Stability and Performance Analysis for SISO Incremental

Flight Control
Zhidong Lu Florian Holzapfel
Institute of Flight System Dynamics, Technische Universität München, Garching, 85748, Germany

ABSTRACT
Incremental Nonlinear Dynamic Inversion (INDI) control has attracted increasing research attention
for it retains the high-performance of NDI and has enhanced robustness. However, when actual
elements of the flight control system and real-world phenomena (such as actuator dynamics, sensor
noise, time delay, etc.) are considered, the INDI control may have degraded performance or even lose
stability. This paper analyzed the stability and performance of an incremental controller for a SISO
linear plant based on transfer functions. Besides, the theoretical analysis results are verified through
numerical evaluations for the incremental controller of the short-period aircraft model using
comprehensive metrics.

Keywords: Incremental Nonlinear Dynamic Inversion; Flight Control; Stability and Performance; Time delay; PCH

1 Introduction
Nonlinear Dynamic Inversion (NDI) is a widely used multivariable control technique in the aerospace
industry [1]. The major advantage of NDI is avoiding gain-scheduling by directly canceling nonlinearities
with feedback linearization. But it requires accurate knowledge of the nonlinear dynamics and is thus
sensitive to model uncertainty [2]. Incremental Nonlinear Dynamic Inversion (INDI) is a variation of NDI
with better robustness because it uses the (angular) acceleration feedback to reduce model dependency [3].
Successful applications of INDI flight control have been conducted on MAVs and multi-rotors platforms
[4]-[6], where good robustness and disturbance rejection were shown. Stability and robustness analysis for
ideal INDI control has been conducted in Ref. [7].
However, there are still challenges for practical applications of INDI. For example, the time delay can
cause significant performance degradation of the flight control system [8], and INDI is very susceptible to
sensor time delay [9]. Other issues, including actuator dynamics [10], input delay [11], noise filtering [12],
and model uncertainties [13] also pose challenges. Besides, the strategy of Pseudo Control Hedging (PCH)
was first developed for approximate dynamic inversion based model reference adaptive control to deal with
input dynamics[14], and it has also been directly transplanted to INDI control[15][16]. However, the
application of PCH to INDI still needs further investigation. Therefore, it is of interest to know about the
stability and performance of incremental flight control under practical circumstances.
1
 In this paper, we discussed the stability and performance of the incremental controller for Single-input-
single-output (SISO) LTI flight dynamics using classical control theories. Elements of the flight control
system, including actuator, sensor, filter, error control gain, and some real-world phenomena like time
delay and model mismatch, are taken into account. This paper is structured as follows: Sec. 2 derives
control law and analytical transfer functions of the incremental controller. Sec. 3 and Sec. 4 address the
stability and performance issues. Numerical evaluations are conducted for a short-period model
incremental controller in Sec. 5. Finally, Sec. 6 gives conclusions and recommendations.

2 Incremental Control Law and Transfer Functions

Consider the generic form of a SISO linear system,

𝑥̇ = 𝐴𝑥 + 𝐵𝑢
{ (1)
𝑦 = 𝐶𝑥
where 𝑥 ∈ ℝ𝑛 is the state vector, 𝑢 is the input, 𝑦 is the controlled variable. Two forms of incremental
control law, the conventional incremental dynamic inversion, and the modified incremental dynamic
inversion, will be discussed below.

2.1 Conventional incremental dynamic inversion

The conventional incremental dynamic inversion method is based on the computation of the control
increment at the current time step with respect to the system’s condition in the past time step.
𝑥̇ = 𝑥̇ 0 + 𝐴(𝑥 − 𝑥0 ) + 𝐵(𝑢 − 𝑢0 ) = 𝑥̇ 0 + 𝐴 ∙ ∆𝑥 + 𝐵 ∙ ∆𝑢 (2)
It is assumed that the state changes significantly slower than the control input in a minimal time
increment. By assuming ∆𝑥 = 0, Equation (2) can be simplified as:
𝑥̇ = 𝑥̇ 0 + 𝐵 ∙ ∆𝑢 (3)
Combining Equation (3) with (1), we have:
𝑦̇ = 𝑦̇ 0 + 𝐶𝐵 ∙ ∆𝑢 (4)
Inverting Equation (4) and setting a pseudo control 𝜈 = 𝑦̇ , the following control law is obtained:
∆𝑢 = (𝐶𝐵)−1 (𝜈 − 𝑦̇ 0 ) (5)
The total input will be the sum of the current input 𝑢0 and the calculated input increment ∆𝑢:
𝑢 = 𝑢0 + ∆𝑢 (6)
In practical applications, either the input measurement value 𝑢0𝑚𝑒𝑎 or the estimation 𝑢0𝑚𝑑𝑙 from an
actuator model will be used as 𝑢0 .

2.2 Modified incremental dynamic inversion

The conventional incremental control law builds a direct connection between the commanded pseudo
control and the command input increment:
2
 ∆𝜈𝑐 = 𝐶𝐵 ∙ ∆𝑢𝑐 (7)
However, this design regards the expected bandwidth of the pseudo control variable and the actuator
bandwidth as consistent, but the two are likely and usually expected to be inconsistent. The expected
bandwidth of the pseudo control variable could be controlled by:
𝜈̇𝑐 = 𝐾𝜈 ∆𝜈𝑐 (8)
and the achieved bandwidth of a first-order actuator model is determined by its time constant:
1 1
𝑢̇ 𝑐 = (𝑢𝑐 − 𝑢) = ∆𝑢 (9)
𝑇𝑎𝑐𝑡 𝑇𝑎𝑐𝑡 𝑐
Based on Eq. (7) and assume the time increment is infinitely small, we have:
𝜈̇𝑐 = 𝐶𝐵 ∙ 𝑢̇ 𝑐 (10)
Combine Eq.(8)~(10), one obtains the modified incremental control input as:
∆𝑢𝑐 = 𝑇𝑎𝑐𝑡 (𝐶𝐵)−1 𝐾𝜈 ∆𝜈𝑐 (11)
And the final control input will be:
𝑢𝑐 = 𝑢0 + ∆𝑢𝑐 (12)
It can be seen that the conventional incremental control law is a particular case of the modified
incremental control law when 𝑇𝑎𝑐𝑡 𝐾𝜈 = 1. Therefore, the modified control law will provide an additional
degree of freedom for control system tuning.
With the following first-order command reference model, the Pseudo Control Hedging (PCH) strategy is
used [15][16].
𝑟̇ = 𝐾𝑟 (𝑐 − 𝑟) (13)
The hedging signal is designed as the difference between the commanded pseudo control and the
achieved pseudo control:
𝜈ℎ = 𝜈𝑐 − 𝜈𝑎 = 𝐶𝐵(𝑢𝑐 − 𝑢̂) (14)
The hedging signal is then introduced into the reference model to modify the state response, while the
instantaneous pseudo-control output of the reference model remains unchanged:
𝑟̇ = 𝑣𝑟 − 𝜈ℎ , 𝑣𝑟 = 𝐾𝑟 (𝑐 − 𝑟) (15)
The block diagram illustrates the modified incremental control law is given by Fig. 1. Other elements of
the incremental controller are also shown, including:
1) The error controller 𝐾𝑃 , which is a scalar value;
2) The filter to estimate the angular acceleration from measurements is simplified as a combination of
a first-order low pass filter 𝐹(𝑠) = 1/(𝑇diff 𝑠 + 1) and the differential operator s (This form of
representation is for the convenience of subsequent analysis);
3) The control allocation module 𝐵̂ −1 = (𝐶𝐵)−1, which is a scalar value for a SISO plant;

