 0.25 0.22 7 4
172 4/20/98 C disc 16 LB 0.75 -22 0.42 0.19 7 4
173 4/20/98 C disc 17 LB 1.1 -22 0.25 0.17
174 4/20/98 C disc 17 LB 1.1 -22 0.22 0.18 5 2
175 4/20/98 C disc 2DVR15 LB 0.75 -22 0.4 0.19 7 4
176 4/20/98 C disc 2DL4 LB 0.75 -22 0.55 0.2 7 4
177 4/20/98 C disc 2A LB 0.75 -22 0.25 0.19 4 3
178 4/20/98 C disc 2DR30 LBR60 0.75 -22 0.29 0.19 7 3
179 4/20/98 C disc 2DVR10 LB 0.75 -22 Lost on initial input
180 4/20/98 C disc 2DVR10 LB 0.75 -22 1.34 0.21 9 5
181 4/20/98 C disc 2DR30 LB 0.75 -22 0.21 0.18 1 1
182 4/20/98 C disc 1DL3 LB 0.5 -22 0.61 0.2 7 4
183 4/20/98 C disc 1716R10 LB 1.1 -22 0.3 0.18 5 3
184 4/20/98 C disc 2DR30 LT33R20 0.75 -18 0.69 0.34 8 4
185 4/20/98 C disc 2DL1 LB 0.75 -22 0.26 0.18
186 4/20/98 C disc 2DL1 LB 0.75 -22 0.25 0.18 4 2
187 4/20/98 C disc 2H LB 0.75 -22 0.31 0.18 6 3
188 4/20/98 C disc 1DL2 LB 0.5 -22 0.34 0.18 5 3
189 4/20/98 C disc 5A LB 0.75 -22 0.35 0.2 7 4
190 4/21/98 E disc 2DR30 LB 0.75 -22 0.3 0.19 2 1
191 4/21/98 E disc 17 LB 1.1 -22 0.27 0.2 3 1

192
 Table A-2. Run Log for Piloted Simulation
Run Conf. ID Stick gearing RMS errors Ratings
No. Date Pilot Task pitch roll KSE KSA pitch roll HQR PIOR Comments
192 4/21/98 E disc 1716R10 LB 1.1 -22 0.34 0.19
193 4/21/98 E disc 1716R10 LB 1.1 -22 0.33 0.19 4 2
194 4/21/98 E disc 2DL4 LB 0.75 -22 0.43 0.2 7 3
195 4/21/98 E disc 2DR30 LBR80 0.75 -22 0.41 0.22
196 4/21/98 E disc 2DR30 LBR80 0.75 -22 0.31 0.23 3 1
197 4/21/98 E disc 2DR30 LB 0.75 -22 0.3 0.2 2 1
198 4/21/98 E disc 2DR30 LBR40 0.75 -22 1.41 0.29 5 1
199 4/21/98 E disc 2DVR15 LB 0.75 -22 0.33 0.21 3 1
200 4/21/98 E disc 1DL3 LB 0.5 -22 0.61 0.21
201 4/21/98 E disc 1DL3 LB 0.5 -22 0.63 0.19 8 3
202 4/21/98 E disc 15 LB 0.75 -22 0.55 0.22 7 3
203 4/21/98 E disc 2DVR10 LB 0.75 -22 10 5 Lost control
204 4/21/98 E disc 2DL1 LB 0.75 -22 0.33 0.2 3 2
205 4/21/98 E disc 2DR30 LBR20 0.75 -22 1.93 0.47 8 4
206 4/21/98 E disc 2DR10 LB 0.75 -22 0.39 0.19
207 4/21/98 E disc 2DR10 LB 0.75 -22 0.32 0.21 4 1
208 4/21/98 F disc 2DR30 LB 0.75 -22 0.28 0.19 4 2
209 4/21/98 F disc 17 LB 1.1 -22 0.27 0.19 2 1
210 4/21/98 F disc 1716R10 LB 1.1 -22 0.55 0.2 8 6
211 4/21/98 F disc 2DL4 LB 0.75 -22 0.55 0.18 5 4
212 4/21/98 F disc 2DR30 LBR80 0.75 -22 0.63 0.22 4 1
213 4/21/98 F disc 2DVR20 LB 0.75 -22 0.41 0.21 6 4
214 4/21/98 F disc 2DR30 LBR40 0.75 -22 0.96 0.25 6 1
215 4/21/98 F disc 2DVR15 LB 0.75 -22 Lost control
216 4/21/98 F disc 2DVR15 LB 0.75 -22 0.39 0.19 10 6
217 4/21/98 F disc 1D LB 0.5 -22 0.26 0.19 3 1
218 4/21/98 F disc 16 LB 0.75 -22 0.43 0.23 8 4
219 4/21/98 F disc 2DR30 LBR20 0.75 -22 1.43 0.39 9 4
220 4/21/98 F disc 2DUR20 LB 0.75 -22 10 6 Lost control
221 4/22/98 E disc 2P LB 0.75 -22 0.32 0.18 4 2
222 4/22/98 E disc 2A LB 0.75 -22 0.43 0.21 5 3
223 4/22/98 E disc 2DUR30 LB 0.75 -22 0.31 0.19 2 1
224 4/22/98 E disc 18 LB 0.5 -22 0.53 0.2
225 4/22/98 E disc 18 LB 0.5 -22 0.58 0.18 8 3
226 4/22/98 E disc 2DR30 LBR40 0.75 -22 0.86 0.3 6 3
227 4/22/98 E disc 2DUR20 LB 0.75 -22 10 5 Lost control
228 4/22/98 E disc 2DR30 LB 0.75 -22 0.84 0.21 3 1 Got behind pitch
229 4/22/98 E disc 16 LB 0.75 -22 0.75 0.27 10 5
230 4/22/98 E disc 2H LB 0.75 -22 0.52 0.18 5 1
231 4/22/98 E disc 2DR30 LBTR60 0.75 -22 0.86 0.27 8 1
232 4/22/98 E disc 2D16R10 LB 0.75 -22 0.32 0.19 3 1
233 4/22/98 E disc 2DG LB 1.5 -22 0.29 0.2 3 1
234 4/22/98 E disc 2F LB 0.75 -22 0.29 0.19 3 1
235 4/22/98 E disc 5A LB 0.75 -22 0.45 0.21 8 4
236 4/22/98 E disc 2DR30 LBTR60 0.75 -22 0.92 0.29 7 1
237 4/22/98 E disc 1716R10 LB 1.1 -22 0.28 0.19 3 1
238 4/22/98 E disc 2DL4 LB 0.75 -22 0.45 0.18 7 4
239 4/22/98 E disc 2DR30 LB 0.75 -22 0.3 0.18 4 2
240 4/22/98 E disc 2DUR30 LB 0.75 -22 0.3 0.19 2 1
241 4/22/98 E disc 2DUR30 LBTR60 0.75 -22 10 5 Lost control
242 4/22/98 E disc 1DL3 LB 0.5 -22 0.53 0.21
243 4/22/98 E disc 1DL3 LB 0.5 -22 0.53 0.19 8 4
244 4/22/98 E disc 2DR10 LBR60 0.75 -22 0.55 0.24 6 2
245 4/22/98 F disc 2F LB 0.75 -22 0.29 0.19 3 1
246 4/22/98 F disc 18 LB 0.5 -22 0.38 0.2 7 3
247 4/22/98 F disc 2A LB 0.75 -22 0.26 0.2 7 4
248 4/22/98 F disc 2DR10 LB 0.75 -22 0.3 0.2 4 2
249 4/22/98 F disc 2DR30 LBTR60 0.75 -22 0.41 0.21 6 4
250 4/22/98 F disc 15 LB 0.75 -22 0.27 0.2 4 2
251 4/22/98 F disc 2H LB 0.75 -22 0.3 0.17 3 1
252 4/22/98 F disc 2D16R10 LB 0.75 -22 0.23 0.19 2 1
253 4/22/98 F disc 1716R10 LB 1.1 -22 0.28 0.19 3 2
254 4/22/98 F disc 2DR30 LBR60 0.75 -22 0.47 0.24 8 4
255 4/22/98 F disc 2DR30 LBTR20 0.75 -22 3.6 0.67 10 4
256 4/22/98 F disc 2DR30 LB 0.75 -22 0.21 0.18 1 1

