Structured H∞ Synthesis in MATLAB

P. Gahinet ∗ P. Apkarian ∗∗
∗
MathWorks, 3 Apple Hill, Natick, MA 01760, USA (Tel: +1 508-647-7368;
e-mail: pascal@mathworks.com).
∗∗
ONERA-CERT, Centre d’études et de recherches de Toulouse, Control System
Department, 2 av. Edouard Belin, 31055 Toulouse, France - and - Institut de
Mathématiques, Université Paul Sabatier, Toulouse, France (Tel: +33
5.62.25.27.84 ; e-mail: apkarian@cert.fr)

Abstract: This paper presents new MATLAB-based tools for tuning fixed-structure linear
control systems using the H∞ methodology. Unlike traditional H∞ synthesis, these tools can
directly tune arbitrary control architectures consisting of one or more feedback loops and one or
more fixed-order, fixed-structure control elements. This makes them ideally suited for deploying
H∞ techniques in real-world applications.

Keywords: Structured controllers, decentralized control, H∞ synthesis, robustness, MATLAB.

1. INTRODUCTION results on par with full-order H∞ synthesis but for much

simpler controller structures. Finally, the optimization-
H∞ theory and its refinements Stein and Doyle [1991], based nature of these tools makes them well suited to
McFarlane and Glover [1992], Zhou et al. [1996] pro- tackle challenging extensions such as multi-model, multi-
vide a powerful frequency-domain framework for capturing objective H∞ synthesis with time-domain specifications
standard control design requirements such as speed of Bompart et al. [2008], Simoes et al. [2009].
response, bandwidth, disturbance rejection, and robust The paper is organized as follows. Section 2 discusses
stability. This framework extends classical control tech- the standard formulation of structured H∞ synthesis and
niques such as Bode loop shaping or gain/phase margin the representation of tunable control elements. Section
analysis to multi-loop and MIMO control architectures. 3 reviews the basics of expressing design requirements
Design requirements are expressed in terms of H∞ gain as H∞ constraints. Section 4 gives a brief overview of
constraints, and efficient algorithms are available to syn- the nonsmooth solvers at the heart of this approach.
thesize MIMO controllers that meet these requirements Section 5 highlights the key features of related software
Doyle et al. [1989], Robust Control Toolbox [2011]. Yet tools available in Robust Control Toolbox [2011]. Finally,
traditional H∞ synthesis has practical limitations that Sections 6 illustrates the potential of this technology on a
have slowed its adoption in industry. In particular, H∞ realistic example.
controllers are monolithic whereas most embedded control
architectures are decentralized collections of simple control
elements such as gains and PID controllers. In addition, 2. FRAMEWORK FOR TUNING FIXED
the order (complexity) of H∞ controllers tends to be high CONTROL STRUCTURES
whereas embedded controllers tend to have low complex-
ity. As a result, engineers must either abandon familiar As mentioned earlier, our starting point is any linear con-
decentralized architectures, or painstakingly re-interpret trol architecture with one or more fixed-structure blocks to
the results of H∞ synthesis in terms of their specific tune. An example of such architecture is shown in Figure
architecture. Neither option being easy, hand-tuning or 6 where the shaded blocks are tunable. Since there are in-
optimization-based tuning tend to remain the norm.
finitely many possible architectures, we use a standardized
This paper presents new tools for Structured H∞ Synthesis representation called Standard Form that is both general
that overcome the limitations listed above. These tools and simple enough to work with. The Standard Form is
leverage state-of-the-art nonsmooth optimizers Apkarian depicted in Figure 1 and consists of two main components:
and Noll [2006a], Apkarian et al. [2007], Burke et al. • An LTI model P (s) which combines all fixed (non
[2006] to directly and efficiently tune arbitrary control tunable) blocks in the control system
architectures. By ”arbitrary,” we mean any single- or • A structured controller C(s) = Diag(C1 (s), . . . , CN (s))
multiple-loop block diagram arrangement containing any which combines all tunable control elements. Each
number and type of linear control elements, from simple control element Cj (s) is assumed to be linear time
gains and PIDs to more complex notch filters and observer- invariant and to have some prescribed structure.
based controllers. Note that H∞ synthesis of fixed-order,
fixed-structure MIMO controllers is just a special case. External inputs to the system such as reference signals and
While the underlying optimization programs are typically disturbances are gathered in w and performance-related
non-convex and NP-hard, the nonsmooth optimizers fare outputs such as error signals are gathered in z (see Section
well on a wide range of applications and often produce 3 for more details).
 w z w M M z
11 12

