Professional Documents
Culture Documents
1
Itzhak Barkana
Am
A
the plant behavior would asymptotically repro-
duce the stable model behavior and would ul-
timately achieve the desired performance repre-
sented by the ideal Model Reference. In partic-
ular, the Lyapunov stability technique revealed
the prior conditions that had to be satisfied in
Fig. 1. Schematic representation of MRAC order to guarantee stability and allowed getting
rigorous proofs of stability of the adaptive control
would follow the model exactly. If, however, not system. Because along with the dynamics of the
all gains are correct, the measured plant output state or the state error, adaptive control systems
differs from the output of the Model Reference. have also introduced the adaptive gains dynamics,
The resulting “tracking error” the positive definite quadratic Lyapunov function
had to contain both the errors and the adaptive
ey (t) = ym (t) − y(t) (8)
gains and usually had the form
Positive Realness or, more precise, Passivity of is SPR, namely, it satisfy the passivity conditions
systems is a useful systems property that has (16)-(17). Because in this case the original system
been first introduced in networks (Cauer, 1958), (1)-(2) was only separated by a simple constant
and probably introduced to dynamic systems by output feedback from strict passivity, it was called
Kalman (Kalman, 1964) in the context of opti- “Almost Strictly Positive Real (ASPR)” or “Al-
mality. It has also shown its usefulness in the most Strictly Passive (ASP)” (Barkana and Kauf-
context of “absolute stability” (Popov, 1962). As man, 1985), (Barkana, 1987). Note that such ASP
we already mentioned, Positive Realness has also systems are sometimes called (Fradkov, 2003),
been shown to be useful for the proof of stability (FradkovHill, 1998) “feedback passive” or “passi-
with adaptive controllers. Here, we present the fiable.” However, as we will show that any stabiliz-
state-space representation of the SPR conditions able system is also passifiable via parallel feedfor-
which seems to be the most useful for successful ward, those systems that are only at the distance
proofs of stability using Lyapunov stability theory. of a constant feedback gain from Strict Passivity
deserve a special name.
Definition 1. A linear time-invariant system with
a state-space realization {A, B, C}, where A ∈ At the time, this “mitigation” of the passivity con-
Rn∗n , B ∈ Rn∗m , C ∈ Rm∗n , with the m*m ditions did not make a great impression, because
transfer function T (s) = C(sI − A)−1 B, is called it was still not clear what systems would satisfy
“strictly passive (SP)” and its transfer function the new conditions. (Some even claimed that if
“strictly positive real (SPR)” if there exist two SPR seemed to be another name for the void class
positive definite symmetric (PDS) matrices, P of systems, the “new” class of ASPR was only
and Q, such that the following two relations are adding the boundary.) Nonetheless, some ideas
simultaneously satisfied: were available. Because a constant output gain
feedback was supposed to stabilize the system, it
seemed apparent that the original plant was not
required to be stable. Also, because it was known
P A + AT P = −Q (16)
that SPR systems were minimum-phase and that
P B = CT (17) the product CB is Positive Definite Symmetric
(PDS), it was intuitive to assume that minimum-
The relation between the strict passivity condi- phase systems with Positive Definite Symmet-
tions (16)-(17) and the strict positive realness ric CB were natural ASPR candidates (Barkana
of the corresponding transfer function has been and Kaufman, 1985). Indeed, simple Root-locus
treated elsewhere (Ioannou and Tao, 1987), (Wen, techniques were sufficient to proof this result in
1988). Relation (16) is the common algebraic Lya- SISO systems, and many examples of minimum-
punov equation and shows that an SPR system phase MIMO systems with CB product PDS were
is asymptotically stable. One can also show that shown to be ASPR (Barkana and Kaufman, 1985),
conditions (16)-(17) also imply that the system (Barkana, 1987). However, it was not clear how
is strictly minimum-phase, yet simultaneous satis- many of such MIMO system actually were ASPR.
faction of both conditions (16)-(17) is far from be- Because the ASPR property can be stated as a
ing guaranteed even in stable and minimum-phase simple condition and because it is the main con-
dition needed to guarantee stability with adaptive the time it seemed to be absolutely necessary for
controllers, it is useful to present here the ASPR the ASPR conditions, the required CB symmetry
theorem for the general multi-input-multi-output proved to be rather difficult to fulfill in practice, in
systems: particular in adaptive control systems where the
plant parameters are not known.
Theorem 1. Any linear time-invariant system with
After many attempts that have ended in failure,
the state-space realization {A, B, C}, where A ∈
a recent publication has managed to eliminate
Rn∗n B ∈ Rn∗m ,C ∈ Rm∗n , with the m*m
the need for a symmetric CB. First, it was easy
transfer function T (s) = C(sI − A)−1 B, that is
to observe that the Lyapunov function remains
minimum-phase and where the matrical product
positive definite if the gain term is rewritten as
CB is PDS, is “almost strictly passive (ASP)” and
follows:
its transfer function “almost strictly positive real
(ASPR).”