3
 𝑒 −𝜏𝑎 𝑠
4) The actuator dynamics, modeled as a first-order transfer function with time delay 𝐺𝑎 (𝑠) = (𝑇 ;
act 𝑠+1)

𝑒 −𝜏𝑠 𝑠
5) The sensor dynamics 𝐻, modeled as a transfer function with time delay 𝐻(𝑠) = (𝑇 ;
sensor 𝑠+1)

6) The actuator measurements dynamics 𝐺𝑎𝑚 .

B̂
h
Reference Model
d
r
c r e c  c c uc uc uc u y
Kr 1/s KP K B̂ −1 Tact Ga P
Control
Input dynamics
Error Allocation û
Controller
ˆ
Gam
sF Filter Input
Measurements

Sensors
n

Fig. 1 Block diagram of the SISO incremental control system.

2.3 Transfer functions

The open-loop and closed-loop transfer functions for both the conventional and modified incremental
controller are derived in this section.
A. Conventional incremental controller
When PCH is off, the open-loop transfer function broken at input u is:
(𝐾𝑃 + 𝑠𝐹(𝑠))
𝐿𝑢 = 𝐺𝑎 (𝑠)𝐶̄ (𝑠)𝐻(𝑠)𝑃(𝑠) = 𝐺𝑎 (𝑠) 𝐻(𝑠)𝑃(𝑠) (16)
𝐵̂ [1 − 𝐺𝑎 (𝑠)𝐺𝑎𝑚 (𝑠)]
where 𝐶̄ (𝑠) is an equivalent INDI controller. And the closed-loop transfer function from command to
roll rate output is:
𝐺𝑎 (𝑠)𝑃(𝑠)(𝐾𝑃 + 𝑠)
𝑇𝑦𝑐 = (17)
(𝐾1𝑟 𝑠 + 1) [𝐵̂ (1 − 𝐺𝑎 (𝑠)𝐺𝑎𝑚 (𝑠)) + 𝐺𝑎 (𝑠)𝐻(𝑠)𝑃(𝑠)(𝐾𝑃 + 𝑠𝐹 (𝑠))]

When PCH is on, the open-loop transfer function is given by:

(𝑠 + 𝐾𝑟 )(𝐾𝑃 + 𝑠𝐹(𝑠))
𝐿𝑢,𝑃𝐶𝐻 = 𝐺𝑎 (𝑠)𝐶̄PCH (𝑠)𝐻(𝑠)𝑃(𝑠) = 𝐺𝑎 (𝑠) 𝐻(𝑠)𝑃(𝑠) (18)
𝐵̂ (𝐾𝑃 + 𝑠)(1 − 𝐺𝑎𝑚 (𝑠)𝐺𝑎 (𝑠))
And the closed-loop transfer function from command to output is:

4
 𝐺𝑎 (𝑠)𝑃(𝑠)(𝐾𝑃 + 𝑠)𝐾𝑟
𝑇𝑦𝑐,PCH = (19)
𝐵̂ [(𝐾𝑃 + 𝑠)(1 − 𝐺𝑎 (𝑠)𝐺𝑎𝑚 (𝑠))] + 𝐺𝑎 (𝑠)𝐻(𝑠)𝑃(𝑠)(𝐾𝑟 + 𝑠)(𝐾𝑃 + 𝑠𝐹(𝑠))
B. Modified incremental controller
Based on Fig.1, one can find that the modified incremental controller can be derived from the
conventional one by extending the control allocation module in the following way:
𝐵̂ −1 ⇐ 𝐵̂ −1 𝑇𝑎𝑐𝑡 𝐾𝜈 (20)
Similarly, the open-loop and closed-loop transfer functions can be obtained:
𝑇𝑎𝑐𝑡 𝐾𝜈 (𝐾𝑃 + 𝑠𝐹(𝑠))
𝐿𝑢 = 𝐺𝑎 (𝑠)𝐶̄ (𝑠)𝐻(𝑠)𝑃(𝑠) = 𝐺𝑎 (𝑠) 𝐻(𝑠)𝑃(𝑠) (21)
𝐵̂ [1 − 𝐺𝑎 (𝑠)𝐺𝑎𝑚 (𝑠)]
𝑇𝑎𝑐𝑡 𝐾𝜈 𝐺𝑎 (𝑠)𝑃(𝑠)(𝐾𝑃 + 𝑠)
𝑇𝑦𝑐 = (22)
(𝐾1 𝑠 + 1) [𝐵̂ (1 − 𝐺𝑎 (𝑠)𝐺𝑎𝑚 (𝑠)) + 𝑇𝑎𝑐𝑡 𝐾𝜈 𝐺𝑎 (𝑠)𝐻(𝑠)𝑃(𝑠)(𝐾𝑃 + 𝑠𝐹(𝑠))]
𝑟

𝐺𝑎 (𝑠)𝑇𝑎𝑐𝑡 𝐾𝜈 (𝑠 + 𝐾𝑟 )(𝐾𝑃 + 𝑠𝐹(𝑠))𝐻(𝑠)𝑃(𝑠)

𝐿𝑢,𝑃𝐶𝐻 = 𝐺𝑎 (𝑠)𝐶̄PCH (𝑠)𝐻(𝑠)𝑃(𝑠) = (23)
𝐵̂ [𝐾𝑃 𝐾𝜈 𝑇𝑎𝑐𝑡 + 𝑠 + 𝐾𝑟 (1 − 𝐾𝜈 𝑇𝑎𝑐𝑡 )][1 − 𝐺𝑎 (𝑠)𝐺𝑎𝑚 (𝑠)]
𝑇𝑦𝑐,PCH
𝑇𝑎𝑐𝑡 𝐾𝜈 𝐺𝑎 (𝑠)𝑃(𝑠)(𝐾𝑃 + 𝑠)𝐾𝑟
= (24)
𝐵̂ [𝐾𝑃 𝐾𝜈 𝑇𝑎𝑐𝑡 + 𝑠 + 𝐾𝑟 (1 − 𝐾𝜈 𝑇𝑎𝑐𝑡 )][1 − 𝐺𝑎 (𝑠)𝐺𝑎𝑚 (𝑠)] + 𝐺𝑎 (𝑠)𝑇𝑎𝑐𝑡 𝐾𝜈 (𝑠 + 𝐾𝑟 )(𝐾𝑃 + 𝑠𝐹(𝑠))𝐻(𝑠)𝑃(𝑠)
Only the modified incremental controller will be analyzed in the following.