193
 Table A-2. Run Log for Piloted Simulation
Run Conf. ID Stick gearing RMS errors Ratings
No. Date Pilot Task pitch roll KSE KSA pitch roll HQR PIOR Comments
257 4/22/98 F disc 2DUR30 LB 0.75 -22 10 6
258 4/22/98 F disc 5A LB 0.75 -22 0.42 0.21 7 4
259 4/22/98 F disc 1DL3 LB 0.5 -22 0.67 0.2 9 4
260 4/22/98 F disc 2DL1 LB 0.75 -22 0.22 0.2 2 1
261 4/22/98 F disc 2H LB 0.75 -22 0.3 0.18 4 1
262 4/23/98 D disc 18 LB 0.5 -22 0.37 0.22 4 3
263 4/23/98 D disc 2A LB 0.75 -22 0.39 0.22 4 2
264 4/23/98 D disc 2DR10 LB 0.75 -22 0.34 0.19 2 1
265 4/23/98 D disc 2DR30 LBTR60 0.75 -22 0.77 0.22 4 2
266 4/23/98 D disc 2DR30 LBTR20 0.75 -22 1.23 0.44 7 4
267 4/23/98 D disc 2DL4 LB 0.75 -22 0.61 0.2 7 3
268 4/23/98 D disc 1DL3 LB 0.5 -22 0.64 0.21 7 4
269 4/23/98 D disc 15 LB 0.75 -22 0.46 0.22 4 2
270 4/23/98 D disc 2DUR30 LB 0.75 -22 0.27 0.2 2 1
271 4/23/98 D disc 2DR30 LBTR60 0.75 -22 0.32 0.2 2 1
272 4/23/98 D disc 2DUR20 LB 0.75 -22 0.25 0.2 2 1
273 4/23/98 D disc 2DUR20 LBR40 0.75 -22 0.26 0.22 2 1
274 4/23/98 D disc 2DUR20 LBTR20 0.75 -22 10 5 Lost control
275 4/23/98 D disc 2D16R10 LB 0.75 -22 0.26 0.19 2 1
276 4/23/98 D disc 16 LB 0.75 -22 0.47 0.24 7 4
277 4/23/98 D disc 1D LB 0.5 -22 0.3 0.2 4 2
278 4/23/98 D disc 1DL2 LB 0.5 -22 0.43 0.2 6 3
279 4/23/98 D disc 5A LB 0.75 -22 0.37 0.21 5 3
280 4/23/98 D disc 17DVR10 LB 1.1 -22 0.27 0.19 2 1
281 4/23/98 F disc 2DR10 LB 0.75 -22 0.28 0.2 1 1 Fixed Base
282 4/23/98 F disc 2DUR30 LB 0.75 -22 0.3 0.2 1 1 Fixed Base
283 4/23/98 F disc 2DUR20 LB 0.75 -22 10 5 Fixed Base
284 4/23/98 F disc 2H LB 0.75 -22 0.34 0.21 5 1 Fixed Base
285 4/23/98 F disc 2DR30 LBR60 0.75 -22 0.3 0.24 5 4 Fixed Base
286 4/23/98 F disc 17DVR10 LBR60 1.1 -22 10 6 Fixed Base
287 4/23/98 F disc 2DR30 LB 0.75 -22 0.24 0.19 4 2 Fixed Base
288 4/23/98 F disc 2DR30 LBR40 0.75 -22 0.39 0.24 4 3 Fixed Base
289 4/23/98 F disc 2DR30 LBR20 0.75 -22 1.1 0.47 8 4 Fixed Base
290 4/23/98 F disc 17 LB 1.1 -22 0.22 0.19 1 1 Fixed Base
291 4/23/98 F disc 2DL4 LB 0.75 -22 0.69 0.23 9 4 Fixed Base
292 4/23/98 F disc 2DVR20 LB 0.75 -22 0.33 0.2 3 2 Fixed Base
293 4/23/98 F disc 2DVR15 LB 0.75 -22 10 6 Fixed Base
294 4/23/98 F disc 1DL3 LB 0.5 -22 0.8 0.2 8 4 Fixed Base
295 4/23/98 F disc 1D LB 0.5 -22 0.23 0.18 4 3 Fixed Base
296 4/23/98 F disc 15 LB 0.75 -22 0.35 0.21 7 4 Fixed Base
297 4/24/98 D disc 2P LB 0.75 -22 0.29 0.2 3 1
298 4/24/98 D disc 18 LB 0.5 -22 0.28 0.21 4 2
299 4/24/98 D disc 17DVR10 LB 1.1 -22 Lost control
300 4/24/98 D disc 17DVR10 LB 1.1 -22 0.29 0.18 9 4
301 4/24/98 D disc 2DG LB 0.75 -22 0.26 0.2 2 1
302 4/24/98 D disc 2PR10 LB 0.75 -22 0.31 0.18 4 1
303 4/24/98 D disc 2DR30 LBR20 0.75 -22 0.4 0.31 7 4
304 4/24/98 D disc 2DR30 LB 0.75 -22 0.22 0.19 2 1
305 4/24/98 D disc 2PR20 LB 0.75 -22 0.27 0.19 3 1
306 4/24/98 D disc 2DR30 LT33R20 0.75 -22 1.13 0.42 7 4
307 4/24/98 D disc 2DVR20 LB 0.75 -22 0.25 0.18 2 1
308 4/24/98 D disc 2DL2 LB 0.75 -22 0.24 0.19 3 1
309 4/24/98 D disc 17DVR20 LB 1.1 -22 0.22 0.18 2 1
310 4/24/98 D disc 2DL1 LB 0.75 -22 0.23 0.19 3 1
311 4/24/98 D disc 2DR30 LBR60 0.75 -22 0.25 0.19 2 1
312 4/24/98 D disc T33 LB 0.5 -22 0.24 0.21 3 1
313 4/24/98 D disc 2DTR10 LB 0.75 -22 0.24 0.18 2 1
314 4/24/98 D disc 2DR30 LT33 0.75 -18 0.24 0.18 2 1
315 4/24/98 D disc 2DVR15 LB 0.75 -22 0.2 0.19 2 1
316 4/24/98 D disc2 2DVR15 LB 0.75 -22 0.27 0.23 2 1
317 4/24/98 D disc 2DR30 LBR10 0.75 -22 0.92 0.44 7 3
318 4/24/98 D disc 2DL4 LB 0.75 -22 0.42 0.18 6 3
319 4/24/98 D disc 18 LBR40 0.5 -22 0.35 0.24 4 1
320 4/24/98 D disc 2DV(o.l.) LB 0.75 -22 0.57 0.22 7 3
321 4/24/98 D disc 2DVR15 LBR40 0.75 -22 0.27 0.22 3 1

194
 Table A-2. Run Log for Piloted Simulation
Run Conf. ID Stick gearing RMS errors Ratings
No. Date Pilot Task pitch roll KSE KSA pitch roll HQR PIOR Comments
322 4/24/98 D disc2 2DVR15 LBR40 0.75 -22 0.29 0.25 5 3
323 4/27/98 D disc 2DUR30 LB 2.358 -18 0.26 0.21 3 2 Stiff stick
324 4/27/98 D disc 2DR10 LB 2.358 -18 0.28 0.2 4 2 Stiff stick
325 4/27/98 D disc 2P LB 2.358 -18 0.26 0.19 3 1 Stiff stick
326 4/27/98 D disc 2DUR20 LB 2.358 -18 10 5 Stiff stick
327 4/27/98 D disc 2PR10 LB 2.358 -18 0.28 0.19 4 2 Stiff stick
328 4/27/98 D disc 2DR30 LBR40 2.358 -18 0.41 0.23 5 2 Stiff stick
329 4/27/98 D disc 2A LB 2.358 -18 0.28 0.21 5 3 Stiff stick
330 4/27/98 D disc 2DVR15 LB 2.358 -18 10 5 Stiff stick
331 4/27/98 D disc 2DL4 LB 2.358 -18 0.44 0.21 6 4 Stiff stick
332 4/27/98 D disc 5A LB 2.358 -18 0.29 0.23 6 4 Stiff stick
333 4/27/98 D disc 2DR30 LB 2.358 -18 0.2 0.18 2 1 Stiff stick
334 4/27/98 D disc 2DR30 LBR20 2.358 -18 0.3 0.29 7 4 Stiff stick
335 4/27/98 D disc 2DUR30 LB 2.358 -18 0.2 0.18 2 1 Stiff stick
336 4/28/98 C disc 1D LB 0.5 -22 0.3 0.18 3 2
337 4/28/98 C disc 2F LB 0.75 -22 0.3 0.17 5 3
338 4/28/98 C disc 2DR20 LB 0.75 -22 0.24 0.18
339 4/28/98 C disc 2DR20 LB 0.75 -22 0.21 0.18 1 1
340 4/28/98 C disc 2D16R10 LB 0.75 -22 0.23 0.19 2 2
341 4/28/98 C disc 2DR30 LBR20 0.75 -22 0.34 0.34 7 4
342 4/28/98 C disc 17DVR10 LB 1.1 -22 0.42 0.19 6 4
343 4/28/98 C disc 2DL2 LB 0.75 -22 0.26 0.18 4 1
344 4/28/98 C disc 2PR10 LB 0.75 -22 0.29 0.19 6 4
345 4/28/98 C disc 2DR30 LBTR60 0.75 -22 0.28 0.19 7 4
346 4/28/98 C disc 2DUR20 LB 0.75 -22 10 6 Lost control
347 4/28/98 C disc 2P LB 0.75 -22 0.27 0.17 2 1
348 4/28/98 C disc 2DG LB 0.75 -22 0.23 0.19 3 1
349 4/28/98 C disc 2PR20 LB 0.75 -22 0.27 0.17 5 2
350 4/28/98 C disc 2DUR30 LB 0.75 -22 0.22 0.18 3 1
351 4/28/98 C disc T33 LB 0.5 -22 0.23 0.18 3 1
352 4/28/98 C disc 2D18R20 LB 0.75 -22 0.23 0.17
353 4/28/98 C disc 2D18R20 LB 0.75 -22 0.25 0.18 4 2
354 4/28/98 G disc 2DR10 LB 0.75 -22 0.56 0.17 4 2
355 4/28/98 G disc 2DL4 LB 0.75 -22 0.66 0.17 5 4
356 4/28/98 G disc 17 LB 1.1 -22 0.43 0.17 3 2
357 4/28/98 G disc 1716R10 LB 1.1 -22 0.84 0.18 6 4
358 4/28/98 G disc 2DR30 LBR40 0.75 -22 0.5 0.27 5 4
359 4/28/98 G disc 16 LB 0.75 -22 0.68 0.21 8 5
360 4/28/98 H disc 2DR10 LB 0.75 -22 0.52 0.21 4 2
361 4/28/98 H disc 2DL4 LB 0.75 -22 0.58 0.23 5 4
362 4/28/98 H disc 17 LB 1.1 -22 0.5 0.27
363 4/28/98 H disc 17 LB 1.1 -22 0.36 0.2 3 2
364 4/28/98 H disc 1716R10 LB 1.1 -22 0.53 0.23 8 5
365 4/28/98 H disc 2DR30 LBR40 0.75 -22 0.45 0.36 Ratings based on this
366 4/28/98 H disc 2DR30 LBR40 0.75 -22 0.58 0.28 7 4 Would be 6/3 for this
367 4/28/98 H disc 16 LB 0.75 -22 0.82 0.29 9 5
368 4/28/98 H disc 2A LB 0.75 -22 0.56 0.25
369 4/28/98 H disc 2A LB 0.75 -22 0.47 0.22 6 3
370 4/28/98 H disc 1D LB 0.5 -22 0.4 0.24 7 5
371 4/28/98 H disc 2DVR20 LB 0.75 -22 0.61 0.25
372 4/28/98 H disc 2DVR20 LB 0.75 -22 0.48 0.23 5 3
373 4/28/98 H disc 2DR30 LB 0.75 -22 0.29 22 2 1
374 4/29/98 G disc 2P LB 0.75 -22 0.51 0.18 3 2
375 4/29/98 G disc 1D LB 0.5 -22 0.48 0.19 4 3
376 4/29/98 G disc 2A LB 0.75 -22 1.5 0.25 5 3
377 4/29/98 G disc 2DR30 LB 0.75 -22 1.2 0.21 3 2
378 4/29/98 G disc 1DL3 LB 0.5 -22 0.78 0.23 6 4
379 4/29/98 G disc 2DVR20 LB 0.75 -22 0.44 0.21 4 2
380 4/29/98 G disc 2DR30 LBR80 0.75 -22 0.53 0.22 3 2
381 4/29/98 G disc 2PR10 LB 0.75 -22 0.54 0.19 4 2
382 4/29/98 G disc 2DVR15 LB 0.75 -22 10 5
383 4/29/98 G disc 15 LB 0.75 -22 0.54 0.22 5 3
384 4/29/98 G disc 2DR30 LBR20 0.75 -22 0.6 0.35 6 4
385 4/29/98 G disc 2DL1 LB 0.75 -22 0.46 0.2 4 2
386 4/29/98 G disc 16 LB 0.75 -22 1.2 0.25 7 4

