You are on page 1of 5

2650

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 55, NO. 11, NOVEMBER 2010

A Bode Plot Characterization of All Stabilizing Controllers


L. H. Keel and Shankar P. Bhattacharyya
Fig. 1. Unity feedback system. AbstractIn this technical note, we consider continuous-time control systems and present a new characterization of the Nyquist criterion in terms of Bode plots of the plant and the controller. This gives a nonparametric, and model independent characterization of arbitrary order stabilizing controllers. The result shows that the frequency response of any stabilizing controller must satisfy constraints on its magnitude, phase, and rate of change of phase at certain frequencies that are imposed by the frequency response of the inverse plant. Index TermsAll stabilizing controllers, Bode plots, Nyquist criterion.

I. INTRODUCTION The Nyquist criterion [1] provides a powerful test for closed-loop stability in terms of open-loop measured data. When applied to a plantcontroller pair however, it requires the testing of the combined transfer function. This is not convenient in some synthesis and design problems, where explicit conditions are required on the controller to be designed, in terms of given plant data. In this technical note, we develop new criteria for controller design to precisely address and x the above problems. This is done by interpreting the Nyquist criterion via separate Bode plots [2] of the plant and the controllers. This result shows that the frequency response of the inverse plant imposes constraints on the magnitude, phase, and rate of change of phase of the controller at certain frequencies. We also show that such conditions can easily be extended to performance requirements such as gain and phase margin specications. These provide useful results for controller design as shown, by examples. Some recent related results on this problem are as follows. The design of xed order controllers for discrete-time systems was discussed in [3]. In [4], the use of quantier elimination (QE) techniques to deal with the xed order controller design problem was proposed. In [5], the D-decomposition technique [6] was applied to design xed order controllers. We mention the related works of Hara [7], Ikeda [8], [9] and Jayasuriya [10]. In Hara [7] a frequency dependent version of the KYP Lemma is developed and used for synthesis. Ikeda [8], [9] advocates a model free approach to design. In Shafai [11] qualitative robust control (QRC) was suggested as a procedure for designing compensators from a qualitative model of the plant. In Jayasuriya [10] a quantitative feedback theory (QFT) approach to design is discussed, wherein robustness bounds are imposed at various frequencies that are relevant to loop shaping. Parameter space methods advocated by Ackermann [12] and Siljak [13] are practical and effective methods in industrial practice. Indeed, dealing with a family of plants in parameter space rather than

a single plant was the rst step in robust parametric control [14][17]. In [18], a method to design three term controllers based directly on frequency domain test data was introduced. The attraction of data based methods is that direct design based on test data is at least as reliable as that based on models. In Richardson, Anderson, and Bose [19], it was shown that controllability indices in a minimal state space realization of a real rational transfer function matrix may be calculated from evaluations of the transfer function matrix at a sufcient number of discrete points in the frequency domain. It will be seen that the new methods developed here also involve various frequencies where specic conditions must hold. The method presented assumes the frequency response data of the plant is available for a sufcient range of frequencies. The result is applicable to stable and unstable systems just as the Nyquist criterion is. If the open loop transfer function G(s) is known, the required Bode plot data can be obtained by evaluating G(j! ). If the open loop system is stable, the Bode plot data can be determined experimentally as the steady state sinusoidal frequency response over a wide range of frequencies. However, it is impossible to directly measure the frequency response of an unstable system. It must be obtained after stabilizing it with a known controller, measuring the corresponding frequency response and then computing it by dividing out the known controller. We believe that it is impossible to develop measurement-only based identication or synthesis methods for unstable systems unless one such controller is known by experimental or other methods and this may be regarded as a limitation of such methods. It is worthwhile to mention that the results presented are admittedly equivalent to model based methods in a mathematical sense because they are solutions to the same problem. Nevertheless, it is useful to know that identied models need not be produced as an intermediate step in the design process. This is especially signicant in view of the fact that identication involves knowledge of the order of the system whereas our methods do not require any knowledge of the system order. II. NOTATION AND PRELIMINARIES Let us begin by considering a continuous-time, linear time-invariant nite dimensional single-input single-output plant G described by a rational proper transfer function G(s) with p+ open right half plane (RHP) poles, in a unity feedback conguration as shown in Fig. 1. Let G(j! ) be the frequency response of the plant and let !i , i = 0; 1; 2; 1 1 1 ; k + 1 with !0 = 0 and !k+1 = 1 denote the frequencies where the Nyquist plot of G(s) cuts the negative real axis of the complex plane. In other words, these frequencies are the solutions of the following equation:

Manuscript received February 16, 2010; revised July 06, 2010; July 20, 2010, and July 21, 2010; accepted July 27, 2010. Date of publication August 16, 2010; date of current version November 03, 2010. This work was supported in part by DOD Grant W911NF-08-0514, NSF Grant CMMI-0927664, and NSF Grant CMMI-0927652. Recommended by Associate Editor P. Tsiotras. L. H. Keel is with the Department of Electrical and Computer Engineering, Tennessee State University, Nashville, TN 37203 USA (e-mail: keel@gauss. tsuniv.edu). S. P. Bhattacharyya is with the Department of Electrical and Computer Engineering, Texas A&M University, College Station, TX 77843 USA (e-mail: bhatt@ece.tamu.edu). Color versions of one or more of the gures in this technical note are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TAC.2010.2067390

G(j!) = n; for n = 1; 3; 5; : (1) ; ! ; ! +1 where 0 =: !0 < !1 < Dene the set
= !0 ; !1 ; !2 < < ! < ! +1 := and !0 and ! +1 are included only
6
6 6 6 111 f 111

111

if they satisfy the above angle condition. Introduce the corresponding sequence of integers fi0 ; i1 ; i2 ; 1 1 1 ; ik ; ik+1 g where

:= 0

if

G(j! ) < 1
t j

(2)

0018-9286/$26.00 2010 IEEE

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 55, NO. 11, NOVEMBER 2010

2651

and otherwise
it
:= sign

d d!

6 G(j!)

for

= 0; 1; 2; 1 1 1

;k

+ 1:

(3)

III. BODE EQUIVALENT OF THE NYQUIST CRITERION Suppose rst that the plant G has no imaginary axis poles. We assume as usual that the Nyquist contour is traversed in the clockwise direction, that is with ! increasing. Lemma 1: Under the assumption that the plant G has no imaginary axis poles, let N denote the net number of counterclockwise encirclement of 01 + j 0 by the Nyquist plot of G(s). Then
k

Fig. 2.
2it +

i0

t=1

ik+1

=:

i(G):

(4)

0contour for Nyquist plot.

Proof: It is easy to see that the Nyquist plot of G(s) cuts the negative real axis at the frequencies where (1) holds. The cuts are to the left of 01 + j 0 when jG(j!t )j > 1. The conditions in (3) along with jG(j!t )j > 1 indicate that it = +1 when the Nyquist plot cuts the negative real axis to the left of 01 + j 0 downward, corresponding to a counterclockwise encirclement, and it = 01 when the plot cuts the negative real axis to the left of 01 + j 0 upward, corresponding to a clockwise encirclement of 01 + j 0. The factor 2 accounts for the negative values of ! where these real axis cuts occur. Thus, the formula (4) is the count of the net number of counterclockwise encirclements of 01 + j 0 by the Nyquist plot of G(s). We can now state the well-known Nyquist criterion for stability of the feedback system. The proof is omitted. Theorem 1: Under the assumption that the plant has no imaginary axis poles and p+ open RHP poles (Re > 0), the unity feedback system in Fig. 1 is stable iff
i(G)
=

1) m0 is odd: In this case, the Nyquist plot starts from the negative or positive imaginary axis as ! increases from zero, depending upon the values of m0 . Furthermore, the Nyquist plot turns 180 clockwise for every pole at the origin. The rst clockwise half circle is located in the LHP or RHP depending upon the sign of G(0+ ). Let us consider, for illustration, the case when m0 = 1. If G(0+ ) < 0, the clockwise half circle is located in the LHP and it results in a negative real axis cut to the left of 01 + j 0. Since this cut is upward, we have i0 = 01. On the other hand, the clockwise half circle is located in the RHP for + G(0 ) > 0 and this results in no negative real axis cut, that is, i0 = 0. From such considerations, we derive the following general formulas: For m0 odd
i0
= 0 0

m m

01 2 +1 2

if (01) if (01)

G(0 G(0

) )

> >

0 0:

(7)

p :

(5)

