Dynamic performance of a two-axis gimbaled

pedestal in keyhole gimbal-lock conditions

James DeBruin*
IJK Controls, 7557 Rambler Road, Suite 700, Dallas, TX 75231

ABSTRACT

The common two-axis azimuth-elevation gimbaled pedestal has a full-hemispheric, horizon-to-zenith field of regard.
This pedestal has no kinematic difficulties at low elevation angles. In this position, the line-of-sight of the mounted
sensor is perpendicular to both the azimuth and elevation gimbal axes, which thus provide two orthogonal degrees of
freedom. However, as the line-of-sight approaches zenith, the sensor axis nears alignment with the azimuth axis. The
azimuth axis thus loses its ability to move the line-of-sight orthogonally to the sweep of the elevation axis. This
condition is known as gimbal lock and the position range in which dynamic difficulties occur is called keyhole. The
keyhole region is a solid cone centered around the zenith axis. The onset of dynamics difficulties is a continuum from
horizon to zenith, and as such defining the keyhole region is arbitrary. However, dynamic difficulties become rapidly
pronounced somewhere between 70 and 80 degrees, so it is generally agreed that the keyhole region starts in this range.
This paper provides a comprehensive analysis of the keyhole region. While performance problems at keyhole are well
known (high torque, acceleration, and speed requirements), certain dynamic effects actually reduce in keyhole, such that
for some systems the range of worst-case performance is actually outside the keyhole region. Gimbal geometry is
introduced and pointing equations derived using vector methodology. Kinematic equations are then developed, with a
focus on the requirements of line-of-sight stabilization for vehicle-mounted systems. Examples and simulation results are
provided to illuminate the issue.
Keywords: gimbals, pedestals, keyhole, gimbal lock, gyroscope, pointing, stabilization, tracking

1. INTRODUCTION
The two-axis, elevation-on-azimuth, “az-el” gimbaled pedestal is widely used to pointing and stabilize a wide variety of
optical and RF systems, including telescopes, cameras, night-vision systems, lasers, and communication and radar
antennas. Az-el pedestals are found in all environments, including ground-fixed, ground-mobile, shipboard, airborne,
and spacecraft applications. The kinematic and dynamic difficulties of these pedestals at high-elevation angles, known as
the keyhole or gimbal lock is well known. While az-el pedestals are feasible in some shipboard applications, such as
littoral water craft or low-performance consumer products, the wide-use of three-axis pedestals in shipboard applications
is in fact a direct consequence of the keyhole limitations of the two-axis az-el gimbal. The keyhole problem is so well
known that it is tempting to assume that the worst-case performance of a two-axis gimbal will always be in the keyhole
region. Surprisingly, there are circumstances in which the worst-case performance of a system using an az-el gimbaled
pedestal will occur away from the keyhole region.
The kinematic, dynamic, and control equations for a two-axis gimbal are presented in this paper. Performance in
stabilized mobile applications is investigated, as it is the stabilization application that provides the most variety of
performance scenarios. The effects of pointing geometry, vehicle motion, friction, motor current and voltage limits,
direct and indirect drive configurations, and noise is investigated and discussed. Results are shown analytically or with
simulation results as best illuminates the issue.

2. AZ-EL GIMBAL KINEMATICS

The development of the kinematics for a vehicle-mounted, two-axis pedestals is found in the literature in a number of
formulations1-3. The development outlined here follows the methodology of Kane and Levinson in their landmark
textbook on dynamics4 and makes use of line-of-sight (LOS) reference frames to tie the payload line-of-sight to the
pedestal kinematics5.
*[email protected]; phone 1 214-458-0550 fax 1 214-257-3247; ijkco.com

Acquisition, Tracking, Pointing, and Laser Systems Technologies XXV,

edited by William E. Thompson, Paul F. McManamon, Proc. of SPIE Vol. 8052, 80520J
© 2011 SPIE · CCC code: 0277-786X/11/$18 · doi: 10.1117/12.884438

Proc. of SPIE Vol. 8052 80520J-1

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/02/2013 Terms of Use: http://spiedl.org/terms

