Joint 48th IEEE Conference on Decision and Control and

28th Chinese Control Conference

Shanghai, P.R. China, December 16-18, 2009

FrB01.1

Nonlinear Adaptive Flight Control Law Design and Handling Qualities

Evaluation
L. Sonneveldt*, E.R. van Oort, Q.P. Chu and J.A. Mulder
Aerospace Software and Technology Institute, Delft University of Technology
P.O. Box 5058, 2600 GB Delft, The Netherlands

Abstract This paper considers the design of a stability and

control augmentation system for a modern fighter aircraft.
The aim of the flight control system is to offer the pilot
consistent good flying and handling qualities over a specified
flight envelope and to provide robustness to model uncertainties.
A nonlinear adaptive backstepping method is proposed to
directly deal with the nonlinearities and the uncertainties of the
system. B-spline neural networks are used to partition the flight
envelope into multiple connecting regions. In each partition a
locally valid linear-in-the-parameters nonlinear aircraft model
is defined, of which the unknown parameters are approximated
online by Lyapunov based update laws. These update laws take
aircraft state and input constraints into account so that they
do not corrupt the parameter estimation process. The desired
aircraft response characteristics are enforced with command
filters and verified by applying conventional handling qualities
analysis techniques to numerical simulation data. Simulation
results show that the controller is capable of giving desired
closed-loop nominal and robust performance in the presence of
aerodynamic uncertainties.

I. INTRODUCTION
Nowadays most modern fighter aircraft are designed statically relaxed stable or even unstable in certain modes to
allow for extreme maneuverability. As a result these aircraft
have to be equipped with a stability and control augmentation system (SCAS) that artificially stabilizes the aircraft
and provides the pilot with desirable flying and handling
qualities. Briefly stated, the flying and handling qualities of
an aircraft are those properties which describe the ease and
effectiveness with which it responds to pilot commands in
the execution of a flight task [1]. Flying qualities can be seen
as being task related, while handling qualities are response
related.
SCAS design is traditionally based on linear control theory
and well-established gain-scheduling methods. The lack of
flexibility and the time consuming process of designing gainscheduled linear controllers has led to a lot of research into
nonlinear control techniques over the last decades. The most
popular of these techniques are based on dynamic inversion
[2]. Dynamic inversion based controllers cancel the real
aircraft dynamics using an accurate onboard aircraft model
and replace them with desired dynamics over the entire flight
envelope.
The main drawback of dynamic inversion is the need for
an accurate onboard aircraft model, since model mismatch
*Corresponding author. Email: [email protected]

978-1-4244-3872-3/09/$25.00 2009 IEEE

will require a more robust outer loop, often resulting in a

more conservative controller. An aircraft aerodynamic model
is usually acquired from wind tunnel experiments and computational fluid dynamics calculations, followed by a very
costly and extensive flight testing program. Furthermore, in
the case of (battle) damage or an actuator failure the real
aircraft dynamics can undergo a large, sudden change leading
to a critical onboard model mismatch.
In this paper the constrained adaptive backstepping approach
[3], [4] is used to design a SCAS for a nonlinear, highfidelity F-16 model which satisfies the handling qualities
requirements [5] across the entire flight envelope of the
model. It is assumed that the aerodynamic force and moment
functions of the model are not known exactly and that they
can change during flight due to structural damage or control
surface failures. There is plenty of literature available on
adaptive backstepping designs for the control of aircraft and
missiles, see e.g. [6], [7]. However, none of these publications considers the flying qualities during the controller
design phase or performs a handling qualities evaluation
after the design is finished. An exception is [8], where
a longitudinal adaptive backstepping controller is designed
for a simplified supersonic aircraft model. The controller
parameters are tuned explicitly via short period handling
qualities specifications [5]. The work in this paper considers
a full six degrees-of-freedom high-fidelity aircraft model and
enforces the handling qualities requirements with command
filters during the control design process.
The outline of this paper is as follows: First, the flight dynamics of the high-fidelity F-16 model are briefly discussed
in Section II. In Section III the flight control law for the F-16
model is derived and the identification process with the Bspline neural networks is explained. Numerical simulations
are performed and handling qualities are evaluated in Section
V. Finally, Section VI contains the conclusions.
II. AIRCRAFT MODEL DESCRIPTION
The aircraft model used in this study is that of an F16 fighter aircraft with geometry and aerodynamic data as
reported in [9]. The aerodynamic data in tabular form have
been obtained from wind tunnel tests and are valid for
subsonic speeds up to Mach 0.6 for the wide range of -20 deg
90 deg and -30 deg 30 deg. The wind tunnel
tests were conducted on sufficiently close points to capture
the nonlinear behavior of the aerodynamic force and moment

