Computer-Controlled Systems

Third Edition

Solutions Manual

Karl J. Åström Björn Wittenmark

Department of Automatic Control Lund Institute of Technology October 1997

Preface
This Solutions Manual contains solutions to most of the problems in the fourth edition of str m, K. J. and B. Wittenmark (1997): Computer controlled Systems { Theory and Applications, Prentice Hall Inc., Englewood Cli s, N. J. Many of the problems are intentionally made such that the students have to use a simulation program to verify the analytical solutions. This is important since it gives a feeling for the relation between the pulse transfer function and the time domain. In the book and the solutions we have used Matlab/Simulink. Information about macros used to generate the illustrations can be obtained by writing to us. It is also important that a course in digital control includes laboratory exercises. The contents in the laboratory experiments are of course dependent on the available equipment. Examples of experiments are ? Illustration of aliasing ? Comparison between continuous time and discrete time controllers ? State feedback control. Redesign of continuous time controllers as well as controllers based on discrete time synthesis ? Controllers based on input-output design ? Control of systems subject to stochastic disturbances. Finally we would like to thank collegues and students who have helped us to test the book and the solutions. Karl J. str m Bj rn Wittenmark

Department of Automatic Control Lund Institute of Technology Box 118 S-220 00 Lund, Sweden

i

Solutions to Chapter 2
Problem 2.1

The system is described by

; b u(kh) x(kh + h) = e;ah x(kh) + 1 ; e;ah a y (kh) = cx(kh) The pole of the sampled system is exp(;ah). The pole is real. For small values of h the pole is close to 1. If a > 0 then the pole moves towards the origin when h increases. If a < 0 then the pole moves along the positive real axis.
Problem 2.2
a. Using the Laplace transform method we nd that

x _ = ;ax + bu y = cx Sampling the system using (2.4) and (2.5) gives

8s ;; 1 ; 1 Ah ; 1 ; 1 = e = L sI ; A = L s2 + 1 > : ;1 8 cos h sin h 9 => : ; sin h cos h > 9 8 1 ; cos h 9 Zh Zh 8 sin s > > As ; = e B ds = : cos s ds = > : sin h >
0 0

1> = s

9!

b. The system has the transfer function s+3 = s+3 2 ; 1 G(s) = s2 + = 3s + 2 (s + 1)(s + 2) s + 1 s + 2 Using the Table 2.1 gives

; e;h ; ;1 ; e;2h H (q) = 2 1 q ; e;h 2 q ; e;2h
c. One state space realization of the system is 819 8 0 0 09 > > > > > > > 0 x _ => 1 0 0 x + > > > :0>u :0 1 0 8 9 y = :0 0 1 x 1

1. ). 1)2 : h3=6 > 3 q 2 + 4q + 1) = h (6( q . 0:5y = 6u x(kh + h) = 0:5x(kh) + 6u(kh) x _ (t) = ax(t) + bu(t) y (kh) = x(kh) y (t) = x(t) (discrete-time system) (continuous time system) 2 ah = ln 0:5 ) a = .1u qy . 1) y . 1 = 6 eas b ds = a a 0 12 ln 2 ) b = eah6a = . 0:5q. ) 1 ln A= h y(k) . 1) (q . 1)3 8 9 :0 0 1 Problem 2. 1) = 6u(k .3 = eAh a. =0 easb ds = 6 Zh 0 b eas h = b eah . 1)3 8 x 8 h 9 x x 9 > > > > > > > 2 =2 > x x x h > > > : h2(q + 1)=2 h(q .= and eAs B ds = Zh 8 9 : 1 s s2=2 9T ds = 8 : h h2=2 h3=6 T 0 0 09 > 1 0> > h =2 h 1 H (q) =C (qI .Now 80 0 09 > > > A2 = > 0 0 0 > > A3 = 0 : 1 0 0 then 8 1 > Ah 2 2 e = I + Ah + A h =2 + : : : = > > : 2h Zh 0 .1 y = 6q .1 h 2 . 0:5y(k . = (q . ln h 8 =eah = 0:5 < Rh : .

= : 1 8 8 9. ) No equivalent continuous system exists. cos 2 k (k .1 8 y = :1 Pulse transfer operator 9 8 9 9 H (q) = C (qI . 1) = 6u(k .1 8 1 9 98 q .0:5 1 9 8 0:5 9 8 > > > < x(kh + h) = : x(kh) + > : 0:7 > u(kh) 0 .1 . 8 9> : = q2 1 + 1 1 0 : .1 9 det(sI .u(k) y (k ) = Cx(k) a. 1) = 1 . y(k) + 0:5y (k . Problem 2.0:5 8 s + 0:5 . ) = > : 0 s + 0:3 > = 0 .(h) = > : sin h 1> . Both eigenvalues of on the negative real axis ) No corresponding continuous system exists. 0 : 3 9 > : y(kh) = 8 : 1 1 x(kh) Eigenvalue to : 1 = . = >1> :1 0 0 x 8 0 h= 2 ) => : . 1 z2 + 1 z . x(k + 1) = x(k) + . 8 .0:3 c. A. 8 9 cos h sin h > (h) = : . 1) 3 . 1) . 1) (k .1 q ).b.0:5y (k) + 6u(t) H (q) = q +60:5 one pole on the negative real axis. (s + 0:5)(s + 0:3) = 0 2 = . cos h 9 > .3 and 3. 1 > 0 :1 q> > :1> 98 9 q 1>>1> = q + 1 : 1 q2 + 1 1 z = 1 + 1 Y (z ) = H (z)U (z ) = zz2 + + 1 z .2a).4 Harmonic oscillator (cf. 1 y(k) = sin 2 (k . 1) + (k . y(k + 1) = . sin h cos h > 8 1 .

1 1 G(s) = s2(s + 2)( = s + 3) 6 s2 36 s 4 s + 2 9 s + 3 1 q + 1 . cos t.3 Problem 2.u(kh) y(kh) = Cx(kh) and the controllable form realization is ~ u(kh) z (kh + h) = ~ z (kh) + .3 H (q) = 12 (q . 1 > : G(s) = see Probl.1: 1 1 1 . e. and y (kh) = 1 . 1) is a step at k = 1. kh)u(kh)ds Problem 2. cos 2 k s Y (s) = s(s21+ 1) = 1 s . 1)2 36 q . ) y (t) = 1 . The same way as a.7) is x(kh + h) = x(kh) + .6 Integrating the system equation x _ = Ax + Bu gives x(kh + h) = eAh x(kh) + kh Z +h kh Ah = e x(kh) + Bu(kh) eAs B (s . 2.5 Do a partial fraction decomposition and sample each term using Table 2.2 27 q . 5 1 + 1 1 .2] = 1 0 : 1 s > > : 1 > = s2 + 1 b.2 . cos 4 k Notice that the step responses of the continuous time and the zero-order hold sampled systems are the same in the sampling points. e. Problem 2.7 The representation (2.where (k . 1 1 . 5 1+1 1 . cos t y (kh) = 1 . ~ (kh) y(kh) = Cz 4 . 1 8 q . s2 + 1 9. y (t) = 1 .1 8 0 9 1 8 98 s . e. e.

t22 ht21 + t22 = t12 h2 t + ht = 1 12 2 11 h2 t + ht = 0 22 2 21 h2 t + h2 t = 1 2 11 2 21 h2 t + h2 t = 0 2 12 2 22 Equations (1){(4) now give (1) (2) (3) (4) t11 = t21 = 1=h2 t12 = .9 The system is described by 8 .8 The pulse transfer function is given by 8 98 9 : 1 0 > z .1 C or or ~T = T ~ =C CT > ~=> . = z(z . ) . 0:1) = z2 (z . . 0:5) : 0 z . :1 0 8 9 From Section 2.where 81 => :0 8 C = :1 h> 1 9 0 9 8 h2=2 9 8 2 . ~ = CT . 0:2 2(z .d > x + > : g > u = Ax + Bu 5 .=: h => :1 0> 8 9 ~ = : h2 =2 h2 =2 C ~ = T T .1 9 > > ~ .5 we get This gives the following relations t11 = 2t11 . 0:2 9 8 2 > . 0:5) = z (z . 0:5 : 1 > = z .h 8 9 Problem 2. 0:5) Problem 2. t21 t21 = t11 ht11 + t12 = 2t12 . 1 > H (z ) = C (zI .a b 9 8 f 9 x _ => : c .t22 = 1=(2h) or >2 h> T = 21 h2 : 2 .1 ~ = T.

Using the result in Appendix B we nd that eAh = 0I + 1Ah and e 1 h = 0 + 1 1h e 2 h = 0 + 1 2h This gives 1e 2 h . and d are nonnegative implies that the eigenvalues. 2e 1 h = 8 .9 gives e 1 h = 0:7895 1 = . 1 and 2. 2 1.e 1h . dh 1 0 1 e . 2 1 . we notice that 0= 1= 1 2 Zh 0 0 and 1 1 2 depend on h.The eigenvalues of A is obtained from ( + a)( + d) . ah 9 0 bh 1. 2 .e 2h . bc = 2 + (a + d) + ad . 1 e 2 2h .e 1 = ( . = ( 0 I + 1 A) B = > : 00::0296 6 8 9 . )h To compute . d)2 + 4bc a + d =. 1 . bc = 0 which gives r (a . 2 > > 0 1 1 =: h h 1 2 ch .10 a. are real. 2 4 The condition that a b c. 1 1 1 1h Zh 0 s 1 (s)ds = 1 1 . Using the result from Problem 2. .1 .e . Introduce 0 (s)ds = 1.0:0197 e 2 h = 0:8566 2 = . There are multiple eigenvalues if both a = d and bc = 0.0:0129 Further 0 = 0:9839 1 = 0:8223 and 8 0:790 0 9 = 0I + 12 1 A = > : 0:176 0:857 > 0 = 11:9412 1 = 63:3824 281 > . = 0 B + 1AB Problem 2.1 then .

1 C . e 0 > > > > : = 0 1 : : .s 9 e > > : 1 . ). e. A). 0 : 790 0 > 1 : . e.h .1 9 (sI .h .h 9 Zh 8 . e. e. e.eAh k = eAkh 7 . = : 0 8 0:030q + 0:026 = q2 .18 0:281 9 98 q .s ds = > : h + e.h => : 1 .h > > : h + e.h ) = (h + e z 2 .b.1 . e. where 0> A=> : . 1 > b. e.1 1 >s 0 > s(s + 1) : 1 s + 1 = 8 9! eAs Bds = 8 1 . The pulse response is k=0 h(k) = 0 k .h . The pulse transfer operator is now given by H (q)= C (qI .1 9 8 1 .h )(z . a. he. 1) .h .h )z + e.1 8 9 8 z. k 1 Now k = .2. e.1 8 1 .h Zh 0 = exp(Ah) = L. 0:857 > > : 0:0297 > which agrees with the pulse transfer operator given in the problem formulation. e.h z .h 9 8 98 . 1 h+e .1 + e z . e.1 1 0 8 9 > B=> :1 0 0> 1 0 8 9 C = :0 1 = L. Problem 2.h z .1. 1) : 1 . h . 1)z + (1 .h 9 :0 1 0 > = (z . he.11 The motor has the transfer function G(s) = 1=(s(s + 1)). e.= 8 e.h )(z .h 9.0:176 q . 1)z + (1 . h . 1 > This gives the sampled representation given in Example A. = .h . 1:65q + 0:68 9.h ) = (h + e ( z .2. The pulse transfer function is H (z ) = C (zI .h . ). A state space representation is given in Example A.1 8 9 . (1 + e.

1 > = 1 . = 1 . . (1 + e.12 Consider the following transfer operators of in nity order.h y (kh) = = (h + e. ) = 0 x + 1 u(t .1)h e 0 > >> 1 : : h + e. 1)u(kh + h) + (1 .(k. e.1 to 0 when h goes from zero to in nity.1)h + h + e. e.h = f (h) z = . A di erence equation is obtained from H (q ) y (kh + 2h) . ) 8 .h 1 . 1 The function f (h) goes from .(k. he.h . . Problem 2. The second pole will move from 1 to the origin when h goes from zero to in nity.h . = : 0 9 8 1 . 2. c.h )u(kh) d.(k.h).d as function of the sampling period.1 This gives C k.(k.kh An alternative way to nd h(k) is to take the inverse z -transform of H (z ).h )y (kh + h) + e. h + e.1)h + e. e. e.h .1.1 . e.h . e. The zero is in e. See Fig. The poles are in z = 1 and z = exp(. Figure 2.h .h .1)h 1 .h 9 98 .1 .s G(s) = 1 se h=1 = 0:5 ( < h) (in nite order) x _ = u(t .1 Pole and zero 0 −1 0 10 Sampling period h 20 The pole (solid) and zero (dashed) in Problem 2. he.11. 1 = 8 = h .

Discrete time system is given by (cf. 1) z.13 a.1 h Z. The pulse-transfer function is ).1 z. 1): 0 z . 1) Invers transform of H (z ) using Table 2. 1) z 0:5 > :1> 8 :5(z + 1) = 0z (z . = : 1 9.1 u(k .1 .1 } | {z } | {z k=0 ) h(kh) = 0:5 k = 1 1 k>1 ! (0 =1 k 1 =1 k 2 c. 1) (Notice that < h) = eAh = e0 = 1 . CCS p.3 gives :5(z + 1) = 0:5 z .6 but with = 1:5. 1) + .1 Problem 2. 0 eAs ds B = Z0:5 0 1 ds 1 = 0:5 = eA(h. 1) State space model: 8 x(k + 1) 9 8 1 > : u(k) > = > :0 8 y (k) = : 1 0:5 > > x(k) > + > 0:5 > u(k)= x(k)+ . 1 .0u(k .1 z + z . 1 : 1 = z (z .a. This is the same system as in Example 2.1 u(k . The poles are z = 0 z = 1 and the zeros: z = . 1) > 98 9 8 9 The system is of second order. x(k + 1) = x(k) + . b. ) Z 0 eAs ds B = Z0:5 0 ds = 0:5 ) x(k + 1) = x(k) + 0:5u(k) + 0:5u(k . In this case we have d = 2 and 0 = 0:5 and we get x(k + 1) = x(k) + .0u(k) + .2 z H (z) = 0z (z .1 8 0:5 9 98 z . 38{40).u(k) : u(k . 2) 9 .0 = . 0 : 0 : z 1 8 0:5 9 8 9 8 9 8 9 8 9 1 z 0 : 5 0 : 5 1 > : > > > : = 1 0 z (z . 1) : 1 0 8 9 > x(k) 9 0 : u(k . 0 : 5 > > > > H (z) =C (zI .

sb ds = b 1 .14 Sampling the given di erential equation using the same procedure as in Problem 2. b = e.12) 8 x(k + 1) 9 8 ..1 = 0:39z + 0:24 =z 2 (z .0 k=2 .= e. e.27) we can also get the pulse response by expanding H (z ) in z . e. 3) .0 9..0:5 . ( k .6 + : : : The pulse response is now given by the coe cients in the series expansion.1 . (h. H (z) = 0:39z.1 8 0 9 > > > 8 9> > > > 0 z .2 + 0:38z .4 and the de nition of the z -transform (2.3 + 0:14z. ( k . 2) + 0 > > > > > > > : u(k) : 0 0 0 : u(k . h Using the de nition of 0 and with h = 1 it follows that d = 4 and = 3+ 0 Further h Z. where Problem 2. ) z 2 (z .0 9 8 x(k) 9 8 0 9 > > > > > > > > > > > > > > > > u ( k . 2) > : u(k . There are three poles p1 = p2 = 0 and p3 = 0:37 and one zero z1 = .0:62. ) .0:5 = 0:39 . 0:37) Some calculations give 80 k 1 < h(k) = : . ) 0 3 0 0 b4 = e.1 = 0:24 A state representation is obtained from (2. (h. c. The pulse transfer function is 8 z .1 e k 3 Using Theorem 2.1 . 0 ) Z 0 0 e.1 = 0:37 .0 z + . 1 0 H (z ) = : 1 0 0 > > > > :1> = : 0 0 z . 10 h .0 e + . 1) = 0 0 1 u ( k . (h.1 = e. s b ds 0 = b e.. 2) . e. e.13 gives a = e.0 = 1 .4 + 0:05z. 1) : 1 > u(k) 8 x(k) 9 8 9> > > y (k ) = : 1 0 0 > u ( k .e. i.5 + 0:02z.1 . . 1) > b.