 In the Kane formulation, vector triads are defined within rigid bodies to define the orientation, angular velocity, and
angular acceleration of the rigid body, such that the triad and the rigid body are kinematically interchangeable. Four
right-handed unit vector triads are defined:
Triad A[a1, a2, a3]: Fixed in the local-level reference frame A. Vectors a1, a2, and a3 are aligned in the north, east,
and down directions respectively.
Triad B[b1, b2, b3]: Fixed in the host vehicle B. Vectors b1, b2, and b3 are in the “NASA standard” alignment, so that
they are aligned in the forward, right-wing, and down directions respectively. Triad B is aligned
with Triad A when the vehicle yaw, pitch, and roll angles are zero. The pedestal bas is fixed
within the host vehicle, such that triad B represents both the vehicle and the pedestal base.
Triad C[c1, c2, c3]: Fixed in the azimuth gimbal C. Triad C is aligned with triad B when the azimuth gimbal angle θ
is equal to zero. Angle θ is defined as positive so that the azimuth gimbal C moves relative to
the vehicle / pedestal base B with a simple right-hand rotation about vectors b3 and c3, which are
aligned (parallel). Angle θ increases when the azimuth gimbal moves clockwise when viewed
from above.
Triad D[d1, d2, d3]: Fixed in the elevation gimbal D. Triad D is aligned with triad C when the elevation gimbal
angle φ is equal to zero. Angle φ is defined as positive so that the elevation gimbal D moves
relative to the azimuth gimbal C with a simple right-hand rotation about vectors c2 and d2,
which are aligned. The payload is mounted within D such that the payload line-of-sight is
aligned with vector d1. Angle φ increases when the line-of-sight vector d1 moves up toward
zenith from the horizon-viewing, zero-angle orientation.
Line-of-sight stabilization is concerned with the time-rate-of-change of the line-of-sight vector d1 within the local-level
reference frame (which serves as the inertial reference frame for most terrestrial stabilization applications). The time-
rate-of-change for a vector can be calculated in a number of ways, but the following formulation is particular convenient
due to the simplicity by which the components are formulated and the equation executed:

A
d& 1 = ω × d 1
A D
(1)

The first term in the vector cross product is the angular velocity of elevation gimbal D within local-level reference frame
A. This quantity can be generated using the addition theorem of angular velocity:

A
ω D = Aω B + B ω C + C ω D (2)

The first term on the right is the angular velocity of the vehicle B in the local-level reference frame A, which can be
expressed as follows:
A
ω B = ω 1 b1 + ω 2 b 2 + ω 3 b 3 (3)

The scalar components ω1, ω2, and ω3 are often referred to as the body rates of the vehicle. These angular rates,
sometimes designated in the literature as P, Q, and R, are the quantities measured by a strap-down triad of gyroscopes as
part of an attitude and heading reference system (AHRS), inertial reference unit (IRU), inertial navigation system (INS),
or other inertial measurement unit (IMU).
The last two terms on the right side of equation 2 are the angular velocities of the azimuth and elevation gimbals C and
D in the base and azimuth reference frames B and C respectively. As both gimbals move about axes fixed in the
referenced reference frame, these quantities are simple angular velocities, readily expressed as:
B
ω C = θ&c 3 (4)

and:
C
ω D = ϕ& d 2 (5)

Proc. of SPIE Vol. 8052 80520J-2

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/02/2013 Terms of Use: http://spiedl.org/terms

 Substituting equations 3, 4, and 5 into 2 yields
A
ω D = ω 1 b1 + ω 2 b 2 + ω 3 b3 + θ&c 3 + ϕ& d 2 (6)

The cross-product of equation 1 is readily handled by expressing equation 2 in the vectors of the line-of-sight reference
frame D. The substitutions are found by inspection of the vector triad geometry:

b1 = cos θ cos ϕ d 1 − sin θ d 2 + cos θ sin ϕ d 3 (7)

b 2 = sin θ cos ϕ d 1 + cos θ d 2 + sin θ sin ϕ d 3 (8)

b 3 = c 3 = − sin ϕ d 1 + cos ϕ d 3 (9)

Substituting equations 7 through 9 into equation 6 produces:

A
ω D = ( cosϕ (cosθ ω 1 + sinθ ω 2 ) − sinϕ (ω 3 + θ&) ) d 1
+ (− sinθ ω 1 + cosθ ω 2 + ϕ& ) d 2 (10)
(
+ sinϕ (cosθ ω 1 + sinθ ω 2 ) + cosϕ (ω 3 + θ&) d 3 )
Using equation10, the vector cross-product in equation 1 is calculated as:
A
d& 1 = (sinϕ (cosθ ω 1 )
+ sinθ ω 2 ) + cosϕ (ω 3 + θ&) d 2 + (sinθ ω 1 − cosθ ω 2 − ϕ& ) d 3 (11)

Stabilization is achieved when the time-rate-of-change of the line-of-sight vector, given by equation 11, is equal to zero.
Equation 11 is uniformly zero only when the d2 and d3 scalar components are zero.