Remark 1: The above result is the Bode plot equivalent of the Nyquist criterion. It is applicable to stable and unstable open loop systems. If G(j! ) data is available, i(G) may be computed and the above relationship veried, provided p+ is known. For unstable systems, G(j! ) may be obtained if the transfer function G(s) is known. If G is stable G(j! ) may be experimentally measured. Direct experimental measurement of G(j! ) for an unstable G is not possible and in this case G(j! ) can only be obtained if a known stabilizing controller is available and the corresponding closed loop frequency response can be measured. We emphasize that the knowledge of p+ is also required for the Nyquist criterion. Moreover, knowledge of one stabilizing controller, which is required anyway for an unstable system, is sufcient to obtain p+ . This is easy to show and was also established in Theorem 1 of [18]. Now consider the case when the plant G has imaginary axis poles. This includes the important class of systems with one or more integrators that is required for a system to track, say, steps and ramps. A. Plants with Poles at the Origin Let m0 be the number of poles at the origin, and let i0 denote the corresponding number of encirclements in the counterclockwise direction by the Nyquist plot of G(s) at s = 0. Note that here
G(0

2) m0 is even: For the case of a plant with an even number of poles at the origin, the Nyquist plot begins from the negative or positive real axis as ! increases from zero, depending upon the value of m0 . With considerations similar to the previous case we derive the following general conditions: For m0 even
0

i0

2 2

0 1

if (01)
d 6 d!

01

G(0

> >

0 0

and (8)

= 0

G(j! )

otherwise.

! =0

B. Plants With Poles on the Imaginary Axis Let the denominator of the plant transfer function be
2 2
m

u1

u2

111

uk

(9)

that is, the plant has mi pairs of poles at 6jui for i = 1; 2; 1 1 1 ; k . Dene integer quantities ji for i = 1; 2; 1 1 1 ; k which denote the number of corresponding counterclockwise encirclement by the Nyquist plot of the point 01 + j 0 as follows. The verication of these is left to the reader and is based on arguments outlined in the previous cases. 0 G(jui ) is complex and mi is odd
0(mi 0 1) 0(mi + 1)

ji

) 6=

G(j 0

if 6 if 6

G jui G jui

0 0

2 (0;  ) 2 (; 2 )

(10)

):

(6)

As typically done in Nyquist theory, we use right indentation of the Nyquist 0-contour when the contour approaches imaginary axis poles to avoid singularities (see Fig. 2).

G(jui

is complex and mi is even


ji
= 0mi

(11)

2652

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 55, NO. 11, NOVEMBER 2010

G(ju0) is real and m is odd, 0(m 0 1) if G ju0 > 0 and 6 G(j!) = > 0 j= 0(m + 1) otherwise
i i i i d d! i ! u i

A. Gain and Phase Margins The performance of a controller is often determined by the closedloop stability margins it provides. The gain margin is such a performance measure. To compute it, dene the distinct frequency set (phase crossover frequencies)

( ) :=

(12)

G(ju0) is real and m


i

is even,

u : 6 C (ju ) = 6 P 01 (ju ) 6 n; n = 1; 3; 5; 1 1 1


i i i

0m
=

if G
i

0(m

d d!

+ 2)

ju0 > 0 or G ju0 < 0 and 6 G(j!) = > 0


i i ! u

for i = 1; 2; 1 1 1 ; n with u0 = 0 and um+1 = magnitudes measured in decibels and dene (13)

1. Let us denote
k) k)

(19)

otherwise/

u 2
() : jC (u )jdb >  0 () := u 2
() : jC (u )j <
db

( ) :=
k k k k

P 01 (u P 01 (u

db db

Theorem 1 can now be restated without restrictions on the location of poles of the plant. Theorem 2: The unity feedback system in Fig. 1 is stable iff
l k

The upper (lower) gain margin is the smallest increase (decrease) in gain measured in decibels that destabilizes the closed-loop. Remark 2: If C is a stabilizing controller, the upper gain margin + denoted by Kdb is

i(G) := i0 +
k

=1

i +1 +
l r

=1

jr = p+

(14)

where p+ is the number of open RHP poles of the plant G. IV. STABILIZING CONTROLLER CHARACTERIZATION We now consider a nite dimensional rational proper controller with frequency response C (j! ) and with c+ RHP poles and ask when it can stabilize a nite dimensional rational, proper plant with frequency response P (j! ). Dene the set of distinct non-negative frequencies

+ Kdb = 2
( ) C (u )db 0 P 01 (u min 0 The lower gain margin denoted by Kdb is 0 Kdb = 2
( ) P 01 (u )db 0 C (u min
u  k u  k

k)

db

: :

(20)

k)

db

(21)

( ) =

The phase margin is also an important performance measure. Let g fv1; v2 ; 1 1 1 ; vm g be the set of gain crossover frequencies