49).10 = = A(q) q10 .q.1 A q . CCS p. 0:5q.1 . 9) 8 . )> . b3 + b4 Thus and 0 ) 0 = 1 ln ab3 + b4 + h b3 + b4 ! 1 ln ab3 + b4 = 4 . 0:5q 1 1 . d) : B q. 0:5y(k + 9) = u(k + 1) + 0:2u(k) (q 10 .1 d=9 Remark Pole excess = deg A(q ) . 1 . 9 . .1 y(k) = . d . . =q =q . (h. 0:5q .10 q 10 . 0:5q9 System order= deg A(q ) = 10 B (q) = q + 0:2 Also .1 u(k .1 = 1 + 0:2q. 0:5y (k .1 y (k) = B q. 1) = u(k . ln a b3 + b4 where it has been used that = . q .9 + 0:2q.1 . 0:5q .1 11 .1 u(k .1 + 0:2q.Straightforward calculations give ab3 + b4 = e. deg B (q ) = 9 = d (cf..1 .1 > < A q = 1 .1 .1 + 0:2q.15 y(k) . 10 (q + 0:2) B (q) = q + 0:2 = q. 10) . 0:5q 9)y (k) = (q + 0:2)u(k) ( A(q) = q10 . ln a ! Problem 2. 0:5q9 q.1 B . 0:5q. y(k + 10) . 9) + 0:2u(k . A q.

a 8 n C = :1 0 Problem 2.j . 1:5z Y .1) = 1 Z qnf = zn F .1 + : : : + bn q .n u(k) = = b0u(k) + b1 u(k .2 (page 57): y (0) = 0:5 y (. y(0) + 0:5Y = z U .1 : : : + bn ) H (q) = b0q + b1q qn + : : : + bn = b0 + b1q + qn (q ) H (q) = D + B A(q) n:th order system b. > 1> : > bn 0 y(k + 2) . 1) + : : : + bnu(k) ) qny(k) = b0qn + b1qn. y(0) .n y(k) = b0 + b1q.Problem 2.1 X j =0 f (jh)z. ) z2 Y . 1) + : : : + bn u(k .1 + : : : + bn u(k) n n. H q. F1 12 . 1:5y (k + 1) + 0:5y(k) = u(k + 1) .17 80 1 0 ::: 1 0 ::: 09 > > > 0 1 ::: 0> 0 0 1 > > = > > 1> > :0 0 0 9 ::: 0 D = b0 8 09 > b1 9 > > > 0>. y(1)z.1 = b0 + b1q.1 n. > > > . = > > . .a2 => > > : . Observable canonical form: 8 . 1:5qy (k) + 0:5y(k) = qu(k) Use the z -transform to nd the output sequence when 0 k<0 u(k) = 1 k 0 Table 2.16 FIR lter: . u(0) F1 (z ) = n. . .1 . . q2y(k) .a1 > > . n) ) y(k + n) = b0u(k + n) + b1u(k + n . .1 + : : : + bnq.

1) (z . 1 = 21 = 1+ 3 2 4 4 8 13 . ln 2 = 8 2 8 y (2) = u(1) + 1:5y (1) . 1:5z + 0:5 U (z ) = z :5z (z . 0:5 Table 2.k=T Z .1=T = e 2 Z .z2 .Compute y(1) =u(0) . 1)2 = Z z (z . 1) + e. 1 :5z + z2 0:5z + 2 + 1 = Y (z ) = z 0 2 .1 z. 0:5z + z Y (z) = z 20:. e e. 1:25z.(k. 0:5) U (z) z (step) ) U =z . 1:25z + 0:75z = zU . 1:5z(Y .1 .z0:5 = e. 1)2 z .1) =1 Checking the result: 1 + 2 + 1 = 21 y (2) = 0:5e. 1:5z + 0:5 Y . 0:5z2 . 0:5 .1 z .1 (z . 1) + = (z 0 . 1) . 2 1:5z + 0:5 z . z 5z 2 .1 z .2ln2 + 2 + e. 0:52 5 . z 1 .1 z Z . 1)(z .1=T = 0:5 ) T = ln . 0:5) z . 1 (h = 1) ) y(k) = 0:5e.25 < (0) =0. 0:5y(.1) + 1:5y(0) = 1 . 0:5 + 0:75 = 1:25 ) z2 Y .3 (page 59) gives via inverse transformation. 0:5 (z . 1)2 = k . 0:5) (z .1) ln2 1 . 0:5y (0) = 1 + 1:5 1:25 .5 :y y(.1 .(k. 1)(z . 0:5 (z . 0:5) + 0:5Y = z(U .k ln2 + 2(k .1) ln2 8 y(1) =1.

k 0 h2 2 z (z + 1) ) F (z) = 2 f (z. 1)2 cf.3) 2 1 F (z ) = 1 1 X d dz z . 1)3 z 2 (z . 1)3 1 X .k z )(multiply by z) kz = (z . Double integrator: The rst step is to translate G(s) to the corresponding pulse transfer operator H (z ). 1 f (z ) = k2z . The sampled system.. ).k = h 2f z X z z.k = d z = (z . 1) .k. In that case there exists a power series expansion and 2 z (z + 1) (Table 3:3) ) Y (z ) = h 2 (z . . multiplication by z 2 gives ) (k2 + k)z .Problem 2. termwise di erentiation is allowed. 1)2 (z .1 k 2 z . z ) U (s) = 1 u(k) = step ) U (z ) = z . Table 2. 1)3 Remark A necessary condition for convergence of f (z. 1)3 z2 . 1 = h2 (z + 1) H (z ) = Y U (z) 2 (z . 1)3 (z .1) = h 2 (z .k.1 (z .k. 1)3 Y (z) = H (z )U (z ) ) (z ) = h2 z (z + 1) z .1 .2 = P(k2 + k)z. 1) = 2 > : dz 2 z.1 (z . 1)2 X . 1)z.2 d2 z = 2(z . 1 s 1 1 Y (s) = G(s) s = s3 14 .k = h ( kh ) 2 2 X k=0 2 .k z = k=0 2 z .k = P .k(. > < dz2 z.18: Verify that 2 2 = h z (z + 1) Z 1 ( kh ) 2 2(z .k = (z 2 .kz. z = . 1)3 (cf.1 di erentiate twice 8 d2 . z = z (z + 1) = (z 2 . 1)4 (z .1 = dz z. example A1: H (z ) = C (zI . 1)2 (z . 1)3 z 2 . Use the method of page 58.1) is that f is analytic for jz j > 1. 1)2 k=0 1 X .k .1.

21 s+ is lead if < s+ Consider the discrete time system z+b z+a ei!h + b = arg cos !h + b + i sin !h = arg e i!h + a cos !h + a + i sin !h sin !h . e.h .19 a. The normalized motor has the transfer function G(s) = s(s 1 + 1) and we get X 1 esh .1 e.1e. 1 + (z .ah . e.a 1 .ah This is the same result as obtained by using Table 2.h )(z . 1 bc = = z . The transfer function of the continuous time system is G(s) = (s bc + a) Equation (2. 1) Problem 2.1.1) 2 + z. e.ah s=. 1)=s and get . 1 1 H (z ) = Res sh s s(s + 1) s=si z . 1 Res e.h . e.11.h ) = (e .ah = bc a z . arctan sin !h = arctan b + cos !h a + cos !h 15 . e To compute the residue for s = 0 we can use the series expansion of (esh .ah . 1) + h(z . 1 h H (z ) = z .a s s+a e.Problem 2. 1 bc H (z ) = z . e. 1)( (z . 1) Compare Problem 2.30) gives esh . he.h e(. b.h . e.h h)z + (1 .1 = . e.h ) = = (z .h )(z .

e. and 1 (lower dashed). 0 (dotted). a)u y =x+u The assumption that the system is stable implies that a > 0.23 A state space representation of G(s) is x _ = . Sampling this system gives a u(kh) x(kh + h) = e.0:4 9 8 1 9 x(k + 1) = > : 1 0 >x+> :0>u 9 1 8 : 1 b x(k) y (k ) = b + 1 Simulation of the system for b = . Phase lead if 0 < !h < A state space representation of the system is 8 1:1 .0:50 (dash-dotted). .0:75 . 10 Time 20 sin!h > arctan sin !h arctan b + cos !h a + cos !h sin !h > sin !h b + cos !h a + cos !h We thus get phase lead if b < a.ah x(kh) + (1 .22 for = .0:9 .22 Problem 2. .0:75 (upper dashed). a y (kh) = x(kh) + u(kh) 16 .2 b Simulation of the step response of the system in Problem 2.0:9 (upper solid). Problem 2.10 Output 0 0 Figure 2.2.ax + (b . 0:5 (lower solid).ah ) b .0:5 0 0:5 and 1 is shown in Fig. 2.

ah)=a + 1 = H (q ) = (b .ah + = qq .ah ) > 0 then a 0 < b < 1 . e.ah where b . e. 1 < 1 a or b (1 . 1) + y (k .ah .The pulse transfer operator is . e. e. aq)(1 . a 1.ah .e . 2) y (0) = y (1) = 1 The equation has the characteristic equation z2 .1 .2e .e = b .1 .ah . z . e. 1) c2 = p 2 5 17 .ah = b.28 y (k) = y(k . e. e. b 2a b > 2a The inverse is stable independent of h: Stable inverse if ah < ln(b=(b . 1 =a .ah For b > 0 we have the cases The inverse is stable if a . Problem 2. 1 = 0 which has the solution The solution has the form z12 = 1 2 5 p y (k) = c1 1 +2 5 Using the initial values give p k + c2 1 .2 5 p k 1 (p5 + 1) c1 = p 2 5 1 (p5 .ah ) < 2 0< a Since a > 0 and (1 . 2a)).

1 . Sampling G2 gives H (z ) = 0. 0:8 H4 (q ) = 2 + 0 q(q . This implies that H3 cannot be obtained by sampling a continuous time system. (see Table 2.29 Problem 2.The system is given by .2q. 0:8) The second part can be obtained by sampling a rst order system with a time delay.11 u(k) Multiply by q 11. 0:5q.e. 0:5q + 1 y (k) = (2q + 1) u(k) The system is of order 11 (= deg A(q )) and has the poles p 1 i p12 = 4 15 p3:::11 = 0 (multiplicity 9) and the zero z1 = . This gives . q 9 q 2 . The only way to get two poles on the negative real axis as in H3 is when the continuous time system have complex conjugate poles given by 2=0 s2 + 2 !0 s + !0 Further we must sample such that h = =! where p ! = !0 1 .1) )(q + ) = 1 + H (z ) = (1 + (q + )2 q+ where = e.0:5 The system H1 has a pole on the positive real axis and can be obtained by sampling a system with the transfer function G(s) = s K +a H2 has a single pole on the negative real axis and has no continuous time equivalent. 2 We have two possible cases 2 0 i) G1(s) = s2 + 2 ! 2 !0 s + !0 s ii) G2(s) = s2 + 2 ! s + ! 2 0 0 Problem 2. we get a pole zero concellation. H4 can be rewritten as :9q . 18 .8.1 + q.30 Sampling G1 with h = =! gives.. Compare Example 2. !0 h i.10 + q.2 y(k) = .

Problem 2.31

We can rewrite G(s) as 1 + 1 G(s) = s + 1 s+3 Using Table 2.1 gives 1 ; e;h + 1 1 ; e;3h H (q ) = q ; e;h 3 q ; e;3h With h = 0:02 we get

:0392q ; 0:0377 H (q ) = (q ;00 :9802)(q ; 0:9418)

Problem 2.32 8

The pulse transfer operator is given by where

9 819 1 0 > x _ = :1 1>x + > = 0:2 h = 0:3 : 0 > u(t ; ) 9;1 8; 9 ;1 8 s ; 1 0 > ; 1 ; 1 : =L = L : ;1 s ; 1 > = sI ; A 8 1 9 > 8 eh 0 9 0 > > > s ; 1 ; 1 > > =L > 1 => : heh eh > > 1 > > : (s ; 1)2 s ; 1 h 8 es 9h; h 9 Z; 8 Z; 8 s 9 e 0 > > > > > ;0 = : ses ds = : ses ; : es > ds = 0 0 0 8 9 8 9 h ; h; e ; 1 ; 1 + e > > > > =: =: h ; h ; h ; (h ; )e ; e + 1 1 + (h ; ; 1)e 8 eh; 9 8 ;1 + e 9 Z8 ;s 9 e 0 > > > A ( h ; ) ;1 = e : se;s ds = : (h ; )eh; eh; > > : 1 + ( ; 1)e > = 8 0 eh; (;1 + e ) 9 > > = : h; h e (1 ; h + ) + (h ; 1)e
H (q) = C (qI ; );1(;0 + ;1q ;1)

8 1:350 0 9 => : 0:405 1:350 >

8 0:105 9 ;0 = > : 0:005 >

8 0:245 9 ;1 = > : 0:050 >

19

Solutions to Chapter 3
Problem 3.1

Jury's criterion is used to test if the roots are inside the unit circle. a. 1 ;1:5 0:9 0:9 ;1:5 1 2 = 0:9 0:19 ;0:15 ;0:15 0:19 1 = ;0:15=0:19 = ;0:79 0:07 The roots are inside the unit circle since the underlined elements are positive. (The roots are 0:75 0:58i.) b. 1 ;3 2 ;0:5 ;0:5 2 ;3 1 3 = ;0:5 0:75 ;2 0:5 0:5 ;2 0:75 2 = 2=3 5=12 ;2=3 ;2=3 5=12 1 = ;8=5 ;13=20 One of the underlined elements is negative and there is thus one root outside the unit circle. (The roots are 2:19 and 0:40 0:25i.) c. 1 ;2 2 ;0:5 ;0:5 2 ;2 1 3 = ;0:5 0:75 ;1 1 1 ;1 0:75 2 = 1:33 ;0:58 0:33 0:33 ;0:58 1 = ;0:57 ;0:39 There are two roots outside the unit circle. (The roots are 0:35 and 0:82 0:86i.)

20

d.

One root is outside the unit circle. (The roots are ;5 ;0:5, and 0:5.) e. 1 ;1:7 1:7 ;0:7 ;0:7 1:7 ;1:7 1 3 = ;0:7 0:51 ;0:51 0:51 0:51 ;0:51 0:51 2=1 0 0 0 0 The table breaks down since we can not compute 1 . There is one of the underlined elements that is zero which indicates that there is at least one root on the stability boundary. (The roots are 0:7 and 0:5 0:866i. The complex conjugate roots are on the unit circle.)

1 5 ;0:25 ;1:25 ;1:25 ;0:25 5 1 ;0:56 4:69 6 6 4:69 ;0:56 63:70 54:92 54:92 63:70 16:47

3 = ;1:25 2 = ;10:71 1 = 0:86

Problem 3.2

The characteristic equation of the closed loop system is

z(z ; 0:2)(z ; 0:4) + K = 0

K>0

The stability can be determined using root locus. The starting points are z = 0 0:2 and 0:4. The asymptotes have the directions =3 and ; . The crossing point of the asymptotes is 0:2. To nd where the roots will cross the unit circle let z = a + ib, where a2 + b2 = 1. Then (a + ib)(a + ib ; 0:2)(a + ib ; 0:4) = ;K Multiply with a ; ib and use a2 + b2 = 1.

a2 ; 0:6a ; b2 + 0:08 + i(2ab ; 0:6b) = ;K (a ; ib)
Equate real and imaginary parts.

a2 ; 0:6a ; b2 + 0:08 = ;Ka b(2a ; 0:6) = Kb
If b 6= 0 then

a2 ; 0:6a ; 1 ; a2 + 0:08 = ;a(2a ; 0:6) 4a2 ; 1:2a ; 0:92 = 0
21

;

The root locus may also cross the unit circle for b = 0. A root at z = .1 is obtained when a12 = 0:15 p . 1 The regulator is Problem 3. 0 Re 1 The solution is n :652 0:0225 + 0:23 = 0:15 0:502 = . ) . .1 Im 0 −1 −1 Figure 3. 0:4) + K = 0 K = . 0:2)(1 . 0:6 = 0:70 and .3 u(kh) = K uc (kh .1.1 . a = 1.2. ) where K > 0 and = 0 or h. 22 . ) = Ke(kh . Sampling the system G(s) gives the pulse transfer operator h H (q ) = q .1:30.1 The root locus for the system in Problem 3. y (kh .e.0:48 The closed loop system is thus stable for K 0:70 The root locus for K > 0 is shown in Fig. 0:4) + K = 0 or There is a root at z = 1 when or K = 1:68 1(1 . i.1(. 0:2)(.0 0:352 This gives K = 2a .1 . 3.

!c > . 23 . The system is thus stable if K < 1=h b.s Go (s) = K se The phase function is and the gain is arg Go (i! ) = . The transfer function for the open loop system is . When = 0 then regulator is u(kh) = Ke(kh) and the characteristic equation of the closed loop system becomes Kh + z . =2 . z + Kh = 0 The constant term is the product of the roots and it will be unity when the poles are on the unit circle.a. 1) = z2 . =0 K < =2 = 1 =h 2h . This value is about 50% larger than the value obtained for the sampled system in b. 1 = 0 The system is thus stable if K < 2=h When there is a delay of one sample ( = h) then the regulator is u(kh) = K q e(kh) and the characteristic equation is Kh + z (z . w The system is stable if the phase lag is less than frequency jGo(i!c)j = 1 ) !c = K The system is thus stable if jGo(i!)j = K ! at the cross over The continuous time system will be stable for all values of K if = 0 and for K < =(2h) when = h. 2 . Consider the continuous system G(s) in series with a time delay of seconds.