3. STABILIZATION CONTROL
Comparison of equations 10 and 11 yields an alternate definition of stabilization: The line-of-sight d1 is stabilized when
the d2 and d3 scalar components of the angular velocity of line-of-sight reference frame D (which here is the elevation
gimbal) are zero. This definition is suggestive of the most common servo configuration for stabilization. Specifically,
gyroscopes are mounted to the elevation gimbal so that they are aligned in the d2 and d3 directions. Control loops are
then implemented using the gyroscopes as feedback, with motors on the elevation and azimuth axis. The control loops
act to hold the gyroscope outputs to zero, thus satisfying the kinematic conditions for stabilization.
Vector d2 is aligned with the elevation axis, and the gyroscope aligned in this direction is called the elevation gyroscope
with no ambiguity. The vector d3, which is perpendicular to both d1 and d2, is aligned with the azimuth gimbal axis only
when the elevation angle is zero. The gyroscope aligned in the d3 direction is commonly called the azimuth gyroscope. A
more technically accurate but less commonly used name is cross-elevation gyroscope. While gyroscope names are not
critical, it is important however to distinguish that angular rate sensed by this gyroscope (specifically, cross-elevation
rate, which is the d3 component of equation 10) is not the same as the inertial rate of the azimuth gimbal.
3.1 Elevation
The block diagram for the kinematics of elevation stabilization control is shown in Figure 1. The primary, “T=Jα”
dynamics of the elevation gimbal are included. Secondary dynamics, such as imbalance, centripetal or product-of-inertia
terms, are not shown on the diagram, but would be included if significant in the torque term. The dynamic equation is
formulated here in inertial terms, such that the output of the acceleration-to-rate integrator is inertial rate. For the
elevation axis, this term, designated in the figure as “Elevation LOS Inertial Rate”, is sensed directly by the elevation
gyroscope. The other terms in the d2 component of equation 10 are shown at the bottom of the diagram and are used
within the model to generate the elevation angle as shown.

Proc. of SPIE Vol. 8052 80520J-3

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/02/2013 Terms of Use: http://spiedl.org/terms

 Figure 1. Block diagram model of elevation kinematics and dynamics

3.2 Azimuth / Cross-Elevation

The block diagram for the kinematics of azimuth stabilization control is shown in Figure 2. The primary dynamics of the
azimuth gimbal are again included, and the output of the acceleration-to-rate integrator is again inertial rate. As
discussed, however, the cross-elevation (azimuth) gyroscope does not directly sense this rate. As observation of equation
10 and the block diagram below indicate, what the cross-elevation gyroscope does sense is dependent on the elevation
and azimuth gimbal angles φ and θ. This term is designated “Cross-Elevation LOS Rate” in the figure.

Figure 2. Block diagram model of azimuth and cross-elevation kinematics and dynamics

Proc. of SPIE Vol. 8052 80520J-4

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/02/2013 Terms of Use: http://spiedl.org/terms

 The effect of the elevation angle is particularly pronounced, as best understood by a term-by-term review of the cross-
elevation LOS rate, which is the d3 coefficient in equation 10:
sin ϕ (cos θ ω 1 + sin θ ω 2 ) + cos ϕ (ω 3 + θ&)
The term in parentheses at the end is the sum of the azimuth gimbal relative rate and the inertial rate of the vehicle about
the azimuth. By the addition theorem this is the inertial rate of the azimuth gimbal about the azimuth axis. This term is
multiplied by the cosine of the elevation angle. These terms together indicate that as the elevation angle increases, the
azimuth gimbal becomes less and less effective at creating LOS cross-elevation rate. This is sometimes described as loss
of mechanical gain. The loss is complete at keyhole, as azimuth gimbal motion produces no motion in the cross-
elevation direction, and thus no rate in the cross-elevation gyroscope. This is the classic keyhole problem for stationary
pedestals: The azimuth gimbal simply loses its ability to move the line-of-sight as the elevation gimbal nears keyhole.
The term in parentheses on the left represents line-of-sight roll. The term can be explained as follows. Stand on the deck
of a ship and face any arbitrary point on the horizon. Line-of-sight roll is the angular velocity component of the ship in
the direction of observation. As the equation indicates, if you are facing the front of the ship (θ = 0), then line-of-sight
roll is the same as ship’s roll ω1. If facing starboard (θ = 90), then line-of-sight roll is ship’s pitch ω2.
The line-of-sight roll term is multiplied by the sine of the elevation angle. These two terms together indicate that as the
elevation angle increases, the cross-elevation gyroscope becomes increasingly sensitive to line-of-sight roll. As it is, the
stabilization loop works to keep the cross-elevation rate at zero. The only way to do this is for the azimuth gimbal to
move in an opposite direction to negate the sensed line-of-sight roll rate (The azimuth gimbal moves because the sensed
line-of-sight roll is fed back to through the control loop and effectively becomes a motion command for the azimuth
gimbal. Negative feedback ensures that the gimbal moves in the opposite direction). This is the classic keyhole problem
for mobile stabilized azimuth-elevation gimbals: As elevation angle increases, the sensitivity to line-of-sight roll
increases, but the mechanical gain of the azimuth gimbal needed to create offsetting cross-elevation motion
correspondingly decreases. The result is the universally observed “washing machine” motion of the azimuth gimbal
during line-of-sight roll in keyhole conditions.