P(s) M M
21 22

u y
.. . . 0.
a ...
u C (s) ... 0 y
..
1
..
. . ..
.
..
. .
0 ... a
0 . . . CN(s)

C(s) Fig. 2. LFT Model of Rational Expressions in a

Fig. 1. Standard Form for Structured H∞ Synthesis

More generally, we can model any rational function of
the scalar- or matrix-valued parameters a1 , . . . , aM by
It is well known from Robust Control theory that any recasting arithmetic operations between these parameters
block diagram can be rearranged into this Standard Form as interconnection operations between LFT models [Zhou
by isolating the tunable blocks and collapsing the rest and Doyle, 1998, p. 165]. This results in an LFT model
of the diagram into P (s). The resulting model has the similar to Figure 2 where a ⊗ I is replaced by
same structure as the uncertain models used in µ analysis
with the uncertainty blocks ∆j (s) replaced by the control Diag(a1 ⊗ Ik1 , . . . , aM ⊗ IkM ). (4)
elements Cj (s) Redheffer [1960], Doyle et al. [1991], Varga
and Looye [1999], Varga and Magni [2005]. Figure 1 is also Since any interconnection of LFT models is an LFT model,
reminiscent of standard H∞ synthesis Doyle et al. [1989] and the Standard Form of Figure 1 is itself an LFT model,
but differs in one key aspect, namely, the special structure it is easily seen that if some tunable element Cj (s) is a
of the controller C(s). Since systematic procedures and rational function of the parameters a1 , . . . , aM , the Stan-
automated tools are available to transform any architec- dard Form can be rearranged so that Cj (s) is replaced
ture to the Standard Form of Figure 1, we assume that the by a1 ⊗ Ik1 , . . . , aM ⊗ IkM along the diagonal of C(s). In
control architecture is specified in this form. other words, we can absorb the specific structure of Cj (s)
into P (s) and keep only the low-level tunable parameters
The next challenge is to describe the tunable control a1 , . . . , aM in C(s). The resulting block-diagonal controller
elements. Again we are faced with a wide range of pos- C(s) is then a mix of predefined blocks (e.g., PID) and pos-
sible structures, from simple gains and PIDs to more sibly repeated tunable parameters aj . Note that unlike µ-
complex lead-lag compensators and observer-based con- analysis, ”repeated” blocks do not affect the optimization
trollers. Since our approach is based on optimization, it outcome and only incur some negligible overhead.
is natural to use parametric models of such components.
For example, a PID can be parameterized by four scalars To illustrate how this comes together in the software tools,
a
Kp , Ki , Kd , Tf as first consider the lowpass filter F (s) = s+a where a is
tunable. This tunable element is specified by the MATLAB
Ki Kd s commands:
Cj (s) = Kp + + . (1)
s Tf s + 1
a = realp('a',1); % a is initialized to 1
Similarly, a state-space model with fixed order n can be F = tf(a,[1 a]); % creates a/(s+a)
parameterized by four matrices A, B, C, D of suitable sizes. This automatically builds the following LFT parameteri-
For convenience we provide a library of basic tunable zation of F (s):
blocks such as static gains, PIDs, and fixed-order transfer
0 0 1
!  !
functions. Yet this library leaves out many useful control a 0
elements, e.g., it cannot model the lowpass and notch F (s) = Fl 1/s −1/s 0 , . (5)
0 a
filters 1/s −1/s 0
a s2 + 2ζ1 ω0 s + ω02
, (2) Next consider the observer-based controller
s+a s2 + 2ζ2 ω0 s + ω02
x̂˙ = Ax̂ + Bu + L(y − C x̂ − Du)
because of the couplings between numerator and denomi- = (A − BK − LC + LDK)x̂ + Ly (6)
nator parameters. u = −K x̂
Rather than expanding the library of pre-defined tunable where the gain matrices K, L are tunable. This observer
blocks, a more scalable approach is to let users dynamically structure is specified as:
create tunable structures that suit their needs. To do this,
we introduce a basic building block called ”real parameter” % For a plant with nx states, nu controls,
(realp). If a is a real parameter, then any well-posed % and ny measurements:
rational function R(a) can be written as the LFT: K = realp('K',zeros(nu,nx));
L = realp('L',ones(nx,ny));
R(a) = Fl (M, a ⊗ I) (3) OBC = ss(A-B*K-L*(C-D*K),L,-K,0);
where M is a fixed matrix and a⊗I := Diag(a, . . . , a) Bett The resulting LFT parameterization OBC of the observer
and Lemmon [1997]. The LFT expression (3) corresponds structure has three copies of the parameter K and two
to the following static block diagram: copies of the parameter L. Note that in both examples,
it is possible to construct more efficient parameterizations • The open-loop gain |L(jω)| approximately matches
using a single copy of a, K, L. the target loop shape |λ(jω)|
• The sensitivity functions satisfy kSk∞ < 1 + δ and
3. H∞ FORMULATION OF DESIGN kT k∞ < 1 + δ at all frequencies, which guarantees
REQUIREMENTS adequate gain/phase margins and small overshoot.