7333

FrB01.1
coefficients. Aerodynamic data for the intermediate points
are linearly interpolated. The control inputs of the model are
the elevator, ailerons, rudder and leading edge flaps, as well
as the throttle setting. The leading edge flaps are controlled
separately and will not be used for the control design. The
control surface actuators are modeled as first-order low pass
filters with rate and magnitude limits as given in [4]. The
relevant dynamic equations of motions used for the control
design can be written as [10]:
VT

1
(D + FT cos cos + mg1)
m
(L FT sin + mg3)
qs ps tan +
mVT cos
(Y FT cos sin + mg2)
rs +
mVT

p =
q =
r =

(c1 r + c2 p) q + c3L + c4 (N + he q)


c5 pr c6 p2 r2 + c7 (M he r)
(c8 p c2r) q + c4L + c9 (N + he q)

(1)
(2)
(3)

(4)
(5)
(6)

where
VT

p, q, r
ps , qs , rs
L,Y, D
M,
N
L,
FT
he
ci , i = 1, ..., 9
g1 , g2 , g3

total airspeed (m/s)

angle of attack (rad)
sideslip angle (rad)
body-axis angular rates (rad/s)
stability-axis angular rates (rad/s)
lift, side and drag force components (N)
rolling, pitching
and yawing moments (N.m)
engine thrust force (N)
engine angular momentum
about x-body-axis (kg.m2/s)
moment of inertia terms (-)
gravity components
in wind-axis reference frame (m/s2 )

qS
c Cm ( , , e ) + CZT xcgr xcg

III. F LIGHT C ONTROL D ESIGN

The goal of this study is design a SCAS that tracks
pilot commands with responses that satisfy the handling
qualities, across the entire flight envelope of the aircraft,
in the presence of uncertain aerodynamic parameters. The
pilot commands should control the responses as follows:
0 ,
longitudinal stick deflection commands angle of attack com
lateral stick deflection commands stability-axis roll rate
0 . The
p0s,com and the pedals command the sideslip angle com
0
total velocity command VT,com is achieved with the total
engine thrust FT , which is in turn controlled with the throttle
lever deflection. The commanded signals are fed through
command filters to produce the signals com , com , ps,com ,
VT,com and their derivatives. The command filters are also
used for specifying the desired aircraft handling qualities.
The SCAS design is based on the constrained adaptive
backstepping approach which is detailed in [3], [4].
A. Outer Loop Design
The control design procedure starts by defining the new
tracking error states as

VT
VT,com
Z1 = com = X1 X1,com
(8)

com

ps,com
ps
(9)
Z2 = qs qs,des = X2 X2,com ,
rs,des
rs

The transformation matrices from body axes to stability

axes Ts/b and from body to wind axes Tw/b are specified
in [10]. The thrust model of [9] is implemented, which
calculates the thrust as a function of altitude, Mach number
and throttle setting t . This model is given in tabular form.
M,
N
The total aerodynamic forces L,Y, D and moments L,
are summations of the various aerodynamic contributions
stored in lookup tables. As an example the total pitch
moment M is given by
h

where q denotes the dynamic pressure, S the wing area and c

the mean aerodynamic chord. Other aerodynamic forces and
moments are given in similar form, a detailed overview can
be found in [9].

with qs,des and rs,des the intermediate control laws that will
be defined by the adaptive backstepping controller. The time
derivative of Z1 can be written as

FT
(10)
Z 1 = A1 F1 + H1 + B11X2 + B12 0 X1,com
0

where
A1

H1


LEF 
(7)
CmLEF ( ) 1
25

h
i
qc
LEF
+
Cmq ( ) + CmqLEF ( ) 1
2VT
25
!