V (t) = eTx (t)P ex (t) (21)
· ³ ´ ³ ´T ¸
Although the original plant is not SPR, a (ficti- + tr S K(t) − K e Γ−1 K(t) − K
e ST
tious) closed-loop system satisfies the SPR condi-
tions, or in other words, there exist two positive
Here, S is any nonsingular matrix. This new for-
definite symmetric (PDS) matrices, P and Q, and
mulation allowed the extension of useful passivity-
a positive definite gain such that the following two
like properties to a new class of systems that was
relations are simultaneously satisfied:
called W-ASPR, where W = S T S, through the
following definition:
e e C) + (A − B K
P (A − B K e e C)T P = −Q (19)
Definition 2. Any linear time-invariant system
P B = CT (20)
with state-space realization {A, B, C}, where
A ∈ Rn∗n , B ∈ Rn∗m , C ∈ Rm∗n , with the
As a matter of fact, a proof of Theorem 1 had been m*m transfer function T (s) = C(sI − A)−1 B, is
available in the Russian literature (Fradkov, 1976) called “W-almost strictly passive (WASP)” and
since 1976 yet it was not known in the West. Here, its transfer function “W-almost strictly positive
many other works have later independently redis- real (WASPR),” if there exist three positive defi-
covered, reformulated, and further developed the nite symmetric (PDS) matrices, P, Q, and W, and
idea (see (Barkana, 2004) and references therein e e such that the following
a positive definite gain K
for a brief history and for a simple and direct, al- two conditions are simultaneously satisfied:
gebraic, proof of this important statement). Even
as late as 1999, this simple ASPR condition was
still presented as some algebraic condition (Huang
et al., 1999) that might look obscure to the con- e e C) + (A − B K
P (A − B K e e C)T P = −Q (22)
trol practitioner. On the other hand, (Huang et
al., 1999) managed to add an important con- P B = CT W T (23)
tribution and emphasize the special property of
ASPR systems by proving that if a system cannot This new definition can be used with the following
be made SPR via constant output feedback, no theorem (Barkana, Teixeira and Hsu, 2006):
dynamic feedback can render it SPR.
Theorem 2. Any minimum-phase LTI system with
Theorem 1 has thus managed to explain the rather
a state-space realization {A, B, C}, where A ∈
obscure passivity conditions with the help of new
Rn∗n , B ∈ Rn∗m , C ∈ Rm∗n , with the m*m
conditions that could be understood by control
transfer function T (s) = C(sI −A)−1 B, where the
practitioners. It is useful to notice an important
positive definite and not necessarily symmetric
property that may makes an ASPR system to be
matrical product CB is diagonalizable is WASP
a good candidate for stable adaptive control: if a
in accord with Definition 2.
plant is minimum-phase and its input-output ma-
trical product CB is Positive Definite Symmetric
(PDS) it is stabilizable via some static Positive Thus, (Barkana, Teixeira and Hsu, 2006) had
Definite (PD) output feedback. Furthermore, if managed to mitigate a result that has been around
the output feedback gain is increased beyond some for more than 40 years. Nevertheless, it was very
minimal value, the system remains stable even if tempting to try to eliminate any restriction be-
the gain increase is nonstationary. The required sides the positivity of CB. This new result was ”al-
positivity of the product CB could be expected, most” made possible by observing that, although
as it seemed to be a generalization of the sign of the product of two PD matrices is not necessarily
the transfer function that allows using negative PD, the trace of the product is PD if at least
feedback in SISO systems. However, although at one of the two matrices is PDS. Therefore, the
Lyapunov function remains positive definite if the 5. SIMPLE ADAPTIVE CONTROL (SAC), OR
second term in it is again rewritten as follows: THE SIMPLIFIED APPROACH TO MODEL
REFERENCE ADAPTIVE CONTROL
V (t) = eTx·(t)P ex (t)
³ ´ ³ ´T ¸ (24)
+ tr W K(t) − K e Γ−1 K(t) − K
e Various kinds of additional prior knowledge have
been used and many solutions and additions have
been proposed to overcome some of the various
even if W is only positive definite yet not necessar-
drawbacks of the basic MRAC algorithm. How-
ily symmetric. However, we will show that in order
ever, this paper sticks to the very basic idea of
to allow the Lyapunov derivative to be negative
Model Following. Next sections will show that
definite or semidefinite, the proof of stability does
those basically ingenious adaptive control ideas
require the symmetry of W.
and the systematic stability analysis they intro-
As we showed above, new developments have duced had finally led to adaptive control sys-
simplified the (sufficient?) conditions that, along tems that can guarantee stability robustness along
with limits on the “unmodeled dynamics” and with superior performance when compared with
with “sufficient excitation,” would be sufficient to alternative, non-adaptive, methodologies. In this
allow rigorous and successful proofs of stability section we will first assume that at least one of
with adaptive controllers. Still, it appears that the passivity conditions presented above holds
the ASPR condition was not sufficient to always and will deal with a particular methodology that
guarantee the stability of MRAC. Besides, al- managed to eliminate the need for the plant order
though successful proofs of stability usually ended and therefore can mitigate the problems related
showing that the following errors vanish asymp- to “unmodeled dynamics” and “persistent excita-
totically, it is rather commonly accepted that the tion.” Subsequent sections will then extend the
adaptive gains may not converge to any specific feasibility of the methodology to those real-world
limit at all, even if they are guaranteed to be systems that do not inherently satisfy the passiv-
bounded. Furthermore, some recent counterexam- ity conditions.
ples seem to show that MRAC systems may di-
The beginning of the alternative adaptive con-
verge even when all previously assumed sufficient
trol approach can be found in the intense activ-
conditions are satisfied (Hsu and Costa, 1999).