Now that we have a framework for describing arbitrary In other words, the H∞ constraint kH(s)k∞ < 1 + δ
control architectures and linear control elements, we turn enforces the desired loop shape along with good stability
to the question of recasting the design requirements as margins.
H∞ constraints. At the heart of H∞ synthesis is the H∞ In general, this procedure leads to one or more normalized
norm, which measured the peak input/output gain of a H∞ constraints kHj (s)k∞ < 1 on frequency-weighted
given transfer function: closed-loop transfer functions H1 (s), . . . , HM (s). Introduc-
kH(s)k∞ := max σ(H(jω)). (7) ing
ω
H(s) := Diag(H1 (s), . . . , HM (s)), (8)
In the SISO case, this norm is just the peak gain over
frequency. In the MIMO case, it measures the peak 2-norm we can consolidate all requirements into the single con-
of the frequency response H(jω) over frequency. straint kH(s)k∞ < 1. Consequently, the tuning of a par-
ticular control architecture can be recast as the following
From classical control, we know that most control design program:
requirements are equivalent to gain constraints on the
open- or closed-loop response. For example, good tracking Structured H∞ Program: Tune the free parameters
in a particular frequency band is equivalent to small gain of C(s) := Diag(C1 (s), . . . , CN (s)) to enforce closed-loop
from reference to error signals in this band. Similarly, (internal) stability and kH(s)k∞ < 1.
bandwidth and roll-off requirements can be enforced by
Note that structured H∞ synthesis can handle constraints
shaping the open-loop response gain. For example, con-
on several independent transfers H1 (s), . . . , HM (s). To
sider the elementary feedback loop of Figure 3 and sup-
see this, observe that for Hj (s) := Fl (Pj (s), C(s)), the
pose we want integral action with bandwidth ωc . Then
Standard Form for H(s) looks like
λ(s) = ωc /s is a suitable shape for the open-loop response
L(s) = G(s)C(s) as depicted in Figure 4. The zero dB H(s) = Fl (P (s), Diag(C(s), . . . , C(s))) (9)
crossover frequency ωc dictates how fast the system re-
sponds and is limited by factors such as stability, delays, where P (s) is some rearrangement of the input and output
actuator bandwidth, and modeling errors. channels of Diag(P1 (s), . . . , PM (s)). So constraining two
or more closed-loop transfer functions Hj (s) leads to re-
ew peating the controller C(s) multiple times in the Standard
Form. The resulting block-diagonal controller structure is
l(s) beyond the scope of standard H∞ algorithms but poses
+ e no problem in our framework since this merely amounts
r C(s) G(s) y
- to repeating the tunable blocks along the diagonal (see
Section 2 for a discussion of repeated blocks). This is an
+
n nw
important difference with traditional H∞ synthesis where
1/l(s)
+ all requirements must be expressed in terms of a single
closed-loop transfer function, and this additional capabil-
Fig. 3. Loop Shaping Configuration ity greatly simplifies the H∞ formulation as illustrated in
Section 6.
Target loop shape