Cm ( ) + Cmds ( , e ) ,

B11

0
1
cos1
mV
0

0
0
1

V
0 ,
0

mg1
1 p tan + FT sin +mg3
s

,
VT cos
m
FT cos sin +mg2

0 0
0 1
0 0

VT

0
cos cos
0 , B12 =
0
1
0

0 0
0 0 ,
0 0

are known (matrix) functions, and F1 = [L,Y, D]T is a vector

containing the uncertain aerodynamic forces. Furthermore,

7334

FrB01.1
let

FT0


q0s,des=B1 C1 Z1 K1 1 A1 F1 H1 +X1,com B11 2 ,
1
0
rs,des

(11)

Rt

where B1 = B11 + B12 and 1 = 0 Z 1 (t)dt be a feedback

control law with C1 = C1T > 0, K1 = K1T 0, F1 the estimate
of F1 , 2 and Z1 to be defined later. The estimate of the
aerodynamic forces F1 is defined as
F1

F
TF1 (X,U)
1

TF1 (X,U)F1

(13)

F =
This means the estimation error can be defined as
1

F1 F1 . Part of the feedback control law (11) is now fed

through second order low pass filters to produce the signals
FT , qs,des , rs,des and their derivatives. These filters can also
be used to enforce rate and magnitude limits on the signals,
see the appendix of [3]. The effect that the use of these
command filters has on the tracking errors can be captured
with the stable linear filter

FT FT0


0
. (14)
0
+ B12
1 = C1 1 + B11 X2,com X2,com
0
Define the modified tracking errors as
Z i = Zi i ,

i = 1, 2.

(15)

B. Inner Loop Design

Taking the derivative of Z2 results in
Z2 = A2 (F2 + G2U) + H2 X 2,com

(16)

where U = [e , a , r ]T ,

c3 0 c4
A2 = Ts/b 0 c7 0 ,
c4 0 c9

(c1 r + c2 p) q +
c4he q
rs
H2 = 0 + Ts/b c5 pr c6 p2 r2 c7 he r ,
ps
(c8 p c2r) q + c9he q

are known (matrix) functions, and

L 0
L e

M0 , G2 = M e
F2 =
N 0
N e

L a
M a
N a

A2 G 2U 0 = C2 Z2 K2 2 BT11 Z 1 A2 F2 H2 +X2,com , (18)

where 2 = 0t Z2 (t)dt with C2 = C2T > 0, K2 = K2T 0 and
where F2 and G 2 are the estimates of the unknown nonlinear
aerodynamic moment functions F2 and G2 , respectively. The
estimates are defined as
R

F2

G2i

(12)

F is a
where TF1 is the known regressor function and
1
vector with unknown constant parameters. It is assumed that
there exists a vector F1 such that
F1

where the higher order control surface dependencies are

still contained in M 0 (X,U). To stabilize the system (16) the
desired control U 0 is defined as

(19)
(20)

where TF2 , TG2i are the known regressor functions and

G are vectors with unknown constant parameters, also
F ,

2
2i
note that G 2i represents the ith column of G 2 . It is assumed
that there exist vectors F2 , G2i such that
F2

TF2 (X,U)F2

G2i

TG2i (X)G2i .

(21)

F =
This means the estimation errors can be defined as
2

F2 F2 and G2i = G2i G2i . The actual control U is

found by again applying command filters as in the outer loop
design. Finally, with the definition of the stable linear filter


2 = C2 2 + A1 G 2 U U 0 ,
(22)
the static part of the control design is finished.
C. Update Laws and Stability Properties
In this section the stability properties of the control law
are discussed and dynamic update laws for the unknown
parameters are derived. Define the control Lyapunov function



1
1 2
F
TF 1
trace
V = Z Tj Z j + Tj K j j +
1
1 F1
2 j=1
2
!