ities at Rensselaer (RPI) during 1978-1983. At
that time, such researchers as Kaufman, Sobel,
4. COUNTEREXAMPLES TO MODEL Barkana, Balas, Wen, and others (Sobel, Kauf-
REFERENCE ADAPTIVE CONTROL man and Mabus, 1982), (Kaufman et al., 1981),
(Barkana and Kaufman, 1982), (Barkana, Kauf-
In the examples of (Hsu and Costa, 1999), a 2*2 man and Balas, 1983), (Barkana, 1983), (Wen and
stable plant with CB positive definite is required Balas, 1989) were trying to use customary adap-
to follow the behavior of a stable model of same tive control techniques with large order MIMO
order. In fact both the plant and the model systems, such as planes, large flexible structures,
have the same diagonal system matrices with etc. It did not take long to realize that it was
negative eigenvalues, and only the input-output impossible to even think of controllers of the same
matrix differentiates between the two. The plant, order as the plant, or even of the order of a “nom-
that appears in a 2D adaptive robotic visual inal” plant. Besides, those were inherently MIMO
servoing with uncalibrated camera, is defined by systems, while customary MRAC techniques at
the system matrices the time were only dealing with SISO systems.
· ¸ · ¸ Because now the very reduced-order model could
−a 0 cosϕ sinϕ not be considered to be even close to the plant, one
A= ; B= (25)
0 −a −hsinϕ hcosϕ could not consider full model state following, so
· ¸ this aim was naturally replaced by output model
10
C= (26) following. Furthermore, as the (possibly unstable)
01
large-order plant state could not be compared
with the reduced-order model state, the model
It is shown (Hsu and Costa, 1999) that standard
could not be thought to guarantee asymptotic
MRAC systems become unstable even though the
stability of the plant any longer.
MRAC system was supposed to be stable because
there was no “unmodeled dynamics,” there was In order to allow stability of the reduced order
“sufficient excitation,” and the assumably “suf- adaptive control system, new adaptive control
ficient” passivity conditions were also satisfied. components that were not deemed to be needed by
We note that (Hsu and Costa, 1999) shows ways the customary MRAC had to be considered. We
to avoid the problem and, using various kinds of will show that this “small” addition had an as-
prior knowledge, other solutions have also been tonishing effect towards the successful application
proposed. of the modified MRAC. In brief, as it was known
that stability of adaptive control systems required We will show how this adaptive gain addition
that the plant be stabilizable via a constant gain is able to avoid some of the most difficult in-
feedback, the natural question was why not using herent problems related to the standard MRAC
this direct output feedback. and to add robustness to its stability. Although
it was developed as a natural compensation for
Following this idea, an additional adaptive output
the low-order models and was successfully applied
feedback term was added to the adaptive algo-
at Rensselaer as just one element of the Sim-
rithm that otherwise is very similar to the usual
ple (Model Reference) Adaptive Control method-
MRAC algorithms, namely,
ology, it is worth mentioning that, similarly to
the first proof of the ASPR property, the ori-
u(t) = Ke ey (t) + Kx xm (t) + Ku um (t) gins of this specific adaptive gain can again be
= K(t)r(t) (27) found in an early Fradkov’s work (Fradkov, 1976)
in the Russian literature. Besides, later on this
where we denote the reference vector gain received a second birth and became very
£ ¤
K(t) = Ke (t) Kx (t) Ku (t) (28) popular after 1983 in the context of adaptive
control “when the sign of high-frequency gain
Subsequently in this paper, it will be shown that is unknown.” In this context (Nussbaum, 1983),
the new approach uses the model as a Command (Morse, 1984), (Heyman, Lewis and Meyer, 1985)
Generator and therefore it is sometime called and after a very rigorous mathematical treatment
Adaptive Command Generator Tracker. Because (Byrnes and Willems, 1984), it also received a
it also uses low-order models and controllers, it new name and it is sometimes called the Byrnes-
was ultimately called Simple Adaptive Control Willems gain. Its useful properties have been thor-
(SAC). Before we discuss the differences between oughly researched and some may even call this one
the new SAC approach and to adaptive control adaptive gain Simple Adaptive Control as they
classical MRAC, it is useful to first dwell over the were apparently able to show that it can do “al-
special role of the direct output feedback term. most” everything (Ilchman, Owens and Pratzel-
If the plant parameters were known, one could Wolters, 1987), (Mareels and Polderman, 1996).
choose an appropriate gain K e e and stabilize the Indeed, if an ASPR system is high-gain stable,
plant via constant output feedback control it seems very attractive to let the adaptive gain
increase to even very high values in order to
e e y(t)
u(t) = −K (29) achieve good performance that is represented by
small tracking errors. However, although at first
As we already mentioned above, it was known thought one may find that high gains are very
that an ASPR system (or, as we now know, a attractive, a second thought and some more en-
minimum-phase plant with appropriate CB prod- gineering experience with the real world applica-
uct) could be stabilized by a positive definite tions make it clear that high gains may lead to
output feedback gain. Furthermore, it was known saturations and may excite vibrations and other
that ASPR systems are high-gain stable, so sta- disturbances. These disturbances may not have
bility of the plant is maintained if the gain value appeared in the nominal plant model that was
happens to go arbitrarily high beyond some mini- used for design and may not be felt in the real-
mal value. Whenever one may have sufficient prior world plant unless one uses those very high gains.
knowledge to assume that the plant is ASPR, yet Furthermore, as the motor or the plant dynamics
does not have sufficient knowledge to choose a would always require an input signal in order to
good control gain, one can use the output itself keep moving and tracking the desired trajectory,
to generate the adaptive gain by the rule: it is quite clear that the tracking error cannot
K̇y (t) = y(t)y T (t)Γy (30) be zero or very small unless one uses very high
gains indeed. Designers of tracking systems know
and the control that feedforward signals that come from the de-
sired trajectory can help achieving low-error or
u(t) = Ky (t)y(t) (31) even perfect tracking without requiring the use
of dangerously high gains (and, correspondingly,
In the more general case when the plant is re- exceedingly high bandwidth) in the closed loop.