4. STABILIZATION PERFORMANCE IN KEYHOLE CONDITIONS

The effect of line-of-sight roll in keyhole conditions is so demonstratively dramatic that it is easy to conclude that this is
the beginning and the end of the keyhole effect. Further investigation reveals what experience also shows, which is that
the effects of keyhole are far more subtle, and these effects are sometimes for better and other times for worse.
4.1 Body Rate Isolation to Line-of-Sight Roll
The simplified model shown in Figure 3 is used to further investigate the effect of line-of-sight roll on stabilization
performance. The gyroscope is modeled without dynamics to simplify the analysis.

Figure 3. Simplified model of cross-elevation (azimuth) rate loop with roll-coupling input ω1.

Proc. of SPIE Vol. 8052 80520J-5

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/02/2013 Terms of Use: http://spiedl.org/terms

 The controller and plant dynamics are modeled as the general transfer function G1(s). The rate command ωCMD is used
for slewing, tracking, and scanning, but does not appear in the analysis of roll coupling. The presence of the cosine term
within the loop indicates that the loop gain is a decreasing function of elevation angle. The secant gain term shown is
added to the loop compensation to offset the gain loss. The secant correction technique brings complications, but these
do not have to be considered for the investigation of roll coupling.
The effect of the roll coupling input ω1 on stabilization is revealed by generating the transfer function from ω1 to cross-
elevation line-of-sight rate input ωXEL. This transfer function, here designated G2(s), is a measure of the body-rate
isolation characteristics of the loop. By observation of the loop in Figure 3 and use of Mason’s rules for block diagram
manipulation, G2(s) is calculated as:
ω XEL 1
G 2 ( s) = = sin(ϕ ) (12)
ω1 1 + G1 ( s )
Roll coupling is an undesirable effect, so we would like the gain of the transfer function G2(s) to be as low as possible.
Disregarding the sine term, the gain of G2(s) is made small by simply making the gain of G1(s) large. That is, increasing
the gain of the stabilization loop improves body rate isolation.
The effect of the sine term is more subtle. Define the baseline roll coupling gain GRC(ω) as the gain value of
1/(1+ G1(s)) at a given frequency ω:
1
G RC (ω ) ≡ (13)
1 + G1 (ω )
The roll coupling gain for any particular elevation angle (which is the gain of the body rate isolation transform of Eq. 12)
can thus be expressed as:

G 2 (ω ) = sin(ϕ ) G RC (ω ) (14)

Equation 13 indicates that, for any given frequency, the roll coupling gain is zero at horizon (φ = 0) and maximum in the
keyhole region (φ approaching 90°). However, note that sine increases quickly from zero as elevation angle increases
from horizon. The rise is rapid enough that roll coupling becomes a significant error source at angles well below the
keyhole region.
The following example is provided for clarification of this effect. A rate loop with PI compensation and a 15 Hz open-
loop crossover has an loop gain of approximately 60 at 1 Hz, which corresponds to a gain GRC(ω) of -36 dB. The roll
coupling gain as a function of elevation angle for this system is calculated using Eq. 14 and shown in Figure 4. The
curve rises quickly, with roll coupling gain within 3 dB of its maximum value at 45 degrees and within 1 dB at 60
degrees.

Figure 4. Roll coupling gain as a function of elevation angle for example system.