0:67 0 The Nyquist curve is shown in Fig. ! ReHo ImHo 0 2 0 =6 0:97 . 0:5 . 0:5 .2 The Nyquist curve for the system in Problem 3. 24 .1:16 =2 .1 Im 0 −1 −1 0 Re 1 Figure 3. cos ! . 0:5 The argument is .2=3. 0:5). i sin ! = cos ! 1:25 .2.0:80 2 =3 . The following table gives some values for the real and imaginary parts. cos ! For ! = 0 then Ho (1) = 2 and for ! = then Ho (. 3.0:57 . Problem 3. 0:1 5 + i sin ! cos ! .1) = .arctg cossin ! .0:40 . The argument is ! arg Ho = .4.0:50 . =2 for ! = =3(cos ! .1:32 =3 0 . i sin ! cos2 ! + sin2 ! + 0:25 . 0:5 = = cos ! . In this case 1 Ho(ei! ) = ei! .4 The Nyquist curve is Ho (ei! ) for ! = 0 ].

Problem 3. The observability matrix is 8 C 9 8 2 .5 Consider the system 81 x(k + 1) = > :1 8 y(k) = : 0 0 > x(k) + > 1 > u(k) :0 1 9 1 x(k) 9 8 9 We have y (1) = x2 (1) = 0 y (2) = x2 (2) = x1(1) + x2(1) = 1 Further ) x1(1) = 1 1 1 1 0 3 The possibility to determine x(1) from y (1).4 9 Wo = > :C > = > : 1 .2 > The system is not observable since det Wo = 0. From the input u0 we get the system 81 0 9 809 > > x(k + 1) = : 0 0:5 x(k) + > : 1 > u0 (k ) In this case rankWc = 1 and the system is not reachable from u0 . = : 4 1 > det Wc = 2 and the system is reachable. .u(1) = > :0> 1 = > :1> :1 1>> :0>+ > 81 09829 819 819 > > > > > > x(3) = : : + : (. = > :1 1 0 0:5 9 80 0 9 Wc = > : 1 0:5 > . the rst two colums are linearly independent. 81 09819 819 829 x(2) = x(1) + . b.6 a.1) = > : > Problem 3. . y (2) and u(1) is called observability.Problem 3. This implies that Wc has rank 2 and the system is thus reachable. 25 8 9 8 1 1 Wc = : . The controllability matrix is 9 8 9 8 6 1 > : Wc = .7 The controllability matrix is 1> 0 For instance.

8 a. 8 not full 9rank T : 1 1 1 is not in the column space of Wc and can therefore not be reached from the origin. Two steps. 0K :5q + K uc(k) Using the conditions in Example 3. c. 0:5) + Kq = q(q2 .Problem 3. 80 1 09 8 9 > > > 2 : Wc = . With Hc = K where K > 0 we get y (k) = q2 . = > 0 0> > :1 0 0 0 ) not reachable. 80 > x(k + 1) = > > :0 0 80 > x(1) = > > :0 0 80 > x(2) = > > :0 1 0 0 1 0 0 1 0 0 0 809 29 > > > 3> 1> > x(k) + > > > u(k) : 0 0 9 8 9 8 8 3 9 2>>1> > 0 9 > > > > > > > > u (0) = 3 + u (0) 3> 1 + >> > > > > > > : : : 0 0 0 1 8 3 9 8 0 9 8 3 + u(0) 9 29 > > > > > > > > > > > > > 3> 3 + u (0) + u (1) = u (1) >> > > > > > : : : 0 0 0 0 ) 8 9T : ) x(2) = 0 0 0 u(0) = . Problem 3. 1:5q + 0:5 + K ) 26 . It is easily seen from the state space description that x3 will be 0 for all k > 0.2 we nd that the closed loop system is stable if K < 1. in general it would take 3 steps since it is a third order system. .3 u(1) = 0 b. With an integral controller we get Kq Hcl (q) = q(q .11 The closed loop system is HcH u (k) y(k) = Hcl(q )uc (k) = 1 + HH c c a. The steady state gain is K Hcl (1) = K + 0:5 b. . 1)(qKq . but may be controlled.

and we get z Uc (z) = (z .3 shows the continuous time and the sampled step responses when h = 2 =! = 6:3148.3. 27 Consider the system . y(k) lim e ( k ) = lim k!1 k!1 c . z . The frequency ! will then have the alias ! 0 = 0. 1)2 Using the pulse transfer functions from Problem 3. 1 z(z2 . where = exp(. )2 = q . 0:5z + K . 1) K + 0:5 The derivative is positive in steady-state and the reference signal and the output are thus diverging. 2 = 0:99 rad/s The sampled system will not detect this oscillation if h = k2 =! . To formalize the analysis we can sample the system with h = 2 =!. 1 (1 . Fig. (q .13 1 G(s) = s2 +s0+ :2s + 1 The damping of the system is = 0:1 and the undamped natural frequency is !0 = 1. also compare Example 3. 1:5z + 0:5 + K ) (z . ) 1 . 3. 1)2 K Problem 3. The pulse transfer function is (Table 2. The step response of the system will have an oscillation with the frequency p p ! = !0 1 . 2 0 (k) = lim z . 1:5z + 0:5 + K . 0:5z + K (z .0:1h).e. H (q) = (1 . To use the nal value theorem in Table 2. K ) z = 0:5 lim e ( k ) = lim z!1 z k!1 z (z 2 .1)F (z ) does not have any roots on or outside the unit circle. For the integral controller we get z . b. K z = 0:5 lim e z !1 z k!1 z 2 .The system is stable if 0:5 + K < 1 and we get the condition and 0:5 + K > . Problem 3.11 and the nal value theorem gives .13.u (k) .12 0 < K < 0:5 The steady state gain is Hcl (1) = 1. i.3 it is required that (1 . H (z) U (z) = zlim c c !1 z if K is chosen such that the closed loop system is stable. a. For the proportional controller we must then look at the derivative of the error. z . There is a pole zero cancellation and only the rst order exponential mode is seen at the sampling points.1 + 1:5 The z -transform for a ramp is given in Table 2. 1 z .1) )(q .

1:65)z + 0:68 + 0:026K = 0 The closed loop system is stable if 0:68 + 0:026K < 1 0:68 + 0:026K > .3 25 Time 50 Problem 3. 0:030K This gives the stability region .2 Output 1 0 0 The continuous time (solid) and sampled step (dots) responses for the system in Problem 3. 1:65 + 0:030K 0:68 + 0:026K > . Hc(1) = 0:03 + 0:056K The minimum error is about 4%. 28 . Figure 3. 1:65qK + 0:68 + K (0:030q + 0:026) The characteristic equation is z2 + (0:030K .1 + 1:65 .0:54 < K < 12:31 The steady state gain is 0:056K Hc (1) = 0:03 + 0:056K The steady state error when the input is a unit step is 0:03 e(1) = 1 .1 .14 a.13. When Hr = K then the pulse transfer function for the closed loop system is (0:030q + 0:026) Hc (q ) = q2 .

Figure 3. 1:65z + 0:68)(z .6 show the step responses when the proportional controller with K = 0:5.2 1 Im 0 −1 −2 −3 −2 −1 Re 0 1 The root locus when the tank system in Problem 3. and 2 and the integral controller with K = 0:01. 1 (dashed).5 250 Time 500 b. 1. Fig. Figure 3. If the closed loop system is stable then the steady state error is zero if the integral controller is used. 0:025. 3. and 0:05 are used on the tank system.14 when the proportional controller is used with K = 0:5 (solid).5 and 3. 1) + Kz (0:03z + 0:026) = 0 A scetch of the root locus is shown in Fig. 3.14 is controlled with an integral controller.4 1 Output 0 0 Step response of the system in Problem 3. 29 . c. The characteristic equation is then (z 2 .4. and 2 (dash-dotted).

1 Output 0 0

Figure 3.6

Step response of the system in Problem 3.14 when the integral controller is used with K = 0:01 (solid), 0:025 (dashed), and 0:05 (dash-dotted).

1000 Time

2000

Problem 3.16

The pulse transfer function for the closed loop system is

Ho (z) Hc (z) = 1 + H (z)
o

and the characteristic equation is

z 2 + z + 0:16 + K (4z + 1) = z2 + (1 + 4K )z + 0:16 + K = 0
Using the conditions in Example 3.2 give 0:16 + K < 1 0:16 + K > ;1 + 1 + 4K 0:16 + K > ;1 ; 1 ; 4K This implies

Since it is assumed that K > 0 we get the condition 0 < K < 0:053 for stability.

K < 0:84 16 = 0:053 K < 0:3 16 = ;0:432 K > ; 2:5

30

Problem 3.17

Using (3.17) we nd that the transformation is given by ~ c Wc;1 Tc = W ~ c is the where Wc is the controllability matrix of the original system and W controllability of the transformed system.

9 8 9 8 3 11 > : Wc = ; ; = : 4 11 >
];1; = : 5

The pulse transfer function of the system is

H (z ) = C zI ; z+4 = 39 z 2 ; 3z

8

9;1 8 3 9 98 z ; 1 ; 2 > 6 : ;1 z ; 2 > > :4>

Thus the transformed system is given by 0> ~ => :3 1 0 and

8

9

> ~=> ; :1 0 8

8 9

~ = : 39 4 C

8

9

8 9 1 3> ~c = : ; ~ ~; ~ => W :0 1 The transformation to give the controllable form is thus 8 Tc = > :1 0 98 9 8

9

3 > > 3 11 >;1 = 1 > 1 2 > 1 : 4 11 11 : 4 ;3

9 9

In the same way the tranformation to the observable form is given by
;1 5 6 1 0 6> > > > ; 1 ~ To = Wo Wo = : 3 1 : 11 22 > = > : ;5 4 4

8

9 8

9 8

Probem 3.18
a) i b) c) d) e) f) g) i iii iii ii ii iii

Poles are mapped as z = esh . This mapping maps the left half plane on the unit circle. see a) When a system with relative degree > 1, \new" zeros appear as described on p. 63 CCS. These zeros may very well be outside the unit circle. See Example 2.18 p. 65 CCC Sample the harmonic oscillator, 8 1 0 9Example A.3 p. 530 CCS. For h = 2 =! = > : 0 1 > not controllable See e) See p. 63, CCS 31

Problem 3.19
a. The controllability matrix is

8 Wc = : ; ;

80 1 09 9 > > 2; = > 2 0> > > :1 2 3 0

Since one column is zero we nd that the system is not reachable (det Wc= 0), see Theorem 3.7. b. The system may be controllable if the matrix is such that n x(0) = 0. In this case 80 0 09 > > > 3=> 0 0 0 > :0 0 0> and the origin can be reached from any initial condition x(0) by the control sequence u(0) = u(1) = u(2) = 0.

Probem 3.20

Ho = q2 +10:4q Hr = K

a.

K Hr Ho = Htot = 1 + 2 Hr Ho q + 0:4q + K The system is stable if, see p. 82,

) K<1 K > ;1 + 0:4 K > ;1 ; 0:4
b.

) ;0:6 < K < 1

e(k) = uc ; y ; E (z ) = 1 ; Htot(z) Uc(z) If K is chosen such that the closed loop system is stable (e.g. K = 0:5) the nal-value theorem can be used. z ; 1 E (z) = lim z ; 1 1 ; K z lim e(k) = zlim !1 z z!1 z k!1 z 2 + 0:4z + K z ; 1 = z2 + 0:4z = 1:4 = 0:74 K = 0:5 = zlim !1 z 2 + 0:4z + K 1:4 + K

Problem 3.21

:4q + 0:8 u(k) y (k ) = q 2 0 ; 1:2q + 0:5

32

Fig.7 The root locus for Problem 3.2 Im 0 −2 −2 Re 0 2 Figure 3. Hr Ho = (0:4q + 0:8)K Htot = 1 + Hr Ho q2 .0:25 . HrHo = (0:4q + 0:8)K Htot = 1 + 2 Hr Ho q(q .7. see p. a.1 + 0:4K . 0:5 + 0:8K < 1 ) 0:5 + 0:8K > . 1:2) + 0:5 + 0:8K The system is stable if.21b. 82 CCS.0:25 < K < 16 Hr = K q 33 . 1:2q 2 (0 + q (0:5 + 0:4K ) + 0:8K Using root locus. 1:2q + 0:5) + K (0:4q + 0:8) = :4q + 0:8)K = q 3 . we can determine that the closed loop system is stable if p .0:25 < K < 0:625 b. 0:4K + 1:2 ) K < 0:625 K > . 1:2 ) 0:5 + 0:8K > .1 . 1:2q + 0:5 + K (0:4q + 0:8) = :4q + 0:8)K = q 2 + q (0:4(0 K .6:75 K > . 17 + 489 . 3.

For the double integrator with h = 1 and dead beat response we have a11 = a12 = a22 = b2 = 1.2 In this case the desired characteristic equation is (z . a12a21 + (a12b2 . 0:1)(z . a12b2 + b1b2 (a22 . (a11 + a22 .1 : `2 : . ( .b2 > > p1 + tr : . b1`1 )z + a11 a22 . a12 a21. 0:35z + 0:025: Using the result from Problem 4. a11 ) 6 0 we form the controllability matrix of the system = = det Wc There exists a solution to the system of equations above if the system is controllable. 1 >> 2> = > 1 > > 1 > = (. a11b2 > : `1 : p . det where tr = a11 + a22 and det = a11a22 . a11b2 .4.a12b2 + a22b1 b1 : p2 . .1 : 1:5 This is the same as given in Example 4.L)) = z 2 . a22b1)`1 + (a21b1 . b2`2 .1 we nd that = 0:5 34 .1) > . Problem 4. 0:25) = z 2 . = : b a b + a b > 2 21 1 22 2 2 = a21b2 1 . . b1 = 0:5. and p1 = p2 = 0. det 2 2 8` 9 > >= : `1 2 1 > a21b1 .Solutions to Chapter 4 Problem 4. The solution is 8 98 9 > where To check when 9 8 9 8 + a b b a b > 1 11 1 12 2 : Wc = .1 The characteristic equation of the closed loop system is det(zI . a22b1 a21b1 . This gives and we nd that 8` 9 8 98 9 8 9 . a21 = 0.0:5 : . a11 b2)`2 Identifying with the desired characteristic equation gives 8 b 9 8 ` 9 8 p + tr 9 b2 > > >=> > 1 1 : a12b2 .0:5 .

) = 2(1 . 1 + ) = h(1 . h .0:025 98 9 8 9 To check the result we form 8 1 0:1 9 8 0:75 0:1 9 8 0:25 0 9 . 8 2 (1 + ) + 2 (1 . 2 + 1 2e. h . 1+ > where = e. )3 + (1 . 0> 1 9 8 1. ) 1 > > = : 1. 2e. h (1 . = 1 h(1 . ) 9 (1 .1. The dead beat controller is characterized by p 1 = p 2 = 0: From Problem 4. e.h .=> :h. )2 then .L thus has the desired eigenvalues 0.25 and 0.3 For the motor in Example A. 1 T L = > : 1. 81. 1 + ) 1 . 8 9T = (1 .1. 1 . )2 or .2 we have 8 => :1.L = > : 0:5 0:1 > . .h + 1 This is a nonlinear equation of the form h = f (h) 35 . 9 .h 2 +1. Problem 4. 1. 4.and L is obtained from 8` 9 1 >= T L => : `1 2 8 0:5 > : 0:1 0 > > 0:75 > = > 0:75 > : 0:1 1 : . )2 This function is shown in Fig. )2 (h . (h . . We thus want to nd h such that 1. .h 2 +1.h 29 1 = > : 1. > where If x(0) = : 1 1 981+ 9 >> > : . )2 . > : 0 0 >=> : 0:5 0:1 > The matrix . u(0) = .h ) h = 2 2(1 2 .2h . `2 = .`1 .1 it follows that 8 (1 . .

0:0880 .63 0.0 Output u(0) −2 0 1 Figure 4.P) 36 .0:1595 = : 3:11 8 98 9 8 9 The problem is easily solved using Matlab by giving the commands fi= 0.12 0 0.3. 2 3 Sampling period h 4 5 One way to nd h is to use the iterative scheme hk+1 = f (hk ) Starting with h0 = 2 gives k 0 1 2 3 4 hk+1 2 2:26 2:20 2:21 2:21 with h = 2:21 we get L = : 0:49 0:51 .2 shows the response of the motor controlled with a deadbeat controller when h = 2:21.0:1600 > > 0:5900 > > 9:22 > = .1 gives LT > . We see that the system settles after two samples and that the constraint on the control signal is ful lled.55 0.16] P=roots( 1 -0.21]) L=place(fi.4 a.ga. Fig.1 u (0) s function of h in Problem 4. 8 9 Problem 4.0:0125 0:0100 : .01 0. Using the results in Problem 4.67] ga= 0.0:1 0029 : . 4.