quired to follow the output of the model, one In the non-adaptive world, feedforward could be
would use the tracking error to generate the adap- problematic because unlike the feedback loop, any
tive gain errors in the feedforward parameters are directly
and entirely transmitted to the output tracking
K̇e (t) = ey (t)eTy (t)Γe (32)
error. Here, the adaptive control methodology can
demonstrate an important advantage on the non-
and the control adaptive techniques, because the feedforward pa-
u(t) = Ke (t)ey (t) (33) rameters are finely tuned by the very tracking er-
flexible structures, and therefore was quite imme-
diately adopted by many researchers and practi-
tioners, the SAC approach got a cold reception
and for a long time has been largely ignored by
Ku the mainstream adaptive control. In retrospective
Kx
(besides some lack of good selling) at the time this
cold reception had some good reasons. Although
1
Um
Bm
1/s
Xm
Cm
Ke B
X it was called “simple” as it was quite simple to
implement, the theory around SAC was not simple
U 1/s C 1
Yp
ey=ym-y
0.4
Imaginary Axis
0
−0.4
V̇ (t) = −eTx (t)Qex (t) + eTx (t)P [f (x) − f (x∗ )] (46) −0.6
−0.8
0.04
0.02
1.5
1
−0.02
0.5 −0.04
−0.06
−0.08 −0.06 −0.04 −0.02 0 0.02 0.04 0.06
0
0 10 20 30 40 50 60
Time [sec]
−0.005
1 2.5
0.8
0.6 2
0.4
−0.2
1
−0.4
0.5
−0.6
−0.8
0
−1
−3.5 −3 −2.5 −2 −1.5 −1 −0.5 0
0 20 40 60 80 100 120 140 160 180 200
Real Axis
Adaptive Control Gain
Imaginary Axis
recall that a minimum-phase plant with relative 0
0.8
selected to provide a counterexample for the sta-
0.6
bility with assumably “constant” gains. Although
0.4
the stability of the augmented system with adap-
0.2
tive control is guaranteed, the plant output may
Imaginary Axis
0
not behave very well, even with the added paral-
−0.2
lel feedforward. In any case, even in those cases
−0.4
when the parallel feedforward is too large to al-
−0.6
low good performance as monitored at the actual
−0.8
plant output, the behavior of the, possibly both
−1
−3.5 −3 −2.5 −2
Real Axis
−1.5 −1 −0.5 0 unstable and non-minimum phase, plant within
the augmented system is stable and it was shown
Fig. 11. Augmented Plant with strictly causal to allow stable identification schemes (Johansson
PFC and Robertsson, 2002) and thus, lead to better un-
derstanding of the plant towards better, adaptive
augmented output. The example here is a very or non-adaptive, control design. Still, as recently
bad system and was only used to illustrate the shown with a non-minimumphase UAV example
problems one may encounter using constant gain (Barkana, 2005b) and with many other realistic
in changing environments and cannot be expected examples (Kaufman, Barkana and Sobel, 1998),
to result in good behavior without performing prior knowledge usually available for design allows
much more study and basic control design. The using basic preliminary design and then very small
examples above have been used to present a additions to the plant that not only result in
simple principle: if the system can be stabilized robust stability of the adaptive control system
by the controller H(s), then the augmented sys- even with originally non-minimum phase plants,
tem Ga (s) = G(s) + H −1 (s) is minimum-phase. but that also lead to performance that is ulti-
Proper selection of the relative degree of H −1 (s) mately superior to other control methodologies.
will thus render the augmented system ASPR A recent publication uses the parallel feedforward
(Barkana, 1987). This last statement implies that compensator for safe tuning of MIMO Adaptive
“passivability” of systems is actually dual to sta- PID Controllers (Iwai, Mizumoto, Nakashima and
bilizability. If a stabilizing controller is known, Kumon, 2007) and another (Ben Yamin, Yaesh
its inverse in parallel with the plant can make and Shaked, 2007) shows how to implement Sim-
the augmented system ASPR. When sufficient ple Adaptive Controllers with guaranteed H∞
prior knowledge is available to design a stabiliz- performance.