Proc. of SPIE Vol. 8052 80520J-6

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/02/2013 Terms of Use: http://spiedl.org/terms

 4.2 Acceleration, Torque, and Current
Operation at keyhole has a dramatic effect on the acceleration, torque, and current requirements of the system. To show
this, the stabilization loop shown in Figure 3 is expanded in Figure 5 to explicitly show the secant correction, azimuth
gimbal inertia, and the plant integrator from azimuth acceleration αAZ to azimuth rate ωAZ. The inertia is shown as
normalized, which means that units have been chosen so that the gimbal inertia is equal to unity. This simplifies the
equations without loss of generality. The controller, now without the inertia and secant correction, is relabeled G3(s).

Figure 5. Model of cross-elevation (azimuth) rate loop expanded to show secant correction
term, normalized azimuth gimbal inertia, and azimuth gimbal acceleration αAZ.

The increase in acceleration requirements at keyhole in roll coupling conditions can be shown by generating the transfer
function from ω1 to azimuth acceleration αAZ. Observation of the loop in Figure 5 shows this transfer function,
designated here as G4(s), to be:

α AZ ⎛ s G 3 ( s) ⎞
G 4 (s) = = tan(ϕ )⎜ ⎟ (15)
ω1 ⎜ s + G 3 ( s) ⎟
⎝ ⎠
At any given frequency the second term is constant, indicating that torque requirements to stabilize in roll coupling
conditions increase with the tangent of the elevation angle. Similar analysis shows that the torque requirements in
response to a commanded rate ωCMD increase with the secant of the elevation angle. These two curves are shown in
Figure 6.

Figure 6. Increase in azimuth acceleration requirements as function of elevation angle φ. The

tangent curve is for roll coupling. The secant curve is for commanded rate motion.

Proc. of SPIE Vol. 8052 80520J-7

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/02/2013 Terms of Use: http://spiedl.org/terms

 Acceleration is proportional to the applied torque, so the torque requirements of the azimuth motor increases with the
acceleration demands. The curve indicates that it takes roughly twice the acceleration (and thus torque) to operate at 60
degrees elevation as it does at 45 degrees. The acceleration and torque double again at approximately 75 and yet again at
83 degrees.
The load inertia seen by the azimuth motor will usually far exceed that of the elevation motor, as azimuth is the lowest
gimbal in the chain. As such, the torque requirements for azimuth motor is usually greater than the requirements for the
elevation motor, even for systems that operate at nominally low elevation angles. Add to that the additional acceleration-
torque multiplier needed to operate at keyhole and the result is an azimuth motor that is far bigger than the elevation
motor. Systems with azimuth peak torque capacity twenty to thirty times that of elevation are not uncommon.
For most high-performance systems, the large acceleration requirements exceed the capabilities of stepper motors and
thus lead to brush or brushless DC motors. In these motors, the output torque is proportional to motor current. Thus
acceleration, torque, and current requirements all increase in keyhole in the same proportion. While nominal RMS
current draw varies greatly depending on operating conditions, it is quite common for the peak current requirements of
the azimuth motor in keyhole to completely dominate the current budget for sizing of servo-amplifiers, power supplies,
circuit breakers, wiring, and connectors.
4.3 Effect of Current Limits
Figures 4 and 6 indicate that while torque requirements to operate increase rapidly in the keyhole region, the resultant
error to roll coupling does not. That is, as long as the system does not run out of acceleration, torque, and current
capacity, there is no substantial increase at all in stabilization error in the keyhole region. This assumes of course that the
elevation gimbal travel stays away from near-zenith angles where the loop becomes unstable, but it is otherwise a
remarkable and counter-intuitive conclusion. The “washing machine” motion just keeps getting worse and worse, but it
causes no corresponding increase in line-of-sight error.
Real systems of course have limits on acceleration, torque, and current. These result from voltage supply limits, current
supply limits, and motor saturation. Some of these limits, such as motor saturation are “soft”, meaning they come into
play gradually. Others, such as “foldback” protection within some servo amplifiers (which severely limits current
availability after being triggered), have a major and immediate impact on performance once applied.
The effects of current limits can be shown using a simulation. The system modeled is a gyroscope-stabilized shipboard
pedestal with a 15 Hz rate loop crossover for both azimuth and elevation. Ship’s motion is as follows:
Pitch: ±7° @ 0.33Hz Roll: ±15° @ 0.2Hz
The response of the system to ship’s motion is simulated first with no current limits (Figure 7) and with current limits
(Figure 8).