0:63 p2 = e. . b. > = > : 0:03 > Wo.5 a.4 we nd that the continuous-time characteristic polynomial s2 + 2 !s + ! 2 corresponds to z 2 + p1z + p2 with p1 = . From Example 4.1 = > : . since !h 1:1 > 0:6 8 9 c.2 !h = 0:21 This has the solution 0:7 and ! 5:6.4:55 4:55 9 8 0 9 8 0:114 9 >> > > > => : 0:03 > .41:55 8 4:55 > 0 9 = .63 0. In Matlab you can do rd=roots( 1-0.3 when x(0) = : 1 0 . > : 0:22 1 > > : 1 0 : 0:03 = : 0 37 .1 Output 0 −1 0 1 5 10 Input 0 −1 0 5 Time Figure 4. Wo. Problem 4.21]) rc=Log(rd)/h Ac=poly(rc) p The chosen sampling interval is higher than recommended by the rule of thumb. In this case 9 8 9 8 C 0 1 > > > Wo = : C = : 0:22 1 > 809 8 0 9 => : C.2 Deadbeat control of the motor when x 8 9 T (0) = : 1 1 and 10 h = 2:21. !h cos(!h 1 . 2 ) = . The closed loop system when using L = : 9:22 3:11 is shown in 8 9T Fig. 4.2e.1 8 0:22 9 8 0:78 0 9 8 . so the characteristic polynomial becomes s2 + 7:8s + 31:7.

1)) + Ky (k): In this case we want to nd K such that i. The dynamical observer (4.u(k) + Ky (k): In this case we choose K such that the eigenvalues of . Further 8 1 .3:55 3:55 9 8 y(k . KC )( x ^(k .28) has the form x ^(k + 1jk) = ( .1 but with T and C T instead of and .32) has the form x ^(kjk) = (I .1 Output 0 0 5 0 Input −5 −10 −15 2 4 0 Figure 4. The reduced order observer (4. gives 8 2:77 9 K=> : 1:78 > c. 1jk .u(k . 1) => : 0 1 : y (k) + : 0 k . KC ) = > :0 01>> : 0:22 0 > = > 0:78 . KC ) has eigenvalues in the origin. 1) + . L : : 2 Time 4 We get x ^(k) = 8 . 1) + .4 (0) = : 1 0 and = : 9 22 3 11 .k1 > : 0 0 0 9 8 9 38 .3 when x The response 8 9T and the 8 control signal 9 of the system in Problem 4. The rst condition implies that k2 = 1.k 9 8 0:78 (I . KC are in the origin. KC )^ x(kjk . 1) b. 0:22k1 . (I .1 > : y(y (k) 8 9 u(k . Using the results from Problem 4. CK = 1 ii. 1) > + Wo. 1) 9 8 0:114 9 >> > > > u(k .

1)+> : 1 > y(k) Since x ^2 (kjk) = y (k) we get 8 .0:0197 and . KC ) x ^(k . KC is thus z2 .6 From Problem 2.1 gives :139 > K=> :0 0:407 8 9 8 9T Fig. 1jk .0:0394 12 = 0:62 The desired characteristic equation of . 1)+ : 0 u(k . 4.10 we get for h = 12 8 0:790 0 9 8 0:281 9 > > x(kh + h) = : 0:176 0:857 x(kh) + > : 0:0296 > u(kh) 8 9 y(kh) = : 0 1 x(kh) The continuous time poles of the system are . Problem 4.u(k .3:55 3:55 9 8 y(k . 1) + (I .0:0129. We thus choose the poles of the observer in z = e. 1) + Ky (k) u(k) = . Ky (k) 39 . 1) x ^(kjk) = > : 0 1 : y (k ) + : 0 which is the same as the observer obtained by direct calculation. 1) 9 8 0:114 9 >> > > > u(k .3:55 9 8 0:114 9 8 3:55 9 > > > > x ^(kjk) = : 0 0 x ^(k .Lx ^(kjk): In the state equation both x ^ and y have the time argument k. Introduce (k) = x ^(kjk) .7 The observer and the controller are described by x ^(kjk) = (I . Problem 4. 1jk . KC ).4 shows the states and the estimated states when x(0) = : 1 1 and when u(kh) is zero up to t = 250 and one thereafter. 1:24z + 0:38 = 0 Using the results from Problem 4. The observer should be twice as fast as the fastest mode of the open loop system.The eigenvalues will be in the origin if k1 = 0:78=0:22 = 3:55: The observer is then 8 0 .

L)Ky (k .5 0 −0.43) with w =1 819 > > xw = : 0 40 .5 0 250 500 x2 and xe2 1 0.8 The constant disturbance v (k) can be described by the dynamical system w(k + 1) = w(k) v (k) = w(k) The process can thus be described on the form given in (4.5 0 −0.L (k) . KC ) Ky (k . KC ).o y (k . Problem 4. KC ) (k . KC )( .5 0 Figure 4.L) (k . . LKy(k) = Co (k) + Do y(k): The observer and the regulator can thus be written in the form given in the formulation of the problem. 1) + . KC ) (k . 1)] = (I . KC ). 1) + Ky (k . 1) + (I . 1) + (I . 1)] + (I . 1) = o (k . 1) + Ky(k . 1) = (I .L (k .4 Problem 4. 1) .u(k .6 The states (solid) and their estimates (dots) for the tank system in 250 Time 500 then (k) = (I .x1 and xe1 1 0. 1) The output of the regulator can be written u(k) = . (I . KC )( . .

x(k .Lw )w(1) = Hw (1)w(1) The in uence of w (or v ) can be zero in steady state if Hw (1) = 0 This will be the case if . 1). . 1) . The following control law can now be used u(k) = . Lw w(k): This gives the closed loop system x(k + 1) = x(k) + xw w(k) . 1)) Since w(k) is constant and x(k) is measurable it is possible to calculate w ^(k) = w(k . 1) = x(k) .0:114 9 . . Lw w ^(k) where Lw is the same as in (a). .Lw w(k) = ( . This is only possible if xw . .. ( . We will therefore only consider the situation at the output in steady state y (1) = C I . In this case is the state but not the disturbance measurable. x(k .L = > : 0:179 0:142 > Lw = 5:356 b. .L)]. 1) = 1 0](x(k) .Lx(k) .L)x(k) + ( xw . 'c22 Lw = (1 . If the state and v can be measured then we can use the controller u(k) = . .Lw )w(k) y(k) = Cx(k) In general it is not possible to totally eliminate the in uence of w(k). The disturbance can now be calculated from the state equation xw w(k .u(k .1 'c22 ) + 2'c12 1 where 'cij is the (i j ) element of . . 1): The rst element in this vector equation gives w(k .0:142 .Lx(k) .Lw is the zero matrix. Compared with the controller in (a) there is a delay in the detection of the disturbance. 41 .u(k . . .a. 1) . .Lx(k) .1 ( xw . Assume that L is determined to give a dead beat regulator then 8 9 : L = 3:21 5:57 and and 8 .L and i is the i:th element of . .

8 for the regulators in a) (upper left).0:6433 kw = 3:3333: The controller now has to be u(k) = .2 2 Output 0 −0. Lw w ^ (k ) where L and Lv are the same as in (a). Fig. KC xw > > x > > :w : . (b) and (c) are used. The error equation is 8 9 8 9 8 ~(k) 9 x ~(k + 1) > = > . If only the output is measurable then the state and the disturbance can be estimated using an observer of the form (4.5 show that the output when the controllers in (a).2 0 2 5 10 15 Output 0 −2 0 5 10 15 Estimate ve 0 5 Time 10 15 Output 0 −2 2 0 0 5 Time 10 15 The output of the system in Problem 4. C x ^(k) The gain vector can now be determined such that the error goes to zero provided the augmented system is observable. 1:7k1 + k2 + kw )z +0:7k1 +0:15 . Notice the di erence in scale in the upper left curve. b) (upper right) and c) (lower left and right).5 c. The estimate of v is also shown for case c).Lx ^(k) . 4.Kw C 1 : w ~(k + 1) ~(k) The characteristic equation of the system matrix for the error is z3 +(k1 . 2:2)z2 +(1:05 .0. The solutions above have the drawback that there may be an error in the output due to the disturbance if there are small errors in the model. 0:7kw . 42 . k2 = 0: The eigenvalues are in the origin if k1 = 2:2 k2 = . 9 8K9 ^(k + 1) > = > > > > > > > xw > > x :w : 0 1 :w ^(k + 1) ^(k) + : 0 u(k) + : Kw "(k) "(k) = y(k) .41) 8x 9 8 9 8 ^(k) 9 8 . Figure 4.

An integrator can be incorporated as shown in Section 4.01:0296`2 .0:0261 . 0:0035`2 .0:281`3 9 > > > > `3 > > : x3(k) > > : 0:176 . 0 : 5218 . 0:281`1 8 x(k) 9 .1:617 + 1:617p 9 > > > 43 . Cx(kh) and using the control law u(kh) = . 1:55 . 8 0:281 0:0296 0 9 8 `1 9 8 2:647 . 0:64p 3 This gives 8 `1 9 8 3:279 . The state equation for the tank system when h = 12 was given in the solution to Problem 4. We get the following system of equations to determine the state feedback vector.0:0296 1 8 0:281`c 9 > > > +> 0 : 0296 ` > : 1 c > uc (k) z 3 + (.2:647 + 0:281`1 + 0:0296`2)z 2 + (2:3240 . 3:028p > > > > > > ` = > > > 2 : : 5:934 .Lx(kh) . 0:0296`3)z + + (.00:0296`1 0:857 . 0 : 0035 . 0:0261`3) = 0 The characteristic equation is Assume that two poles are placed in the desired location and that the third pole is in p.6. p 9 > > > > > > > > > > > > . 0:0261`2 . 5:038p `3 .1 give L = : 0:251 0:8962 8 9 b.0:0261 > > : `2 : 0:6770 . 0 : 0296 ` = . 2 : 3240 + 1 : 55 p + 0 : 64 > > > > : 0:2408 .0:6770 + 0:2408`1 . The desired characteristic equation is z2 .. 1:55z + 0:64 = 0 Using the result in Problem 9.5 by augmenting the system with x3 (kh + h) = x3 (kh) + uc (kh) .0:281`2 . 0:5218`1 .Problem 4.9 a. `3 x3(kh) + `cuc(k) The closed loop system is then 8 x(k + 1) 9 > : x3(k + 1) > = 8 0:790 .

4.11 CCS.7 The step response for the tank process when p = 0 and when `c = 0 (solid). Figure 4. 250 Time 500 c.6 shows the step response for some values of p.6 250 Time 500 1 Output 0 0 Figure 4. 44 . The parameter `c will not in uence the characteristic equation. but it is a feedforward term from the reference signal. Fig. and when `c = 0. Fig.7 shows the in uence of `c when p = 0.5 (dashed).1 Output 0 0 The stepresponse for the tank process when p = 0 (solid) and 0. 3 (dashed) and 6 (dash-dotted). 4. see Fig. 4.

2:9022 1 0:5 0 > 6 > > > > > > 2:9022 .0:4756 . a dead beat controller for the states.Kw C 1 : w ~(k + 1) ~ (k ) 8 1 . w ( kh ) 0 :0 0 : 0 where = cos !o h and = sin !o h.4 3 .8. Further if Lw = : 1 0 we would totally eliminate v since v and u are in uencing the system in the same way.35) where 8 h2=2 0 9 8 .33) is now 8 1 h h2=2 0 9 8 h2=2 9 > > > 8 x(kh + h) 9 > 8 9 > > > > 0 1 h 0 x ( kh ) h > > > >+> > u(kh) : w(kh + h) > = > > > : > > > > > > 0 0 .. Since x(kh) and (kh) cannot be measured we use the observer of the structure (9. Assume rst that the control law is u(kh) = .1:9022 0:0245 . Compare the discussion in the solution of Problem 4.kw1 0 w2 Let the desired characteristic equation of the error be (z . )4 = 0 If h = 1 and !o = 0:1 then for = 0:5 we get the following system of equations 8 1 8 k1 9 8 . k 0 .0:1545 kw2 45 8 9 . k1 h h2=2 0 9 > > 8x 9 > > .Lx(kh) . k 1 h 0 ~ ( k ) > > > > 2 => > : > > . 9 > > > > and xw = : h w=: 0 The error equation is then 8x 9 8 9 8 ~(k) 9 ~(k + 1) > = > .Problem 4. 5:8044 > k2 > . I.1 1 . sin ! h 9 > w(kh) w(kh + h) = > : sin !ooh cos !oo h 8 9 v (k) = : 1 0 w(kh) 81 x(kh + h) = > :0 9 8 9 8 9 The augmented system (9. w ~ ( k ) : . 3:9022 > > > > > >> > : : .4 + 3:9022 9 0 0 0 9 > > > > > > > > > 2 .e.0:1545 > > kw1 > = > .1 : 4.e. Lw w(kh) where 8 9 L = : h12 23h i. it can be described by 8 cos ! h .10 The process is h > x(kh) + > h2=2 > u(k) + > h2=2 > v (k) : : 1 h h where v (k) is a sinusoidal. KC xw > > x > > :w : .

x1 and xe1 2 0 −2 0 10 20 30 x2 and xe2 2 0 −2 0 10 Time 20 30 The states of the double integrator (solid) and 8their estimates 9 (dots) when v(t) = sin(!o t). 9. or 46 2 + (. . ` .11 We have to determine the feedback vector L such that a. 8 0:9 .8 shows the states of the double integrator and their estimates when v(t) = sin(!ot).L = > : 1 1 0:72 > det( I . 8 k1 = 1:9022 > > < k2 = 1:1018 > > :kw1 = 0:2288 Problem 4. 0:7`1 + `2 = 0 L = : 1:6 0:49 8 9 ..1:6 + ` 2 1) has all eigenvalues in the origin. Figure 4. 0:7`1 + `2 .8 which has the solution kw2 = 0:1877 Fig.` 9 .e. This gives the condition + 0:63 .1:6 + `1 = 0 0:63 . + . It is seen that the controller is able to eliminate the disturbance. The controller is de ned by L = : 1 1:5 and Lw = 8 9 :1 0 .L) = = I.

0:25 + ` ` .0:03 The stability area is shown in Fig. + .1:6 + `1) 0:63 .1:7`1 + `2 > .1:6 + `1) This gives .L) = det > : 4`1 . one is L = : 2:25 0 8 9 47 . 0:7`1 + `2 < 1 0:63 .. but the system will remain stable if x2 is disconnected (if `2 = 0). 0:7`1 + `2 > .9. = m To get unit steady state gain we choose m = 1 b.1 + (. a. The closed loop system will be unstable if the feedback from x1 is disconnected (i. 0:5 9 > 2 2 = det( I .e. (. is used. 2 + 4`2 2 2 = + (`1 + 4`2 . 4.0:7`1 + `2 < 0:37 .l2 3 2 1 Deadbeat 1 −1 2 3 l1 Figure 4. The closed loop characteristic equation is stable if (See Example 3.L).9 The stability area for L in Problem 4.2) 0:63 .3:23 0:37`1 + `2 > . 2:25) There are in nitely many solutions.1 .11. Assume that the deadbeat control in a. The deadbeat requirement implies that the characteristic equation of the closed loop system is 8 . + .1 . Problem 4. 1 1 . 0:67`1 + `2 > . if `1 = 0).12 The stationary gain of the closed loop system is given by stationary gain of the m C (I .

u(0) + . 2k1 Identifying coe cients give the system k1 . 1 .2 = 2 + (k 1 . 0:2)2 = 2 .0:5 9 > det( I . 2k1 = 0:04 which has the solution 8 1:85 9 K=> : 7:48 > 48 . = : 4 9 > det Wc = 0 implies that the system is not reachable and 8 arbitrary 9T states : cannot be reached from the origin. 0:25 .u(1) 8 0:5625 1:125 9 8 2:25 9 819 > > > > = : 2:25 4:5 x(0) + : 9 u(0) + > : 4 > u(1) With x(2) = : 2 8 8 9T and x(0) = : 0 0 8 9T we get 2 = 2:25u(0) + u(1) 8 = 9u(0) + 4u(1) u(0) = 0 u(1) = 2 c. + KC ) = det > : k21. The observer should have the characteristic equation One solution is 8 + k . 2:25) ( . 2:25 = . x(k) = 2 8 is in the column space of Wc and the point can thus be reached. . x(2) = 2x(0) + .0:4 0:5k2 .b. However. The controllability matrix is 9 8 9 8 1 2 : 25 > : Wc = . 0:4 + 0:04 = 0 + 0:5k2 .

0:5. Since the desired pulse-transfer function from uc to y is Hm (z ) = (1 + )=(z + ). with A and Acl being rst order polynomials and B a scalar. 0:6 A2 = . . 0:55)=0:1775 A4 = 0 The greatest common factor of A and B is thus z .1 Euclid's algorithm de nes a sequence of polynomials A0 : : : An where A0 = A.0:4(z2 . you can solve the equation for the closed loop system using scalars R(z ) = r0 and S (z ) = s0 : A(z)R(z ) + B(z)S (z) = Acl (z ) (z + a) r0 + 1 s0 = z + Identi cation of coe cients gives the following equation system: ( r0 = 1 ar0 + s0 = 49 with the solution r0 = 1 and s0 = . For the polynomials in the problem we get A0 = z4 .1 we must know the pulse-transfer function B (z )=A(z ) and the desired closed-loop characteristic polynomial Acl (z ). You easily see that. we know at least that (z + ) must be a factor in Acl (z ). 0:95)=(.2 a. 0:5) Finally Q3 = (z . To use Algorithm 5.0:4) A3 = 0:1775z . 0:08875 = 0:1775(z . a. Problem 5. 0:35: This gives Q1 = z . 0:11 = . Step 1. 1:05z + 0:275) The next step of the algorithm gives Q2 = (z .Solutions to Chapter 5 Problem 5. A1 = B and Ai = Ai+1Qi+1 + Ai+2 If the algorithm terminates with An+1 then the greatest common factor is An . 2:6z3 + 2:25z 2 . 0:8z + 0:1 A1 = z3 .0:4z 2 + 0:42z . 2z 2 + 1:45z .