195
 Table A-2. Run Log for Piloted Simulation
Run Conf. ID Stick gearing RMS errors Ratings
No. Date Pilot Task pitch roll KSE KSA pitch roll HQR PIOR Comments
387 4/29/98 G disc 2D16R10 LB 0.75 -22 0.66 0.25 3 2
388 4/29/98 G disc 2DR30 LBTR60 0.75 -22 0.35 0.23 3 2
389 4/29/98 G disc 1DL2 LB 0.5 -22 0.46 0.25 5 4
390 4/29/98 G disc 2DUR30 LB 0.75 -22 9 5
391 4/29/98 G disc 18 LB 0.5 -22 0.46 0.25 7 4
392 4/29/98 G disc 2DL3 LB 0.75 -22 0.48 0.21 4 3
393 4/29/98 G disc 2DR30 LB 0.75 -22 0.31 0.22 3 2
394 4/29/98 G disc 2DR30 LBR60 0.75 -22 0.33 0.24 3 2
395 4/29/98 G disc 5A LB 0.75 -22 0.49 0.27 6 4
396 4/29/98 H disc 2P LB 0.75 -22 0.42 0.22 5 3
397 4/29/98 H disc 1DL3 LB 0.5 -22 0.51 0.23 7 5
398 4/29/98 H disc 2DR30 LBR80 0.75 -22 0.49 0.24 6 3 PIOR due to pitch!
399 4/29/98 H disc 2PR10 LB 0.75 -22 0.48 0.2 4 2
400 4/29/98 H disc 2DVR15 LB 0.75 -22
401 4/29/98 H disc 2DVR15 LB 0.75 -22 0.46 0.24 9 5
402 4/29/98 H disc 2DR30 LB 0.75 -22 0.33 0.19 5 3
403 4/29/98 H disc 15 LB 0.75 -22 0.47 0.23 7 5
404 4/29/98 H disc 2DR30 LBR20 0.75 -22 1.28 0.52 8 4
405 4/29/98 H disc 2DL1 LB 0.75 -22 0.38 0.2 6 3
406 4/29/98 H disc 2D16R10 LB 0.75 -22 0.31 0.19
407 4/29/98 H disc 2D16R10 LB 0.75 -22 0.26 0.2 3 2
408 4/29/98 H disc 2DR30 LBTR60 0.75 -22 0.29 0.22 3 2
409 4/29/98 H disc 1DL2 LB 0.5 -22 0.43 0.21 6 4
410 4/29/98 H disc 2DUR30 LB 0.75 -22 0.28 0.2 4 2
411 4/29/98 H disc 2DUR20 LB 0.75 -22 10 5
412 4/29/98 H disc 2DR30 LB 0.75 -22 0.34 0.21 5 4
413 4/29/98 H disc 18 LB 0.5 -22 0.33 0.22 6 1
414 4/30/98 G disc T33 LB 0.5 -22 0.33 0.2 4 2
415 4/30/98 G disc 2D LB 0.75 -22 0.61 0.2 4 2
416 4/30/98 G disc 2F LB 0.75 -22 0.43 0.19 4 2
417 4/30/98 G disc 17DVR20 LB 1.1 -22 0.27 0.18 3 2
418 4/30/98 G disc 2DR30 LT33 0.75 -18 0.35 0.21 3 2
419 4/30/98 G disc 17DVR10 LB 1.1 -22 0.6 0.28 7 4
420 4/30/98 G disc 2D16R10 LB 0.75 -22 0.45 0.21 5 3
421 4/30/98 G disc 2DR30 LBTR20 0.75 -22 0.34 0.28 6 4
422 4/30/98 G disc 2DR30 LB 0.75 -22 0.24 0.2 3 2
423 4/30/98 G disc 17DVR10 LB 1.1 -22 0.61 0.23 6 4
424 4/30/98 G disc 2PR20 LB 0.75 -22 0.35 0.22 3 2
1 2/1/99 I disc 2DR30 LB 0.75 -22 0.37 0.20 4 3
2 2/1/99 I disc 1D LB 0.50 -22 0.46 0.19 7 4
3 2/1/99 I disc 2P LB 0.75 -22 0.54 0.19 5 3
4 2/1/99 I disc 1D LB 0.25 -22 0.33 0.19 6 3
5 2/1/99 I disc 2DR30 LB 0.25 -22 1.60 0.23 7 2
6 2/1/99 I disc 2DR30 LBR40 0.75 -22 1.08 0.30 8 5
7 2/1/99 I disc T33 LB 0.50 -22 0.36 0.19 6 3
8 2/1/99 I disc 2DL2 LB 0.75 -22 0.31 0.21 3 2
9 2/1/99 I disc T33 LB 0.25 -22 0.34 0.21 4 2
10 2/1/99 I disc 2DVR20 LB 0.75 -22 0.49 0.27 8 5
11 2/1/99 B sos 2D LB 0.75 -22 Data for Pilot model
12 2/1/99 B sos 2H LB 0.75 -22 Data for Pilot model
13 2/1/99 B disc 2DR30 LB 0.75 -22 0.26 0.19 4 2
14 2/1/99 B disc 1D LB 0.50 -22 0.29 0.20 5 3
15 2/1/99 B disc 2P LB 0.75 -22 0.41 0.22
16 2/1/99 B disc 2P LB 0.75 -22 0.31 0.19 8 3
17 2/1/99 B disc 18 LB 0.50 -22 0.38 0.24 9 4
18 2/1/99 B disc 2DL2 LB 0.75 -22 0.37 0.20
19 2/1/99 B disc 2DL2 LB 0.75 -22 0.37 0.22 7 4
20 2/1/99 B disc 18 LB 0.25 -22 0.36 0.21 9 4
21 2/1/99 B disc 2DR30 LBR40 0.75 -22 0.60 0.23
22 2/1/99 B disc 2DR30 LBR40 0.75 -22 0.30 0.20 9 4
23 2/1/99 B disc 2DVR20 LB 0.75 -22 0.30 0.19 4 3
1 2/2/99 B disc T33 LB 0.50 -22 0.28 0.18
2 2/2/99 B disc T33 LB 0.50 -22 0.27 0.19 4 3
3 2/2/99 B disc 2DR30 LB 1.50 -22 0.29 0.20 3 3
4 2/2/99 B disc 18 LB 0.15 -22 0.43 0.19 8 4

196
 Table A-2. Run Log for Piloted Simulation
Run Conf. ID Stick gearing RMS errors Ratings
No. Date Pilot Task pitch roll KSE KSA pitch roll HQR PIOR Comments
5 2/2/99 B disc 2DR30 LBR60 0.75 -22 0.36 0.20 7 2
6 2/2/99 B disc T33 LB 0.25 -22 0.31 0.17 3 2
7 2/2/99 B disc 2DVR15 LB 0.75 -22 0.29 0.20 5 4
8 2/2/99 B disc 2DR10 LB 3.00 -28 0.32 0.20 Stiff Stick
9 2/2/99 B disc 2DR10 LB 3.00 -28 0.33 0.20 6 3 Stiff Stick
10 2/2/99 B disc 2DVR20 LB 3.00 -28 0.31 0.20 Stiff Stick
11 2/2/99 B disc 2DVR20 LB 3.00 -28 0.28 0.20 6 4 Stiff Stick
12 2/2/99 B disc 2DUR20 LB 3.00 -28 10 6 Stiff Stick
13 2/2/99 B disc 2DR30 LB 3.00 -28 0.29 0.18 5 3 Stiff Stick
14 2/2/99 I disc 2DR10 LB 3.00 -28 1.00 0.25 7 3 Stiff Stick
15 2/2/99 I disc 2DVR20 LB 3.00 -28 0.85 0.24 8 5 Stiff Stick
16 2/2/99 I disc 2P LB 3.00 -28 0.33 0.19 4 3 Stiff Stick
17 2/2/99 I disc 2DUR30 LB 3.00 -28 Stiff Stick
18 2/2/99 I disc 2DUR30 LB 3.00 -28 8 5 Stiff Stick
19 2/2/99 I disc 2DR30 LB 3.00 -28 0.40 0.20 4 2 Stiff Stick
20 2/2/99 I disc 2DL2 LB 3.00 -28 0.43 0.22 6 4 Stiff Stick
21 2/2/99 I disc 18 LB 0.50 -22 0.54 0.23
22 2/2/99 I disc 18 LB 0.50 -22 0.80 0.20 8 4
23 2/2/99 I disc 2DR30 LB 1.50 -22 0.30 0.20 5 3
24 2/2/99 I disc 18 LB 0.15 -22 0.68 0.22 7 4
25 2/2/99 I disc T33 LB 0.75 -22 0.44 0.20 7 4
26 2/2/99 I disc 2DL2 LB 0.50 -22 0.44 0.19 7 2
27 2/2/99 I disc 18 LB 0.25 -22 0.48 0.20 7 4
28 2/2/99 I disc 2DUR30 LB 0.75 -22 Lost control
29 2/2/99 I disc 2DUR30 LB 0.75 -22 9 5 Lost control
30 2/2/99 D sos 2D LB 0.75 -22 Data for Pilot model
31 2/2/99 D sos 2D LB 0.75 -22 Data for Pilot model
32 2/2/99 D sos 2H LB 0.75 -22 Data for Pilot model
33 2/2/99 D disc 2DR30 LB 0.75 -22 0.27 0.17 2 1
34 2/2/99 D disc 15 LB 0.50 -22 0.25 0.19 3 1
35 2/2/99 D disc 2DUR20 LB 0.75 -22 0.40 0.17 2 1
36 2/2/99 D disc 2DR30 LBR20 0.75 -22 2.50 0.42 7 4
37 2/2/99 D disc 2DVR15 LB 0.75 -22 0.34 0.18
38 2/2/99 D disc 2DVR15 LB 0.75 -22 10 5 Lost control
39 2/2/99 D sos 2D LB 0.75 -22 FB data for Pilot model
40 2/2/99 D sos 2D LB 0.75 -22 FB data for Pilot model
41 2/2/99 D sos 2H LB 0.75 -22 FB data for Pilot model
42 2/2/99 D sos 1G LB 0.50 -22 FB data for Pilot model
43 2/2/99 D disc 2DR30 LB 0.75 -22 0.40 0.19 5 1 Fixed Base
44 2/2/99 D disc 15 LB 0.50 -22 0.28 0.18 3 1 Fixed Base
45 2/2/99 D disc 2DUR20 LB 0.75 -22 0.27 0.17 2 1 Fixed Base
46 2/2/99 D disc 2DR30 LBR20 0.75 -22 0.29 0.29 5 1 Fixed Base
47 2/2/99 D disc 2DVR15 LB 0.75 -22 Fixed Base
48 2/2/99 D disc 2DVR15 LB 0.75 -22 0.31 0.19 Fixed Base
49 2/2/99 D disc 2DVR15 LB 0.75 -22 0.32 0.21 8 5 Fixed Base
50 2/2/99 D disc 2DR10 LB 0.75 -22 0.32 0.20 3 1 Fixed Base
51 2/2/99 D disc 17 LB 0.75 -22 0.25 0.18 2 1 Fixed Base
52 2/2/99 D disc 2DL4 LB 0.75 -22 0.51 0.20 6 3 Fixed Base
53 2/2/99 D disc 1DL3 LB 0.50 -22 0.76 0.25 7 4 Fixed Base
54 2/2/99 D disc 2H LB 0.75 -22 0.34 0.19 4 1 Fixed Base
55 2/2/99 D disc 2DUR20 LB 0.75 -22 10 5 Fixed Base
1 2/3/99 B disc 15 LB 0.50 -22 0.35 0.18 Fixed Base
2 2/3/99 B disc 15 LB 0.50 -22 0.31 0.19 4 3 Fixed Base
3 2/3/99 B disc 2DUR20 LB 0.75 -22 0.26 0.17 2 1 Fixed Base
4 2/3/99 B disc 2DR30 LBR20 0.75 -22 0.38 0.29 8 4 Fixed Base
5 2/3/99 B disc 2DVR15 LB 0.75 -22 0.31 0.19 3 1 Fixed Base
6 2/3/99 B disc 2DR10 LB 0.75 -22 0.25 0.17 Fixed Base
7 2/3/99 B disc 2DR10 LB 0.75 -22 0.24 0.18 3 1 Fixed Base
8 2/3/99 B disc 2DL4 LB 0.75 -22 0.40 0.21 8 3 Fixed Base
9 2/3/99 B disc 2DR30 LB 0.75 -22 0.23 0.18 3 1 Fixed Base
10 2/3/99 B disc 1DL3 LB 0.50 -22 0.56 0.24 Fixed Base
11 2/3/99 B disc 1DL3 LB 0.50 -22 0.49 0.24 9 4 Fixed Base
12 2/3/99 B disc 2DUR20 LB 0.75 -22 0.32 0.19 Fixed Base
13 2/3/99 B disc 2DUR20 LB 0.75 -22 0.35 0.20 7 2 Fixed Base
14 2/3/99 B sos 2D LB 0.75 -22 FB data for Pilot model