( ) :=

v : jC (jv)j = P 01 (jv)
k k

(22)

and dene

+
() := f!0 ; !1 ; 1 1 1 ; !l ; !l+1 g
with 0 =: condition
l l

(15)

!0 < !1 < 1 1 1 < ! < ! +1 := 1 satisfying the phase C (j!) = 6 P 01 (j!) 6 n; for n = 1; 3; 5; 7; 1 1 1
(16)

and the magnitude condition

v 2
(g) : 6 C (jv ) > 6 P 01 (jv ) + n ; v 2
(g) : 6 C (jv ) < 6 P 01 (jv ) + n : Remark 3: If C is a stabilizing controller, the phase margin is 8 := + 0 minf8 ; 8 g where + 6 C (jv ) 0 6 P 01 (jv ) 0 n ; min 8 := min odd 2
( ) 0 6 P 01 (ju ) 0 6 C (ju ) + n : min 8 := min odd 2
( )
0
(g ) :=

( ) :=

k k

k k

jC (j! )j > P 01 (j! ) :


k k

(17)

V. EXAMPLES In this section, we illustrate the usefulness of the previous results in controller design by examples. Example 1: Consider a plant with 2 RHP poles and known frequency response P (j! ) as shown in Fig. 3. In this example, we examine whether the plant under consideration can be stabilized by integral controllers with various structures. For convenience we write G(s) = C (s)P (s). For any integral controller, it is important to note that jC (j 0)j ! 1. Thus, we have jP 01 (j 0)j < + + + jC (j 0)j. On the other hand, 6 C (j 0) = (zc 0 pc ) where pc and + are numbers of RHP poles and zeros of the controller, respectively. zc However, 6 C (j 0+ ) quickly jumps to  (z + 0p+ ) 0 (=2) because of the pole at the origin (an integrator). This results in

Note that these are the frequencies where the Nyquist plot of P (s)C (s) intersects the negative real axis to the left of 01 + j 0. For the case when P (s)C (s) has no imaginary axis poles, we introduce the integers ik for k = 0; 1; 1 1 1 ; l + 1

+1 = 0

; 01;

if if if

d d! d d! d d!

C (j!) = > 6 C (j!) = = C (j!) = < 6


6
! ! ! ! ! !

d d! d d! d d!

P 01 (j!) = 6 P 01 (j!) = P 01 (j!) =


6
! ! !

! !

! :

The case when P (s)C (s) has imaginary axis poles can be treated by using the formulas (7), (8), and (10)(13). The Nyquist criterion can now be restated as follows. Theorem 3: The controller C stabilizes the plant P if and only if

i(C ) := i0 +
k

=1

i +1 +
l r

=1

p+ + c+ :

(18)

d 6 01 d6 d! C (j!) =0  01 < d! P (j!) =0 : Note that 6 P 01 (j 0) =  , so we have + + 0 i0 = 0;1; when z 0 p = n + 1; n = 61; 63; 1 1 1 otherwise.
! ! c c

(23)

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 55, NO. 11, NOVEMBER 2010

2653

. Thus, it cannot intersect P 1 (j! ) 6  . Thus, i(G) = 2 is not possible. + c) PID Control with two nonminimum phase zeros (zc = 2): from (23), we have i0 = 01. Thus, i(G) = 2 is not possible. Therefore, the plant cannot be stabilized by a PID controller. 4) PI control with a real pole
1

6 0

C s

+K ( ) = K(pss+ ):i s
0

(27)

a) When
i0

b)

Fig. 3. Frequency response of the plant considered (Example 1).

to the behavior of these two phase functions, it is impossible. Therefore, the plant cannot be stabilized by a controller of the given structure. 5) PID with a real pole
2 K ( ) = Kd s s(+ + p s)+ Ki s :

> 0 (a stable pole) closed-loop stability requires + = 0, equivalently zc = 1. The phase of a such controller decreases monotonically from =2 to (=2) for ! (0; ). Such a phase plot can cut 6 P 01 (j!) only downward. Thus, i(G) = 2 cannot be achieved. When < 0 (an unstable pole) closed-loop stability requires + i(G) = 3. Since i0 can only be 1 with zc = 1, it requires 6 C (j!) cuts 6 P 01 (j!) n at least three times. Due that
1 0 6

We also note that i = 0 because jP 1 (j1)j > jC (j 1)j for any proper C (s). We now consider integral controllers with various structures. 1) Pure Integral Control

C s

(28)