20

15
4. NONSMOOTH H∞ SOLVERS
10 ω
c

5
The Structured H∞ Program of Section 3 can be solved
Magnitude (dB)

0
using specialized nonsmooth optimization techniques. This
−5
section gives a high-level overview of the structured H∞
−10
solver at the heart of our software tools. The full details
−15
can be found in Apkarian and Noll [2006a, 2007] and a
−20
10
0
10
1 2
10
variety problem studies as well as applications can be
Frequency (rad/sec)
found in Apkarian [2010]. Note that related nonsmooth
techniques and software for control applications are dis-
Fig. 4. Target Loop Shape |λ(jω)| = ωc /ω cussed in Burke et al. [2006], Michiels and Niculescu [2007]
and references therein.
Once we express requirements in terms of open- or closed- Structured H∞ synthesis requires solving semi-infinite,
loop gains, there are well-established procedures for de- nonconvex, and nonsmooth programs of the form
riving the corresponding H∞ constraints Skogestad and
Postlethwaite [1996]. For example, if the closed-loop trans- minimize kFl (P (s), C(s))k∞ ⇐⇒
C(s)
fer function H(s) from (r, nw ) to (y, ew ) in Figure 3 satis-
  (10)
minimize max σ Fl (Pb(jω), x) ,
fies kH(s)k∞ < 1 + δ, then for small δ: x∈Rk ω∈[0, ∞]
where C(s) is the structured controller of Section 2 and the σ(Fl (P
b(jω), x)
vector x gathers all low-level tunable parameters in C(s). Hamiltonian
method
While this is a challenging mathematical programming kFl (P
b(s), x)k∞
f (ωb , x) := 0.9kFl (Pb(s), x)k∞ ** * *
problem, it is important to note that the right-hand-side bracketing frequencies ωb

function in (10) is the composition of the convex but

nonsmooth function maxω ◦ σ(.) with the nonconvex but
differentiable mapping x → Fl (Pb(jω), x). Such composite
functions are Clarke regular Clarke [1983], which means
0
that a complete description of the Clarke subdifferential log ω (rad.s−1 )

is accessible. To simplify the discussion, rewrite program

(10) as Fig. 5. Frequency selection to build tangent program

minimize f∞ (x) := max f (ω, x) . (11)

x∈Rk ω∈[0, ∞] 5. SOFTWARE TOOLS

Clarke regularity ensures that critical points x∗ (typically This section gives a brief overview of new software tools to
local minima) are characterized by the condition 0 ∈ formulate and solve structured H∞ problems for arbitrary
∂f (x). linear control architectures. These tools are available in
Robust Control Toolbox [2011]. Note that some functional-
To solve (11), we construct a tangent model around the ity discussed here is only available in the R2011a version.
current iterate x that constitutes a ”quadratic first-order”
local approximation of the original problem. An adequate As discussed in Section 2, the structure of individual
descent direction h is then computed by solving a convex tunable elements is described in terms of parameterization.
quadratic program of the form: The software provides pre-defined parameterizations for
gains, PIDs, fixed-order transfer functions, and state-space
minimize fˆ∞ (x + h) := models. For example
h∈Rk
1 (12) C1=ltiblock.tf('C1',2,3) % 2 zeros, 3 poles
max f (ω, x) − f∞ (x) + ΦTω h + hT Qh
ω∈Ωf 2 parameterizes C1 (s) as a strictly proper third-order trans-
fer function. In addition, users can create their own pa-
where Ωf is some finite set of frequencies and Φω ∈ rameterization using the realp building block (real pa-
∂f (ω, x) is a subgradient of f (ω, x). There are many rameter) and standard arithmetic operations, see Section
possible implementations of this scheme Apkarian and Noll 2 for examples.
[2006a], but a minimal requirement is that Ωf contain the The H∞ formulation of design requirements remains a
active frequencies ωa achieving the peak value in (11): manual process. This is the same process as for standard
f∞ (x) = f (ωa , x) . H∞ synthesis, somewhat simplified by the fact that we
can independently constrain several closed-loop transfer
This simple requirement is enough for the algorithm to functions (see Section 3). Tools are available to derive
converge. Yet by adding a few extra well-chosen frequen- the Standard Form of Figure 1 in both MATLAB and
cies, we can often improve the quality of the tangent model Simulink. Note that the objects used to parameterize
(12) and take longer steps at each iteration. In particu- C1 (s), . . . , CN (s) can be combined with regular LTI ob-
lar, including frequencies bracketing the active peaks can jects to build ”parametric” models of the relevant closed-
dramatically accelerate convergence Apkarian and Noll loop transfer functions. For example, consider the simple
[2006b,a]. See Figure 5 for an illustration of this strategy. scenario where the requirements for the feedback loop of
Note that multiple active frequencies typically arise as the Figure 3 can be expressed as
optimization progresses due to the ”waterbed effect”. The
corresponding loss of differentiability can often spell trou- kwS Sk∞ < 1, kwT T k∞ < 1 (13)
ble for smooth nonlinear programming algorithms Bert-
sekas [1995] which may fail to find a joint descent direction where S = 1/(1 + L), T = L/(1 + L), and wS , wT
for all active peaks. are suitable frequency-weighting functions. Assuming the
tunable block C(s) is a PID controller, the aggregate
The resulting algorithm is guaranteed to converge to a transfer function H(s) = Diag(wS S, wT T ) is obtained by:
critical point (a local minimum in practice) and has proved
very effective on a wide range of test problems including G = tf([1 2],[1 5 10]); % plant model
large scale systems Simoes et al. [2009]. Arguably, the C = ltiblock.pid('C','pid'); % define PID
function f∞ (.) is non convex and there is no guarantee of S = feedback(1,G*C);
reaching the global optimum. But ”convex” formulations T = feedback(G*C,1);
(e.g., in terms of LMIs Boyd et al. [1994]) often rely on H0 = blkdiag(wS * S, wT * T);
conservative and expensive relaxations or resort to bi- The variable H0 contains a MATLAB representation of the
convex schemes like BMIs Safonov et al. [1994] and D-K (untuned) Standard Form for H(s) and depends on the
iterations [Zhou and Doyle, 1998, p. 381]. As a result, our tunable PID block C.
approach is very competitive with such formulations and
has the advantage of tuning the original control structure Once the Standard Form is available, the hinfstruct
and controller parameters. command invokes the solver outlined in Section 4 to
 reference model
optimize the free parameters of C1 , . . . , CN . For example, tracking error
the PID gains in the simple example above are tuned by Gref (s)
−
feedforward
H = hinfstruct(H0);
Kf
The output H contains the tuned Standard Form of H(s)
w
and you can access the tuned PID controller C with roll−off filter aircraft
Nz
Nzc Ki + δm
s+ Fro(s) G(s) q
H.Blocks.C − −