3 Stability Analysis for Incremental Controller

This section analyzes the incremental controller's stability, especially the influence of time delay,
control gain, actuator dynamics, sensor dynamics, filter, and Pseudo-Control-Hedging on stability.

3.1 Ideal case

Ideally, assuming 𝐻(𝑠) = 1, 𝐹(𝑠) = 1, 𝐺𝑎𝑚 (𝑠) = 1, then the equivalent incremental controller becomes:
𝑇𝑎𝑐𝑡 𝐾𝜈 (𝐾𝑃 + 𝑠)(𝑇𝑎𝑐𝑡 𝑠 + 1) (𝐾𝑃 𝑇𝑎𝑐𝑡 + 1)𝐾𝜈 𝑇𝑎𝑐𝑡 𝐾𝜈 𝐾𝜈 𝐾𝑃 1
𝐶̄ (𝑠) = = + 𝑠+ (25)
𝐵̂ 𝑇𝑎𝑐𝑡 𝑠 𝐵̂ 𝐵̂ 𝐵̂ 𝑠
In this condition, the incremental controller is reduced to a PID controller, and the proportional, integral,
(𝐾 𝑇 +1)𝐾𝜈 𝐾 𝐾 𝑇 𝐾
and derivative coefficients are respectively 𝐶̄𝑃 = 𝑃 𝑎𝑐𝑡 , 𝐶̄𝐼 = 𝜈 𝑃 , 𝑎𝑛𝑑 𝐶̄𝐷 = 𝑎𝑐𝑡 𝜈 . When
𝐵̂ 𝐵̂ 𝐵̂
𝐾𝑃 , 𝐾𝜈 and 𝑇𝑎𝑐𝑡 increase or 𝐵̂ decreases, all three coefficients will increase.
The open-loop transfer function is given by:
𝐾𝜈 (𝐾𝑃 + 𝑠)
𝐿𝑢 = 𝐺𝑎 (𝑠)𝐶̄ (𝑠)𝐻(𝑠)𝑃(𝑠) = 𝑃(𝑠) (26)
𝐵̂ 𝑠

5
 We can find that the actuator dynamics are canceled in the open-loop transfer function. And the
incremental controller is reduced to a PI controller for the plant 𝑃(𝑠). The open-loop frequency response
goes:

𝐾𝜈 √𝜔 2 + 𝐾𝑃2 𝜔 𝜋
|𝐿𝑢 (𝑗𝜔)| = |𝑃(𝑗𝜔)|, ∠𝐿𝑢 (𝑗𝜔) = ∠𝑃(𝑗𝜔) + 𝑎𝑟𝑐𝑡𝑎𝑛 − (27)
𝐵̂ 𝜔 𝐾𝑃 2
It can be inferred that a larger 𝐾𝑃 leads to increased magnitude and decreased phase; thus the gain
margin, phase margin, and time delay margin are all reduced; 𝐾𝜈 and 𝐵̂ only affect the magnitude and a
larger 𝐾𝜈 or smaller 𝐵̂ will increase the magnitude and result in reduced gain margin and time delay margin.
Example 1. Take the aircraft roll mode dynamics for example, which is 𝑝̇ = 𝐿𝑝 𝑝 + 𝐿𝛿𝑎 𝛿𝑎 . The stability
margins of the incremental controller for the roll mode dynamics can be derived as follows:
𝜔𝑐 𝐿𝑝 1 𝜔𝑐 𝐿𝑝
𝐺𝑀 = ∞, 𝑃𝑀 = arctan − arctan , 𝑇𝐷𝑀 = (arctan − arctan ) (28)
𝐾𝑃 𝜔𝑐 𝜔𝑐 𝐾𝑃 𝜔𝑐
where 𝜔𝑐 is the magnitude cross-frequency:

2 2 2 2
𝐿̂
𝛿𝑎 𝐿̂
𝛿𝑎 𝐿̂
𝛿𝑎
𝐾𝜈 − (𝐿𝑝 𝐿 ) + √(𝐾𝜈 − (𝐿𝑝 𝐿 ) ) + 4 (𝐿 ) 𝐾𝑃 2 𝐾𝜈2
2 2
𝛿𝑎 𝛿𝑎 𝛿𝑎

𝜔𝑐 = (29)
√ 2
Assume 𝐿̂ 𝛿𝑎 = 𝐿𝛿𝑎 , the time delay margin of the roll mode incremental controller with different values
of [𝐾𝑃 𝐾𝜈 𝐿𝑝 ] are calculated according to Eq.(28), and the results are shown in Fig. 2. It can be
inferred that gain 𝐾𝜈 has the most significant effect on the time delay margin, while 𝐾𝑃 and 𝐿𝑝 only have
apparent influence at large values.

Fig. 2 Time delay margin of the incremental controller for roll dynamics.

3.2 Time delay

Based on the ideal case described above, we additionally consider the time delays in the actuator, sensor,
𝑒 −𝜏𝑎 𝑠
and input measurement, i.e. 𝐻(𝑠) = 𝑒 −𝜏𝑠 𝑠 , 𝐺𝑎 (𝑠) = 𝑇 , 𝐺𝑎𝑚 (𝑠) = 𝑒 −𝜏𝑎𝑚 𝑠 .
𝑎𝑐𝑡 𝑠+1
6
 The open-loop transfer function and closed-loop transfer function become:
𝑇𝑎𝑐𝑡 𝐾𝜈 (𝐾𝑃 + 𝑠)𝑒 −(𝜏𝑎+𝜏𝑠 )𝑠
𝐿𝑢 = 𝑃(𝑠) (30)
𝐵̂ [𝑇𝑎𝑐𝑡 𝑠 + 1 − 𝑒 −(𝜏𝑎+𝜏𝑎𝑚)𝑠 ]
𝐾𝑟 𝑇𝑎𝑐𝑡 𝐾𝜈 (𝐾𝑃 + 𝑠)𝑃(𝑠)𝑒 −𝜏𝑎𝑠
𝑇𝑦𝑐 = ∙ (31)
𝑠 + 𝐾𝑟 𝐵̂ [𝑇𝑎𝑐𝑡 𝑠 + 1 − 𝑒 −(𝜏𝑎+𝜏𝑎𝑚)𝑠 ] + 𝑇𝑎𝑐𝑡 𝐾𝜈 (𝐾𝑃 + 𝑠)𝑃(𝑠)𝑒 −(𝜏𝑎+𝜏𝑠 )𝑠
Define 𝜏1 = 𝜏𝑎 + 𝜏𝑠 , 𝜏2 = 𝜏𝑎 + 𝜏𝑎𝑚 and use the second-order Padéapproximant [18]:
12 − 6𝜏𝑠 + (𝜏𝑠)2
𝑒 −𝜏𝑠 ≈ (32)
12 + 6𝜏𝑠 + (𝜏𝑠)2
Consider the following two situations:
a) 𝜏1 = 𝜏2 , which means the time delays 𝜏𝑠 and 𝜏𝑎𝑚 are well synchronized.
In this case, the open-loop transfer function becomes:
𝐾𝜈 (𝐾𝑃 + 𝑠)(12 − 6𝜏1 𝑠 + (𝜏1 𝑠)2 )
𝐿𝑢 = 𝜏 𝑃(𝑠) (33)
𝐵̂ 𝑠 (12 + 12 𝑇 1 + 6𝜏1 𝑠 + 𝜏12 𝑠 2 )
𝑎𝑐𝑡