197
 Table A-2. Run Log for Piloted Simulation
Run Conf. ID Stick gearing RMS errors Ratings
No. Date Pilot Task pitch roll KSE KSA pitch roll HQR PIOR Comments
15 2/3/99 B sos 2H LB 0.75 -22 FB data for Pilot model
16 2/3/99 B sos 1G LB 0.50 -22 FB data for Pilot model
17 2/3/99 I disc 15 LB 0.50 -22 0.77 0.25 7 3 Fixed Base
18 2/3/99 I disc 2DUR20 LB 0.75 -22 Fixed Base
19 2/3/99 I disc 2DUR20 LB 0.75 -22 10 5 Fixed Base
20 2/3/99 I disc 2DR30 LBR20 0.75 -22 1.46 0.54 7 5 Fixed Base
21 2/3/99 I disc 2DR30 LB 0.75 -22 0.29 0.20 3 2 Fixed Base
22 2/3/99 I disc 2DVR15 LB 0.75 -22 Fixed Base
23 2/3/99 I disc 2DVR15 LB 0.75 -22 1.53 0.30 8 5 Fixed Base
24 2/3/99 I disc 2DR10 LB 0.75 -22 0.33 0.21 4 2 Fixed Base
25 2/3/99 I disc 2DL4 LB 0.75 -22 1.01 0.22 7 4 Fixed Base
26 2/3/99 I disc 1DL3 LB 0.50 -22 1.10 0.25 7 5 Fixed Base
27 2/3/99 I disc 17 LB 0.75 -22 0.27 0.20 3 1 Fixed Base
28 2/3/99 I disc 2DVR20 LB 0.75 -22 10 6 Fixed Base
29 2/3/99 I sos 2D LB 0.75 -22 FB data for Pilot model
30 2/3/99 I sos 2D LB 0.75 -22 FB data for Pilot model
31 2/3/99 I sos 2H LB 0.75 -22 FB data for Pilot model
32 2/3/99 I sos 1G LB 0.50 -22 FB data for Pilot model
33 2/3/99 H disc 2DR30 LB 0.75 -22 0.28 0.20 5 2
34 2/3/99 H disc 18 LB 0.50 -22 0.52 0.22 8 5
35 2/3/99 H disc 2DR30 LBR60 0.75 -22 0.60 0.28
36 2/3/99 H disc 2DR30 LBR60 0.75 -22 0.31 0.24 7 4
37 2/3/99 H disc T33 LB 0.50 -22 0.39 0.21 6 4
38 2/3/99 H disc 18 LB 0.15 -22 0.66 0.23 8 5
39 2/3/99 H disc 2DR30 LB 1.50 -22 0.37 0.25
40 2/3/99 H disc 2DR30 LB 1.50 -22 0.29 0.22 6 3
41 2/3/99 H disc 1D LB 0.50 -22 0.38 0.19
42 2/3/99 H disc 1D LB 0.50 -22 0.36 0.21 7 5
43 2/3/99 H disc 17 LB 0.75 -22 ? ?
44 2/3/99 H disc 17 LB 0.75 -22 0.30 0.19 4 1
45 2/3/99 H disc T33 LB 0.25 -22 0.62 0.22
46 2/3/99 H disc T33 LB 0.25 -22 0.39 0.24 5 4
47 2/3/99 H disc 2DVR15 LB 0.75 -22 0.69 0.23 9 5
48 2/3/99 H disc 2DR30 LB 0.75 -22 0.34 0.20 2 1
49 2/3/99 H disc 1D LB 0.25 -22 0.37 0.20 6 5
50 2/3/99 H disc 18 LB 0.25 -22 0.55 0.22
51 2/3/99 H disc 18 LB 0.25 -22 0.44 0.23 8 5
52 2/3/99 H sos 2D LB 0.75 -22 Data for Pilot model
53 2/3/99 H sos 2H LB 0.75 -22 Data for Pilot model
54 2/3/99 H sos 1G LB 0.50 -22 Data for Pilot model

K. REFERENCES FOR APPENDIX A

1
Kish, Brian A., Capt., USAF, et al., A Limited Flight Test Investigation of Pilot-Induced
Oscillation due to Elevator Rate Limiting, AFFTC-TR-97-12, June 1997.
2
Klyde, David H., and David G. Mitchell, Handling Quality Demonstration Maneuvers for
Fixed-Wing Aircraft, Volume II: Maneuver Catalog, WL-TR-97-3100, Oct. 1997.
3
Neal, T. Peter, and Rogers E. Smith, An In-Flight Investigation to Develop Control System
Design Criteria for Fighter Airplanes, AFFDL-TR-70-74, Volume I, Dec. 1970.
4
Iloputaife, Obi, “Minimizing Pilot-Induced-Oscillation Susceptibility During C-17
Development,” AIAA Atmospheric Flight Mechanics Conference, New Orleans, LA, Aug.
1997, pp. 155-169.

198
 APPENDIX B. AN EXAMINATION OF THE RESULTS OF HAVE CAP

A. BACKGROUND
The HAVE CAP flight experiment1 was conducted by students of the USAF Test Pilot School in
the Fall of 1995 with sponsorship from the Flight Dynamics Laboratory, Wright-Patterson AFB,
Ohio. Hoh Aeronautics, Inc. (HAI) was involved in an observation and advisory role.
As with most handling-qualities flight data, the information provided by HAVE CAP is certain
to be reanalyzed, dissected, interpreted, and generally scrutinized for many years to come. Since
it is clear that the project was well-conducted and generated valid data, it is critical that the
documentation be as thorough and correct as possible. This Appendix presents a brief
examination of the project. A portion is devoted to expanding the documentation of the
configurations flown, and especially to noting several minor errors discovered in Ref. 1. It is
based on a combination of a review of the Ref. 1 documentation, discussions with participants,
and most importantly, assessment of data files provided to HAI by Capt. David A. Kivioja,
Project Manager for HAVE CAP.
HAVE CAP illustrated some fundamental shortcomings with a time-domain version of the pitch
attitude Dropback/pitch rate overshoot criterion, developed by Gibson as a design criterion2 and
later expanded for specification of flying qualities. 3 Because of these shortcomings, inherent to
any time-domain-based criterion, a search for an alternative definition was begun. Since the
most important portion of this criterion is pitch rate overshoot (the Dropback portion is important
only in that it reconciles the apparent effect of flight condition on acceptable values of
overshoot3), a frequency-domain measure of overshoot was developed, as defined in the main
document for this report.

B. ERRATA FOR THE HAVE CAP REPORT

Following are recommended corrections for the HAVE CAP report.1 They consist of
typographical errors, suspected errors in measurement for the dynamics of several of the HAVE
CAP configurations, and new measurements of handling-qualities parameters. That any of these
errors could be detected is due entirely to the documentation, in the form of data files, provided
by Capt. Kivioja. Errors such as those described below are probably not unusual for a flight test
report; the opportunity to detect them, and report on the appropriate corrections, is unusual.

1. Summary of Corrections
The interested reader may wish to read the remainder of this section to learn the details of each
error. If not, here is a brief summary of the changes that should be noted in Ref. 1. (The most
concise location for making these changes is in Table A1, page 35 of Ref. 1.)
• Configuration A: /2(6 WLPH GHOD\ 2 ≈  VHF QRW  &BWg may be near zero
• Configuration D: &BWg is approximately zero (not 6.1)
• Configuration H: Magnitude Bode plot in Ref. 1 (Figure J62, page 172) is incorrect (a
FRUUHFWHG ILJXUH LV VKRZQ EHORZ &BWg  UDGVHF QRW  &BW = 3.8 rad/sec (not
2.3)
• Configuration I: Magnitude Bode plot in Ref. 1 (Figure J74, page 184) is incorrect (a
FRUUHFWHG ILJXUH LV VKRZQ EHORZ &BWg &BW = 4.6 rad/sec (not 3.0)
199
• Configuration K: Typographical error in Table A1 of Ref. 1 &BWp = 1.9 rad/sec, not 1.4
RYHUDOO &BW is correct)