Note that hinfstruct can be configured to automatically Kp Kq

run multiple optimizations from randomly generated start-
ing points. This helps mitigate the local nature of the
optimizer and increases the likelihood of finding parameter nq nNz
values that meet the design requirements.
Fig. 6. Aircraft Autopilot
6. AIRCRAFT AUTOPILOT EXAMPLE ωn2
Gref (s) :=
+ 2ζωn s + ωn2 s2
This section illustrates the use of hinfstruct to tune the
longitudinal autopilot for a supersonic transport aircraft with ζ = 0.7 and ωn = 1.7. This reference model captures
flying at Mach 0.7 and altitude 5000 ft. Further details on the desired time response characteristics by keeping the
the model used in this application can be found in Boiffier natural frequency of the α-oscillation mode and increasing
[1998]. The aircraft model is fairly conventional and given its damping to a suitable value. We formulate this require-
in state-space form as ment as kW1 (s)T1 (s)k∞ < 1 where T1 (s) is the closed-loop
transfer function from Nzc to e and
ẋ = Ax + Bu 2
(14)

y = Cx + Du , s 5
W1 (s) := 15 (16)
s + 0.05 s + 5
where the state vector is x := [ V γ α q H ] and the state
variables are the aerodynamic speed V (m/s), climb angle is a bandpass filter emphasizing the frequency band where
γ (rad), angle of attack α (rad), pitch rate q (rad/s), good tracking is desired
and altitude H (m). The elevator deflection δm (rad) is High-frequency roll-off is required to attenuate noise,
used to control the vertical load factor Nz (m/s2 ). Two increase robustness to unmodeled dynamics, and pre-
measurements are used for feedback: Nz and the pitch vent high-frequency control activity that would satu-
rate q. Typical of such models, the open-loop dynamics rate the actuator physical limits.This is expressed as
include the α oscillation with frequency and damping kW2 (s)T2 (s)k∞ < 1 where T2 (s) is the closed-loop transfer
ratio ωn = 1.7 (rad/s) and ξ = 0.33, the phugoid mode from the noise inputs nNz , nq to the control signal δm , and
ωn = 0.64 (rad/s) and ξ = 0.06, and the slow altitude √
mode, λ = −0.0026. See the corresponding demo in Robust s (1/82 )s2 + ( 2/8)s + 1
W2 (s) := √ (17)
Control Toolbox [2011] for the model data. s + 8 (1/8002 )s2 + ( 2/800)s + 1
The control structure considered here is the conventional is a high-pass filter chosen to enforce second-order roll-off
longitudinal control law depicted in Figure 6. The autopi- past 8 rad/s.
lot is comprised of four tunable elements:
Stability margins can be viewed as imposing some
• PI controller with proportional and integral gains Kp minimum distance between the open-loop response and
and Ki the critical point, or equivalently an upper bound on the
• Static feedback gain Kq on the pitch rate q. This gain gain of the sensitivity function T3 (s) from w to δm (see
is used to improve damping of the responses Section 3). Here we impose a minimum distance of 0.8,
• Feedforward gain Kf , which helps achieve better or equivalently kW3 (s)T3 (s)k∞ < 1 with W3 (s) = 0.8. In
performance and reduce the burden on the feedback terms of classical gain and phase margins, this guarantees
controller gain margins of at least −5.10dB and 13.97dB and phase
• Second-order roll-off filter 2 2
−0.82
margins of at least ±47 = ± arccos( 1 +12×1×1 ) degrees.