Compare Eq.(26) and Eq.(33),the effect of synchronized-delay is equivalent to the transfer function:
12 − 6𝜏1 𝑠 + (𝜏1 𝑠)2
Γ1 (𝑠) = 𝜏 (34)
12 + 12 𝑇 1 + 6𝜏1 𝑠 + 𝜏12 𝑠 2
𝑎𝑐𝑡

This transfer function Γ1 (𝑠) has a similar “phase-lagging” effect to the Padéapproximant of the pure
delay 𝑒 −𝜏1𝑠 . Besides, it will also reduce the magnitude |𝐿𝑢 (𝑗𝜔)| in the low-frequency range by
𝜏
−20log(1 + 𝑇 1 ) dB. Therefore, both the phase margin and gain margin of the closed-loop is reduced, as
𝑎𝑐𝑡
illustrated in Fig. 3, where the red line represents an increased synchronized-delay.

Fig. 3 Bode diagram and Nyquist diagram of the open-loop with synchronized delay.
To determine how much synchronized-delay the system can tolerate, the closed-loop transfer function
with synchronized-delay is investigated:
7
 𝐾𝑟 𝐾𝜈 (𝐾𝑃 + 𝑠)𝑒 −𝜏𝑎𝑠
𝑇𝑦𝑐 = ∙ (35)
𝑠 + 𝐾𝑟 𝐵̂ 𝜏
[𝑠(12 + 6𝜏1 𝑠 + 𝜏12 𝑠 2 ) + 12 𝑇 1 𝑠] + 𝐾𝜈 (𝐾𝑃 + 𝑠)(12 − 6𝜏1 𝑠 + 𝜏12 𝑠 2 )
𝑃(𝑠) 𝑎𝑐𝑡

𝐵 ̂
Assume 𝐵̂ = 𝐶𝐵, then 𝑃(𝑠) = |𝑠𝐼 − 𝐴|, and the characteristic polynomial can be obtained as:

𝜏1
det(s) = |𝑠𝐼 − 𝐴| [𝑠(12 + 6𝜏1 𝑠 + 𝜏12 𝑠 2 ) + 12 𝑠] + 𝐾𝜈 (𝐾𝑃 + 𝑠)(12 − 6𝜏1 𝑠 + 𝜏12 𝑠 2 ) (36)
𝑇𝑎𝑐𝑡
Take the roll-mode dynamics in Example 1 for instance. The characteristic polynomial in this case is:
𝜏1
det(s) = 𝜏12 𝑠 4 + (6𝜏1 + 𝐾𝜈 𝜏12 − 𝐿𝑝 𝜏12 )𝑠 3 + (12 − 6𝐿𝑝 𝜏1 − 6𝐾𝜈 𝜏1 + 𝐾𝑃 𝐾𝜈 𝜏12 + 12) 𝑠 2
𝑇𝑎𝑐𝑡
𝜏1
+ (12𝐾𝜈 − 12𝐿𝑝 (1 + ) − 6𝐾𝑃 𝐾𝜈 𝜏1 ) 𝑠 + 12𝐾𝑃 𝐾𝜈 (37)
𝑇𝑎𝑐𝑡
According to the Routh criterion, the necessary and sufficient condition for closed-loop stability is:
𝑎𝑖 > 0, 𝑖 = 0,1,2,3,4
{ (38)
𝛥 = 2 𝑎3 − 𝑎4 𝑎1 )𝑎1 − 𝑎32 𝑎0 > 0
(𝑎
Considering that 𝐾𝜈 does not exceed 1/𝑇𝑎𝑐𝑡 in practice, the solution to the above inequality is:
2𝐾𝜈 − 2𝐿𝑝
0 < 𝜏1 < , 𝐾𝑃 𝐾𝜈 > − 2𝐿𝑝 ⁄𝑇𝑎𝑐𝑡
{ 𝐾𝑃 𝐾𝜈 + 2𝐿𝑝 ⁄𝑇𝑎𝑐𝑡 (39)
0 < 𝜏1 , 𝐾𝑃 𝐾𝜈 < − 2𝐿𝑝 ⁄𝑇𝑎𝑐𝑡
Eq.(39) suggests that when the product of control gain 𝐾𝑃 𝐾𝜈 is large enough, the system can only
tolerate limited synchronized delay.
b) 𝜏1 ≠ 𝜏2 , which denotes the time delays between 𝜏𝑠 and 𝜏𝑎𝑚 are not synchronized.
In this case, the open-loop transfer function becomes:
𝐾𝜈 (𝐾𝑃 + 𝑠)(12 − 6𝜏1 𝑠 + 𝜏12 𝑠 2 )(12 + 6𝜏2 𝑠 + 𝜏22 𝑠 2 )
𝐿𝑢 = 𝜏 𝑃(𝑠) (40)
𝐵̂ 𝑠(12 + 6𝜏1 𝑠 + 𝜏12 𝑠 2 ) (12 + 12 𝑇 2 + 6𝜏2 𝑠 + 𝜏22 𝑠 2 )
𝑎𝑐𝑡

Compare Eq.(40) and Eq.(33), it can be inferred that the effect of asynchronized-delay is identical to
the transfer function defined as:
(12 − 6𝜏1 𝑠 + 𝜏12 𝑠 2 )(12 + 6𝜏2 𝑠 + 𝜏22 𝑠 2 )
Γ2 (𝑠) = 𝜏 (41)
(12 + 6𝜏1 𝑠 + 𝜏12 𝑠 2 ) (12 + 12 2 + 6𝜏2 𝑠 + 𝜏22 𝑠 2 )
𝑇𝑎𝑐𝑡
Similar to Γ1 (𝑠), transfer function Γ2 (𝑠) also has both “phase-lagging” and “low-frequency magnitude
reducing” effects while they vary with the relation between 𝜏1 and 𝜏2 . Set 𝑇𝑎𝑐𝑡 to 1/50s, the bode diagram
of Γ2 (𝑠) with different values of 𝜏1 and 𝜏2 is shown in the left figure of Fig.4. And the corresponding open-
loop bode diagram for the roll dynamics incremental controller is plotted in the right figure, where 𝐾𝜈 =
1⁄𝑇𝑎𝑐𝑡 , 𝐾𝑃 = 5, 𝐿𝑝 = 5, 𝐿̂ 𝛿𝑎 = 𝐿𝛿𝑎 . Compared with the case when 𝜏1 = 𝜏2 , more phase lag and fewer
magnitude reduction will be brought to the open-loop when 𝜏1 > 𝜏2 , both of which contribute to a reduced

8
gain margin. On the other hand, when 𝜏1 < 𝜏2 , ω∠𝐿𝑢=180° increase, but the magnitude at this frequency
increase, so the gain margin is reduced again as a result.