2. Configuration A
VSS Configuration A in Ref. 1 is different from the others in several ways. First, the Lower-
2UGHU (TXLYDOHQW 6\VWHP /2(6 WLPH GHOD\ 2, of 0.040 sec is about half that of the other
configurations, especially those nearest in frequency and damping (Configurations C2 and D).
Second, the coherence for the frequency-response data (Figure J3) is significantly poorer than
that for any other configuration (though there is still a region over which the coherence is
reasonably good), with a rapid falloff in coherence at frequencies above about 8 rad/sec. Third,
it is highly unusual to have an airplane with such a wide frequency separation between 1/T2 and
&sp and not have a pitch attitude Bandwidth that is significantly gain-margin-limited.
For the equivalent time delay, it appears that the value obtained was too low. For example,
Figure B-1 shows a comparison Bode plot of the frequency-response data (diamond symbols)
and the resulting LOES (solid lines). While the amplitude ratio match is quite good, and phase is
good at frequencies near the short-period, there is a definite difference in phase at higher
frequencies as the LOES exhibits too little rolloff. There is a nonlinear effect, since any added
equivalent delay would degrade the match at the lower frequencies. Based solely on the high-
frequency data, however, it appears that a more reasonable equivalent time delay would be
around 0.08 sec, or near the values for Configurations C2 and D.
30
25
20
15
|q/Fes| (dB)

10
5
0
-5
-10
-15
-20
0.1 1.0 10.0 100.0
frequency (rad/sec)

200
150
100
Φ(q/Fes) (deg)

50
0
-50
-100
-150
-200
0.1 1.0 10.0 100.0
frequency (rad/sec)

Figure B-1. Frequency-Response Data and LOES for Configuration A

200
The apparent nonlinearity in Figure B-1 may be due to the lack of good coherence for this
configuration. There is evidence that, even though there is some high-coherence data, overall the
coherence was not very good and the entire frequency response may be questionable. The
coherence plot in Figure J3 of Ref. 1 shows significant variation in coherence for adjacent
frequency points, which is usually evidence of poor input or output power at these points. In
addition, the time history of the pilot-generated frequency sweeps indicates that the pilot stopped
the sweeps before reaching very high frequencies. Figure B-2 is an example time history from
the sweeps; the upper two plots are stick force and pitch rate for the first sweep and the lower are
for the second sweep. The highest frequency reached by the pilot was on the order of one cycle
per second, or barely up to the short-period mode, which was at 5.7 rad/sec. Any data above this
frequency comes entirely from other inputs made during the sweep and thus have little input
power. Hence all the frequency-response data for Configuration A above about 6 rad/sec is
suspect and is probably no good.
By way of contrast, Figure B-3 is an example of a sweep (for Configuration G) that has much
higher input frequencies (probably higher than necessary). This sweep, too, has some
shortcomings, as it, like those in Figure B-2, is completed too quickly, with not enough input at
low frequencies, and possibly with input amplitudes that are marginally too small (the pitch rate
signal at the higher frequencies is not much larger than sensor noise).
Based on a review of the frequency-sweep time histories from HAVE CAP, two
recommendations can be made: 1) a sweep of Configuration A should have been run again, and
2) pilots who are to perform sweeps should always be coached on proper technique. For the
latter, the general rules for frequency-sweeping listed in Table A-1, Appendix A, of Ref. 4 would
be a good starting point.
As documented in Ref. 1, the pitch attitude Bandwidth for Configuration A is around 7.8 rad/sec,
with the gain margin and phase margin values almost equal. These numbers are taken from the
flight-derived frequency responses. Aircraft with a long "shelf" in the attitude response, as
exhibited by Configurations A, C2, and D, typically have very low gain margin Bandwidths as a
result. There is no evidence of this in the frequency responses for this case, however.
Given the low overall coherence for the frequency response, and the apparent nonlinear
response, it seems appropriate to look at the theoretical linear dynamics for this case. A transfer
function model for the theoretical Configuration A was assembled by combining the LOES
short-period mode (which is close to the mode originally requested from Calspan, as shown in
Figure 2 of Ref. 1) with the documented VISTA feel system and actuator dynamics (page 99 of
Ref. 1 7KH )es transfer function for this model is as follows:
θ 7.9 x1010 (0.45)
=
Fes (0)(35)(114)[0.38, 5.68][0.7, 30][0.57,70.7]
This theoretical model is highly gain-margin-limited, with a Bandwidth frequency of 0.49
rad/sec, compared to 7.8 for the flight model. The Phase Delay is 0.085 sec, consistent with the
other configurations but not with the flight-identified value. If there is an error in the flight data,
Configuration A may have been gain-margin-limited, as it was designed to be.

201
 )(6B08;







         









        





)(6B08;







         









        





Figure B-2. Time Histories of Frequency Sweeps for Configuration A

)(6B08;







         









        





Figure B-3. Time History of a Sweep with High-Frequency Inputs (Configuration G)

202
3. Configuration D
Reference 1 reports both the gain and phase Bandwidths of Configuration D as 6.1 rad/sec.
While the scatter in the frequency-response data allows some room for individual interpretation,
closer inspection of the Bode plots for this case in Ref. 1 suggests that it was gain-margin-limited
– with essentially zero Bandwidth.
Figure B-4 is a plot of the magnitude Bode data and corresponding LOES. If we measure up 6
G% IURP WKH JDLQ DW &180, we find that the LOES is below 6 dB at its peak. Thus, the gain margin
Bandwidth was unmeasurable, and had to be below 1 rad/sec (the lower limit of good
coherence). In reality this configuration probably had a Bandwidth around one-half radian per
second (due to the phugoid mode), and the Bandwidth for the LOES is 0.4 rad/sec, so it is well
below the Level 1 limit on Bandwidth.

4. Configurations H and I
The Bode magnitude plot for Configuration H, Figure J62 of Ref. 1, is not correct. This was
discovered while comparing the reported LOES models with the flight-derived frequency
responses.
Figure B-5 shows the magnitude data for Configuration H as plotted in Ref. 1 (only the data
between 1 and 20 rad/sec are shown for convenience). Also shown on this figure are the data
points used for obtaining the LOES match (the match was actually done on pitch rate; the data
have been corrected to attitude for comparison purposes) and the resulting LOES model. The
LOES model is a good match, but the data from Figure J62 look very different. By inspecting
the data files used to generate Figure J62 it was learned that the conversion from pitch rate to
attitude for that figure was incorrect. For most plots in the report the attitude Bode was created
E\ DGMXVWLQJ WKH UDWH PDJQLWXGH IRU V LH *DLQdB = Gain(q)dB + 20log10M& *DLQTdB -
20log10M& )RU WKLV FRQILJXUDWLRQ WKH IDFWRU RI  ZDV LQDGYHUWHQWO\ RPLWWHG ZKHQ )LJXUH -
was created.

20

10
|θ/Fes| (dB)

-10

-20
0.1 1.0 10.0 100.0
frequency (rad/sec)

Figure B-4. Magnitude Bode Plot for Configuration D

203
 30
25
20
15
10
|θ/Fes| 5 Flight data
(dB) 0 LOES
-5 Figure J62

-10
-15
-20
-25 1 10 100
frequency (rad/sec)

Figure B-5. Comparison of Pitch Attitude Magnitude Plots for Configuration H

Figure B-6 is a corrected version of Figure J62 from Ref. 1. From this figure a gain-margin
Bandwidth of 5.0 is obtained. Thus the actual Bandwidth for Configuration H is 3.8 rad/sec
(phase-margin Bandwidth).
As with Configuration H, the amplitude ratio of pitch rate for Configuration I was not correctly
converted to attitude for the Bode plot (Figure J74 in Ref. 1). Figure B-7 is a corrected version,
from which gain-margin Bandwidth is 4.6 rad/sec and the overall Bandwidth for this
configuration is also 4.6 rad/sec (gain-margin Bandwidth).

C. MEASUREMENTS OF PITCH RATE OVERSHOOT AND PITCH ATTITUDE

DROPBACK
The HAVE CAP project was the first attempt at applying the time-domain Dropback/overshoot
criteria3 in a flight program. The parameters for these criteria, like all time-based metrics,
proved to be difficult to obtain in flight. The primary problem is in defining "steady state" pitch
rate, especially if a normal phugoid mode exists. This appears to have been the case for the
HAVE CAP configurations. The problems exposed by HAVE CAP served as the catalyst for the
development of a more robust, frequency-domain version of Dropback and overshoot. The first
results of this effort are documented in the main text of this report and applied to the HAVE CAP
configurations in this appendix.
Dropback as a handling-qualities criterion was found to have only minor effects on data
correlation. Pitch rate overshoot, however, has a more fundamental influence and is the only
parameter currently used by the criteria. The frequency response measurement for pitch rate
overshoot is defined in Figure B-8.

204
 25
20
15
10
5
0
Magnitude (dB)

-5
Pitch Attitude

-10
-15
-20
-25
-30
-35
-40
-45
-50
0.1 1.0 10.0 100.0
Frequency (rad/sec)

Figure B-6. Corrected Bode Magnitude Plot for Configuration H (Ref. 1 Figure J62)
20

15

10

0
Magnitude (dB)
Pitch Attitude

-5

-10

-15

-20

-25

-30

-35

-40
0.1 1.0 10.0 100.0
Frequency (rad/sec)

Figure B-7. Corrected Bode Magnitude Plot for Configuration I (Ref. 1 Figure J74)

205
 ∆G (q )
q
F es

(d B )

0. 0 1 0. 1 1 10

F re qu en cy, ω (rad /sec)

Figure B-8. Definition of Pitch Rate Overshoot in the Frequency Domain

Application of the frequency-domain overshoot parameter requires a fairly complete frequency

response plot, covering the frequency range between the phugoid and short-period mode for
conventional airplanes. Unfortunately, the flight-derived frequency responses from HAVE CAP
are not useful for this measurement. The coherence plots in Ref. 1 show that little information
can be obtained below about one radian per second; thus, while it is possible to get a reasonable
estimate of the short-period frequency and the related magnitude peak on a frequency response
of pitch rate from the flight data, it is not possible to determine the associated "shelf" in the
response below the short-period.
For this analysis, it was necessary to use reconstructed transfer-function models of the HAVE
CAP configurations. The models were devised by combining the LOES models of short-period
frequency with an assumed set of low-frequency dynamics. Based on other data sources, in
addition to the analysis of the pitch rate responses, a reasonable phugoid mode for the F-16 is
approximately [0.2, 0.2]. This phugoid mode, along with a zero 1/Tθ2 = 0.03 rad/sec, was added
to each of the LOES short-period modes and time delay values. In reality the frequency and
damping of the phugoid mode probably varied between configurations, but relatively large
variations in phugoid would result in only small changes in the measured values of pitch rate
overshoot.
Using the analytical models, it was found that Configurations A, C2, D, and J exhibited
excessive overshoot. This is in contrast to the time-domain measures reported in Ref. 1, for
which all 10 configurations had excessive overshoot.