Figures 7a and 7b. X-Y error plots for example system in ship’s motion with no current limits. Figure 7a on the left is at
low elevation angles (30 degrees) and the Figure 7b on the right is for elevation angles near keyhole (80 degrees).

Proc. of SPIE Vol. 8052 80520J-8

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/02/2013 Terms of Use: http://spiedl.org/terms

 Figure 8a (φ = 70.5°, secφ = 3) Figure 8b (φ = 75.5°, secφ = 4)

Figure 8c (φ = 78.5°, secφ = 5) Figure 8d (φ = 80.4°, secφ = 6)

Figures 8a-8d. X-Y error plots for example system in ship’s motion with nominal current limits. No current
limiting is apparent at 70.5° (Figure 8a), but by 75.5° (Figure 8b) effects of limiting are apparent. Performance
is clearly compromised at 78.5° (Figure 8c) and the line-of-sight is all but uncontrollable by 80.4° (Figure 8d).
Current limits have been arbitrarily chosen to produce just these results, but they nonetheless demonstrate the
importance of providing adequate drive capacity if keyhole operation is part of the system specification.

4.4 Friction and Torque Disturbance Rejection

The primary coupling path for vehicle yaw motion to the cross-elevation rate is through friction in the azimuth bearing.
A simple Coulomb friction model is added to the system and shown in Figure 9. The relative gimbal rate, often called
the tachometer rate and shown here as ωTACH, is formed by subtracting the vehicle yaw rate ωYAW from the azimuth
inertial rate ωAZ. In the Coulomb model, friction reverses with the sign of the tach rate. This reversal is modeled by
routing the tach rate into a sign function, which produces an output of plus or minus one for positive and negative inputs
respectively. The output of the sign function and the negative sign on the torque summing junction has the effect of
applying a negative or positive friction torque for positive and negative relative rate.

Proc. of SPIE Vol. 8052 80520J-9

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/02/2013 Terms of Use: http://spiedl.org/terms

 Figure 9. Model of cross-elevation (azimuth) rate loop with friction included.

The sign function makes the response of this system definitely non-linear, but under any conditions friction is an
unwelcome torque disturbance that degrades the level of stabilization performance. Though the system itself is non-
linear, the degree to which friction disturbs the cross-elevation rate can be investigated by ignoring (for now) the non-
linear feedback path from ωAZ to friction and using the same linear analysis technique as before. In this way, the transfer
function G5(s) from the friction input TF to the cross-elevation rate ωXEL can be derived from observation of the block
diagram in Figure 9:

ω XEL cos(ϕ )
G 5 (s) = = (16)
TF s + G 3 (s)
Transfer function G5(s) characterizes the torque disturbance rejection capabilities of the system. Like roll coupling,
torque disturbances produce an undesirable effect, so we would like the gain of the torque disturbance transfer function
G5(s) to be as low as possible. Disregarding the cosine term, the gain of G5(s) is made small by making the gain of G3(s)
large. That is, increasing the control gain of the stabilization loop improves torque disturbance rejection by driving the
gain of the torque disturbance rejection curve down.
The effect of the cosine term in Eq. 16 is similar but opposite to the effect of the sine term in Eq. 15. Using the same
formulations as Eq. 13 and 14, and the baseline torque disturbance gain GTD as the gain value of 1/(1+ G3(s)) at a given
frequency ω yields:
1
GTD ≡ (17)
1 + G3 (ω )
The gain of the torque disturbance rejection curve is expressed as:

G 5 (ω ) = cos(ϕ ) GTD (18)

Equation 18 indicates that the magnitude of the torque disturbance rejection curve (for any given frequency) is maximum
at horizon (φ = 0) and diminishes rapidly in the keyhole region (φ approaching 90°). This means that the sensitivity of
the system to torque disturbances decreases with increasing elevation angle. Note however that cosine does not decrease
rapidly from unity as elevation angle increases from horizon. As such, the sensitivity to torque disturbances is relatively
flat until the keyhole region is approached.

Proc. of SPIE Vol. 8052 80520J-10

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/02/2013 Terms of Use: http://spiedl.org/terms

 The previous example system (PI compensation with 15 Hz crossover) is used again here for clarification of this effect.
The baseline torque disturbance gain GTD for this system is approximately -53.5 dB. The torque disturbance gain as a
function of elevation angle for this system is calculated using Eq. 18 and shown in Figure 10. The curve falls slowly,
with torque disturbance gain down less than 1 dB of its maximum value at 30 degrees and only 3 dB at 60 degrees.