Fig. 4 Bode diagram of 𝚪𝟐 and the open-loop with asynchronized delays.

Fig. 5 Stability area of [𝜏1 , 𝜏2 ] for the incremental controller with different gains.

9
 A detailed illustration of the relationship between stability and asynchronized-delay [ 𝜏1 , 𝜏2 ] is
demonstrated in Fig.5, where the colored zone represents stability. Three findings can be drawn from Fig.5:
Firstly, closed-loop stability is independent of 𝜏2 as long as 𝜏1 is small enough, which means actuator
measurements delay will not drive the closed-loop unstable when sensor delay is below the delay margin.
Secondly, stability could be maintained with a considerable value of 𝜏1 under the condition that 𝜏2 is close
enough to 𝜏1 , that is, delay synchronization. Thirdly, larger control gain 𝐾𝑃 and 𝐾𝜈 will contract the
stability area. Significantly, they will impair the capability of delay synchronization.
Combining the analysis results of a) and b), a conclusion can be drawn that the incremental controller is
very sensitive to sensor time delay, but the tolerance to delay could be enhanced by introducing a
synchronization delay in the actuator measurement loop. However, as the synchronized delays increase
altogether, the stability margin will decrease, and the benefits of synchronization will be infringed by high
control gains.

3.3 Sensor and filter dynamics compensation

The sensor and filter dynamics themselves also have a “time delay” effect. This part investigates the
impact of compensation for the sensor and filter dynamics into the actuator measurement loop.
When neither sensor nor filter dynamics compensation is incorporated in the actuator measurement loop,
the open-loop transfer function is given by:
𝐾𝜈 (𝐾𝑃 + 𝑠 + 𝐾𝑃 𝑇𝑑𝑖𝑓𝑓 𝑠)
𝐿𝑢1 = 𝑃(𝑠) (42)
𝐵̂ 𝑠(𝑇𝑑𝑖𝑓𝑓 𝑠 + 1)(𝑇𝑠𝑒𝑛𝑠𝑜𝑟 𝑠 + 1)
When filter dynamics compensation alone is incorporated in the actuator measurement loop, one obtains:
𝐾𝜈 (𝐾𝑃 + 𝑠 + 𝐾𝑃 𝑇𝑑𝑖𝑓𝑓 𝑠)
𝐿𝑢2 = 𝑃(𝑠) (43)
𝑇𝑑𝑖𝑓𝑓
̂
𝐵 𝑠 (𝑇𝑑𝑖𝑓𝑓 𝑠 + 1 + 𝑇 ) (𝑇𝑠𝑒𝑛𝑠𝑜𝑟 𝑠 + 1)
𝑎𝑐𝑡

When both the sensor and filter dynamics compensation are introduced in the actuator measurement
loop, the open-loop transfer function is:
𝐾𝜈 (𝐾𝑃 + 𝑠 + 𝐾𝑃 𝑇𝑑𝑖𝑓𝑓 𝑠)
𝐿𝑢3 = 𝑃(𝑠) (44)
𝑇𝑑𝑖𝑓𝑓 𝑇𝑠𝑒𝑛𝑠𝑜𝑟
̂
𝐵 𝑠 [(𝑇𝑑𝑖𝑓𝑓 𝑠 + 1 + 𝑇 ) (𝑇𝑠𝑒𝑛𝑠𝑜𝑟 𝑠 + 1) + 𝑇 ]
𝑎𝑐𝑡 𝑎𝑐𝑡

Therefore, the relation between these three loop-transfers can be obtained:

|𝐿𝑢1 (𝑗𝜔)| > |𝐿𝑢2 (𝑗𝜔)| > |𝐿𝑢3 (𝑗𝜔)|
(45)
∠𝐿𝑢3 (𝑗𝜔) > ∠𝐿𝑢2 (𝑗𝜔) > ∠𝐿𝑢1 (𝑗𝜔)
It suggests that incorporating sensor and filter dynamics compensation into the actuator measurement
loop will increase the gain margin, phase margin, and time delay margin. Based on the above analysis, we
can infer that introducing compensation for any additional linear dynamics aimed at measurements noise
attenuation into the 𝐺𝑎𝑚 loop will improve stability.

10
3.4 Pseudo-control-hedging
Although pseudo-control-hedging was originally used to cope with the nonlinear characteristics of the
actuator like saturation [14], only its linear dynamics are considered here. The open-loop functions with
PCH and without PCH are Eq.(21) and Eq.(23) respectively. The ration function between these two is:
𝐿𝑢,𝑃𝐶𝐻 𝑠 + 𝐾𝑟
𝑅(𝑠) = = (46)
𝐿𝑢 𝑠 + 𝐾𝑟 + 𝐾𝜈 𝑇𝑎𝑐𝑡 (𝐾𝑃 − 𝐾𝑟 )
We can infer that in the case of 𝐾𝑃 > 𝐾𝑟 , PCH will reduce the gain and increase the open-loop phase in
the low-frequency band; As a result, the stability margins increase. And the effect of PCH will be more
prominent with a higher value of 𝐾𝜈 𝑇𝑎𝑐𝑡 . Conversely, when 𝐾𝑃 < 𝐾𝑟 , PCH will reduce stability, and this
reduction will be more severe with a higher value of 𝐾𝜈 𝑇𝑎𝑐𝑡 . However, the actual situation for flight control
is generally the former.

4 Performance Analysis for Incremental Controller

This section derives the closed-loop transfer function that characterizes the closed-loop system's
performance, including command tracking, robustness, disturbance suppression, and noise suppression, and
connects them with the open-loop transfer function, and analyzes the relationship between them.