S (q)y (k) ) . . Solution with higher order observer: The solution above is not the we must increase the order of R by one. and the observer dynamics are cancelled in the pulse-transfer function from uc to y . 1) The controller thus is a dynamical system. 50 . though. a)y (k) i. This gives and the equation system becomes only one solving the original problem. there are no observer dynamics. S (q)y (k) u(k) = (1 + )uc (k) . This could have been avoided by choosing deg S = 1. . for example. Step 1. 1) uc (k) + uc (k . a u(k . decide to have another closed loop pole in z = . a( + . a) Step 2. With this choice of T .Step 2. Choose c (1) T (z) = t0 Ao (z) = A B(1) Ao(z) = 1 + deg R = 0. Factor Acl (z ) as Ac (z )Ao(z ) where deg Ao Ao = 1 and Ac = z + . + . The resulting control law becomes R(q)u(k) = T (q )uc(k) . R(q)u(k) = T (q )uc(k) . To solve the equation for the closed loop characteristic polynomial (z + a)(r0z + r1 ) + 1 s0 = (z + )(z + ) 8 r0 = 1 > < ar + r1 = + > :ar0 1 + s0 = () 8r = 1 > <0 r1 = + . ( . We can. Thus. . 1) + 1 + . say.e. Splitting Acl into factors Ao = z + and Ac = z + gives c (1) A (z ) = (1 + )(z + ) T (z) = t0 Ao(z ) = A B (1) o The resulting control law becomes u(k) = . a) y(k . a (static) proportional controller. a( + . . In this case. a > :s0 = . In this case there is a delay of one sample from the measurements y to the control signal u.. the static gain from uc to y is set to 1 (Hm (1) = 1). since deg Ao = 0.

3 or through the following discussion.1 shows the response when the two di erent controllers are used. The dots corresponds to the zero order controller. Cancellation of process poles and zeros is handled by Algorithm 5. 5. the process numerator is factored as B = B + B .e.3 a.1 Output 0 −1 0 25 Time 50 The output of the process with y(0) = 1. and the crosses to the rst order controller. First. The closed loop characteristic polynomial is given by AR + BS . (z + ) in the rst solution. and (z + )(z + ) in the second one. so (z + 0:7) is not a factor of Bm .0:7 and = . . Figure 5. = 1.1 for k >= 25.. B + must be a part of the R polynomial as 51 . By choosing Bm = 0:2z we make the steady state gain Hm (1) = 1.0:5. The desired closed loop pulse-transfer function is m (z ) Hm(z ) = z2 . i. and the pole excess in Hm equals the pole excess in H . but not in the response to a set point change.e. i. Problem 5. In both cases we get the same closed loop pulse-transfer function from uc to y since the observer polynomial is cancelled by T (z ): (z )T (z ) t0 B(z) = 1 + Hm(z) = A(z)RB = (z ) + B (z )S (z ) A (z ) z + c Fig.0:99.1 b. It is assumed in the simulations that a = . and that the design parameters are = .B 1:5z + 0:7 In this case. where B + is the part of the numerator which should be cancelled by the controller. B + = z + 0:7 and B. You can see the e ect of the observer polynomial when regulating a nonzero initial state. and uc (k) is 0 for k < 25 and . the process zero is cancelled by the controller.

0:11 Thus. 0:3y (k) + 0:11y(k . Now.1:5 r1 = 0:0875 () > > s0 = 0:2125 1 + 0:7s0 + s1 = 0:7 > > :0:81r0 . 1:5z 2 + 0:7z . 1:8z + 0:81) r0 + (s0 z + s1 ) = z 2 . the causality condition (deg S deg R = 1) leads us to set S (z ) = s0 z + s1 .1:8r0 + s0 = . 1) + 0:2uc(k) . 1:8z + 0:81)R(z ) + S (z ) = Acl (z ) If R(z ) is a constant r0. 1:5z + 0:7 in order to get Hm (1) = 1. This gives the Diophantine equation A(z )B+ (z )R(z) + B+ (z)B. will not su ce in this case since it would only give 2 parameters r0 and s0 to select the 3 parameters in the second order polynomial z 2 . Letting R = r0z + r1 S = s0z + s1 the identity then gives the system of equations 8 8r = 1 r0 = 1 0 > > > > < < . s0 and s1 ) and 3 coe cients to set: (z 2 . we can solve the Diophantine equation above. 1): b. To obtain the desired B+ B. so 0:2 (z + 0:7) m (z ) 1:7 Hm (z ) = z 2 .1:5 () >s0 = 0:3 > : 0:81r0 + s1 = 0:7 :s1 = . a zero order controller. T 0:2z Hm(z ) = ARBT = = + 2 2 + BS B (z . since we have 3 indeterminates (r0. i. add an observer pole which is placed at the origin.1:8r0 + r1 + s0 = . 1:8r0 : s1 = . consequently. 1:5z + 0:7 ) 8 8 > r = 1 > 0 < <r0 = 1 . Ao = z and Acl = z 3 . (z )S (z ) = B + (z )Acl(z) (z 2 .e. and so must Acl be.0:1012 :81r1 + 0:7s1 = 0 52 . 1:5z + 0:7) z . The closed loop characteristic polynomial is now given by the identity (z 2 . In this case we do not want to cancel the process zero.well as Acl . we have R(z ) = B + (z )R(z ) = z + 0:7 and S (z ) = 0:3z . 1:5z + 0:7 we must select T (z ) = 0:2z The controller is now u(k) = . we must increase the order of the controller by one and. 1:8z + 0:81)R(z ) + (z + 0:7)S (z ) = Acl (z ) The simplest choice.B = 1:5z + 0:7 z 2 . the left hand side is of second order.0:7u(k . With this choice of R. Thus. 1:5z + 0:7. 0:11.

Problem 5.10 gives one solution to the problem R = B(Am .1 0 −0.1 0 20 Time 40 Control 0. 1) c. y (k)) u(k) = R (q ) c gives the closed loop system BS = Bm AR + BS Am Section 5. Further the pole of the process is cancelled. Fig. 53 . The ringing in the control signal in Case a is due to the cancellation of the process zero on the negative real axis.Output 1 Output 0 20 40 1 0 0. ) = z .0:0875u(k . The controller is thus u(k) = . 1 .4 a. 0:2125y (k) + 0:1012y (k .2 :2 Further T = t0 Ao = 0 1:7 z .1 0 −0.2 0 0.3. 1) + 0:1176uc(k) . 5. Case a should probably be avoided because of the ringing in the control signal. 1 S = (z + a)(1 + ) The controller contains an integrator. Bm ) S = ABm With the given system and model we get R = 1(z + .2 0 20 40 Control 0.2 shows the output and the control signal for the controllers in Case a and Case b.1 0 20 Time 40 The output and the control signal for the controllers in Case a (left) and Case b (right) in Problem 5. Figure 5. Using the controller S (q ) (u (k) .

Now. By inspection of the transfer function from v to y we see that we must make R(1) = 0 in order to remove the steady state error after a load disturbance step. The design problem is solved by using the general Algorithm 5. ii. Bm From this we can conclude that in order to get a stable closed loop we must ful ll the following constraints. Bm must contain the zeros of B that are outside the unit circle. 1) into R(z ) we thus have obtained integral action in the controller. 1)R(z ) + 1 S (z ) = Acl (z ) If R(z ) is a constant r0. the causality condition (deg S deg R = 1) leads us to set S (z ) = s0 z + s1 . Equation (5.b. 1)R(z ) the closed loop characteristic equation becomes We thus get A(z)(z . Problem 5. Bm must contain the poles of the process that are outside the unit circle. With R(z ) = (z . the left hand side is of second order.3 or through a discussion like the one below.2 gave R=1 S = . i. Am . The controller can be written S = A Bm R B Am . The rst constraint is the same as for the polynomial design discussed in Chapter 5. b. and so must Acl be. By forcing the factor (z . 1)R(z ) + B(z )S (z) = Acl (z ) (z + a)(z .5 a.a BR = 1 AR + BS z + If v (k) is a step there will thus be a steady state error 1=(1 + ) in the output. With this choice of R. The characteristic polynomial of the closed loop system is AR + BS = (z + )(z + a) The closed loop system will contain an unstable mode if jaj > 1.33) gives the pulse transfer operator from uc and v to y : BR m y(k) = B A uc (k) + AR + BS v(k) m The design in Problem 5. we can solve the Diophantine 54 .

a + 1)y(k) .0:5.3 2 Output 1 0 0 The output of the process in Problem 5. 1) .0:7 and = . 1) u(k) = u(k .2 and the controller with an integrator. and the design parameters are chosen as = . 1)r0 + s0 = + > : . a + 1 : s1 = + a (z )T (z ) T (z) 1+ Hm(z) = A(z)RB = = (z ) + B (z )S (z ) (z + )(z + ) z + we must select The controller is now T (z) = (1 + )(z + ) )y (k . 1) Fig. (a + + (1 + )uc (k) + (1 + )uc (k . At t = 25 a constant load disturbance is introduced.5 when the controller does not contain an integrator (dots) and when an integrator is introduced (crosses).0:99. 5. s0 and s1 ) and 3 coe cients to set: 8 > r0 = 1 < (a . It is assumed that a = . Figure 5.3 shows the controllers in Problem 5. The reference value is zero and there is an initial value of the state in the process.3 25 Time 50 equation above. 1) r0 + (s0 z + s1 ) = (z + )(z + ) ) 8 > <r0 = 1 () >s0 = + .ar0 + s1 = To obtain the desired (z + a)(z . ( + . 55 . since we have 3 indeterminates (r0.

6 Problem 5. B B.1:4 < a0 < 0:6 56 . 1 R B ! ! = H = Bm m Am Ao B. 1 1 1+ R + Am H 0 . H Problem 5.7 a.a The closed loop system is stable if ja0 + .41) gives Hcl = z + a10 + + . 1 1 RB. aj < 1 With the numerical values in the problem formulation we get .It is assumed that the design is based on the model H = B=A while the true model is H 0 = B 0 =A0 . A R BR B This gives 0 Ao 0 Ao Bm Bm R R ! Hcl = 0 = + A B AoAm A AoAm 1 1 + BR .2 gives R=1 S = . R + H 0 . H Ao Am H 0 . The design in Problem 5. H B0 Ao 0 . The pulse transfer operator of the closed loop system is B0T = T=R Hcl = A0 R + B 0 S A0=B 0 + S=R The design gives 0 Ao T = Bm and AR + BS = A0Am B+ or S = B + Am Ao .a T =1+ Assume that the true process is 1 z + a0 Equation (5.

The nal value is 1 + a.7 when z = ei! .6 shows that the control signal is given by (5.4 1 Frequency 2 3 The left hand side of the inequality in Problem 5.6 (dotted). .e. Equation (5.8 . Further the desired model is stable. 0:9 2:5 Fig. 0:9 or 1 z .4 shows for z = ei! the left hand side of the inequality for di erent values of a0 .0:955 (dashed).0:9 (dash-dotted) and 0.52) m (q ) u (k) = (1 + )(q + a) u (k) u(k) = H c H (q ) c q+ We may assume that both the process and the model have a continuous time correspondence. which is a stable rst order system. The magnitude at the rst step can be determined either through the initial value theorem or by using series expansion and the value is 1 + .10 5 0 0 Figure 5. 57 Problem 5. The control signal is now obtained by studying the step response of Hm =H . If j j < jaj then the closed loop system is faster than the open loop system and the control signal is largest at the rst step. 5. i. The largest value is then either at the rst step or the nal value. The right hand side of the inequality is also shown (solid). a m (z ) Hfb (z ) . The right hand side is also shown. j j < 1. b. 0:5 1 z . This implies that a and are less than zero. If the desired response is slower than the open loop system then the nal value is the largest one. Section 5. 0:9 .40) gives the inequality H (z ) Hff (z) = z+ 1+ H ( z ) . z + a0 < z . 0:5 2:5 = z z . 0 < ! < for a0 = . H 0 (z ) < H (1 + )(z + a) .

= A = q 2 .Problem 5.16 we get sampled data characteristic equation z 2 + a1z + a2 = 0 a1 = . Bm = B . 0:6 Identifying with s2 + 2 !s + ! 2 gives ! = 0:1. Bm ) Bm = 0:55=0:4 = 1:375 Degree condition: deg Acl = 2 deg A + deg Am + deg Rd + deg Sd . 1 = = 2 2+2+0+0. Rd = Sd = 1 since no given factors are forced into the controller. Step 1. The desired response to command signals is assumed to be Hm = Bm =Am = Bm =Ac = 0:55=(q 2 . Pole excess condition: where p Problem 5.15 deg Am . B + = q + 0:75 and B . The rule of thumb on p.2e. 0:7q + 0:25.6 b. Acl will at least contain Ac = q 2 . Data: The process is given by A = q 2 .14 a.1 Remark: The fact that we cancel one zero and do not introduce any other Model following condition: zero in Bm causes the delay from the command signal to be one time unit more than the delay of the process.0 2.1:32 a2 = e. other factors may be added later on. = 0:4 achives cancellation of the process zero. 1:6q + 0:65.3. 0:7q + 0:25) (cancelled process zero.1= 5 with Acl = A+ B + Am Acl and Acl = Ac Ao . This solution demonstrates how to use Algorithm 5. Using Example 2. A. 58 . Thus an appropriate sampling interval is h = 1.2 !h = 0:5 The poles are in 0:66 0:25i. 2!h) = . deg B 2. deg Bm deg A . 1:6q +0:65 and B = 0:4q +0:3. Hm (1) = 1). 130 gives !h = 0:1 . !h cos( 1 . but no cancellation of process poles. A+ = 1.

In order to solve the identity we must have two more parameters. Am = Ac . Problem 5. 1:6q + 0:65)r0 + (s0 q + s1 ) 0:4 = q 2 .e. This gives the Diophantine equation A. 1) T = Bm A+ Acl = 1:375 Ac Since. 1)R + 0:4S = (q 2 .42)).16 In this case we want to have an integrator in the controller.2. This will increase the degree of the closed loop by one compared to Problem 5. 1 T = 1:375 The corresponding degree of the closed-loop polynomial AR + BS will thus be 3 instead of 5.45): R = AmB + Rd R = Am(q + 0:75) S = AmA+ SdS = Am (2:25q . Rd R + B. giving R = q + 0:75 S = 2:25q . deg Am . i. this factor can (and should) of course be cancelled in all controller polynomials. in this case. Rd = (q . deg B + . say. deg A+ . say. The controller polynomials are now given by (5.Rd R + B.1 Step 3.. so we let S = s0 q + s1 : (q 2 . 0:7q + 0:25)(q + ao ) R must still be a constant (which as usual will be 1) and S must be of second order: (q 2 .Step 2.2 = 0 The Diophantine equation to solve thus becomes A. 0:7q + 0:25)(q + a0) 59 . SdS = Acl (q 2 . 1:6q + 0:65)(q .15 (see (5. deg Ac = = 5.1:6r0 + 0:4s0 = . 1:6q + 0:65)R + 0:4S = q 2 . 1). SdS = Acl (q 2 . 0:7q + 0:25 This gives the system of equations r0 = 1 . 0:7q + 0:25 Since this is of second order.1. R must be a constant.0. 1:6q + 0:65)(q . which is done by having Ao = (q + ao ). 1) + (s0 q 2 + s1 q + s2 ) 0:4 = (q 2 .0:7 0:65r0 + 0:4s1 = 0:25 () r0 = 1 s0 = 2:25 s1 = . r0 . Using the degree condition above we may conclude that deg Ao = deg Acl .

Using Theorem 5.2:6 + 0:4s0 = a0 .0:65 + 0:4s2 = 0:25a0 S (q) = 2:5 (a0 + 1:9)q2 . Q(z . 4:5625q + 1:46875 T = Bm Ao = 1:375q .This gives and . 1) = q2 . 1. 0:75 S = 4:125q 2 . 0:7 2:25 + 0:4s1 = . 1)(z . Let us instead use deg A0 = 2 and deg S = deg R . 3z S0 = 7z + 6 The controller is causal. 0:25q . T = S respectively are BS0 = z(7z + 6) AR0 + BS0 z4 BS = z(z 2 + 4z + 8) AR + BS z4 The number of zeros are di erent. (0:7a0 + 2)q + 0:25a0 + 0:65 Using a0 = .45) gives (after cancelling the common factor Am ): R = B+ (q .17 The minimum degree solution has deg A0 = 1 and gives a unique solution to the Diophantine equation. z = z 2 . This gives the equation (z + 1)(z + 2)(z 2 + r1z + r2) + z (s1 z + s0 ) = z 2 z 2 with the solution R0 = z2 . 2) where Q is an arbitrary polynomial.1. The closed loop systems when using R0 S0. 3z . 0:34375 Problem 5. T0 = S0 and R S . Choose for instance Q = .0:7a0 + 0:25 .1 we also have the solutions R = R0 + Qz S = S0 .0:25. 60 . (5. 3z + 2) = z2 + 4z + 8 This is also a causal controller. 4z S = 7z + 6 + (z2 . This gives R = z 2 .