ωn2
Fro (s) = 2 . (15) Accordingly, we need to tune the autopilot parameters
s + 2ζωn s + ωn2
Kp , Ki , Kq , Kf , ζ, ωn to enforce closed-loop stability and
Note that pure integral action in the PID controller is not kH(s)k∞ < 1 where
possible because this would cancel the zero at s = 0 in the
Nz channel of the aircraft model G(s). Instead we use a H := Diag(W1 T1 , W2 T2 , W3 T3 ). (18)
pseudo-integrator 1/(s + 0.001). Here steady-state errors
are of no concern here since only the transient response to Each closed-loop transfer function Tj (s) can be written
the acceleration command Nzc matters. in Standard Form as Tj = Fl (Pj , C) where C(s) :=
Diag(Kp , Ki , Kq , Kf , Fro (s)). Starting from a Simulink
There are three main design requirements. Setpoint model of the block diagram in Figure 6, we use the
tracking is expressed as a model following objective and command linlft to compute the plant models P1 , P2 , P3 .
involves minimizing the tracking error e between Nz and Next we specify the tunable components using the realp
the output of the reference model command and form the structured controller C(s):
 Ki = realp('Ki',0); Kp = realp('Kp',0); Christopher J. Bett and Michael Lemmon. On linear fractional
Kq = realp('Kq',0); Kf = realp('Kf',0); representations of multidimensional rational matrix functions.
wn = realp('wn',3); zeta = realp('zeta',0.8); Technical report, Eindhoven, University of Technology, 1997.
J. Boiffier. The Dynamics of Flight, the Equations. John Wiley &
Fro = tf(wn^2,[1 2*zeta*wn wn^2]);
Sons, New York, 1998.
C = blkdiag(Kp,Ki,Kq,Kf,Fro); V. Bompart, P. Apkarian, and D. Noll. Control design in the time-
Note that most gains are initialized to zero (open loop). and frequency-domain using nonsmooth techniques. Syst. Control
Letters, 57(3):271–282, 2008.
Finally, we build a (parametric) model H0 of the transfer
S. Boyd, L. ElGhaoui, E. Feron, and V. Balakrishnan. Linear Matrix
function H(s) in (18) using: Inequalities in Systems and Control Theory, volume 15 of SIAM
H0=blkdiag(W1*lft(P1,C),W2*lft(P2,C),W3*lft(P3,C)) Studies in Applied Mathematics. SIAM, Philadelphia, 1994.
J. V. Burke, D. Henrion, A. S. Lewis, and M. L. Overton. Stabiliza-
We are now ready to tune the parameters with hinfstruct: tion via nonsmooth, nonconvex optimization. IEEE Trans. Aut.
Control, 51(11):1760–1769, November 2006.
[H,gam] = hinfstruct(H0); F. H. Clarke. Optimization and Nonsmooth Analysis. Canadian
Final: Peak gain = 1.06, Iterations = 57 Math. Soc. Series. John Wiley & Sons, New York, 1983.
J. Doyle, K. Glover, P. P. Khargonekar, and B. A. Francis. State-
This returns the tuned H(s) along with the best achieved space solutions to standard H2 and H∞ control problems. IEEE
H∞ norm gam = 1.06. Starting from an open-loop config- Trans. Aut. Control, AC-34(8):831–847, August 1989.
uration with zero gains, this optimization runs in 4 seconds J. Doyle, A. Packard, and K. Zhou. Review of LFT’s, LMI’s and µ.
on a Desktop PC with a 2.4GHz Intel Core2 Quad CPU In Proc. IEEE Conf. on Decision and Control, volume 2, pages
and 6.00Gb of RAM. The tuned values of the controller 1227–1232, Brighton, December 1991.
parameters are as follows: D. McFarlane and K. Glover. A loop shaping design procedure using