D. EVALUATION OF RESULTS USING BANDWIDTH AND REVISED OVERSHOOT

Figure B-9 shows the 10 HAVE CAP configurations on the pitch attitude Bandwidth and Phase
Delay requirements from the main text of this report. The labels identify each case and list the
average Handling Qualities Rating, with average PIO Tendency Rating in parentheses. Solid
206
cases exhibited PIO in flight. Table B-1 also lists the dynamics of the cases. Both Figure B-9
and Table B-1 reflect the new Bandwidth numbers for some of the cases, and Table B-1 includes
WKH SLWFK UDWH RYHUVKRRW SDUDPHWHU û*T IRU DOO WKH FRQILJXUDWLRQV
Table B-1. Bandwidth/Overshoot Values for Have Cap Configurations
Bandwidth, Phase Delay, Overshoot,** Handling Qualities Level
Aircraft
Configuration
ωBWθ τ pθ û*T Predicted Flight Test
(rad/sec) (sec) (dB)
A 0.5* 0.079 19.6 3 3
C2 6.7 0.084 14.1 2 2
D 0.4* 0.077 21.5 3 3
E 2.8 0.079 8.7 1 1
G 3.6 0.071 6.5 1 1
H 3.8* 0.074 4.1 1 1
I 4.6* 0.071 8.3 1 2
J 1.7 0.078 12.6 2 2
K 1.9 0.082 4.8 2 1-2
P 1.4 0.077 5.2 2 3
*
Differs from value reported in Ref. 1
**
Measured from analytical model
Correlation with the criteria, using the new definition of pitch rate overshoot and new values of
pitch attitude Bandwidth, is improved significantly over that reported in Ref. 1. This is primarily
a result of the replacement of time-domain pitch attitude Dropback with frequency-domain pitch
rate overshoot. The main text has the following handling qualities Level definitions. Level 1: in
/HYHO  UHJLRQ RQ %DQGZLGWK DQG û*T   G% /HYHO  LQ /HYHO  UHJLRQ RQ %DQGZLGWK RU LQ
/HYHO  UHJLRQ ZLWK û*T ≥ 9 dB. Level 3: in Level 3 region on Bandwidth, or in Level 2
UHJLRQ ZLWK D FRPELQDWLRQ RI &BW   UDGVHF DQG û*T ≥ 12 dB. Using these definitions, the
handling qualities Levels are correctly predicted for eight of the ten cases. The exceptions are
&RQILJXUDWLRQV , /HYHO  %DQGZLGWK EXW û*T  G% DQG 3 /HYHO  SUHGLFWHG /HYHO 

E. PIO ANALYSIS

1. Occurrence of PIOs
Three of the ten HAVE CAP configurations experienced PIOs. The three configurations with
PIOs (solid symbols on Figure B-9) were A (PIO Tendency Ratings of 4/4,4/4/4 from Pilots
1/2/3/4), D (PIORs of 4/4,4/-/4), and P (PIORs of 5,3/4/4/2). Some pilots saw the PIO on
Configurations A and D as pitch bobble or as "a cross between a pitch bobble and a PIO." One
landing with Configuration A was aborted because of the development of a divergent PIO.
Two other configurations exhibited what was reported to be a pitch bobble, but not a PIO: C2
(PIORs of 4/-/3,3/-), with comments about a "pitch ratchet" that "could not be described as a PIO
but more like a bobble-type effect;" and I (PIORs of 1,2/1/-/2), with comments that "pitch
response has a bobble that's annoying but doesn't compromise task performance." A mild bobble
tendency was noted by one of the two pilots who flew Configuration E (PIORs of -/2/1,1/-):
"Small pitch bobble noted encouraging minimal smoothing of inputs." Similarly, one pilot
reported "slow oscillations in pitch (not divergent)" for Configuration J (PIORs of 1/3/-/1,1).

207
 0.2

0.15

Pitch
Attitude
Phase
Delay, 0.1 A:
K:
τpθ P: 3.8 (1.3) H:
7.0 (4.0)
6.8 (3.6) 2.0 (1.0)
(sec)
C2:
D: J: E:
5.3 (3.3)
7.3 (4.0) 5.1 (1.5) 1.7 (1.3) G: I:
2.5 (1.0) 4.5 (1.5)
0.05
Level 2 Level 2
("Moderate" (No PIO) Level 1 (No PIO)
PIO if Bobble if Bobble if Overshoot
Overshoot Overshoot Excessive
Excessive) Excessive

0
0 1 2 3 4 5 6 7
Pitch Attitude Bandwidth, ωBWθ (rad/sec)
Figure B-9. Bandwidth Characteristics of HAVE CAP Configurations (Solid Symbols
Denote Configurations with PIO in Flight)

A PIOR of 3 was given to Configuration K (PIORs of 3/1/1/1,1,1) because of "undesirable pitch

motions resulting from large, fast inputs." This does not appear to be a pitch bobble.
No tendency to PIO or bobble was found for Configurations G, H, and K.
In summary, in HAVE CAP:
• Configurations G, H, and K had no bobble or PIO tendency;
• Configurations E and J had a mild bobble tendency;
• Configurations C2 and I had pitch bobble;
• Configurations A, D, and P exhibited PIOs.

2. Predictions Based on Pitch Attitude Bandwidth and Overshoot

The PIO susceptibility prediction criteria in this report are based on a combination of pitch
attitude and flight path Bandwidths and pitch rate overshoot. Ignoring the flight path
requirement (since we do not know the flight path Bandwidths of the HAVE CAP
configurations), the PIO requirements may be summarized as follows:
• Not susceptible to PIO if ωBWθ > 1 rad/sec and τ pθ < 0.14 sec;

• Pitch bobble if ωBWθ > 1 rad/sec, τ pθ < 0.14 sec, and û*T !  G%

208
• Susceptible to PIO if ωBWθ   UDGVHF DQG û*T !  G%

• Always susceptible to PIO if τ pθ ≥ 0.14 sec.

On this basis, using the numbers for Bandwidth and overshoot listed in Table B-1, we get the
results listed in Table B-2.

Table B-2. Application of Bandwidth/Overshoot to HAVE CAP for PIO

Bandwidth, Phase Delay, Overshoot, PIO/Bobble?
Aircraft
Configuration
ωBWθ τ pθ û*T Predicted Flight Test
(rad/sec) (sec) (dB)
A 0.5 0.079 19.6 PIO PIO
C2 6.7 0.084 14.1 Bobble Bobble
D 0.4 0.077 21.5 PIO PIO
E 2.8 0.079 8.7 No* Small bobble
G 3.6 0.071 6.5 No No
H 3.8 0.074 4.1 No No
I 4.6 0.071 8.3 No* Small bobble
J 1.7 0.078 12.6 Bobble Slow oscills.
K 1.9 0.082 4.8 No No
P 1.4 0.077 5.2 No PIO
*
Overshoot value is close to bobble limit of 9 dB

Comparing this list with the flight results shows extremely encouraging agreement: the two
"marginal" bobblers got PIORs of either 1 or 2; A and D did indeed PIO; and G, H, and K did
not exhibit either PIO or bobble.
The lone exception – and a serious one – is Configuration P, which "passed" the PIO criteria but
had a real PIO tendency. It is not known why this case was so bad; the pitch attitude Bandwidth
of 1.4 rad/sec may be an indicator, since the low Bandwidth results not from gain-margin
limiting (as is the case with the lowest-Bandwidth cases, A and D), but from a very low short-
period frequency. On the other hand, one may also suspect some influence from the
control/response sensitivity. Configuration P was evaluated five times, with one pilot noting that
"Stick dynamics are fine, but the airplane is lousy" for one evaluation, but the "forces [are] too
high" on a repeat. One pilot did not comment on the forces; another reported "large longitudinal
forces" while a different pilot commented "forces/displacements too light. Stick too sensitive.
Difficult to determine the contribution this makes to the PIO tendencies [emphasis added]." The
confusion about stick forces may be a manifestation of the very poor, PIO-prone nature of this
configuration, or it may be a sign that the forces were not optimum.

F. REFERENCES FOR APPENDIX B

1
Kivioja, David A., Capt., USAF, et al., A Limited Evaluation of Predicting Pilot Opinion of
Aircraft Handling Qualities in the Landing Phase of Flight Using the Control Anticipation
Parameter and Bandwidth Criterion (HAVE CAP), AFFTC-TR-95-78, Jan. 1996.

209
2
Gibson, John C., “Piloted Handling Qualities Design Criteria for High Order Flight Control
Systems,” Criteria for Handling Qualities of Military Aircraft, AGARD CP-333, Apr. 1982,
pp. 4-1 – 4-15.
3
Mitchell, David G., Roger H. Hoh, Bimal L. Aponso, and David H. Klyde, Proposed
Incorporation of Mission-Oriented Requirements into MIL-STD-1797A, WL-TR-94-3162,
Oct. 1994.
4
Hoh, Roger H., David G. Mitchell, Bimal L. Aponso, David L. Key, and Chris L. Blanken,
Background Information and User's Guide for Handling Qualities Requirements for Military
Rotorcraft, USAAVSCOM TR-89-A-008, Dec. 1989.

210
 APPENDIX C. ANALYSIS OF DATA FROM NAVY STUDY OF F-14 DUAL
HYDRAULIC FAILURES*

A. F-14 DUAL HYDRAULIC FAILURE FLIGHT TEST PROGRAM

From October 1990 to March 1991 the U.S. Navy conducted a flying qualities evaluation of the
F-14 with simulated dual hydraulic failure.1 In this study, the back-up flight control module or
BUFCM was evaluated to define areas of operation for in-flight refueling and landing. The
BUFCM has two modes that are available to the pilot. The BUFCM-HIGH mode features a
maximum stabilator rate of 10 deg/sec, while BUFCM-LOW has a 5 deg/sec maximum rate.
Although the aircraft demonstrated good handling qualities using the BUFCM in formation flight
with a tanker, a number of PIOs were encountered during in-flight refueling, drogue tracking,
and offset field landings. Because the F-14 was fully instrumented, a valuable PIO database was
created.

B. ANALYSIS TECHNIQUES
Fast Fourier transform (FFT) analysis techniques were used to obtain vehicle describing
functions and pilot input power spectral densities. The FFT algorithm is embedded in FREDA, a
frequency domain analysis software package proprietary to Systems Technology, Inc. An
important feature of FREDA is the ability to average data in both linear and log spaced bins.
The quality of the data obtained in the FFT is identified by the coherence (ρ2), a measure of how
well the system output is correlated with the input. For the work presented herein, high quality
points are those that have a coherence greater than 0.8. Pitch rate and pitch attitude to stick force
describing functions were used to compute the airplane bandwidth parameters. The pilot input
power spectral densities were used to identify task bandwidths and PIO frequencies.