There are three subsystems each with the transfer function Ki Gi (s) = s + Ki and the total transfer function from uc to y is K1K2K3 1G2 G3 G = s +G K4G1 G2G3 = s(s + K1)(s + K2)(s + K3 ) + K1K2K3K4 If either of the gains Ki is increased su ciently much the closed system will become unstable. In the case of control in the direction opposite to the ow each of the subprocesses Raw material flow u c x2 x3 K2 Σ Σ 1 s K1 Σ 1 s Σ 1 s K3 1 s Final product x 4 =y −1 −1 −1 − K4 Figure 6. and 0. 1 Output 0 0 Figure 6. 0.1 In the rst case it is assumed that we have a control structure as in Fig.75 (dash-dotted). and 0.1 Block diagram for the control in the direction of the ow in Problem 6.Solutions to Chapter 6 Problem 6. 25 Time 50 61 .25 (dashed). 6. A disturbance in the process will propagate in the direction of the ow.2 K 4 Impulse response for the control in the direction of the ow when = 0:1 (solid). 0. Fig.75.1.2 shows the response when uc is an impulse and when K1 = K2 = K3 = 1 and K4 = 0:1. 6.25.1.

An intentional nonlinearity is introduced in the selector. 6. Notice the order of the states.x 4 =y x3 1 s K4 uc x2 K2 1 s Σ 1 s K3 Σ 1 s Σ 1 s K1 −1 −1 −1 Problem 6. Nonlinear elements are used in the ow control loops of the product output and the coolant ow. The square root device is thus used to remove the nonlinearity of the measurement device. Problem 6. The system is then represented with the block diagram in Fig. A disturbance will now propagate in the direction opposite the ow. b. Feedforward is used for the level control loop where the input ow is used as a measurable disturbance.3 Block diagram for the control in the direction opposite of the ow in has a transfer function of the type Gi . Either the temperature or the pressure is used to control the coolant ow depending on the status of the process.2 Fig. a. 6. c.3 in CCS contains several examples of couplings of simple control loops. The system will remain stable for all positive values of Ki .1. The input ow is also used as feedforward for the cooling of the jacket. 62 . Cascade control loops are found for the cooling media ow and for the output product ow. The ow is probably measured using di erential pressure which is proportional to the square of the ow.3. Figure 6. A disturbance in uc will now only in uence the rst subprocess and will not propagate along with the ow. The reader is strongly recommended to compare with the case where the disturbance appears at the nal product storage instead.

2 . All frequencies in the interval (.Solutions to Chapter 7 Problem 7.f1 f1) should be possible to reproduce from the samples of the continuous time signal. ! . The Fourier transform of sin !0 t has its support (i. 3:63 6:37 13:63 16:37 : : : The output of the sampler is composed of the frequencies . More precisely.e. 63 Problem 7. .3) 1 1 X Fs = h F (! + k!s ) k=. 2 8 12 18 22 : : : ! = 20 gives rise to 20 k 10 . The sampled signal has the Fourier transform. We want to eliminate the disturbance with the known and xed frequency f2 5f1. it equals i ! + !0 . . 2.1 !s = 2h where F (! ) is the Fourier transform of the time continuous signal.2 3:63 6:37 8 12 13:63 16:37 : : :) We have the following speci cations on the choice of sampling period and presampling lter: 1.1 Which frequencies will the signal f (t) = a1 sin 2 t + a2 sin 20t give rise to when sampled with h = 0:2? Since sampling is a linear operation we consider each component of f (t) separately. points !0 . !0 . see (7. if the signal sin !0 t is sampled with the sample interval h its Fourier transform will be 6= 0 in the points ! + k!s 2 k = 0 1 2 ::: For !0 = 2 and !s = 2 =0:2 = 10 we get the angular frequencies k 10 = .. Thus. the set where it is 6= 0) in the two .

It can instead be removed using digital lters. the disturbance does not mix with the data signal.1 0 a) h=2π/10 10 0 10 20 30 b) h=2π/20 0 c) h=2π /50 10 20 30 Figure 7. 64 .1 and Fig. Then sample with fs > 2f1 . See Fig.1 Folding for di erent frequencies in Problem 7. So to achieve 20 dB in log f 1 2. Suppose the disturbance should be attenuated 20 dB without e ecting the datasignal. dB/decade. If fs > 6f1 . 7. 0 10 25 50 The sampling theorem states that the rst speci cation will be satis ed if and only if the sample frequency fs is chosen such that fs > 2f1 Moreover. Filter out the disturbance using an antialiasing lter. Problems 7. f1 ) > f2 .6 The magnitude of the spectrum of the sampled signal can be obtained by folding the spectrum of the time continuous signal around the angular frequency !N = !s=2 = =h.2. A n:th order lter gives maximally n 20 f2 = 0:699 decades takes n 2. fs =2 ) fs > f2 + f1 = 6f1 Two cases: 1.5. 7. for the disturbance f2 not to fold on the data signal (fs =2 .5 and 7.

7 when !s < 2!r .− 100 100 a) ω s =120 − 120 − 100 − 60 − 20 20 60 100 120 b) ω s =240 − 140 − 100 100 140 Figure 7. The picture will not move if !r = n !s for integer values n.6. A correct picture will be seen. The wheel then appears to rotate three times slower and in the wrong direction. Problem 7. if !s > 2!r according to the sampling theorem. − ωs − ω r 0 ω r ωs 4 3 1 2 1 4 2 3 t Figure 7. (The eye acts like a low pass lter).3 Folding in Problem 7. Aliasing will give a frequency ! = 1=3 !r. See Fig. If !s = !r the wheel will appear to stand still. For instance let !s = 4=3 !r . The wheel will appear to rotate with a frequency lower than r if !s < 2!r .7 65 .3. Compare the stroboscope.2 Folding for di erent frequencies in Problem 7. The frequency of the camera shutter !s = 2 =h. The rotation frequency of the wheel !r = 2 r. 7.

9 The signal is 1 sin(6! t) + sin(2! t)] u(t) = sin(4!0t) cos(2!0t) = 2 0 0 Sampling the signal with !s = 2h = 6!0 gives the Nyquist frequency !N = 3!0. 66 .Problem 7. Sampling the signal u(t) gives the alias of sin(6!0 t) in ! = 0. We thus get the frequencies f1 = 0 f2 = !0 in the sampled signal.

1a.Solutions to Chapter 8 Problem 8. sh For s = i! we get 1 1 . If a transfer function is stable in the s-plane it will be translated into a stable discrete time system if using the backward di erence or Tustin's approximation. Figure 8.5 and going through the points 0 and 1. see Fig.2 a G(s) = s + a a>0 Forward differences Transformation of the left half s-plane when using a. For Euler's method we have z = sh + 1 This implies that the stability boundary in the sh-plane (the imaginary axis) is translated one unit to the right.6) have di erent stability properties. 8. Finally for Tustin's approximation with s = i! + i!h=2 z=1 1 . This can be seen by nding how the left half s-plane is transformed into the z -plane. !h This represents a circle with radius 0. b. 8. Backward di erence and c. backward di erence (8.1b. Eulers method. Problem 8.5) and Tustin's approximation (8. Tustin's approximation. When the backward di erence is used then 1 z = 1.4)). see Fig. i!h=2 Now jzj = 1 arg z = 2 arctan !h The imaginary axis is thus transformed into the unit circle in the z -plane.1 Backward differences Tustin 67 .1 The three transformations Euler's method (forward di erence (8.

For Tustin's approximation we get or a (z + 1)ah=2 H (z ) = 2 z . 1 +a z+ z+1 a= + 1 where Thus a = tan(ah= 2) tan(ah=2) H (z ) = 1 + tan(ah=2) z+1 tan(ah=2) . 1 z + ah=2 + 1 The pole of the discrete time system will vary from 1 to . The discrete time approximation is always stable when a > 0.3 The lead network +1 Gk (s) = 4 s s+2 should be approximated using di erent methods. 1)y(kh) = ahu(kh): The di erence equation is stable if jah . Using Tustin's approximation with prewarping gives z+1 a a= = 1+ H (z ) = z . be poor even if the di erence equation is stable. 1) +a hz+1 z+1 ah=2 = 1+ ah=2 ah=2 .1 when h vary from 0 to in nity. b. 68 . 1) =h + a z . 1 z + tan(ah=2) + 1 Problem 8. c. 1 a= a= .a.95. Using Euler's method we get ah a = H (z) = (z . At ! = 1:6 rad/s it has the argument 19 and the gain 2. 1 + ah This corresponds to the di erence equation y(kh + h) + (ah . 1j < 1 0 < h < 2=a: The approximation may. however. 1 = (1 + ah= 2)z + (ah=2 .

Euler's method gives z . 0:803 HZOH (z ) = 4 . a) sin(!h) = K 1 + ab + (a + 1 + b2 + 2b cos(!h) (b . Tustin's approximation with prewarping z .2h z . 69 .2h z . (1 .0:0778 :6 2z. Backward di erences (z .i!h + b) b) cos(!h) + i(b . 1 + 2h = 4 z .a. e.2h = 4 z .2h )=2 = 4 z . 1)=h + 1 z . 0:596 +2 z+1 where Within two decimals this is the same as in (c).1 +2 hz+1 d. e. (1 .. 0:607 All ve approximations have all the form +a H (z ) = K z z +b The gain and the phase at ! = 1:6 are obtained from i!h + a (ei!h + a)(e. 1 = 4z z(1 + 2= ) . h=2) HT (z) = 4 = 4 z (1 + h) . e. 0:75 HE (z ) = 4 ( (z . 1)=(zh) + 1 = 4 z (1 + h) .1+h z . 0:775 HTW (z ) = 4 z . 4 2 z . 0:667 c. h) = 3:6 zz. 1= ) = 3:596 z . 1 = 3:333 z . (1 . 0:5 b. Zero order hold sampling gives 1 1 . e. a) sin(!h) arg H (ei!h ) = arctan 1 + ab + (a + b) cos(!h) !1 = tan(! h=2) 1 7:893 jH (ei!h)j = K s 1 + a2 + 2a cos(!h) 1 + b2 + 2b cos(!h) The di erent approximations give at ! = 1:6 rad/s.1 +1 z+1 (1 + 1= ) . e. (1 .i!h + b) H (ei!h) = K e = K ei!h + b (ei!h + b)(e. 0:80 HB (z ) = 4 ( z . 2= ) z . Tustin's approximation 2z.2h . 1)=(zh) + 2 z (1 + 2h) . 1 z . (1 . e.1 h z + 1 + 1 z (1 + h=2) . 1)=h + 2 = 4 z .

4 . 8.3 is the same as Fig. backward and Tustin approximations. The shorter sampling period gives a better approximation. 0:15. 0:5: 70 jH (ei! )j arg H (ei!) 2:97 24 2:92 16 2:96 19 2:96 19 3:25 22 Problem 8. 8. This approximately corresponds to a decrease of the damping by 0:05 .10 1 Gain 10 −2 10 30 0 10 −1 10 0 10 1 10 2 Phase (deg) 20 10 0 −2 10 10 −1 Figure 8.2 shows the Bode diagrams for the continuous time system and for the Euler.3 when h = 0:25 continuous time lter (full) Euler's method (dashed) backward di erence (dotted) Tustin (dash-dotted). A time delay of h=2 seconds gives at the crossover frequency a decrease of !c h = !ch = 5 .2 The Bode diagrams for the lter in Example 8. Fig. 10 Frequency (rad/s) 0 10 1 10 2 Euler Backward Tustin Tustin with prewarping Zero order hold Fig. 0:52 or approximately !c h 0:15 . 8. 15 .2 but with h = 0:05. Further we will allow a decrease of the phase margin of 5 . 15 ' = !c h=2 rad] = 180 2 0:035 This gives !c h = 0:17 . It is assumed that the sample and hold circuit can be approximated by a delay of h=2 seconds.

10 1 Gain 10 −2 10 30 0 10 −1 10 0 10 1 10 2 Phase (deg) 20 10 0 −2 10 10 −1 Figure 8.4) gives HI (z) = T (Kh z . 1) zh KTd (z . 1) 1+ d zhN z . Nh + T d which is the same as the derivative part on page 308. 10 Frequency (rad/s) 0 10 1 10 2 Problem 8. 1) i which is the same integral part in (8. The derivative part of (8.5 The transfer function of the integral part of the PID-controller (8.22) is K GI (s) = T s i Using Euler's approximation (8. Td=N T (z . 71 . 8.22) has the transfer function KTds GD (s) = 1 + T s=N Using backward di erence gives d KTd(z .3 The same as Fig.1 KTd = h+ Td=N Td z .2 but when h = 0:05.23). 1) HD (z ) = = z (h + Td=N ) .

1 Ti . however. Since the D-part of a PIDcontroller sometimes is not used it is necessary that the regulator remains stable when Td = 0. Approximation of the integral part with backward di erence gives Khz HI (z ) = T ( i z . This gives K = 1:85 T = 149 72 . 1 + hN=Td A small value of Td can make Hd unstable.24) with Kd = K 1 + 2h Ti Tid = Ti + h=2 0 1 1 C= K 1+ h + h 1 HT (z) = K B @1 + 2 z A 2Ti Ti z .a.115 We will use a PI controller of the form + 1) Gr (s) = K (Ts Ts and we want the gain 1=0:523 and the phase .6 Using the bilinear transformation gives hz +1 h h = K 1 + 2T 1 + (2T + 2 i i h)(z . 1) Hd(z) = z . 1) This is of the same form as (8. At the desired cross over frequency we have jG(i!c)j = 0:525 arg G(i!c) = . The sampling interval is. 1 Problem 8. b. usually short for digital PID-algorithms. Problem 8.7 The tank process in Problem 2. Euler's approximation for the derivative part gives KN (z . 1) An error will then directly in uence the computation of the integral part. Euler's approximation gives a delay of one sampling interval before an error will in uence the integral part.10 has the transfer function 0:000468 G(s) = (s + 0:0197)( s + 0:0129) a.15 degrees at !c .

c. 73 .4 250 Time 500 b.0:0062.1 1:85 z + 1 + 0:0067 Hr (z ) = z.4 shows simulations of the step response of the system controlled with the continuous time and the approximate discrete time PI-controller when h = 6 and 12 seconds. The complex poles have a damping = 0:43.1 z+1 :0134 = 1:85( + 0:0067) 1 + ( + 0:0 0067)(z . 20 seconds The choice h = 12 seems to be reasonable.2 for choosing the sampling period gives h 6 . 1) Using the rule of thumb from Section 8. This gives = 0:165 and ! Hr (z) = 1:925 1 + 0z:0778 . The zero of the closed loop system is . 8.0:006. Fig. Tustin's approximation with warping gives with = !c = tan(!c h=2) z .1 Output 0 0 Step response of the tank process when controlled with a continuous time (solid) and a discrete time PI controller.1 d. The characteristic equation of the closed loop system is s3 + 0:0326s2 + 0:00112s + 0:00000581 = 0 The roots are s1 2 = . Figure 8. The sampling interval is 6 (dashdotted) and 12 seconds (dashed).0:0135 0:0281i and s3 = .

Using (8. 8.h ) 74 . e.24) and (8. we get u(kh) = Muc(kh) . e > > 2 .h . Using L = `1 `2 gives 8 . Lx(t): A discretization is obtained by sampling uc and x and letting u be constant between the sampling period points i. Figure 8. `1(h .e.5 Problem 8. The continuous time controller is u(t) = Muc(t) . ` (1 .h ) .h .h ) 9 e . `1(1 .h ) 1 .5 shows the stepresponse of the system when using the continuous controller and the controllers in a) and b) when h = 0:25. Lx(kh) b. h 4 . 1 + e. A discrete time 8 state 9 space : representation of the motor is given in (A. BL)h=2) = : 2 4 L 8 8 9 9 98 1 . `2(h .L = : 1 .9 when a continuous time (solid). h) M ~ = L(I + (A .6). Fig.25) give = : 2 . . 3h=2 . a discretized (dash-dotted) and a modi ed discretized state (dashed) feedback controller is used when h = 0:25. It is possible to calculate backwards to nd out the corresponding damping and natural frequency for the controllers in a) and b). 4h ~ = (I .2h > > : h=2 1 c. 1 + e.9 a.1 Output 0 0 1 2 Time 3 4 5 Stepresponses for the motor in Problem 8. e. LBh=2)M = 4(1 .