4.1 Tracking and Robustness

When PCH is off, the closed-loop transfer function from command to output measurement is:
𝐾𝑟 𝑇𝑎𝑐𝑡 𝐾𝜈 𝐺𝑎 (𝑠)𝐻(𝑠)𝑃(𝑠)(𝐾𝑃 + 𝑠)
𝑇𝑦𝑚𝑐 =
(𝑠 + 𝐾𝑟 )[𝐵̂(1 − 𝐺𝑎 (𝑠)𝐺𝑎𝑚 (𝑠)) + 𝑇𝑎𝑐𝑡 𝐾𝜈 𝐺𝑎 (𝑠)𝐻(𝑠)𝑃(𝑠)(𝐾𝑃 + 𝑠𝐹(𝑠))]
𝐾𝑟 𝐾𝑃 + 𝑠 𝐿𝑢
= ∙ ∙ (47)
𝑠 + 𝐾𝑟 𝐾𝑃 + 𝑠𝐹(𝑠) 1 + 𝐿𝑢
And the tracking error function will be:
𝐾𝑟 𝐾𝑃 + 𝑠 𝐿𝑢
𝑇𝑒𝑐 = 𝑇𝑟𝑐 − 𝑇𝑦𝑚𝑐 = [1 − ∙ ] (48)
𝑠 + 𝐾𝑟 𝐾𝑃 + 𝑠𝐹(𝑠) 1 + 𝐿𝑢
Consider an additive uncertainty in the plant dynamics ∆𝑃(𝑠), then the deviation of command tracking
performance due to ∆𝑃(𝑠) can be given by:
𝐾𝑟 𝑇𝑎𝑐𝑡 𝐾𝜈 𝐺𝑎 (𝑠)[𝑃(𝑠) + ∆𝑃(𝑠)](𝐾𝑃 + 𝑠)
∆𝑇𝑦𝑐 =
(𝑠 + 𝐾𝑟 )[𝐵̂(1 − 𝐺𝑎 (𝑠)𝐺𝑎𝑚 (𝑠)) + 𝑇𝑎𝑐𝑡 𝐾𝜈 𝐺𝑎 (𝑠)𝐻(𝑠)[𝑃(𝑠) + ∆𝑃(𝑠)](𝐾𝑃 + 𝑠𝐹(𝑠))]
𝐾𝑟 𝑇𝑎𝑐𝑡 𝐾𝜈 𝐺𝑎 (𝑠)𝑃(𝑠)(𝐾𝑃 + 𝑠)
− (49)
(𝑠 + 𝐾𝑟 )[𝐵̂(1 − 𝐺𝑎 (𝑠)𝐺𝑎𝑚 (𝑠)) + 𝑇𝑎𝑐𝑡 𝐾𝜈 𝐺𝑎 (𝑠)𝐻(𝑠)𝑃(𝑠)(𝐾𝑃 + 𝑠𝐹(𝑠))]
The sensitivity function can be obtained as follows:
∆𝑇𝑦𝑐 ⁄𝑇𝑦𝑐 𝐵̂ (1 − 𝐺𝑎 (𝑠)𝐺𝑎𝑚 (𝑠))
𝑆(𝑠) = lim = (50)
∆𝑃→0 ∆𝑃⁄𝑃 𝐵̂ (1 − 𝐺𝑎 (𝑠)𝐺𝑎𝑚 (𝑠)) + 𝑇𝑎𝑐𝑡 𝐾𝜈 𝐺𝑎 (𝑠)𝐻(𝑠)𝑃(𝑠)(𝐾𝑃 + 𝑠𝐹(𝑠))

11
 Combing Eq.(21) with Eq.(50), the sensitivity function can be written as:
1
𝑆(𝑠) = (51)
1 + 𝐿𝑢
Similarly, when PCH is on
𝐾𝑟 𝐾𝑃 + 𝑠 𝐿𝑢,𝑃𝐶𝐻
𝑇𝑒𝑐,𝑃𝐶𝐻 = [1 − ∙ ] (52)
𝑠 + 𝐾𝑟 𝐾𝑃 + 𝑠𝐹(𝑠) 1 + 𝐿𝑢,𝑃𝐶𝐻
1
𝑆𝑃𝐶𝐻 (𝑠) = (53)
1 + 𝐿𝑢,𝑃𝐶𝐻
It can be inferred that tracking performance and robustness are consistent, and high loop gain will
result in enhanced tracking ability and robustness. According to Eq.(46), when 𝐾𝑃 > 𝐾𝑟 , PCH will
undermine tracking ability and robustness, and vice versa.

4.2 Disturbance rejection and noise attenuation

When PCH is off, the closed-loop transfer function from disturbance to output in Fig. 1 goes:
𝑃(𝑠) 𝑃(𝑠)
𝑇𝑦𝑑 = = = 𝑃(𝑠)𝑆(𝑠) (54)
1 + 𝐺𝑎 (𝑠)𝐶̄ (𝑠)𝐻(𝑠)𝑃(𝑠) 1 + 𝐿𝑢
And the closed-loop transfer function from noise to output is:
−𝐻(𝑠)𝐶̄ (𝑠)𝐺𝑎 (𝑠)𝑃(𝑠) 1
𝑇𝑦𝑛 = = − 1 = 𝑆(𝑠) − 1 (55)
̄
1 + 𝐺𝑎 (𝑠)𝐶 (𝑠)𝐻(𝑠)𝑃(𝑠) 1 + 𝐿𝑢
When PCH is on, the disturbance and noise functions can be derived similarly:
𝑃(𝑠)
𝑇𝑦𝑑,𝑃𝐶𝐻 = = 𝑃(𝑠)𝑆𝑃𝐶𝐻 (𝑠) (56)
1 + 𝐿𝑢,𝑃𝐶𝐻
1
𝑇𝑦𝑛,𝑃𝐶𝐻 = − 1 = 𝑆𝑃𝐶𝐻 (𝑠) − 1 (57)
1 + 𝐿𝑢,𝑃𝐶𝐻
Therefore, for a given plant, the expectation of improving its disturbance rejection ability is consistent
with the enhancement of robustness but contradicts its noise attenuation. According to Eq.(46), when 𝐾𝑃 >
𝐾𝑟 , PCH will enhance noise attenuation but undermine disturbance rejection, vice versa.

5 Numerical Evaluation
In this section, numerical evaluations are conducted for the incremental controller of a short-period
model. Comprehensive evaluation metrics are developed to quantitatively evaluate the stability, tracking
performance, disturbance rejection ability, noise attenuation ability, and robustness to model uncertainty of
the incremental controller. The influence of each element of the control system is thoroughly investigated.