ing controller, some researchers prefer to use this
knowledge and directly design the corresponding
parallel feedforward (Iwai and Mizumoto, 1992) 9. ROBUSTNESS OF SIMPLE ADAPTIVE
or ”shunt” (Fradkov, 1994). When the “plant” is CONTROL WITH DISTURBANCES
a differential equation, it is easy to assume that
the order or the relative degree is available and The presentation so far showed that a simple
then a stabilizing controller or the parallel feedfor- adaptive controller can guarantee stability of any
ward can be implemented. However, in real world, system that is minimum-phase if the CB prod-
where the “plant” could be a plane, a flexible uct is Positive Definite and diagonalizable if not
structure or a ship, the available knowledge is the symmetric. In case these conditions do not inher-
result of some wind-tunnel or other experimental ently hold, basic knowledge on the stabilizability
tests that may result in some approximate fre- properties of the plant, usually known, can be
quency response or approximate modeling, suffi- used to fulfill them via Parallel Feedforward Con-
cient to allow some control design, yet in general figurations. Therefore, the proposed methodology
do not provide reliable knowledge on the order or seems to fit almost any case where asymptotically
relative degree of the real plant. On the other hand perfect output tracking is possible. However, after
(although it may very much want some adaptive we presented the eulogy of the adaptive output
control to help improving performance if it only feedback gain (32), it is about time to also present
could be trusted), the control community actually what could become its demise, if not properly
continues to control real-world systems with fixed treated. When persistent disturbances such as
controllers. Therefore, in our opinion the ques- random noise or very high frequency vibrations
tion “How can you find a stabilizing controller?” are present, perfect tracking is not possible. Even
should not be given any excessive emphasis. In when the disturbance is known and various vari-
any case, if there is sufficient prior knowledge to ations of the Internal Model Principle can be
directly design the feedforward there is definitely devised (Fradkov and Andrievsky, 2007) to filter
sufficient information to design a stabilizing con- them out, some residual tracking error may always
be present. While tracking with small final errors y ∗ (t) = Cx∗ (t) (A.3)
could be acceptable, it is clear that the adaptive
gain term (32) would, slowly but certainly, in- moves along “ideal trajectories” such that
crease without limit. Indeed, theoretically, ASPR
systems maintain stability with arbitrarily high y ∗ (t) = ym (t) (A.4)
gains and in some cases (in case of missiles, for
example) the adaptive system mission could end We assume that the underlying LTI problem is
solvable and thus, that some ideal gains Ke x and
even before problems are even observed. How-
Ke u exist (Barkana, 2005a). Because the plant
ever, allowing the build-up of high gains that do
not come in response to any actual requirement and the model can have different dimensions, the
is not acceptable, because in practice they may “following error” is defined to be the difference
lead to numerical problems and saturation effects. between the ideal and the actual plant state
However, very early we observed how the robust- ex (t) = x∗ (t) − x(t) (A.5)
ness of SAC with disturbances can be guaran-
teed by adding Ioannou’s σ-term (Ioannou and and correspondingly
Kokotovic, 1983) with the error adaptive gain that
would now be
ey (t) = ym (t) − y(t) = y ∗ (t) − y(t)
K̇e (t) = ey (t)eTy (t)Γe − σKe (t) (55) = Cx∗ (t) − Cx(t) = Cex (t) (A.6)
Finally, this new addition is literally making SAC Differentiating ex (t) gives:
an adaptive controller (see (Barkana, 2005a) and
(Kaufman, Barkana and Sobel, 1998) and ref- ėx (t) = ẋ∗ (t) − ẋ(t) (A.7)
erences therein): while the control gains always ∗ ∗
= Ax (t) + Bu (t) − Ax(t) − Bu(t)
perform a steepest descent minimization of the
tracking error, the error gain defined in (55) goes ėx (t) = Aex (t) − Bu(t) + Bu∗ (t) (A.8)
up-and-down fitting the right gain to the right
situation in accord with the changing operational
needs. ėx (t) = Aex (t) − B(Ke ey (t) + Kx xm (t) + Ku um (t))
e x xm (t) + K
+ B(K e u um (t)) (A.9)
Let the linear time-invariant plant (1)-(2) track CAx∗0 (t) ≡ 0 (B.13)
the output of the model (3)-(4). In general, in
the past we have assumed that the model uses and are solutions of the plant differential equation
step inputs to generate the desired command
(Barkana, 2005a). For the more general command ẋ∗0 (t) = Ax∗0 (t) (B.14)
following case, we assume that the command itself
is generated by an unknown input generator that result in
ẋu (t) = Au xu (t) (B.1) ẏ0∗ (t) = C ẋ∗0 (t) = CAx∗0 (t) ≡ 0. (B.15)
u(t) = Cu xu (t) (B.2)
Note that the differential equation (B.14) of the
We want to check what the ultimate adaptive supplementary term x∗0 (t) does not contain con-
control gains that perform perfect tracking could trol terms because those would be included in the
be. When the error is zero, the input control other terms in (B.10). Because CB is nonsingular
to the plant is a linear combination of available one gets from (B.4)
measures. −1
u(t) = (CB) [Cm Am xm (t) + Cm Bm um (t) − CAx∗ (t)](B.16)
u(t) = Kx xm (t) + Ku um (t) (B.3)
or
Assume that the plant moves along such “ideal