8 9 Problem 8. We rst want to compute a state feedback such that A .0:664 9 .3 1 > A . `1 . BL is (s + 3)(s + 2 + `2 ) + `1 = s2 + (5 + `2 )s + 6 + 3`2 + `1 s + 8s + 32: This gives 8 9 L = : 17 3 8 9 75 . BL has the characteristic equation s2 + 8s + 32 = 0: 9 8 Assume L = : `1 `2 then .2 . 4h ) we get = 0:77 !0 = 1:96 The change in the damping is smaller when the modi ed controller is used and the change in the undamped natural frequency is also smaller. h 4 .16 we can calculate the corresponding 8 9 continuous time systems. BL = > :.8 0:336 . For the discretized controller (L = : 2 4 ) we get = 0:71 !0 = 2:31 and for the modi ed controller (L = : 2 . .0:885 9 .L = > : 0:164 0:885 > 8 9 and for h = 0:25 and L = : 1:75 3 we get 8 0:392 . . 1:221z + 0:442 = 0 and z 2 .10 a. `2 The characteristic equation of A . 1:305z + 0:471 = 0: From the equations given in Example 2.L = > : 0:171 0:914 > These two matrices have the characteristic equations For h = 0:25 and L = : 2 4 we get 8 9 z2 .

LBh=2 M = 2 . 1) + 0Bhu(k) + 0 Khy (k) 8 0:81 0:16 9 8 0:03 9 8 0:19 9 > > > > = : .b.1 I x x(k) + Bu(k) + Ky (k) h ^(k) = (A . 2h = 1:6 M b. KC )^ (I .12 a.6 shows the output when using the discrete time controller in a) for di erent values of h.0:16 0:97 x ^(k . 8. Using (8.8 BL)h=2) 8 9 > 1 . Problem 8. 8.7. Modifying L using (8.17h=2 1 .16) and (8.h=2 1 . 3 h = : 0:8 1:7 = :1 2 > : . h 2 . 3h) 3 + h Fig. The response when using the modi ed discrete time controller from b) is shown in Fig. 3h=2 h=2 9 = : 17 3 : .h= 2 1. Ah + KCh)^ x(k) = q. Ah + KCh) = 8 9 1 h > 1 > : 2 1 + h + h .16) gives L = L(I + (A . q . Using the backward di erence approximation gives 1 .1x ^(k) + Bhu(k) + Khy (k) Introduce . BL) h = L > L : 2 .17) give 1 h=2 > ~ = L I + (A .1 0 = (I .h 1 + h 8 9 This gives x ^(k) = 0 x ^(k . h 9 8 98 9 8 9 1 h=2 > = 8 : 1 .h 2 8 9 ~ = : I . 5h=2 > 8 9 = : 17(1 . 1) + : 0:19 u(k) + > : 0:16 > y(k) 76 .

7 2 Time 3 4 77 .10 when the state feedback controller in a) is used with h = 0:1 (dashed) .10 when the modi ed state feedback controller in b) is used with h = 0:1 (dashed).5 Output 0 −0. 0.1 0.5 Output 0 −0. 0.3 (dotted). Figure 8. Figure 8.3 (dotted).2 (dash-dotted) and 0.5 0 1 The response of the system in Problem 8. The response for the continuous-time controller is also shown (solid). The response for the continuous-time controller is also shown (solid).5 0 1 The response of the system in Problem 8.6 1 2 Time 3 4 0.2 (dash-dotted) and 0.

h z .3 qm. e. Hd(z ) = sin(h!)z.1 + 2he.2h z. 2 z cos !h + 1 See Table 2.h )2 Long division gives Hd (z) = he. b.h z Y (z ) = (z he . The time function y (t) = sin(!t) has the z-transform z sin !h Y (z ) = z 2 .1 C (q) = A(q )F (q ) + G(q) 78 .h z Hd(z) = (z he . Temperature control in houses.2 + Using the model of the disturbance gives (q ) w(k + m): y ( k + m) = C A(q ) Introduce the identity Problem 10.2 + sin(3h! )z. e. Problem 10.1 a.3 + b.1.2 a. Consider a system with Hd(z) = Y (z) The impulse response of Hd will thus be sin(kh! ).1 + sin(2h! )z. The desired system thus has the z-transform . Better sensors. for instance a tachometer with less noise. c.t has the z-transform . That this is the correct answer is easily seen by making long division. Flow control with local feedback.h )2 This can be found by looking in a table of z-transforms. The time function t e.Solutions to Chapter 10 Problem 10.

Introduce 8p p 9 12 > P => : p11 p 12 22 79 . 0:5.0:5f1 + f2 0 = .0:5 + f1 0 = . 1.11) we nd that the stationary variance of x ful ls ( stable) P = P T + R1 The stationary covariance exists since the system is stable. Then (q ) qG(q) y (k + m) = F (q )w(k + 1) + qG A(q) w(k) = F (q )w(k + 1) + C (q) y (k) If w(k + 1) : : : w(k + m) are assumed to be zero then the best prediction of y (k + m) is (q ) y (k): y ^(k + m) = qG C (q ) Problem 10. Since R1 is symmetric P is also symmetric.0:5f2 + g0 with solution f1 = 0:5 f2 = 0:25 g0 = 0:125 The predictor at k + 3 given data up to and including k is thus q y ^(k + 3jk) = 0:125 q y (k) = 0:125y (k) Let w(k) be zero except at k = 0 and 5 when it is assumed to be one then k y (k) = 0:5y (k .4 Using (10. 3) .1 0 0 0 1 0 1 0:5 0 2 0:25 0 3 0:125 0:125 4 0:063 0:063 5 1:031 0:031 6 0:516 0:016 7 0:258 0:008 8 0:129 0:129 9 0:064 0:064 10 0:032 0:032 where deg F = m .The operator qG(q )=C (q ) is casual since deg C = deg G + 1. 0:5)(q2 + f1q + f2 ) + g0 This gives the system of equations 0 = . 1) + w(k) y ^(kjk . 1 and deg G = n . Let A(q ) = q . C (q ) = q and m = 3 then q2 q = (q .

then

8 p p 9 8 0:4 0 9 8 p p 9 8 0:4 ;0:6 9 8 1 > 11 12 > = > 12 > > : p12 : ;0:6 0:2 > > : p11 : 0 0:2 > + > :0 p22 12 p22
p11 = 0:16p11 + 1 p12 = ;0:24p11 + 0:08p12 p22 = 0:36p11 ; 0:24p12 + 0:04p22 + 2

0> 2

9

This gives

The solution is

The stationary covariance function is Ex(k + )x(k)T = P It remains to compute . The eigenvalues of are 1 = 0:4 and Using the results on matrix functions in Appendix B we get = 0I + 1 where 0 and 1 are obtained from 1= 0+ 1 1 2= 0+ 1 2 The solution is ; 1 2( 1;1 ; 2;1) = 0 ; and Finally

8 1:19 ;0:31 9 P => : ;0:31 2:61 >

2

= 0:2.

8 9 0 : 4 0 > = : ;3(0:4 ; 0:2 ) 0:2 >
0

1=

1; 2 1; 2

1

2

8 9 1 : 19 0 : 4 ; 0 : 31 0 : 4 > rx( ) = : ;3:57 0:4 + 3:27 0:2 0:93 0:4 + 1:68 0:2 >

From the state space description we get the input-output description of the process y(k) + a1y (k ; 1) + + an y(k ; n) = c1v (k ; 1) + + cnv (k ; n) where ai i = 1 : : : n are the coe cients of the characteristic polynomial of the matrix . Multiply the equation above by y (k ; ) and take the mathematical expectation. y (k ; ) is independent of all the terms on the right hand if > n + 1. Thus ry ( ) + a1ry ( ; 1) + + an ry ( ; n) = 0: This is called the Yule-Walker equation. 80

Problem 10.5

Problem 10.6

There are two noise sources v1 and v2 that is used to generate y (k). Using Theorem 10.2 we nd that the spectral density of y is
y

= H (z ) v H T (z ;1 )
1 z+b

where z = ei!

H (z) = C (zI ;
and Thus

);1; = : 1

8

1 y = : z+a

8

1 2 = (z + a)(z ;1 + a) + (z + b)(z ;1 + b)

2

9;1 8 98 z + a 0 > 1 1 : 0 z + b > 1 = : z+ a 8 2 09 > 1 > v=: 2 0 2 8 11 9 98 2 09 > z +a > > > 1 1 : 11 > z +b : 0 2 > 2
;

9

2

z +b
;

=

Using the spectral factorization theorem (Theorem 10.3) we nd that we can generate the same spectral density by sending white noise through c H1(z) = (z +za+ )(z + b) this gives the spectral density
1

2 ;1 2 ;1 1 (z + b)(z + b) + 2 (z + a)(z + a) (z + a)(z + b)(z ;1 + a)(z ;1 + b)

(z + c)(z ;1 + c) = 2 (z + a)(z ;1 + a)(z + b)(z ;1 + b) gives the relationship
2(1 + c2) = 2(1 + b2) + 2 (1 + a2 ) 1 2 2c = 2b + a 2 1

Identi cation with

y

Problem 10.7
x(k + 1) = ax(k) + v (k) y(k) = x(k) + e(k) This is the same as in Example 10.3 with the exception that Ev (k)e(s) = r12 (k ; s). The covariance function for x will thus be the same but the covariance of y will contain an additional term due to the correlation between v and e. From Example 10.3 we know that r1 rx( ) = aj j 1 ; a2
81 The process is

The covariance of y is

ry ( ) = E fy (k + )y(k)g = E f x(k + ) + e(k + )] x(k) + e(k)]g = = rx( ) + rxe ( ) + rex ( ) + re( ) = rx ( ) + rxe ( ) + rxe(; ) + re ( )
where it has been used that rex( ) = rxe (; ) in stationarity.

rxe ( + 1) = E fx(k + + 1)e(k)g = E f ax(k + ) + v (k + )] e(k)g = = arxe ( ) + rve ( )
The last term is zero except for = 0, where it is r12. The rst term is zero for 0. It is natural that white noise in the future is uncorrelated with the present value of x. This gives

80 < rxe( ) = : r12 a ;1 r12

0 =1 = 0 a ;1 r12 >1

0 1

and

a 1;a2 + a r12 r1 1;a2 + r2 The de nition of spectral density gives
=

8 a; r1 2 + 0 + a; ;1r12 + 0 > < r1 1;a ry ( ) = > 1;a2 + 0 + 0 + r2 : a r1 2 + a ;1r12 + 0 + 0 1;a ( j j r1 j j;1
6= 0
=0
1 X 1 ry ( )e;i! = y (! ) = 2 =;1

<0 =0 = >0

r1 ; r12 1 ; a2 r1 ; r12 + r1 + r = ; 2 i! ; i! 1;a a (e ; a)(e ; a) 1 ; a2 a 1 ; a2 2 2 ei! ; a)(e;i! ; a) + r2a(ei! ; a)(e;i! ; a) = 21 r1a + r12(1 ; a ) ; r12(a (ei! ; a)(e;i! ; a) (1) where it has been used that
= 21

; a2 aj je;i! = (ei! ; 1 a)(e;i! ; a) =;1
The spectral density for is (see Theorem 10.2)

1 X

q ; c "(k) y (k) = q ;a
y (! ) =

(ei! ; c)(e;i! ; c) 2 (ei! ; a)(e;i! ; a)
2

(2)

82

a + r2 12 + r2(1 + a2 ) + z .1 (r12 . a)(z. 2ar = 2(1 + c2) 2 1 12 9 8 98 v ( k ) > : y(k) = H (q ) 1 : e(k) > 8 98 .1 . c)2 r12 = a .1) > 18 : = H ( z ) 1 : r12 r2 : 1 y 2 = 21 r1 H (z )H (z . r12 = 2c A more elegant solution The ouput can be written as 1 . r2a) = 21 z (r12 .Identi cation of (1) with (2) gives the relation 8 > <r1 + r12 1 .1 ) + r2 r1 1 + 1 + r = 21 (z .1) + r12H (z ) + r12H (z . a) ( r2(1 + a2 ) + r1 .a q.7 we nd that r1 = (a . r2a . a2 . r12 1 + a2 + r2a 1 + a2 = 2(1 + c2) a a a > :r12 . a)( 12 . (2zar . c)e(k) Using Problem 10. 1 z . 2c (r (1 + a2) + r . c r2 = 1 Thus aj jr1 + aj j. a z .a where x(k + 1) = ax(k) + (a . a) z . r12 = 2 c 9 which gives the same equations as in the previous method Problem 10. The spectral density of y is given by where H (z ) = z. a 9 > r1 r12 > > H (z.1 . 2ar12 = 2 (1 + c2) ar2 . r2a = . c e(k) = a . c e(k) + e(k) = x(k) + e(k) y(k) = q .8 The process can be written as q .1r ry ( ) = 1 12 . a2 83 . r2a) + r1 .

0:8824 .0:45 0:1129 1:1765 0:1129 This gives :04 1:7 + 0:4 1:5 ry (0) = I2 = (1 .1:5 0:7 1 0:2 0 0:7 ..9 The variance of the process q2 + 0:2q e(k) = b0q 2 + b1q + b2 e(k) y (k) = q2 .1:5 1 2 = 0:7 0:51 . c)2 + 1 = 1:08 ry (0) = 1 . 1:5q + 0:7 a q2 + a q + a 0 1 2 can be determined using Theorem 10. Problem 10. 0:49 0:7 For = 0 r1 + r = (a . a2 = 1:08 Further (10. 0:1 49)1:7 .17) gives 1 (z . The formula for I2 gives B0 = 1 + (0:2)2 + 0 = 1:04 B1 = 2(0:2 + 0) = 0:4 B2 = 0 e1 = 1:7 and The recursive scheme in Section 10. a) 2 1:49 . a)(z .1:5 1 0:7 . a2 The variance can also be obtained from Theorem 10. 84 .4.4 can also be used to get 1 .0:45 1 0:2 . cos ! = y 2 (z .0:45 0:51 0:51 1 = . c)(z . 0:7) = 12:33 2=0 1 = 0:3922 0 = 10:4167 I2 = 10:4167 1:1765 + 0:3922 0:2 12:33 These calculations have been performed in high precision and thereafter rounded.with a = 0:7 and c = 0:5 we get for 6= 0 0:04 + 0:2 (0:7)j j = 0:36(0:7)j j ry ( ) = (0:7)j j 1 . a2 2 1 .4. 2 2ac ry (0) = I1 = 1 +1c.1 . c) ) = 1 1:25 . 1:4 cos ! where z = ei! .1 . 1:5 1:5(1 .

2 .2) + (.2) (. 2a cos ! = 2 1 + a2 b2 1 2a cos ! . 4e(k .10 The process is y (k) = e(k) . a e .i! ) = H (ei! ) 2 H (e. 2e(k . 2) . i! 2 . 2e(k .i! ) 1 b2 b2=2 = = .i! = b2=2 1 + a2 . 2)2 + 16e(6 .i! e . a 2 ei! + e.4 ry (4) = 0 k 4 Problem 10. a 1 y (! ) = H (ei! ) u (! )H (e. 4e(k .4) = . b2 2 Identifying with the desired spectral density gives 2 (1 + a2 ) = b2 1:36 2 (2a) = . 1 = 1:36 + 1 :2 cos ! Using Theorem 10.20 ry (2) = 1 3 + (. 3)2 + crossterms = 1 + 4 + 9 + 16 = 30 The mean value of the crossterms are zero since E fe(k + )e(k)g = 0 In the same way 6= 0 ry (1) = 1 (. 3) ry ( ) = Ey (k + )y (k) This gives ry (0) = E (e(k) . 3))2 = E e(k)2 + 4e(k .Problem 10.2 we get b H (z) = z .11 y a.1:2b2 85 . 1) + 3e(k . 1)2 + 9e(k . . a 1 + a2 .4) = 11 ry (3) = 1 (. 2) .2) 3 + 3 (.4) = . 1) + 3e(k .