Figure 10. Torque disturbance gain as a function of elevation angle for example system.

Note that, while Eq. (16) is formulated as the transfer function from the friction torque TF to the cross-elevation rate
ωXEL, it is in fact independent of the value of TF. As it is, the torque disturbance transfer function is valuable tool to
investigate the effects of any torque disturbance TD that acts on the azimuth gimbal. This fact is used in the next section
to investigate the effect of other torque disturbances.

Note also that, though the torque disturbance transfer function is independent of fiction level, it is in fact the combined
effects of torque disturbance gain and friction level and frequency content that determine actual stabilization
performance. A common measure of the friction level in a system is the friction-to-inertia ratio. The friction-to-inertia
ratio has units of angular acceleration, usually radians-per-second-squared though degrees-per-second-square are
sometimes used. The friction-to-inertia ratio is a measure of the amount of acceleration the friction level can produce,
and in that regard friction-to-inertia ratio is a method of normalized the friction level to allow direct comparisons of the
effect of friction in systems of different sizes and different inertias. Note that in the normalized inertia formulation used
here, the friction level KF is the friction-to-inertia ratio. Large gimbals, even with moment-bearing, preloaded duplex
bearings, typically have friction-to-inertia levels of under 1.0 rad/s2. Small gimbals, such as found in miniature missile
seekers or the polarization mechanisms of stabilized antennas, can have friction-to-inertia ratios in the 20 to 40 range or
higher. This wide range plays into later discussions of total system stabilization performance.

4.5 Other Torque Disturbances: Indirect Drive Coupling and Noise

There are many possible torque disturbances on a gimbal, such as spring torques, windage torques, and imbalance
torques. The effect of these on stabilization is determined by the magnitude and frequency content of torque disturbance
in combination with the torque disturbance rejection characteristics of the system. Two examples, indirect drive coupling
and noise are provided for illustration.

4.5.1 Indirect Drive Coupling

Indirect drives include belt drives, band drives, and gear drives. In each of these configurations, the motor drives the load
indirectly through a drive ratio. The exact dynamical equations for indirect drives vary with the details of the
configuration, such as whether the motor and the gimbal rotate in the same direction or whether they counter-rotate, and
whether the motor is mounted to the moving gimbal or mounted adjacently6,7. In any case, the torque disturbance comes
from the fact that the motor must turn during base motion if the payload is to remain stationary (It is sometimes
explained by stating that the motor has to “get out of the way” or “wind itself up and back down” during base motion).

Proc. of SPIE Vol. 8052 80520J-11

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/02/2013 Terms of Use: http://spiedl.org/terms

 As an example, the indirect-drive coupling for a belt-driven, co-rotating azimuth drive with motor mounted to system
base is shown in Figure 11. The controller produces motor torque TM. The parameter N is the drive ratio, defined here as
the ration of the gimbal pulley to the motor pulley (and thus generally greater than 1). Parameters JM and JAZ are the
motor and azimuth gimbal inertia respectively. Indirectly coupling produces a torque disturbance on the azimuth gimbal
that is proportional to the base (vehicle) yaw acceleration αM. The coupling factor for this system is (N-1)JM. The
resulting torque disturbance varies with elevation angle in the same way as friction. That is, the effects on line-of-sight
stabilization are greatest at lower elevation angles and diminish as keyhole is approached.

Figure 11. Indirect-drive torque disturbance model for a belt-driven, co-rotating, motor-on-base azimuth drive

4.5.2 Noise
By far the largest source of noise in stabilization loops is sourced within the feedback gyroscope. To a great extent, the
cost of gyroscopes is in inverse to the noise level. Generally speaking, any gyroscope used for gimbal stabilization will
have a noise floor far in excess of the noise floors of any other electrical component or process in the loop. A model to
investigate the effects of noise as a function of elevation angle in azimuth loops is shown in Figure 12.

Figure 12. a function of elevation angle for example system.