 3

G
TG 1
F + trace
TF 1
+ trace
2

F2

2i

i=1

G2i

2i

with the update gains matrices F1 = TF1 > 0, F2 = TF2 > 0

and G2i = TG2i > 0. Selecting the update laws

F1

G2i

F1 F1 AT1 Z1
F2 F2 AT2 Z2

ProjG2i G2i G2i AT2 Z 2Ui

(23)

and substituting (11), (14), (22) and (18) reduces the derivative of V along the trajectories of the closed-loop system
to

L r
M r ,
N r

are unknown (matrix) functions containing the aerodynamic

moment components. Note that for a more convenient presentation the aerodynamic moments have been decomposed,
e.g.

M(X,U)
= M 0 (X,U) + M e e + M a a + M r r

F
TF2 (X,U)
2
G
for i = 1, 2, 3
TG2i (X)
2i

(17)

Z 1T C1 Z 1 Z 2T C2 Z 2 ,

(24)

which is negative definite. By using Barbalats lemma [11] it

can be shown that Z 0 as t . When the command filters
are properly designed and the limits on the filters are not in
effect, Zi will converge to the close neighborhood of Zi . If the
limits are in effect the actual tracking errors Zi may increase,
but the modified tracking errors Z i will still converge to zero

7335

FrB01.1
and the update laws will not unlearn, since they are driven by
the modified tracking error definitions. Note that the update
laws for G 2 include a projection operator [12] to ensure that
certain elements of the matrix do not change sign and full
rank is maintained always. For most elements the sign is
known based on physical principles.
IV. M ODEL I DENTIFICATION
To simplify the approximation of the unknown aerodynamic force and moment functions, and thereby reducing
computational load to make real-time implementation feasible, the flight envelope is partitioned into multiple, connecting operating regions called hyperboxes or clusters. This
is done manually using a priori knowledge of the nonlinearities of the system. In each hyperbox a locally valid linearin-the-parameters nonlinear model is defined, which can be
estimated using the update laws of the Lyapunov based
control laws. The aerodynamic model can be partitioned
using different state variables, the choice of which depends
on the expected nonlinearities of the system. In this study
B-spline neural networks are used to interpolate between the
local nonlinear models ensuring smooth transitions.
In the previous section parameter update laws (23) for
the unknown aerodynamic functions (12)-(20) were defined.
Now these unknown vectors and known regressor vectors
will be further specified. The total force and moment approximations are written in the standard coefficient notation,
see e.g. (7). The total nonlinear function approximations
are divided into simpler linear-in-the-parameter nonlinear
coefficient approximations, e.g.
CL0 ( , ) = CTL ( , )CL0 ,
0

(25)

where the unknown parameter vector CL0 contains the

network weights, i.e. the unknown parameters, and CL0 is
a regressor vector containing the B-spline basis functions
[4]. All other coefficient estimates are defined in similar
fashion. In this case a two-dimensional network is used with
input nodes for and . Different scheduling parameters
can be selected for each unknown coefficient. In this study
third order B-splines spaced 2.5 degrees and up to three
scheduling variables, , , e , depending on coefficient are
used. With these approximators sufficient model accuracy
was obtained. Following the notation of (25) the estimates
of the aerodynamic forces and moments can be written as
L , L = T ( , , e )
L,
L = TL ( , , e )

L
T
T

Y = Y ( , , e )Y , M = ( , , e )M,

D =

D , N
TD ( , , e )

M
N ,
= TN ( , , e )

(26)

which is a notation equivalent to the one used in (12)-(20).