C. FLIGHT TEST DATA ANALYSIS

The Naval Air Warfare Center, Aircraft Division (NAWCAD), in Patuxent River, MD provided
selected flight test data from F-14 Dual Hydraulic Failure flight test program to HAI in digital
ASCII and Matlab file formats with corresponding documentation. These high-quality data
included a subset of the flight test runs and a number of frequency sweep runs that were
generated as part of the flight test program. The run numbers and configurations identified in
this appendix correspond to those used in this program for in-flight refueling, drogue tracking,
offset field landings, and frequency sweeps. The altitudes, airspeeds, and Mach numbers
provided here were taken directly from the digital data files. In the work reported herein,
individual runs are characterized by aircraft configuration (wing sweep and gear and flap
positions), flight condition (altitude, knots indicated airspeed, and Mach number as recorded at
the start of a run), and flight control mode (SAS On, SAS Off, BUFCM-HIGH, or BUFCM-
LOW).
The frequency sweep data identified in Table C-1 were used to obtain pitch rate to stick force
(q/Flon), pitch rate to stabilator (q/δstb), pitch attitude to stick force (θ/Flon), and normal
acceleration to stick force (Nz/Flon) describing functions. The quality, duration, and frequency
range of a given sweep all impacted the ability to provide “good” data as indicated by coherence
values of 0.8 or higher.

*
This Appendix was co-authored by David H. Klyde, Systems Technology, Inc.
211
Attempts were also made to obtain describing function data from the high gain aerial refueling
tasks. A run log for the aerial refueling cases is provided in Table C-2.
The hook-up task was considered high-gain relative to normal in-flight refueling in a two-seat F-
14 in which the distance to the basket is called out by the back-seater. In general, the drogue
tracking task forced the pilot into a tighter, that is more closed-loop, operating environment than
the hook-up task. As a result, the drogue tracking cases tend to have more high coherence data.
Similar results were obtained from data collected in a probe-and-drogue refueling study
conducted by the USAF Test Pilot School.2
The data provided by the Navy also contained an offset landing run as indicated in Table C-3. In
general and in this case, the lateral offset landing task does not provide enough excitation of the
longitudinal axis dynamics to generate high-coherence frequency-response data.
Time history plots of stick force and pitch rate for an example frequency sweep, drogue hook-up,
and drogue tracking run are shown in Figure C-1. Note that the time scales are different for each
of the examples.

D. ANALYSIS RESULTS

1. Identification of Stick Dynamics

Longitudinal stick position to stick force describing functions for the BUFCM-HIGH and
BUFCM-LOW 35° wing sweep frequency sweep runs are shown in Figure C-2. As indicated in
the coherence plot, high-coherence data are present in the frequency range from 0.4 to 8.5
rad/sec. In this range no significant differences are noted between the BUFCM-HIGH and
BUFCM-LOW runs. A low order equivalent systems (LOES) fit to the describing function data
is also shown as a solid line on the magnitude and phase plots. To fully define the effects of the
rate limiting nonlinearity, however, a series of frequency sweeps should be conducted for a
complete range of stick force levels.

Table C-1. Run Log of Frequency Sweep Cases

Run No. Configuration* Flight Control Altitude Airspeed Mach No.

System ID (ft) (KIAS)

R001G0.31M CR-35-0 BUFCM-HIGH 15,300 255 0.51

R001G0.31M CR-35-0 BUFCM-LOW 15,500 249 0.50

R012G0.31M CR-20-0 BUFCM-HIGH 15,000 191 0.38

R003G0.31M CR-20-0 BUFCM-HIGH 15,000 240 0.48

R016G0.31M CR-20-0 BUFCM-HIGH 15,400 300 0.60

R016G0.31M CR-20-0 BUFCM-LOW 15,400 300 0.60

R017G0.31M CR-20-0 SAS Off? 15,000 296 0.58

*
Configuration CR-XX-YY denotes landing gear retracted, wing sweep at XX deg, and flaps at YY deg

212
 Table C-2. Run Log of Aerial Refueling Cases

Run No. Configuration* Maneuver Flight Control Altitude Airspeed Mach No.
System ID (ft) (KIAS)

J_07 CR-35-0 Hook-Up SAS On 17,000 256 0.53

J-08 CR-35-0 Hook-Up SAS Off 17,000 252 0.52

J_09 CR-35-0 Tracking BUFCM-HIGH 16,000 256 0.52

J_01 CR-20-0 Hook-Up SAS On 14,000 262 0.51

J_03 CR-20-0 Hook-Up SAS Off 20,000 307 0.67

J_04 CR-20-0 Hook-Up SAS Off 20,000 200 0.44

J_05 CR-20-0 Hook-Up BUFCM-HIGH 19,000 200 0.42

J_06 CR-20-0 Hook-Up BUFCM-HIGH 19,000 200 0.42

J_36 PA-20-10 Hook-Up BUFCM-HIGH 14,400 181 0.37

J_37 PA-20-10 Tracking BUFCM-LOW 14,500 183 0.36

*
Configuration CR-XX-YY denotes landing gear retracted, wing sweep at XX deg, and flaps at YY deg;
Configuration PA-XX-YY denotes landing gear extended, wing sweep at XX deg, and flaps at YY deg

Table C-3. Run Log of Field Offset Landing Run

Run No. Configuration* Maneuver Flight Control Altitude Airspeed Mach No.
System ID (ft) (KIAS)
J_56 PA-20-35 Offset Landing BUFCM-HIGH 720 → 0 143 0.22
*
Configuration PA-XX-YY denotes landing gear extended, wing sweep at XX deg, and flaps at YY deg

2. Effects of Rate Limiting

Magnitude and phase Bode plot and coherence comparisons for all of the available 35° wing
sweep cases are shown in Figure C-3. Note by comparing Table C-1 and Table C-2 that these
five runs were flown at essentially the same flight condition. As indicated by the coherence plot,
only high coherence data points, that is ρ 2 > 0.8 , were included in the comparison. The phase
comparison indicates that the BUFCM cases (triangles and diamonds), as expected, have a
significant phase rolloff when compared to the SAS On (circles) and SAS Off (squares) runs.

In addition, the high frequency dropoff in the magnitude response occurs at lower frequencies for
the BUFCM frequency sweeps compared to the drogue tracking run. A similar dropoff is not
present in the magnitude response of the BUFCM-HIGH drogue tracking run, as it closely
follows the SAS On and Off cases. This apparent discrepancy arises from the nonlinear nature
of rate limiting which is both frequency and amplitude dependent.

213
Figure C-1. Example Stick Force and Pitch Rate Time History Plots

214
 Figure C-2. Longitudinal Stick Dynamics (δLON/FLON) from Frequency Sweeps

Additional insight is gained through a direct comparison of the two BUFCM-HIGH cases. The
Figure C-1 example time traces of stick force and pitch rate include the BUFCM-High drogue
tracking and frequency sweep runs. The drogue tracking stick force and pitch rate traces reveal
two distinct PIOs (12 to 22 seconds and 43 to 55 seconds). The stick force time trace for the
frequency sweep shows a significant increase in amplitude at the higher frequencies, while the
pitch rate signal for the sweep reveals a reduction in output power at the highest attained
frequencies.

215
 Figure C-3. Frequency Responses of q/FLON for 35° Wing Sweep Cases

Figure C-4 repeats the describing function magnitude, phase, and coherence plots for the two
BUFCM-HIGH cases. Here, the different input amplitude features of the rate limiting for the
two runs are clearly evident. The tracking run has a reduced amplitude and larger phase lag in
the frequency range from approximately 1 to 4 rad/sec. At higher frequencies the magnitude and
phase for the sweep cross over the data for the tracking case and continue to drop off at steeper
slopes. The amplitude reduction and additional phase lag resulting from the rate limiting is most
evident in the PIO frequency range for the tracking run and at higher frequencies for the sweep.

216
 Figure C-4. q/FLON Frequency Responses for BUFCM-HIGH 35° Wing Sweep Cases

Thus, as expected, the amplitude dependence of the nonlinearity directly influences the
magnitude and phase characteristics of the resulting describing functions. This influence is not
only at high frequencies, as might be expected, but also at all frequencies in the range shown in
Figure C-4.

E. IDENTIFICATION OF PIO FREQUENCY AND TASK BANDWIDTH

Stick force power spectral density plots of three segments of the J_09 drogue tracking run are
shown in Figure C-5. The three segments consist of two distinct PIO regions and a no-PIO
region. The peaks in input power for the PIO regions are significantly higher than the input
217
power of the no-PIO region. There is little input power for the no-PIO region at frequencies
below about 5 rad/sec, and the peak power occurs at 6.1 rad/sec. This is probably the operating
frequency required to perform the drogue tracking task. The peak power for the PIO regions is
significantly higher at the PIO frequencies, between 3.7 and 4.2 rad/sec. For both PIO regions,
there is a second peak at higher frequencies (around 5-5.5 rad/sec for the first PIO and 6.1
rad/sec for the second), corresponding to the task frequency from the no-PIO region.
Pitch rate to stick force describing functions for the three segments of the J_09 drogue tracking
run are shown in Figure C-6. For the PIO segments, the Bode magnitude and phase plots display
gain reductions and larger phase lags, respectively, in the frequency range of 2.5 to 4.0 rad/sec
(the low-frequency end of the plot). For a high-gain task such as drogue tracking, such results
are expected in the presence of the 10 deg/sec BUFCM-HIGH rate limit.

F. AIRPLANE BANDWIDTH ASSESSMENTS

Airplane bandwidth parameters were obtained for all of the runs identified in the previous
sections. Only high-coherence data points, ρ 2 > 0.8 , were used in determining the bandwidth
frequency, neutral-stability frequency, phase delay, and frequency domain pitch rate overshoot.
Because of their gain-margin-limited nature, the BUFCM-HIGH and BUFCM-LOW cases most
exhibited this lack of good data for the bandwidth frequency measures. This factor was
amplified for the non-frequency-sweep runs that are limited in frequency content by the nature of
the tasks. It is likely, however, that a frequency sweep with significant low-frequency content
could generate the full range of data required for these cases. In the future it would be
appropriate to carry out such sweeps at multiple input amplitudes to more fully expose the
nonlinear nature of the rate limiting.

Figure C-5. PSD Comparisons for BUFCM-HIGH Drogue Tracking Run Segments

218
 Figure C-6. q/FLON Frequency Responses for BUFCM-HIGH Drogue Tracking Run
Segments

In Figure C-7 the phase delay values for the drogue tracking and hook-up tasks are plotted
against the neutral-stability frequency because the bandwidth frequency was not available for all
cases. Several observations can be made from the figure. First, the data confirm the current
phase delay requirement, although the no-PIO points were all generated from the hook-up task
which was significantly less demanding than drogue tracking. On the other hand, the PIO cases
indicate that high phase delay leads to PIO for either task. The data also demonstrate the general
result that handling qualities ratings degrade uniformly with increasing phase delay.