C s

( )= K: s

(24)

From (23), we have i0 = 01. Clearly, it is impossible to achieve ( ) = 2 that is required for stabilization. 2) Proportional-Integral (PI) Control
i G

C s

+ ( ) = Kp s s Ki :

a) When > 0 (a stable pole): The stability condition requires + i0 = 0, equivalently zc = 1. Clearly, C (j! ) for ! 2 (0; 1) moves from =2 to 0 and cannot cut P 1 (j!) 6 . Thus, i(G) = 2 cannot be achieved. b) When < 0 (an unstable pole): The stability condition + requires i0 = 01 (i.e., zc = 1) and i(G) = 3. This requires 1 C (j! ) to cut P (j!) 6 n at least ve times, which is not possible, due to the behavior of the two phase plots Therefore, the plant cannot be stabilized by a PID controller with a real pole, either. 6) Second Order Controllers with an integrator:

6 0

6 0

(25)
C s

+ a) PI Control with a minimum phase zero (zc = 0): from (23), we have i0 = 01. Consequently, i(G) = 2 is impossible. + b) PI Control with a nonminimum phase zero (zc = 1): from (23), we have i0 = 0. Note that C (j! ) runs monotonically from  to 0 for ! 2 (0; 1). Thus, C (j! ) cannot intersect 1 P (j!) 6 n for n = 61; 63; 1 1 1. Thus, i(G) = 2 is not possible to attain. Therefore, the plant cannot be stabilized by a PI controller. 3) PID Control

2 + 1 + 0 ( ) = s(2ss + ss+ ): 2 1 0

(29)

6 0

+ a) When pc + leads zc

b)

C s

2 ( ) = Kd s + Kp s + Ki : s

(26)

Observing the high frequency slope of jP 1 (j! )j, we conclude that the relative degree of the plant is 2. Thus, 1 jC (j 1)j < jP (j 1)j and i = 0. + a) PID Control with two minimum phase zeros (zc = 0): from (23), we have i0 = 01. Thus, i(G) = 2 is not possible. + b) PID Control with one nonminimum phase zero (zc = 1): from (23), we have i0 = 0. In this case, C (j! ) increases from =2 and decreases back to =2 as ! runs from 0+ to

possible. + c) When pc = 2 (two unstable poles): Similarly, the closed+ loop stability requires that zc = 1 and i(G) = 4, and it is not possible to obtain. The above analysis shows that the plant requires a controller with a structure more complicated than (29). To the best of our knowledge, such informative analysis would not be possible without Theorem 3 of this technical note. The following example illustrates the gain and phase margin computations.

= 0 (stable controllers): The condition i0 = 0 = 1 and i(G) = 2. Although 6 C (j!) cuts 6 P 01 (j!) n for ! (0; ) once, the cut will be downward. Thus, i(G) = 2 is not possible. + + When pc = 1 (one unstable pole): We need zc = 1 to have i0 = 1. To obtain i(G) = 3 for closed-loop stability, we (0; ) need that 6 C (j! ) cuts 6 P 01 (j! ) n for ! at least three times. However, it is easy to see that 6 C (j! ) cannot cut 6 P 01 (j! ) n at all. Thus, i(G) = 3 is not
6 2 1 0 6 2 1 6

2654

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 55, NO. 11, NOVEMBER 2010

Fig. 4. Illustrating gain and phase margins (Example 2).

Example 2: Consider a plant with 2 RHP poles and let the frequency response of the system P (j!) be known. Consider the test controller

0 12 : s+9 Since C (0) < P 01 (0) and C (1) < P 01 (1), we have i0 = i1 = 0.
C (s) =
36s

From Fig. 4, we have

i(C ) = 0 + 2(1) + 0 = 2
and

p+ + c+ = 2 + 0 = 2
and therefore the closed-loop system is stable. The gain and phase margins are shown in Fig. 4. VI. CONCLUSION In this note, we have presented new interpretations and alternatives to the traditional Nyquist criterion which are useful for both synthesis and analysis of controllers. The result shows how traditional Bode plot data augmented by rates of change of phase, of the inverse plant and controller are to be related in specic ways to ensure closed loop stability as well as gain and phase margins. The fact that every stabilizing controller regardless of order, must have this relationship to the plant can serve as a useful tool for analysis and synthesis as we have shown by examples. We also note that it is practically impossible to measure the complete frequency response of a system ranging from 0 to innite frequency. Nevertheless frequency response methods based on Bode plots have proved very useful in control engineering. Our results clearly show that detailed frequency response is needed only in a low frequency range determined by the characteristics of the controller which in turn determine the range of crossover frequencies. For most plants, magnitude and rate of change of phase decrease monotonically after a low frequency range. Rough information only is needed outside this range. It is also our hope that these results will ultimately lead to design methods based on measurements at a nite set of frequencies as in [20] and [21].