Therefore, the update laws (23) can be used adapt the Bspline network weights.
V. NUMERICAL SIMULATIONS AND ANALYSIS
This section presents the simulation results from the application of the controller developed in the previous sections
to the high-fidelity, six-degrees-of-freedom F-16 model of

Section II. Both the control law and the aircraft model are
c
written as C S-functions in MATLAB/Simulink
. Sensor
models and transport delays of 20 ms have been added to the
controller to simulate an onboard computer implementation
of the control laws. First however, the controller tuning
and the design of the command filters that enforce desired
handling qualities based on MIL-STD-1797B [5] will be
discussed. Nonlinear simulations of the closed-loop system
have been performed and the small amplitude responses have
been matched to low order equivalent systems (LOES). It
is then verified if the LOES parameters meet all desired
conventional handling quality criteria as expected. Finally,
several maneuvers at a wide range of flight conditions have
been performed with uncertainties in the aerodynamics to
verify the robustness of the control law.
A. Controller Tuning and Command Filter Design
The goal of the control law is to provide the pilot with
Level I handling qualities throughout the whole flight envelope of the aircraft model: altitudes from 0 to 10000 m and
Mach numbers from 0.2 to 0.8. The filters used to convert
the commands of the pilot into smooth reference signals for
the control law also serve as reference models.
1
ps,com
=
,
ps,com,0
Tp s + 1

com
com,0

com
2
= 2
com,0
s + 2 s + 2
2

,
2
s + 2 s + 2

where Tp = 0.5, = = 0.8, = 1.25 and is a linear

function of the dynamic pressure q with value 2.5 for low
q and 6.5 for high q.
The controller gains are selected as
C1 = 0.5I, C2 = I and the integral gains as

0.2 0
0
0.5 0 0
K1 = 0 0.2 0 , K2 = 0 0 0
0
0 0.2
0 0 0

Note that Lyapunov stability theory only requires the control

gains to be larger than zero, but it is natural to select
the gains of the inner loop largest. The update laws (23)
are normalized and robustified against parameter drift with
continuous dead-zones and leakage terms [12]. The update
gains are all selected positive definite and tuned in a trialand-error procedure. The size of the selected update gains
varies with the tracking errors, larger tracking errors result
in bigger update gains to ensure more aggressive parameter
approximation in case of sudden failures.
B. Handling Qualities Analysis
Low-order equivalent systems analysis of frequency responses obtained from frequency sweeps (0.212 rad/s)
performed at twenty flight conditions over the entire operating range were used as the primary means to verify the
handling qualities. The flight conditions used for verification
are shown in Figure 1. The transforming of the time history
data from the sweeps into the frequency domain and the
transfer function fitting was done with the commercially
c
available software package CIFER
. Good fitting results

7336

FrB01.1
Category A Flight Phases

10

(rad/s)

10

sp

Level 2

were achieved at all test flight conditions. The following

LOES are considered:
i
h
K s s2 + 2 s + 2 e p s
p


=
roll
(s + 1/Ts) (s + 1/Tr ) s2 + 2d d s + d2




K s s + 1/T1 s + 1/T2 eq s
q




=
2
pitch
s2 + 2 p p s + p2 s2 + 2spsp s + sp






A s + 1/T1 s + 1/T2 s + 1/T3 e s


 .
=
yaw
(s + 1/Ts) (s + 1/Tr ) s2 + 2d d s + d2

Level 1
0

10

Level 2

Level 3
1

10

For level 1 handling qualities the LOES parameters must

satisfy the ranges

10

Fig. 2.

0.28 CAP 3.6,

Tr 1.0 s
,
sp > 1.0 rad/s,
d 0.4
0.35 sp 1.3, d d 0.4 rad/s

10
nz/a

10

LOES short period frequency estimates.

Pitch Attitude Bandwidth vs. Phase Delay Criterion

Phase Delay p (sec)

0.25

Cat. C

0.2

Cat. A
0.15
Level 2/3

Cat. A

0.1

Cat. C

0.05

Level 1/2

2
3
Pitch Attitude Bandwidth

BW

Roll Mode Time Constant (sec)

Fig. 3.

4
(rad/s)

Pitch attitude bandwidth vs.phase delay.