trajectories” and the nonstationary gains are such u(t) = Kx (t)xm (t) + Ku (t)um (t) (B.17)
that the plant output y ∗ (t) = Cx∗ (t) perfectly
tracks the model output, namely, ey (t) = 0, or Here
∗
Cx (t) = Cm xm (t) (B.4) −1
Kx (t) = (CB) (Cm Am − CASx (t)) (B.18)
Differentiating (B.4) gives Ku (t) = (CB)
−1
(Cm Bm − CASu (t)) (B.19)
∗
C ẋ (t) = Cm ẋm (t) (B.5)
ẋ∗ (t) = Sx (t)ẋm (t) + Ṡx (t)xm (t)
(B.20)
+Su (t)u̇m (t) + Ṡu (t)um (t) + ẋ∗0 (t)
or
∗
CAx∗ (t) + CBu(t) = Cm Am xm (t) + Cm Bm um (t)(B.6) ẋ (t) = Sx (t)Am xm (t) + Sx (t)Bm um (t)
+Ṡx (t)xm (t) + Su (t)Cu Au xu (t) (B.21)
∗
+Ṡu (t)Cu xu (t) + ẋ0 (t)
CAx∗ (t) + CBKx xm (t) + CBKu um (t)
ẋ∗ (t) = Ax∗ (t) + Bu(t) = ASx (t)xm (t)
= Cm Am xm (t) + Cm Bm um (t) (B.7)
+ASu (t)um (t) + Ax∗0 (t) (B.22)
+BKx (t)xm (t) + BKu (t)um (t)
CAx∗ (t) = [Cm Am − CBKx ] xm (t)
ASx (t)xm (t) + ASu (t)um (t) + Ax∗0 (t)
+ [Cm Bm − CBKu ] um (t) (B.8) +BKx (t)xm (t) + BKu (t)um (t)
= Sx (t)Am xm (t) + Sx (t)Bm Cu xu (t) (B.23)
We assume that CA is maximal rank and get
+Ṡx (t)xm (t) + Su (t)Cu Au xu (t)
x∗ (t) = +Ṡu (t)Cu xu (t) + ẋ∗0 (t)
³ ´−1
T T
(CA) CA (CA) [Cm Am − CBKx (t)] xm (t)
³ ´−1 (B.9)
The terms in x∗0 (t) and ẋ∗0 (t) cancel each other
T T
+ (CA) CA (CA) [Cm Bm − CBKu (t)] um (t) and we get
+ x∗0 (t)
ASx (t)xm (t) + ASu (t)Cu Au xu (t)
−1
or +B (CB) (Cm Am − CASx (t)) xm (t)
−1
+B (CB) (Cm Bm − CASu (t)) Cu Au xu (t)
x∗ (t) = Sx (t)xm (t) + Su (t)um (t) + x∗0 (t) (B.10) (B.24)
= Sx (t)Am xm (t) + Sx (t)Bm Cu xu (t)
+Ṡx (t)xm (t) + Su (t)Cu Au xu (t)
Here,
+Ṡu (t)Cu xu (t)
³ ´−1
T T
Sx (t) = (CA) CA (CA) [Cm Am − CBKx (t)](B.11)
and finally
³ ´−1
T T
Su (t) = (CA) CA (CA) [Cm Bm − CBKu (t)](B.12) Mx xm (t) + Mu xu (t) = 0 (B.25)
Here, At t = 0 one gets
in failure. There is no doubt that, in principle, per- finds the final gainn on a different
o hyper-ellipsoid
fect tracking can occur while the bounded time- e e
with the center in Kx2 , Ku2 . Therefore, for the
varying gains keep wandering across some hyper-
same adaptation process, that starts and ends
surface described, for example, by (B.50) or by
with the same values, this thinking experiment
any corresponding equation. However, although
finds the final gains located at the intersection of
such solutions for the perfect tracking exist, one
infinitely many distinct hyper-ellipsoids, so their
may still ask whether those nonstationary gains
common intersection is a point or a ”line” of
can be the ultimate values of the adaptation
Lebesque measure zero. Although this argument
process. Can the steepest descent minimization
may requires more polishing, it points to the fact
end with some ever wandering gains? As we con-
that, ultimately, the adaptive gains do converge
clude below, most certainly, not. First, although
to a limit. In some cases, the rate of convergence
it is hard to translate engineering intuition into
may be slow and simulations may show the gain
rigorous mathematics, it is ”felt” that the lack
varying for a long-long time. Hence, it is impor-
of ”richness” that the perfect following equation
tant to know the gains do not vary at random and
shows does not express the ”richness” of signals
that, even if sometimes slowly, they certainly tend
that exists during the entire process of adaptation
to reach their final bounded constant limit.
up to and until ”just before” perfect tracking. Yet,
somewhat more rigorously, the same argument
that seems to fail the Lyapunov-LaSalle approach
can now be used to redeem it. Along with equation REFERENCES
(B.50) of the ultimate hyper-surface that contains
Artstein, Z. (1977). Limiting equations and sta-
them, the ultimate adaptive gains are also located
bility of nonautonomous systems. Journal of
on the hyper-ellipsoid that corresponds to the final
Differential Equations 25, 184–202.
value of the Lyapunov function. If the initial value
Astrom, K. J. (1983). Theory and Applications
of the Lyapunov function is V(t=0) = V0, its final
of Adaptive Control - A Survey. Automatica
value is given by
19(5), 471–486.
Z∞ Astrom, K. J., and B. Wittenmark (1989).
lim V (t) = V0 − eTx (t)Qex (t) dt = Vf (B.52) Adaptive Control, Addison Wesley, Reading,
t→∞
0 Massachusetts.
Barkana, Itzhak (1987). Parallel feedforward Processing (ALCOSP 2007). Saint Peters-
and simplified adaptive control. Interna- burg, RUSSIA August, 2931, 2007.
tional Journal of Adaptive Control and Signal Bitmead, R., M. Gevers, and V. Wertz (1990).
Processing 1(2), 95–109. Adaptive Optimal Control, The Thinking
Bar-Kana, I. (1983). Direct Multivariable Adap- Man’s GPC, Prentice Hall, Englewood Cliffs,
tive Control with Application to Large Struc- New Jersey.
tural Systems, Ph.D. Dissertation, ECSE Bobtsov A., and A. Nagovitsina (2007), Adaptive
Dept., Rensselaer Polytechnic Institute, Troy, output controlof linear time-varying systems,
New York. IFAC Workshop on Adaptation and Learning
Barkana, I. (2004). Comments on ‘Design of in Control and Signal Processing (ALCOSP
Strictly Positive Real Systems Using Con- 2007). Saint Petersburg, RUSSIA August,
stant Output Feedback’. IEEE Transactions 2931, 2007.
on Automatic Control 49, 2091–2093. Broussard J., and P. Berry (1978). Command
Barkana, I. (2005a). Gain conditions and con- Generator Tracking - The Continuous Time
vergence of simple adaptive control. Interna- Case, TASC, Tech. Rep, TIM-612-1.
tional Journal of Adaptive Control and Signal Byrnes, C. I., and J. C. Willems (1984). Adaptive
Processing 19( 1), 13–40. Stabilization of Multivariable Linear Systems.