12
5.1 Incremental Control for the Short-Period Model
The short-period model with the wind effect is given as follows [19]:
𝑍
𝛼̇ 𝑍𝛼 𝑍𝑞 + 1 𝛼 𝑍𝜂 −𝑍𝑉 − 𝑉𝛼 𝑢𝑔
0
[ ]=[ ] [𝑞 ] + [ ] 𝜂 + [ 𝑀𝛼 ] [𝑤𝑔 ]
𝑞̇ 𝑀𝛼 𝑀𝑞 𝑀𝜂 −𝑀 −𝑉
𝑉
0

The control derivatives of the DA-42 aircraft at the flight condition of 𝑉0=70m/s are used and listed in
Table. 1. These derivatives are taken as nominal values, a ±30% deviation of each parameter will be
considered when performing robustness verifications.
Table. 1 Parameter of the short-period model of DA-42 at 𝑽𝟎 =70m/s[20][21]
Parameter 𝑍𝛼 𝑍𝑞 𝑍𝜂 𝑍𝑉 𝑀𝛼 𝑀𝑞 𝑀𝜂 𝑀𝑉
Value -1.27 0.0037 4.4e-4 -0.003 -17.71 -2.63 -8.18 2.8e-04
The incremental pitch rate control system is illustrated in Fig. 6. It should be noted that because the PCH
modifies the state of the reference model, as well as the error dynamics. Therefore, the absolute error value
is obtained by subtracting the system output from the unmodified reference signal.
h  ug 
M̂ w 
 g
Reference Model r

r c  c uc uc 
( )
c −1
Kr 1/s KP K M̂ Tact Ga ( s ) Short-Period
Dynamics q
Control
Allocation Input dynamics
Error û
Controller
ˆ0
Gam ( s )
sF ( s )
e Sync
Filter
Kr 1/s
Dynamics Input
synchronization Measurements

Reference Model(unmodified)
H (s)

Sensors
n

Fig. 6 Block diagram of the incremental pitch rate control system

The descriptions of each part of the INDI control system are shown in Table. 2
Table. 2 Descriptions of the incremental pitch rate control system
Elements Description Nominal values

Reference Model 𝑣𝑟 = 𝐾𝑟 (𝑐 − 𝑟), 𝑟̇ = 𝑣𝑟 − 𝜈ℎ 𝐾𝑟 = 5

𝑇act = 1/60(𝑠), 𝜏𝑎 = 0
𝑢(𝑡 − 𝜏𝑎 ) 𝑠𝑎𝑡𝜂 [𝜂(𝑡)]
Actuator 𝜂̇ (𝑡) = 𝑠𝑎𝑡𝜂̇ { − } 𝑠𝑎𝑡𝜂̇ = {𝜂̇ ||𝜂̇ | < 100°/𝑠}
𝑇act 𝑇act
𝑠𝑎𝑡𝜂 = {𝜂||𝜂| < 30°}
𝑞(𝑡 − 𝜏𝑠 ) 𝑞𝑚 (𝑡) 𝑇sensor = 1/300(𝑠), 𝜏𝑠 = 0
Sensor 𝑞𝑚̇ (𝑡) = − ,
𝑇sensor 𝑇sensor Noise (𝜎 2 ) = 4.0 × 10−7

13
 𝑠
Filter 𝑞̇ 𝑓 = 𝑞 𝑇diff = 1/30 (𝑠)
𝑇diff 𝑠 + 1 𝑚
𝑒 −𝜏𝑎𝑚 𝑠
Actuator measurements 𝐺𝑎𝑚 (𝑠) = 𝜏𝑎𝑚 = 0
(𝑇diff 𝑠 + 1)(𝑇sensor 𝑠 + 1)

Error gain 𝑞̇ 𝑐 = 𝐾𝑃 (𝑞𝑟 − 𝑞𝑚 ) 𝐾𝑃 = 8

Control effectiveness estimate ̂𝜂

𝑀 ̂𝜂 = 𝑀𝜂
𝑀

𝑢𝑚 𝜋𝑥
𝑢𝑔 = (1 − 𝑐𝑜𝑠 ( )) 0 ≤ 𝑥 ≤ 𝑑𝑥 𝑑𝑥 = 120𝑚，𝑑𝑧 = 80𝑚，
2 𝑑𝑥
Disturbance
𝑤𝑚 𝜋𝑥
𝑤𝑔 = (1 − 𝑐𝑜𝑠 ( )) 0 ≤ 𝑥 ≤ 𝑑𝑧 𝑢𝑚 = 3.5𝑚/𝑠，𝑤𝑚 = 3𝑚/𝑠
2 𝑑𝑧

5.2 Evaluation Metrics

The overall performance of the incremental controller is examined through a series of simulations, and
evaluation metrics are extracted in the meantime, as shown in Fig. 7. For command tracking evaluation, the
q command is designed as a two-way square wave signal with an amplitude of 10°/s and an interval of 3s,
and the root-mean-square (RMS) tracking error and input are recorded. For disturbance rejection evaluation,
both horizontal and vertical gust-winds are activated at t=3s. For noise attenuation evaluation, white noise
with variance 𝜎 2 =4.0×10-7 is added to the sensor measurements, and the RMS of tracking error and input
are recorded. For robustness verification, command tracking simulations are done repeatedly for N=100
samples of the uncertain plant described above. The standard deviation of all N=100 RMS tracking errors is
calculated to characterize the performance deviation due to model uncertainties. The descriptions of all the
evaluation metrics and their values at the nominal condition are shown in Table. 3.
Table. 3 Evaluation metrics for the incremental pitch rate control system
Category Evaluation Metric Description Nominal value
GM Gain Margin 21.89dB
Stability PM Phase Margin 62.74°
TDM Time Delay Margin 0.0485s
RMSer RMS error for tracking 0.9014°/s
Command tracking
RMSur RMS input for tracking 14.3805°
RMSed RMS error due to disturbance 0.0773°/s
Disturbance rejection
RMSud RMS input due to disturbance 1.85°
RMSen RMS error due to noise 0.0325°/s
Noise attenuation
RMSun RMS input due to noise 0.0641°
Robustness σ(RMSer𝑖 ), 𝑖 = 1,2, … , 𝑁 Standard deviation of RMS error 0.115°/s

14
 Fig. 7 Simulation results of the incremental control system under the nominal condition

5.3 Results and Analysis

The evaluation results are shown in Fig. 8. Each subplot represents the effect of one or more specific
parameters on the whole incremental controller. Here is a brief analysis of each:
It can be learned from Fig. 8(a) that a larger gain 𝐾𝑃 will improve the command tracking, robustness,
and disturbance rejection at the cost of stability and noise attenuation. Besides, when 𝐾𝑃 < 𝐾𝑟 , PCH will
improve tracking and disturbance rejection while hurt stability and noise attenuation, and vice versa when
𝐾𝑃 > 𝐾𝑟 . Fig. 8(b) suggests that gain 𝐾𝜈 has similar effects as 𝐾𝑃 while it is much more essential to stability
margins. Fig. 8(c) reveals that it is more favorable to appropriately over-estimate control effectiveness than
under-estimate it. Fig. 8(d) indicates that higher actuator frequencies lead to better tracking performance,
disturbance rejection, and robustness, while decreased stability and noise attenuation. Besides, PCH will
improve stability for slower actuators. Fig. 8(e) and Fig. 8(f) show that the introduction of sensor and filter
dynamics compensation into the actuator measurement loop can significantly enhance system stability and
improve noise attenuation. Still, it will adversely affect command tracking, disturbance suppression, and
15
robustness when the sensor frequency or filter bandwidth is low. At last, Fig. 8(g) demonstrates that the
delay-synchronized system has enhanced stability and improved performances.