219
 Figure C-7. PIO Phase Delay Requirement for Hook-Up and Drogue Tracking Runs

G. OBSERVATIONS FROM THIS ANALYSIS

• Fast Fourier transform techniques can be successfully applied to time history data to
obtain describing functions in the presence of rate limiting. In this example the analysis
results displayed the magnitude reduction, significant additional phase lag, and input
amplitude sensitivity that are associated with rate limiting.
• Because they produce the widest range of high coherence data, frequency sweeps and
drogue tracking tasks allow for best extraction of pilot-vehicle characteristics.
• By examining the power spectral density of the pilot’s stick force input, PIO frequencies
and task bandwidths can be identified.
• It was shown that for the F-14 with simulated dual hydraulic failure, excessive phase
delay led to PIO for the both the drogue hook-up and tracking tasks. The correspondence
between added phase delay and PIO supports recommended criteria for the prevention of
PIO.

H. REFERENCES FOR APPENDIX C

1
Niewoehner, R., LCDR USN, and Minnich, S., “F-14 Dual Hydraulic Failure Flying Qualities
Evaluation,” Thirty-Fifth Symposium Proceedings, Society of Experimental Test Pilots,
Lancaster, CA, Sep. 1991, pp. 4-15.
2
Klyde, David H., David G. Mitchell, and Kelly J. Latimer, “Development of the Probe-and-
Drogue Handling Qualities Demonstration Maneuver,” J. Guidance, Control, and Dynamics,
Vol. 22, No. 4, July-Aug. 1999, pp. 528-535.
220
 APPENDIX D. RECOMMENDATIONS FOR OBTAINING AIRPLANE
LONGITUDINAL BANDWIDTH PARAMETERS

A. RECOMMENDED STEPS FOR GENERATING PARAMETERS

1. Required Data: pitch rate (q),vertical velocity ( h ), and cockpit control force
a. Define dynamics of prefilters and sensors for all signals
b. Use instantaneous data (such as IVSI) as opposed to lagged data
c. Verify correct sequence for data sampling and recording: all samples from same time
frame

2. Frequency Sweeps (General):

a. Frequency range can be narrower than that needed for parameter identification
– Lowest frequency around 0.2 rad/sec or 0.03 Hz (30 seconds per cycle)
– Highest frequency around 12-18 rad/sec or 2-3 Hz (3 cycles per second)
b. Start from and end in trim conditions, sweeping from lowest to highest frequency
c. Total time for the sweep should be 60-90 seconds
d. Attempt to keep input amplitude relatively constant (smaller amplitude will be
necessary for very low-frequency portion of sweep)
e. Pilot-generated sweeps are preferred; pilot should be allowed to assist automated
sweeps to remain near trim conditions
f. Avoid dwelling at frequencies of high aircraft natural response (linear and nonlinear
modes)
g. Repeat sweeps are useful

3. Frequency Sweeps in Simulation:

a. Confirm that model includes all expected nonlinearities
– Control command shaping and control feel dynamics
– Actuator rate and position limits
– Surface effectiveness variations
b. Run sweeps of varying input amplitude
– Essentially linear region of aerodynamics/control system: peak pitch rates of ±10
deg/sec
– Near normal acceleration limits: between 0 and 2g for transports, larger for fighters
– At or near full stick (will probably require automated inputs at higher frequencies)

4. Frequency Sweeps in Flight:

a. Run sweeps of varying input amplitude
221
 – Essentially linear region of aerodynamics/control system: peak pitch rates of ±10
deg/sec
– Normal acceleration range of 0-2g (if peak pitch rates ±10 deg/sec do not cover this
load factor range)
b. Confirm that flight results are consistent with those from simulation
c. Adjust simulation as necessary to improve correlation

5. Determination of PIO Susceptibility:

a. Check both simulator and flight results with requirements
– Generate required frequency responses using reliable software (e.g., CIFER)
– If a “dip” occurs in coherence at any frequency where both input and output powers
are high, use single sinewaves at the amplitude of the corresponding sweep and
around the frequencies of the loss of coherence to verify the results of the sweep
– Convert pitch-rate-to-control-force response to effective pitch attitude (add 1/s)
– Measure Bandwidth parameters
b. Confirm that simulator and flight predictions are in good agreement for small-
amplitude sweeps
– If no PIO is predicted, no further testing is needed
– If PIO is predicted from both simulation and flight, piloted closed-loop tracking
should be performed in flight to test for susceptibility
– If PIO is predicted from simulation for large-amplitude sweeps, further flight
testing is necessary to confirm this
c. Sensitivity to increases in input amplitude may be spot-checked in flight
– Note frequency where pitch attitude is 180 degrees out of phase with stick,
ω180θ (equivalently, where pitch rate is 90 degrees out of phase with stick, ω90 q )

– Apply several cycles of a sinewave at this frequency in flight, at highest control

amplitude used for in-flight frequency sweeps
– Repeat for progressively higher control input amplitudes, as flight safety allows
– Analyze the single frequencies using time-series analysis (measure amplitude ratio
and phase angle directly from time responses)
– Compare the loss in phase with the full frequency sweep at the single frequency
– This will provide a rough measure of the amplitude attenuation and phase rolloff
with increasing amplitude without performing a full frequency sweep

B. REAL-TIME MONITORING OF FREQUENCY SWEEPS

There are inherent risks in performing in-flight frequency sweeps under any circumstance. The
risks are minimal when the sweeps are properly performed and monitored. Running sweeps or
single-sinewave inputs at very large amplitudes increases the level of risk. Clearly, there should

222
always be a real-time monitor for airplane-specific loads and stress parameters, to assure that
structural limits are not exceeded.
The Realtime Oscillation VERifier (ROVER) algorithms introduced in the main text of this
report can also be employed to assist with real-time monitoring. In this case, the purpose is more
for quality assurance rather than flight safety, as the output from ROVER can be observed to
assure that the inputs reach their intended goals in terms of frequency and amplitude of inputs
and response. In this way, ROVER is a supplement to FFT methods that will be used to analyze
the data in detail, after the flight, sometimes days after the flight.
An example application, using an actual output from ROVER for data from a flight-generated
frequency sweep, is shown in Figure D-1. The time histories of stick force (largest-amplitude
signal), pitch rate (smaller amplitude), and load factor (smallest) are plotted in the top sketch.
The bottom sketch shows identified peak-to-peak magnitudes of stick force (circle symbols) and
pitch rate (plus symbols) for each half-cycle of the sweep. Input amplitude is relatively constant
until around 45 seconds, where it begins to increase to a maximum of about ±30 lbs (peak-to-
peak value of 60 lbs). Peak pitch rate occurs around 50-52 seconds, where the largest change in
load factor (not shown) also occurs. Peak load factors were 0.1g and 1.9g at 50.4 and 51.8
seconds, respectively. Beyond 52 seconds, output amplitude decreases as a result of natural
reduction in aircraft response and extensive actuator rate limiting.
Figure D-2 shows the other relevant output parameters from ROVER: derived oscillation
frequency (computed from pitch rate) and derived phase between pitch rate and stick force.
These plots indicate a relatively small change in sweep frequency for the first 45 seconds, then a
relatively rapid increase in frequency and corresponding loss of phase. The lowest frequency for
this sweep is slightly less than 1 rad/sec and the highest frequency is less than 8 rad/sec.
Comparison of Figure D-1 with Figure D-2 instantly reveals several useful pieces of information.
For example, if the goal of this sweep were to apply relatively constant-amplitude stick force
inputs at the higher frequencies (and this was not the goal in this case), it was not done, and
another run would be in order. If, however, the goal were to achieve a peak change in load factor
of roughly 2g, this goal was reached, as noted in the text above (0.1g-1.9g is a change of 1.8g).
In addition, the peak pitch rate and load factor responses were around 52 seconds (Figure D-1),
corresponding to a response frequency of 3 rad/sec and phase angle of roughly 90 degrees
(Figure D-2). Since phase angle here is based on the angle between stick force and pitch rate, an
angle of 90 degrees corresponds to the pitch attitude neutral-stability frequency, ω180θ . So we
know that ω180θ for this sweep was around 3 rad/sec.

Similarly, the pitch attitude Bandwidth frequency (based on phase angle), the frequency where
phase in Figure D-2 was 45 degrees, was around 2 rad/sec (sweep time around 48 seconds). At
twice the neutral-stability frequency, or 6 rad/sec, the phase angle is about 180 degrees, or 90
degrees of added phase from the neutral-stability frequency. Then Phase Delay is roughly
90/(57.3*6) = 0.26 sec. Based on this number alone, this airplane should be susceptible to PIO.
The power of ROVER is exemplified by the frequency response plots in Figure D-3. The time-
history data from the example sweep were analyzed using FFT techniques (program FREDA
from Systems Technology, Inc.), and the high-coherence output data points for q/Fes are plotted
on Figure D-3 as open symbols.

223
 40 Run: f14s wp 1 a

20

Un s m o o th ed I/O 0

-20

-40
0 10 20 30 40 50 60 70 80
T im e (sec )

60
stick fo rce
40
P eak v alu es
pitch rate
20

0
0 10 20 30 40 50 60 70 80
T im e (s ec )
Figure D-1. Example Frequency Sweep and ROVER Output of Peak Values
8

F req u en c y 4
(rad /sec )
2

0
0 10 20 30 40 50 60 70 80
T im e (s ec )
2 00

1 50

P h ase 1 00
(d eg )
50

0
0 10 20 30 40 50 60 70 80
T im e (s ec )

Figure D-2. ROVER-Identified Frequency and Phase for Example Sweep

224
 20 90

10 45

0 0

-10 -45

Gain Phase
-20 -90
(dB) (deg)

-30 -135
gain from FFT
-40 gain from ROVER -180
phase from FFT
-50 phase from ROVER -225

-60 -270
0.1 1 10
Frequency (rad/sec)

Figure D-3. Comparison of Frequency Responses of q/Fes for Example Sweep

For comparison with the FFT data, gain of q/Fes was computed from the ROVER output from the
ratios of the peak-to-peak values in Figure D-1, converted to dB, 20log10(q/Fes), and phase angle
was taken as the negative of the phase angle in Figure D-2. These results are shown in Figure D-
3 by solid symbols. The agreement between the two methods is quite close.
ROVER cannot replace FFT analysis, however. In its current form, ROVER makes no estimates
of data quality (coherence) and does not obtain very low-frequency data. Still, the results of
Figure D-3 show that, for a good frequency sweep, initial estimates of PIO susceptibility can be
made directly off of the raw output from ROVER.
As a footnote, from the FFT-derived data of Figure D-3, it was estimated that the pitch attitude
Bandwidth was about 0.37 rad/sec, based on gain margin; the phase margin Bandwidth
frequency was estimated to be 2.0 rad/sec – in agreement with the eyeball measurements from
ROVER discussed above. Pitch attitude neutral-stability frequency was about 2.6 rad/sec (based
on ROVER it was estimated to be around 3 rad/sec), and Phase Delay was computed to be 0.28
sec (from ROVER it was 0.26 sec). Thus the very rough initial estimates from ROVER were
quite close to the more precise values obtained from the computed frequency responses from
FFT.

225