The torque disturbance TD from the gyroscope noise εXEL is embedded within the motor torque TM and is calculated from
the figure to be:

Proc. of SPIE Vol. 8052 80520J-12

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/02/2013 Terms of Use: http://spiedl.org/terms

 G 3 (s)
TD = ε XEL (19)
cos(ϕ )
Equation 19 indicates that torque disturbance from cross-el gyro noise increases with the secant of the elevation angle as
the elevation angle increases from horizon to zenith. Combining this with the torque disturbance transfer function of
Eq. 16 produces the transfer function from gyroscope noise to line-of-sight rate:

ω XEL G 3 (s)
= (20)
ε XEL s + G 3 ( s)
Equation 20 indicates that the disturbance to the line-of-sight from gyroscope noise is independent of the elevation
angle. This is a somewhat non-intuitive, especially if you have witnessed the audible growling and clearly visible
shaking of the azimuth gimbal of a high-gain system as the elevation angle increases into the keyhole region. However,
the increase in azimuth motion from noise amplification of the secant correction term is indeed offset exactly by the
reduction in mechanical gain of the azimuth gimbal to cross-elevation rate.

Secant amplification of gyroscope noise is, however, anything but benign. The noise amplification increases motor
current requirements, heats the motor, and reduces the voltage and current dynamic range headroom. It can indeed lead
to a reduction in stabilization performance, either through gradual means, such as early saturation of the azimuth motor,
or through hard-limiting such as servo-amplifier foldback.

5. OVERALL SYSTEM PERFORMANCE

In surveying the operation of a two-axis, azimuth-elevation gimbal in keyhole conditions, it is clear that the stabilization
errors from some error sources increase at keyhole, some reduce, and some are mostly insensitive to elevation angle. It is
often heard in planning verification tests for stabilized azimuth-elevation gimbals that are specified to operate near or in
keyhole conditions that “the system must be tested at keyhole in roll coupling conditions”. This is true in the sense that
this is the operation that will most likely exercise the system to maximum current, torques, and acceleration conditions.
It is further true that roll-coupling-at-keyhole tests are necessary to demonstrate that the system will operate without
being limit-constrained. However, it is not necessarily the case that “if it meets the stabilization specification at keyhole
it will meet specification at all other elevation angles”. In fact, if the system is designed with sufficient overhead, it may
be that the stabilization errors are actually lowest at high elevation angles, especially if roll conditions are not severe.

Take, for example, a small satellite communications antenna mounted to an azimuth-elevation pedestal designed to
operate on on-road and off-road ground vehicles within the continental United States. Maximum satellite elevations in
the United States are in the approximate range of 35 to 55 degrees elevation. The roll coupling gains are not excessive
for these angles, so it is quite feasible to provide an azimuth drive with the overhead needed to avoid current limiting and
motor saturation. If this system has a high friction-to-inertia ratio and/or a high-ratio indirect drive, it is very likely that
maximum stabilization errors will occur at the lower elevation angles. This is especially true given that ground vehicles,
especially in off-road conditions, tend to undergo frequent yaw reversals, thus adding high-frequency content to the
azimuth friction and further increasing the effects of friction and indirect-drive coupling on line-of-sight stabilization.

ACKNOWLEDGEMENTS

The author wishes to thank Rohit Murthy of General Dynamics Satcom Technologies, Richardson, Texas, for his
encouragement and contributions during the preparation and review of this manuscript.

Proc. of SPIE Vol. 8052 80520J-13

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/02/2013 Terms of Use: http://spiedl.org/terms

 REFERENCES

[1] Rue, A., “Stabilization of precision electro-optical pointing and tracking systems,” IEEE Trans. Aerosp. Electron.
Syst., AES-5(5), 805–819 (1969). (Errata: Nov., 1970, pp. 855–857).
[2] Hilkert, J. M., “Inertially stabilized platform technology,” IEEE Contr. Syst. Mag., 28(1), 26–46 (2008).
[3] Oslon, A.R., “Shipboard Stabilization of Large Antennas”, ASME Paper 65-WA/AUT-8 (1965).
[4] Kane, T.R. and Levinson, D.A, [Dynamics, Theory and Applications], McGraw-Hill, New York (1985).
[5] DeBruin, J.C. and Johnson, D., “Line-of-sight reference frames: A unified approach to plane-mirror kinematics,”
Proc. SPIE 1697, 111–129 (1992).
[6] Ellison, B. and Richi, J., “Inertial stabilization of periscopic sights band driven three axle gimbal”, Proc. SPIE 389,
107-119 (1983).
[7] Masten, M., “Inertially stabilized platforms for imaging optical systems,” IEEE Contr. Syst. Mag., 28(1), 47–64
(2008).

Proc. of SPIE Vol. 8052 80520J-14

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/02/2013 Terms of Use: http://spiedl.org/terms