a) Influence of error gain 𝐾𝑃 on evaluation metrics

b) Influence of gain 𝐾𝜈 on evaluation metrics

16
c) Influence of control effectiveness scaling on evaluation metrics

d) Influence of actuator frequency evaluation metrics

e) Influence of sensor frequency on evaluation metrics

f) Influence of filter bandwidth on evaluation metrics

17
 g) Influence of sensor delay and input measurement delay on evaluation metrics
Fig. 8 Influence of different elements of the incremental control system on its overall performance

6 Conclusion
This paper has discussed the stability and performance of SISO linear incremental flight control systems
in both analytical and numerical methods. The following conclusions can be drawn:
1) The advantages of incremental control are better tracking performance, robustness, and disturbance
rejection ability, which are mainly attributed to high loop gain caused by the direct correspondence
between the pseudo control and the control input. On the other hand, the major challenges are
reduced stability margins and insufficient noise attenuation. The modified incremental controller
has extra design freedom, the pseudo control gain, which helps to establish a balance between the
above two aspects.
2) Delay synchronization between sensor measurements and actuator measurements can extend
stability. Besides transport delay, linear dynamics of sensor and filter should also be compensated
for in the actuator measurement loop.
3) PCH incorporates the reference model into the closed-loop. When the error control gain is greater
than the reference model gain, PCH will increase stability margins and improve the noise
attenuation ability, but weaken the tracking and disturbance rejection, and the benefits brought by
PCH is more evident under the conditions of high pseudo control gain and slow actuator dynamics.
The contrary is the case when the error control gain is less than the reference model gain.
4) Higher error gain, smaller control effectiveness estimate, higher actuator frequency, higher sensor
frequency, and higher filter bandwidth are beneficial to command tracking, robustness, and
disturbance rejection. At the same time, they may hurt stability and noise suppression.
This work provides some insights on the INDI flight control in a practical context. Further work should
be done for more complex dynamic systems such as MIMO and nonlinear flight systems.

18
References
[1] Balas, G. J. Flight Control Law Design: An Industry Perspective. European Journal of Control, 9(2-3), 2003.
[2] Enns, D., Bugajski, D., Hendrick, R., and Stein, G. Dynamic Inversion: An Evolving Methodology for Flight
Control Design. International Journal of Control, 59(1), 1994. DOI: 10.1080/00207179408923070.
[3] Smith, P. A Simplified Approach to Nonlinear Dynamic Inversion Based Flight Control. 23rd Atmospheric
Flight Mechanics Conference, American Institute of Aeronautics and Astronautics, Reston, VA, 1998.
[4] Smeur, E. J., de Croon, G. C., and Chu, Q. Gust Disturbance Alleviation with Incremental Nonlinear Dynamic
Inversion. 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, 2016.
DOI: 10.1109/IROS.2016.7759827.
[5] FalconíG P, Schatz S P, Holzapfel F. Fault tolerant control of a hexarotor using a command governor
augmentation. 2016 24th Mediterranean Conference on Control and Automation (MED). IEEE, 2016.
DOI: 10.1109/MED.2016.7535981.
[6] Akkinapalli, Venkata Sravan, and Florian Holzapfel. "Incremental Dynamic Inversion based Velocity Tracking
Controller for a Multicopter System." 2018 AIAA Guidance, Navigation, and Control Conference. Kissimmee,
Florida, January 2018. DOI:10.2514/6.2018-1345.
[7] Wang X, Van Kampen E J, Chu Q, et al. Stability analysis for incremental nonlinear dynamic inversion control.
Journal of Guidance, Control, and Dynamics, 42(5), 2019. DOI: 10.2514/1.G003791.
[8] Stéphane Delannoy and Simon Oudin. Longitudinal control law for modern long-range civil aircrafts. In
Proceedings of the 2nd CEAS Specialist Conference on Guidance, Navigation & Control (EuroGNC 2013), Delft,
Netherland, 2013.
[9] Sieberling, S., Q. P. Chu, and J. A. Mulder. "Robust flight control using incremental nonlinear dynamic
inversion and angular acceleration prediction." Journal of guidance, control, and dynamics 33.6 (2010): 1732-
1742. DOI:10.2514/1.49978
[10] Raab S A, Zhang J, Bhardwaj P, et al. Consideration of Control Effector Dynamics and Saturations in an
Extended INDI Approach. AIAA Aviation 2019 Forum. Dallas, USA, 2019. DOI: 10.2514/6.2019-3267.
[11] Koschorke, J., Falkena, W., Van Kampen, E.-J., & Chu, Q. P.. Time Delayed Incremental Nonlinear Control.
AIAA Guidance, Navigation, and Control (GNC) Conference. Boston, USA, 2013. DOI:10.2514/6.2013-4929.
[12] Smith, P. & Berry, A.. Flight Test Experience of a Non-Linear Dynamic Inversion Control Law on the VAAC
Harrier. In Atmospheric Flight Mechanics Conference. Denver, USA, 2000. DOI:10.2514/6.2000-3914.
[13] Jeon B J, Seo M G, Shin H S, et al. Understandings of the Incremental Backstepping Control through
Theoretical Analysis under the Model Uncertainties. 2018 IEEE Conference on Control Technology and
Applications (CCTA). IEEE, Copenhagen, Denmark, 2018. DOI: 10.1109/CCTA.2018.8511479.
[14] Johnson, Eric N., and Anthony J. Calise. Pseudo-control hedging: A new method for adaptive control. Advances
in navigation guidance and control technology workshop. Alabama, USA, 2000. DOI: 10.2514/6.2018-0845.
[15] Raab, Stefan A., et al. Proposal of a unified control strategy for vertical take-off and landing transition aircraft
configurations. 2018 Applied Aerodynamics Conference. Atlanta, USA, 2018. DOI: 10.2514/6.2018-3478.
[16] Keijzer, Twan, et al. Design and flight testing of incremental backstepping based control laws with angular
accelerometer feedback. AIAA Scitech 2019 Forum. San Diego, USA, 2019. DOI: 10.2514/6.2019-0129.
19
[17] Cook, Michael V. Flight dynamics principles: a linear systems approach to aircraft stability and control.
Butterworth-Heinemann, 2012, pp. 198–199.
[18] Vajta, Miklos. Some remarks on Padé-approximations. In Proceedings of the 3rd TEMPUS-INTCOM
Symposium. Veszprém, Hungary, 2000.
[19] Stevens, Brian L., Frank L. Lewis, and Eric N. Johnson. Aircraft control and simulation: dynamics, controls
design, and autonomous systems. John Wiley & Sons, 2015.
[20] Karlsson, Erik, et al. "Automatic flight path control of an experimental DA42 general aviation aircraft." 2016
14th International Conference on Control, Automation, Robotics and Vision (ICARCV). IEEE, 2016.
DOI:10.1109/icarcv.2016.7838566
[21] Krause, Christoph, and Florian Holzapfel. "System Automation of a DA42 General Aviation Aircraft." 2018
Aviation Technology, Integration, and Operations Conference. 2018. DOI: 10.2514/6.2018-3984

20