Barkana, I. (2005b) Classical and Simple Adaptive Proc. 23rd IEEE Conf. on Decision and Con-
Control for Non-Minimum Phase Autopilot trol, 1547–1577.
Design. Journal of Guidance, Control, and Cauer, W. (1958). Synthesis of Linear Communi-
Dynamics 24(4), 631–638 (also in shorter cation Networks, McGraw-Hill, New York.
form in Proceedings of the AIAA GN&C 2004, Feuer, A., and A.S. Morse (1978). Adaptive
paper AIAA2004-4868). Control of Single-Input, Single-Output Lin-
Barkana, I. (2005c) On output feedback stability ear Systems. IEEE Trans. Automatic Control
and passivity in discrete linear systems. Pro- AC-23, 557–569.
ceedings of The 16th Triennial IFAC World Fradkov, A. L. (1976). Quadratic Lyapunov Func-
Congress, Prague, July 2005. tion in the Adaptive Stabilization Problem of
Barkana, I. (2005d) Output Feedback Stability a Linear Dynamic Plant. Siberian Mathemat-
and Passivity in Nonstationary and Nonlinear ical Journal 2, 341–348.
Systems. Proceedings of The 2005 Interna- Fradkov, A. L. (1994). Adaptive stabilization
tional Conference on Control and Automation of minimal-phase vector-input objects with-
(ICCA2005) June 27-29, 2005, Budapest, out output derivative measurements. Physics-
Hungary. Doklady 39(8), 550–552
Barkana, I. (2007), Adaptive model tracking with Fradkov, A. L. (2003). Passification of non-square
mitigated passivity conditions, IFAC Work- linear systems and feedback Yakubovich -
shop on Adaptation and Learning in Control Kalman - Popov lemma. European Journal of
and Signal Processing (ALCOSP 2007). Saint Control 6, 573-582.
Petersburg, RUSSIA August, 2931, 2007. Fradkov, A. L., and B. Andrievsky (2007),
Bar-Kana, I., and H. Kaufman (1982). Model Ref- Passification-Based Adaptive Control with
erence Adaptive Control for Time-Variable implicit reference model IFAC Workshop on
Input Commands. Proceedings 1982 Confer- Adaptation and Learning in Control and Sig-
ence on Informational Sciences and Systems, nal Processing (ALCOSP 2007). Saint Peters-
Princeton, New Jersey, 208–211. burg, RUSSIA August, 2931, 2007.
Barkana, I., H., Kaufman and M. Balas (1983). Fradkov, A. L., and D. Hill (1998). Exponential
Model reference adaptive control of large feedback passivity and stabilizability of non-
structural systems. AIAA Journal of Guid- linear systems. Automatica 34(66), 697–703.
ance 6(2), 112–118. Goodwin, G. C., and K. Sin (1984). Adaptive
Barkana, I., and H. Kaufman (1985). Global Sta- Filtering, Prediction and Control, Prentice
bility and Performance of an Adaptive Con- Hall, Englewood Cliffs, New Jersey.
trol Algorithm. Int. J. Control 46(6), 1491– Goodwin, G. C., P.J. Ramadge and P.E. Caines
1505. (1980). Discrete Time Multivariable Adap-
Barkana, I., M. C.-M. Teixeira and Liu Hsu tive Control. IEEE Trans. Automatic Control
(2006). Mitigation of symmetry condition AC-25, 449–456.
from positive realness for adaptive control. Hahn, W. (1967). Stability of Motion, Springer-
AUTOMATICA 42(9), 1611–1616. Verlag, New York.
Ben Yamin, R., I., Yaesh and U. Shaked (2007), Heyman, M., J. H. Lewis and G. Meyer (1985).
Simplified adaptive control with guaranteed Remarks on the Adaptive Control of Linear
H∞ performance, IFAC Workshop on Adap- Plants with Unknown High Frequency Gain.
tation and Learning in Control and Signal Systems and Control Letters 5, 357–362.
Hsu, L., and R. Costa (1999). MIMO direct adap- Landau, I. D. (1979). Adaptive Control - The
tive control with reduced prior knowledge of Model Reference Approach, Marcel Dekker,
the high-frequency gain. Proceedings of 38th New York.
IEEE Conference on Decision and Control, LaSalle, J. (1981). Stability of non-autonomous
Phoenix, AZ, 3303–3308. systems. Nonlinear Analysis Theory Methods
Huang, C.-H., P.A. Ioannou, J. Maroulas, and and Applications 1(1), 83–90.
M.G. Safonov (1999). Design of S. IEEE Mareels, I. A Simple Selftuning Controller for Sta-
Transactions on Automatic Control44, 569– ble Invertible Systems. Systems and Control
573, 1999. Letters 4, 5–16.
Ilchman, A., D. Owens and D. Pratzel-Wolters Mareels, I., and J. W. Polderman (1996). Adaptive
(1987). Remarks on the Adaptive Control of Systems: An Introduction, Birkhauser.