[3] W. M. Haddad, H. H. Huang, and D. S. Bernstein, Robust stability and performance via xed-order dynamic compensation: The discrete-time case, IEEE Trans. Autom. Control, vol. 38, no. 5, pp. 776782, May 1993. [4] P. Dorato, Quantied multivariable polynomial inequalities: The mathematics of all practical design problems, IEEE Control Syst. Mag., vol. 20, no. 5, pp. 4858, Oct. 2000. [5] E. N. Gryazina and B. T. Polyak, Stability regions in the parameter space: D-decomposition revisited, Automatica, vol. 42, no. 1, pp. 1326, Jan. 2006. [6] D. D. Siljak, Nonlinear Systems: The Parameter Analysis and Design. New York: Wiley, 1969. [7] S. Hara, T. Iwasaki, and D. Shiokata, Robust PID control using generalized KYP synthesis: Direct open-loop shaping in multiple frequency ranges, IEEE Control Syst. Mag., vol. 26, pp. 8091, Feb. 2006. [8] U. S. Park and M. Ikeda, Data-based stability analysis for linear discrete-time system, in Proc. 43rd IEEE Conf. Decision Control, Atlantis, Bahamas, Dec. 1417, 2004, pp. 17211723. [9] F. Yasumasa, Y. Duanm, and M. Ikeda, System representation and optimal tracking in data space, in Proc. 16th IFAC World Congress, Prague, Czech Republic, Jul. 38, 2005, [CD ROM]. [10] W. Wu and S. Jayasuriya, A QFT design methodology for feedback systems under input saturation, J. Dyn. Syst., Meas., Control, vol. 123, pp. 225232, Jun. 2001. [11] S. P. Linder and B. Shafai, Qualitative Robust Fuzzy Controller for the 1992 ACC Robust Control Benchmark, in Proc. Amer. Control Conf., 1997, pp. 12581263. [12] J. Ackermann, Parameter space design of robust control systems, IEEE Trans. Autom. Control, vol. AC-25, no. 6, pp. 10581072, Jun. 1980. [13] D. D. Siljak, Parameter space methods for robust control design: A guided tour, IEEE Trans. Autom. Control, vol. AC-34, no. 7, pp. 674688, Jul. 1989. [14] J. Ackermann, Robust Control: Systems With Uncertain Physical Parameters. New York: Springer-Verlag, 1993. [15] S. P. Bhattacharyya, H. Chapellat, and L. H. Keel, Robust Control: The Parametric Approach. Upper Saddle River, NJ: Prentice-Hall PTR, 1995. [16] R. S. Snchez-Pea and M. Sznaier, Robust Systems: Theory and Applications. New York: Wiely, 1998. [17] A. Vicino and A. Garulli, Eds., Special issue on Robustness in Identication and Control,, Int. J. Robust Nonlin. Control, vol. 11, no. 7, Jun. 2001. [18] L. H. Keel and S. P. Bhattacharyya, Controller synthesis free of analytical models: Three term controllers, IEEE Trans. Autom. Control, vol. 53, no. 6, pp. 13531369, Jul. 2008. [19] D. W. Richardson, B. D. O. Anderson, and N. K. Bose, Matrix-fraction description from frequency samplesm, Linear Algebra Appl., vol. 181, pp. 249266, Mar. 1993. [20] A. Dehghani, A. Lecchini-Visintini, A. Lanzon, and B. D. O. Anderson, Validating controllers for internal stability utilizing closed-loop data, IEEE Trans. Autom. Control, vol. 54, no. 11, pp. 27192725, Nov. 2009. [21] W. Malik, S. Darbha, and S. P. Bhattacharyya, A linear programming approach to the synthesis of xed structure controllers, IEEE Trans. Autom. Control, vol. 54, no. 6, pp. 13411352, Jun. 2008.

REFERENCES
[1] H. Nyquist, Regulator theory, Bell Syst. Tech. J., vol. 11, pp. 126147, 1932. [2] H. W. Bode, Network Analysis and Feedback Amplier Design. Princeton, NJ: Van Nostrand, 1945.

You might also like