Kp = −0.0048, Ki = −0.022, Kq = −0.29, H∞ synthesis. IEEE Trans. Aut. Control, 37(6):759–769, 1992.
W. Michiels and S. I. Niculescu. Stability and stabilization of
Kf = −0.033, ζ = 0.79, ωn = 6.46. time-delay systems. An eigenvalue based approach, volume 12 of
To validate the design, Figure 7 compares the actual Advances in Design and Control. SIAM Publications, 2007.
response Nz to a step command Nzc with the response R. M. Redheffer. On a certain linear fractional transformation. J.
of the reference model Gref (dashed). The plot on the Math. and Phys., 39:269–286, 1960.
right shows the overall deflection δm (solid) and the Robust Control Toolbox. The MathWorks Inc., Natick, MA, 2011.
respective contributions of the feedforward and feedback M. G. Safonov, K. C. Goh, and J. H. Ly. Control System Synthesis
via Bilinear Matrix Inequalities. In Proc. American Control Conf.,
paths. Finally, Figure 8 shows the open-loop response
pages 45–49, 1994.
measured at w and the corresponding stability margins (18 A. M. Simoes, Diego C. Savelli, P. C. Pellanda, N. Martins, and
dB gain margin and 90 degrees phase margin). Note that P. Apkarian. robust design of a tcsc oscillation damping controller
the roll-off constraint is inactive and could be tightened. in a weak 500-kv interconnection considering multiple power flow
scenarios and external disturbances. IEEE Trans. on Power
REFERENCES Systems, 24(1):226–236, 2009.
P. Apkarian. Internet pages. http://pierre.apkarian.free.fr, S. Skogestad and I. Postlethwaite. Multivariable feedback design -
2010. analysis and design. Wiley, 1996.
P. Apkarian and D. Noll. Nonsmooth H∞ synthesis. IEEE Trans. G. Stein and J. C. Doyle. Beyond singular values and loop shapes.
Aut. Control, 51(1):71–86, 2006a. J. Guidance and Control, 14:5–16, 1991.
P. Apkarian and D. Noll. Nonsmooth optimization for multidisk H∞ A. Varga and G. Looye. Symbolic and numerical software tools for
synthesis. European J. of Control, 12(3):229–244, 2006b. LFT-based low order uncertainty modeling. In Proc. CACSD’99
P. Apkarian and D. Noll. Nonsmooth optimization for multiband Symposium, Cohala, pages 1–6, 1999.
frequency domain control design. Automatica, 43(4):724–731, A. Varga and J.F. Magni. Enhanced LFR-toolbox for Matlab.
April 2007. Aerospace Science and Technology, 9(2):173–180, 2005.
P. Apkarian, V. Bompart, and D. Noll. Nonsmooth structured K. Zhou and J. Doyle. Essentials of robust control. Prentice-Hall
control design with application to PID loop-shaping of a process. International, Inc., 1998.
Int. J. Robust and Nonlinear Control, 17(14):1320–1342, 2007. K. Zhou, J. C. Doyle, and K. Glover. Robust and Optimal Control.
D. P. Bertsekas. Nonlinear Programming. Athena Scientific, USA, Prentice Hall, 1996.
Belmont, Mass., 1995.
Step Response Bode Diagram
Step Response
From: Nzc To: deltam 80
From: Nzc To: Nz
1.4 0.01
60 Open−Loop Response
Actual response Total
Feedforward 1/W2
Gref response
0.005
Magnitude (dB)

Feedback 40
1.2
0 20

1 −0.005 0

−0.01 −20
0.8
δm
Nz

−40
−0.015 180
0.6
−0.02 90
Phase (deg)

0.4 −0.025 0

−0.03 −90
0.2
−0.035 −180

0 −0.04 −270
−3 −2 −1 0 1
0 1 2 3 4 5 6 0 1 2 3 4 5 6 10 10 10 10 10
Time (sec) Time (sec) Frequency (rad/sec)

Fig. 7. Response to a Step Command Nzc Fig. 8. Open-Loop Response