Linear Plants with Unknown High Frequency Moir, T., and M. Grimble (1984) Optimal Self-
Gain. Systems and Control Letters 8, 397– Tuning Filtering, Prediction, and Smoothing
404. for Discrete Multivariable Processes. IEEE
Ioannou, P. and P. Kokotovic (1983) Adaptive Sys- Transactions on Automatic Control 29(2),
tems with Reduced Models, Springer-Verlag, 128–137.
Berlin. Monopoli, R. V. (1974). Model Reference Adap-
Ioannou, P., and J. Sun (1996). Robust Adaptive tive Control with an Augmented Error Sig-
Control, Prentice-Hall, Upper Saddle River, nal. IEEE Transactions on Automatic Con-
New Jersey. trol AC-19(5), 474–484.
Ioannou, P. A., and G. Tao (1987). Frequency Morse, A.S. Global Stability of Parameter Adap-
domain conditions for strictly positive real tive Control Systems (1980). IEEE Transac-
functions. IEEE Transactions on Automatic tions on Automatic Control AC-25, 433–439.
Control 32, 53–54. Morse, A. S. (1984). New directions in Parameter
Adaptive Control Systems. Proc. 23rd IEEE
Iwai Z., and I. Mizumoto (1992). Robust and sim-
Conf. on Decision and Control, 1566–1568.
ple adaptive control systems (1992). Interna-
Narendra, K. S., and A. Annaswami (1989). Stable
tional Journal of Control 55, 1453–1470.
Adaptive Systems, Prentice Hall, Englewood
Iwai Z., I. Mizumoto, Y. Nakashima, and M. Ku-
Cliffs, New Jersey.
mon (2007). Adaptive MIMO PID Controller
Narendra, K. S., and L.S. Valavani (1978). Adap-
with Model-based Parallel Feedforward Com-
tive Controller Design - Direct Control. IEEE
pensator (2007). IFAC Workshop on Adap-
Transactions on Automatic Control AC-23,
tation and Learning in Control and Signal
570–583.
Processing (ALCOSP 2007). Saint Peters-
Narendra, K. S. and L.S. Valavani (1979) Direct
burg, RUSSIA, August 2931, 2007.
and Indirect Model Reference Adaptive Con-
Johansson, R., and A. Robertsson (2002).
trol. Automatica 15, 653–664.
Observer-Based Strict Positive-Real (SPR)
Narendra, K. S., Y.H. Lin, and L.S. Valavani
Feedback Control System Design. Automat-
(1980). Stable Adaptive Controller Design -
ica 38(9), 1557–1564.
Part II: Proof of Stability. IEEE Transactions
Kalman, R. (1964). When is a Linear System on Automatic Control AC-25, 440–448.
Optimal? Transactions of ASME, Journal of Nussbaum, R. O. (1983) Some Remarks on a
Basic Engineering, Serries D 86, 81–90. Conjecture in Parameter Adaptive Control.
Kaufman, H., M. Balas, I. Bar-Kana, and L. Systems and Control Letters 3, 243–246, 1983.
Rao (1981). Model Reference Adaptive Con- Ortega, R., and T. Yu (1987). Theoretical Results
trol of Large Scale Systems. Proceedings 20th on Robustness of Direct Adaptive Controllers.
IEEE Conference on Decision and Control, Proc. IFAC Triennial World Conference 10,
San Diego, California, 984–989. 1–15.
Kaufman, H., I. Barkana and K. Sobel 1998. Osborn, P. V., H. P. Whitaker and A. Kezer
Direct Adaptive Control Algorithm, 2nd ed., (1961). New developments in the design of
Springer-Verlag, New York. Model reference Adaptive Control Systems.
Kreiselmayer, G., and B. Anderson (1986) Robust Inst. Aeronautical Sciences, Paper 61-39.
Model Reference Adaptive Control. IEEE Popov, V. M. (1962). Absolute Stability of Non-
Transactions on Automatic Control AC- linear Control Systems of Automatic Control.
31(2), 127–133. Automation and Remote Control 22.
Krstic, M., I. Kanellakopoulos and P. Kokotovic Sastri, S., and M. Bodson (1989). Adaptive Con-
(1995). Nonlinear and Adaptive Control De- trol: Stability, Convergence and Robustness,
sign, John Wiley, New York. Prentice-Hall, Upper Saddle River, New Jer-
Landau, I. D. (1974). A Survey of Model Refer- sey.
ence Adaptive Techniques: Theory and Appli- Sobel, K., H. Kaufman and L. Mabus (1982).
cations. Automatica 10, 353–379. Adaptive Control for a Class of MIMO Sys-
tems. IEEE Transactions on Aerospace 18,
576–590.
Torres, S., and E. Mehiel (2006). Nonlinear Direct
Adaptive Control and Disturbance Rejection
for Spacecraft. Proceedings of the AIAA GNC
2006, paper AIAA2006-6038.
van Amerongen, J., and A. U. Ten Cate (1975).
Model Reference Adaptive Controller for
Ships. Automatica 11, 441–449.
Wellstead, P., and M. Zarrop (1991). Self-Tuning
Control Systems, Wiley, Chichester, UK.
Wen, J. T. (1988). Time-domain and frequency
domain conditions for strictly positive real-
ness. IEEE Transactions on Automatic Con-
trol 33, 988–992.
Wen, J., and M. Balas (1989). Finite-dimensional
direct adaptive control for discrete-time
infinite-dimensional Hilbert space. Journal of
Mathematical Analysis & Applications 143,
1–26.
Whitaker, H. P. (1959). An Adaptive Performance
of Aircraft and Spacecraft. Inst. Aeronautical
Sciences, Paper 59-100.