Roll Requirements

2
Level 3
1.5
Level 2
1
0.5
Level 1
0

6
4

0.25

Level 3

0.2
Level 2

0.15
0.1
0.05
0

C. Nominal Performance Analysis

Level 1
0

Fig. 4.

6
4

Dynamic Pressure (N/m )

The first series of simulations will test the control law on

the nominal F-16 model. The B-spline networks are initialized with the correct values. The simulations are performed
at the evaluation flight conditions defined in Figure 1. The

x 10
Effective Time Delay (sec)

2 /(n / ) is the control anticipation paramwhere CAP = sp

z
eter and the equivalent time delays i must be less than 0.10
seconds. Guidelines for estimating the substantial number
of parameters in the LOES transfer functions are given in
[5], [13]. For the longitudinal response the pitch attitude
bandwidth versus phase delay criterion [5] is also taken into
account as recommended by [14]. Plots of the CAP versus
the short period frequency sp can be found in Figure 2,
while the bandwidth criterion plot appears in Figure 3. It can
be seen that both criteria predict level 1 handling qualities.
Short period damping sp values were between 0.60 and 0.82,
while the largest effective time delay was 0.084 s. The gain
margin was larger than 6 dB and the phase margin larger than
45 deg at all test conditions. Finally, the Neal-Smith criterion
[15] also predicts level 1 handling qualities. The NealSmith method estimates the amount of pilot compensation
required to prevent pilot-in-the-loop resonance. Plots of
the LOES roll mode time constant and effective time delay
requirements can be found in Figure 4 and the LOES Dutch
roll frequency d and damping d requirements in Figure 5.
The figures demonstrate that also for lateral maneuvering all
criteria for level 1 handling qualities are met.

x 10

Roll mode time constant and effective time delay.

Dutch Roll Data
6

d (rad/s)

2
Level 1
1
Level 3
0

0.2

Level 2

0.4

0.6

0.8

Fig. 1.

Fig. 5.

Flight conditions for handling qualities analysis.

7337

Dutch roll frequency vs. damping.

FrB01.1

trim (deg)

r (deg)

5
0
5
0

10

15

e (deg)

10
nominal
nonadaptive
adaptive

10

20

a (deg)

0
5

10

15

r (deg)

(deg)

2
4
5

10

15

20

10

15

20

10

15

20

10
time (s)

15

20

10
time(s)

15

20

20
0

20
asymm weights

symm weights

15

20

2
0

10

10
0

20

6
0

10

100

p (deg/s)

200

100
0

0
10

0
0.05
0.1
0.15
0

1
0
5

0
5

15
0

10

a (deg)

ps (deg/s)

Aircraft responses at Mach 0.6 and 5000 m altitude.

10

10

10
0

10

for a nonlinear high-fidelity F-16 model over a wide flight

envelope. The dynamic update laws of the controller approximate the aerodynamic coefficients of the onboard aircraft
model in case of uncertainties and store them in B-spline
networks at each flight condition. The update laws have been
made robust against actuator saturation and parameter drift.
The proposed controller is relatively easy to construct and
requires a less accurate onboard aircraft model compared to
standard dynamic inversion designs. The implementation in
a fixed-base simulator for further evaluation with test pilots
is the next step in this ongoing research project.

0.2

10

5
time (s)

10

r (deg)

20

0.2
0

20

20
0

Fig. 7.

10

20

(deg)

15

Fig. 6.

trim (deg)

results of a coupled maneuver are shown in Figure 6. A 2

deg angle of attack command lasting 4 seconds is given at 1
second simulation time, followed by a 20 deg/s roll doublet
command lasting 4 seconds at 2 seconds simulation time.
The figure demonstrates that tracking performance is good
and consistent over the entire subsonic flight envelope. It can
also be seen that the effectiveness of the control surfaces
varies a lot with the airspeed and altitude, but the actuators
are never saturated during this maneuver.

10
0
10

5
time (s)

10

Nominal aircraft response at all test flight conditions.

D. Robust Performance Analysis

To analyze the robustness of the controller a number
of maneuvers with 30% uncertainty in all aerodynamic
coefficients, i.e. all stability and control derivatives, are
performed. Figure 7 shows the aircraft responses for one
of these simulations with an initial flight condition at Mach
number 0.6 and an altitude of 5000 meters. The solid lines
represent the nominal performance with correct onboard
model, the dash-dotted lines represent the responses with
all stability and control derivatives of the onboard model
initialized 30% too small and adaptation turned off, and
finally the dashed lines are of the simulation with 30%
uncertainty and adaptation on. The bottom plots of Figure
7 are the B-spline network weights for the longitudinal and
lateral aerodynamic coefficients. As can been seen from the
figures, the adaptive control law manages to restore tracking
performance to an acceptable level within a short amount
of time. Due to the varying update law gains the initial
adaptation to the tracking errors is quite fast, and then
gradually becomes gradually more slowly as the tracking
errors converge to zero. The weight plots show that the
weights converge to constant values, some lateral weights are
still slowly adapting, since there is still a small tracking
error later in the simulation. It can also be observed that not
all weights are adapting at the same time, this illustrates the
local character of the approximation process.
VI. CONCLUSIONS AND FUTURE WORK
In this study, a control and stability augmentation system
for a modern fighter has been developed based on adaptive
backstepping techniques. It is demonstrated in numerical
simulations that desirable handling qualities are achieved

R EFERENCES
[1] M. V. Cook, Flight Dynamics Principles. Butterworth-Heinemann,
1997.
[2] J. J. E. Slotine and W. Li, Applied Nonlinear Control. Prentice Hall,
1991.
[3] J. Farrell, M. Sharma, and M. Polycarpou, Backstepping based
flight control with adaptive function approximation, AIAA Journal of
Guidance, Control and Dynamics, vol. 28, pp. 10891102, Jan. 2005.
[4] L. Sonneveldt, Q. P. Chu, and J. A. Mulder, Nonlinear flight control
design using constrained adaptive backstepping, Journal of Guidance,
Control, and Dynamics, vol. 30, pp. 322336, Mar-Apr 2007.
[5] Military Standard, Flying Qualities of Piloted Aircraft, MIL-STD1797B, 2006, 2006.
[6] S. N. Singh and M. Steinberg, Adaptive control of feedback linearizable nonlinear systems with application to flight control, in AIAA
Guidance, Navigation and Control Conference, July 1996.
[7] K. S. Kim, K. J. Lee, and Y. Kim, Reconfigurable flight control
system design using direct adaptive method, Journal of Guidance,
Control and Dynamics, vol. 26, pp. 543550, July-Aug. 2003.
[8] H. S. Ju and C. C. Tsai, Longitudinal axis flight control law design
by adaptive backstepping, in IEEE Transactions on Aerospace and
Electtronic Systems, 2007.
[9] L. T. Nguyen, M. E. Ogburn, W. P. Gilbert, K. S. Kibler, P. W. Brown,
and P. L. Deal, Simulator study of stall post-stall characteristics of
a fighter airplane with relaxed longitudinal static stability, tech. rep.,
NASA Langley Research Center, 1979.
[10] B. L. Lewis and F. L. Stevens, Aircraft Control and Simulation, pp. 1
54,110115. John Wiley & Sons, 1992.
[11] M. Krstic, I. Kanellakopoulos, and P. Kokotovic, Nonlinear and
Adaptive Control Design. John Wiley & Sons, 1995.
[12] P. A. Ioannou and J. Sun, Stable and Robust Adaptive Control,
pp. 555575. Prentice-Hall, 1995.
[13] T. J. Curry, Estimation of handling qualities parameters of the tu-144
supersonic transport aircraft from flight test data, tech. rep., NASA
CR-2000210290, August 2000.
[14] M. B. Tischler, Advances in Aircraft Flight Control. Taylor & Francis,
1996.
[15] R. E. Bailey and R. E. Smith, Analysis of augmented aircraft flying
qualities through application of the neal-smith criterion, in Guidance
and Control Conference, no. AIAA 81-1776, 1981.

7338