Divide the two equations 1 + a2 = . 1:2 1:2 r 2a p b = . 86 .12 The stationary covariance is given by 8 0:3 0:2 9 809 > > x(k + 1) = : 0 0:5 x(k) + > : 1 > u(k) + v(k) 8p > : p11 12 P = P T + R1 9 8 0:3 0:2 9 8 p p 9 8 0:3 0 9 8 1 0 9 p12 > = > >> >+> > 11 12 > > : : : : p22 0 0:5 p12 p22 0:2 0:5 0 0:5 p11 = 0:09p11 + 0:12p12 + 0:04p22 + 1 p12 = 0:15p12 + 0:1p22 p22 = 0:25p22 + 0:5 8 1:1385 0:0784 9 P => : 0:0784 0:667 > Problem 10.13 Example 10. 2 0:6 0:64 H (z ) = z +20:6 p Problem 10.0:6 1 : 36 a = . 5:40 cos ! = 2 1:81 . 2a cos ! In this case r1 = 1.4 CCS Var y = I1 = 1 . Identify with the desired spectral density gives 3 1 1 1 2 5:43 . Theorem 10. a gives the spectral density r1 b2 (! ) = 2 1 + a2 .4 shows that the lter b H (z ) = z . 1:8 cos ! This gives a = 0:9 and b = 1.1:2 2 = 2 The desired lter is 2 2 = b. 1 = . 1:36 ) a2 + 2:72 a + 1 = 0 2a 1:2 1:2 r 1:36 2 .

a(s.ah )2 e.6)-(11.a(t.kh) ds = 21a (1 . 2ba3 3 .2ah) 2 + % h .kh) )2 + % ds Q2 = a2 (1 . e.kh) a 2 a2 e.ah + e.a(t.a(s.a(s. e = b (1 .2 80 x _ => :0 8 y = :1 = eAh . e.Solutions to Chapter 11 Problem 11.8).= 1>x + >0>u :1 0 9 0 x 9 8 9 Sample the system Zh 0 81 => :0 eA 8 h2=2 9 Bd => : h > 87 h> 1 9 .kh) ) ds = b (1 . 4e. Problem 11.kh) ) e.2ah Notice that there will be a Q12 term even if Q12c = 0.kh) and .s)b ds = a The discrete time loss function is obtained from CCS (11.2a(s. e.a (t kh) dt This di erential equation has the solution (kh kh) = 1 (t kh) = e.1 For the system we have d (t kh) = .a(t. e.(t kh) = Zt kh b (1 . which gives Q1 = Q12 = Z kh+h Zkh kh+h kh b2 a2 Zkh kh+h b2 .

T S 8s s 9 12 > S=> : s11 s 12 22 . The loss function is y 2 + u. ab b2 x(k) = . T S .1 .17) gives + 1)2 = 1 k = N .). 1 :: 1 s(k) = a2s(k + 1) + 1 .Sample the loss function Q1c = I Q2c = 1 Q12c = 0.19) gives which gives the controller L(k) = b. 0> 0 The steady state value of the Riccati equation is obtained from Q1 = C T C 81 => :0 9 S= Let 88 TS + Q1 . Problem 11. a2 b2 bs2(sk(k + 1) and (11. i. Using (11.(Q2 + .L(k)x(k) = .3 The Riccati equation(11. b x(k) min J = x(0)2s(0) = x(0)2 x(k + 1) = 0 The state is zero after one step and the resulting controller is thus a dead-beat controller.5 a.T S .e.8) we get Zh 8 Q1 = > :1 Zh 0 0 0 > > 1 > d =Z 1 :0 1 0 98 9 h 81 > : 81 Q12 = > : Q2 = Zh 0 ! 8 2 98 2 =2 9 > > : =2 +1 d = : h8 98 9 0 > > 2 =2 > d = Z > : 1 : 0 8 h 2=2 9 h > = : h2 =2 h3 =3 + h > 2+1 8 h 3 =6 9 2 =2 9 > d => : h4=8 + h2=2 > 3 =2 + ! Zh 4 h3 h5 0 9 >d 1+ 2 + 4 d = h + 3 + 20 Problem 11.6)-(11.2 ba = a b The minimum loss is given by and the closed loop system is a u(k) = .

K (%).1 The pole in Problem 11.5 as a function of the control weighting %.K (%) > The poles are obtained from ( . % + s The solution is The feedback vector is where 1 + 4% s12 = s22 = 1 + 2 s11 = 1 + s12 p 8 9 L = K (%) : 1 1 p1 + 4% s 1 + 12 K (%) = % + s = 2% + 1 + p1 + 4% 22 b. Fig 11.5 0 −2 10 10 −1 Figure 11. 1)( + K (%)) + K (%) = ( . K (%) = 0 and as % ! 1 then 1 . For % = 0 then 1 . . % + s 22 22 22 s12 = s11 .L = : . % + s s2 12 s2 12 s22 = s11 .K (%) . K (%) ! 1. The dynamics of the closed loop system is 8 1 9 1 > . 89 .1 shows how the pole varies as a function of %. 10 10 10 Control weighting rho 0 1 2 10 3 10 4 For the inventory model we get s2 12 s11 = s11 + 1 .1 Pole 0. 1 + K (%)) = 0 There is one pole in the origin and one in 1 .

u(k) 2 0 −2 0 5 10 b) y(k).T S . z ) + 1 = r(z2 + p1 z + p2 )(z . Figure 11. K ( ). For the inventory system A(z) = q(q . p 90 .1)(z2 . 2% + 1 .p1 = 2% + 1 . u(k) 2 0 −2 0 5 10 c) y(k). u(k) 2 0 −2 0 5 Time 10 d) y(k). u(k) 2 0 −2 0 5 Time 10 The output (solid) and the control signal (dashed) for the system in Problem 11. 2% p The poles of the closed loop system are thus one at the origin and one at 1 + 4% .2 The poles of the closed loop system can also be determined from (11.2 + p1 z.40). 1) B (z ) = 1 and we get %(z . + Q2 6= 0 then the rst equation implies that p2 = 0 and we get ( .% = r(p1 + p1p2) 2 2% + 1 = r(1 + p2 1 + p2 ) Since r = .% = rp1 2% + 1 = r(1 + p2 1) which has the solution 2 2%p 2% + 1 + 1 + 4 % r = 2% + 1 . = 2 1 + 4% p 1 + 4% p1 = . c) % = 5 and d) % = 25. z.2 .5 when a) % = 0.1 + p2) or 0 = rp2 .a) y(k). 2% It is easily seen that this is the pole in 1 . b) % = 0:5.

sampling of the system.3 The closed loop poles in Problem 11. 1:65z + 0:68 z.6 when % is varying.2 .9.1 + 0:026 0:030z + 0:026 % + z2 .6 % + H (z)H (z .1 + 0:68 0:030z + 0:026 0:030z + 0:026z 2 = 0 =% + z 2 . 1:65z + 0:68 Only the output and the control signals are penalized in the loss function.e. sampling of the loss function and solving the Riccati equation. With numerical packages. The system has the transfer function 0:030z + 0:026 H (z) = z2 . i. 1:65z + 0:68z2 or 0:00078z 3 + 0:001576z 2 + 0:00078z + %(0:68z 4 .3 shows the closed loop poles when % is varying.9 Solving LQ-problem for system of higher order than one by hand causes much work. 1:65z + 0:68 1 .1 Im 0 −1 −1 0 Re 1 Figure 11. the design is signi cantly simpli ed for the control engineer. Problem 11. 91 . 2:7720z 3 + 4:1849z2 . 11. The closed loop system has poles that are determined by the stable roots of Problem 11. The following Matlab-macro illustrates the solution of Problem 11. 2:7720z + 0:68) = 0 Fig. 1:65z. like Control toolbox for Matlab.1) = 0 This gives 0:030z .

%Macro for Problem 11.001 k=0.Gam. z + (z2 + p1z .4.Gam.e.h.Lv.Q12.Q2.R1. 11.1 < p1 + .37) may be used to get the exact values of the gain margin.Q1c.2)) %Linear quadratic regulator design for discrete-time system L.10 In Problem 11.1 + 1 + 4% = min max and max are also shown in Fig. the system is stable if .9 CCS alpha=0.08 h=5 A= 0 1 0 -alpha] B= 0 k] Q1c= 1 0 0 0] Q12c= 0 0] Q2c=rho %Transform continuous-time LQG problem to the corresponding %discrete-time LQG problem Phi.Q12) L The design gives L = 3:055 108:7] Problem 11.Q2c.. 1 < 1 ) 4 p min = 0 .Q1.Je] = lqgsamp(A. 1)) = 0 I. z P (z ) = z2 + p1z we get z 2 .Q12c.zeros(2. then %= 2% r 2% + 1 + p1 + 4% Equation (11. 92 .Q1. z 2 + z) = z (z + ( p1 + .B.S] = lqrd(Phi. With A(z) = z2 .0005 rho=0.5 we have determined r.Q2.

11 has a control signal u(k). In the second case P = 1:13r1 and . 1) + K (k)(y (k) . The system is Additional problem Suppose that the system in Problem 11. Q12 = 0 and Q2 = . In the rst case P r1 and .5 shows . 11. i. KC 0:5.11 x(k + 1) = 0:5x(k) + v (k) y (k) = x(k) + e(k) Theorem 11. Finally if r1 r2 then P 1:33r1 and . 0 0 r2 + P (k) The dynamics of the lter is determined by :5r2 . r1 = r2 and r1 r2. Fig.10 Gain margin 5 0 0 5 10 Control weighting rho Figure 11. the system is x(k + 1) = 0:5x(k) + v (k) + u(k) y(k) = x(k) + e(k) Determine a steady-state LQG-controller when Q1 = 1. Problem 11. KC = 0:5 . 1)) x ^(0j . KC for di erent values of r1 when r2 = 1.4 The gain margin min (dashed) and max (solid) from (11. x ^(kjk . r1)P = r1 r2 Consider three cases r1 r2 . 1) = 0 > > > < K (k) = 0:5P (k) r2 + P (k) > > :25P (k)2 P (0) = r > : P (k + 1) = 0:25P (k) + r1 .5 gives that the Kalman lter is de ned by 8x ^(k + 1jk) = 0:5^ x(kjk . 93 . KC = 0:23.e. K (k) = r 0 2 + P (k) The steady state variance is given from P 2 + (0:75r2 . KC 0.37).

1K Y (q) LK Hreg (q) = q . 0:25 +S 0 : 5 S L= +S which has the solution 2 2+4 L = .19) gives S S = 0:25S + 1 . KC )^ x(kjk . C x ^(kjk . (. 1) + Ky (k) and thus or U (q) = .0. . 1) + .5 Pole 0 0 Figure 11. 1)2 + 4 p Using the Kalman lter from Problem 11. Equation (11. P (k)C T (R2 + CP (k)C T ). 5+K +L Problem 11.L + KC ).1CP (k) T . 1)) = ( .47) gives P (k + 1) = P (k) 94 T + R1 . 1) 2:5 + 2 + 2 (0:75 .u(k) + K (y (k) . + . 0:.12 a. 1) + .17) and (11.L . C x ^(kjk .L(qI .11 and the LQ-controller gives x ^(k + 1jk) = x ^(kjk .0:75 + 1 +p (0:75 .Lx ^(kjk . 1)) = x ^(kjk . 1)) + K (y (k) . 5 State weighting r1 10 Solution to the additional problem Equation (11.5 The pole of the Kalman lter in Problem 11.11 for di erent values of r1 when r2 = 1.

e. The steady state value of P is 81 19 P => :1 2> 2 = 0 The x ^(0j . p12(k)(p11 11 p11(k) 2 p22(k + 1) = p22(k) + 1 .12. KC )) = lter has a dead beat response. 95 . 11. with 1> 18 9 0 0 > T R1 = . p12(k) = p11(k)p22 p11(k) (k) + p12(k)) = p (k + 1) p12(k + 1) = p12(k) + p22(k) .v = : 0 1 > 8 9 C = :1 0 2 81 => :0 9 we get ) + p12(k)) p11(k + 1) = p11(k) + 2p12(k) + p22(k) .4 p11(k) 2 0 p12(k) 0 Time p22(k) 5 4 2 0 4 2 0 0 5 0 Time 5 Figure 11. p p12 = 1 + p11(k + 1) 11 (k) Further 8 9 8 9 8 9 k p ( k ) + p ( k ) 2 1 > > > > > > 1 11 12 K (k) = : k = : p (k) p11(k) = : 1 2 12 k>0 The poles of the lter are found from det( I . K is timevarying. 1) = : 1 1 8 9T and assume that P (0) = 3I . Fig.v . The initial values of the lter are For k = 0 K (0) = 1 0]T i. (p11(kp 11(k) 2 (k) .5 shows the elements of the covariance matrix as a function of time. b.6 The elements of the variance matrix in Problem 11. ( .

Using Theorem 10.Problem 11.x(k)]= (1q . v (k +1) and v (k +2) are independent of y (k) then the best estimate of x(k +3) is given by x ^(k + 3jk) = 3x ^(kjk) + ( 2 + 81 + I ). Since v (k). )q e(k) = .1 = : 1 and .4 we get 2 1. x(k) x ^(kjk).2 The best estimate of x(k) given y (k) is determined from (11.15 x(k + 1) = ax(k) + v(k) y (k) = x(k) + e(k) We use the exponential smoothing estimator cov v = 1 cov e = x ^(kjk) = x ^(k . a ) + + 1 96 .2 = > :0 3>x 4:5 > > ^ ( k j k ) + : 1 3 T 9 8 9 The variance of the estimation error is 2 )T + . q.50). .14 Introduce the notation 81 19 809 8 0:5 9 > > > > = :0 1 .a q .2 = 3 x(k) + 2. ! (1 .a q . 8 9 Problem 11.T P (k + 3jk) = 83 P (kjk )( 3)T + 0:01( 2. ) q 1 1 v (k) v ( k ) + e ( k ) .2 + . )y (k) )q y (k) . 1) v (k) + (1 . (a )(q .1 v(k + 2) + . = q.1 v (k + 2) + 2.2 + . ) q. 1jk .2 = > :1> The state at time k + 3 can now be determined x(k + 3) = x(k + 2) + .2 + .T 1( 8 9 8 9 91 1 1 3 1 0 5 3 => : 0 1 > P (kjk) > : 3 1 > + 0:01 > :3 3> If x(0) is known then P (0j0) = 0 and + . (q .1 v (k) + . 1) + (1 .T 1) y ^(3) = : 1 3 x(0) + 4:5 and the variance of the prediction error is 0.1 . .1 v (k + 2) + . var x ~(kjk) = (1 + a)(1 2 + )(1 .1 v (k + 1) + .05.1 .2 = 2 x(k + 1) + .1 v (k + 1) + .

(1 + a)2 + 1 min = a2(1 + a) + a . Kalman variance a2 P 2 P = a2 P + 1 .P + = . 1 . a(1 + a) + 1 .16 The scalar state equations are ( x(k + 1) = ax(k) + u(k) + v (k) y (k) = x(k) + e(k) v (k) = v1(k) + mv Ev1 = 0 e(k) = e1(k) + me Ee1 = 0 Let and 8 9T : X = x mv me 8 8a 1 09 819 819 > > > > > > > > > > > > > <X (k + 1) = > 0 1 0 X ( k ) + 0 u ( k ) + 0 > > > > > :0 0 1 :0 : 0 > v1(k) > 8 9 > : y(k) = : 1 0 1 X (k) + e1 97 . 1jk .(1 . 1jk . a2) + 1 p1 + 2 (1 + a2) + 2(1 . P + This gives . P P = + P+ a K = PaP + Numerical values: = 1 a = 0:5.a2P 2 . Exp. use Maple. a2)2 P = 1 . 1) = (I . 1) + K (y (k) . (1 2 2 The gain in the Kalman lter is 2 P = K var x ~(kjk) = P . 1 Kalman with direct term p x ^(kjk) = x ^(k . smoothing: Kalman: = 0:4222 K = 0:2650 var x ~(kjk) = 0:6181 var x ~(kjk) = 0:5311 Problem 11.Minimize with respect to . KC ) x ^(k . 1jk . C x ^(k . 1) + Ky (k) This will have the same form as the exponential smoothing estimator only when a = 1. a2)P .

The observability matrix is then 8 C 9 81 > > > > >a Wo = > CA = > > : 2 > : 2 CA 0 19 > 1 1> > a a+1 1 rank Wo = 2 This means that me and mv are not both observable and no Kalman. The correct mean value should be obtained. a1 x(k) . possible to design a second order observer with reconstruction of a linear combination of me and mv . 18 = 0 () a1 = 10 The estimator is thus 9 y (k) + 1 y (k) x ^(k) = 10 1 10 2 98 Problem 11. Rede ning the state vector X as 8x 9 1> X=> :m where ( x1 = x + me m = (a . x ^(k))2 = E (x(k) . 2.lter can be designed. Condition 1 gives that a1 + a 2 = 1 The variance of the estimation error is V = E (x(k) . The observability matrix is with 8 8 9 C > > Wo = : C = > :1 a 0> 1 9 from which follows that rank Wo = 2. The variance should be minimized. a2 e2 (k))2 2 2 2 2 = a2 1 1 + a2 9 = a1 + (1 .17 . The constants a1 and a2 are determined by the following two conditions 1. however. 1)me + mv gives 8a 19 819 819 8 > > > > > : 0 > v1(k) <X (k + 1) = : 0 1 X (k) + : 0 u(k) + > 819 > : y (k ) = > : 0 > X (k) + e1(k) Reconstruction of x1 and m is possible if these states are observable. It is. 18a1 Taking the derivative with respect to a1 gives the condition 9 20a1 . a1 ) 9 = 10a1 + 9 . a1 e1(k) . a2 x(k) .

e.0:45 9 819 > > x(k + 1) = : 1 x(k) + > : 0 > u(k) 0 8 9 y(k) = : 0:5 0:38 x(k) 8 0:25 0:19 9 > > T Q1 = C C = : 0:19 0:1444 Q12 = 0 Q2 = 0 The steady state solution is obtained from (11.50) { (11. Using Matlab we get the solution 8 0:25 0:19 9 ) S=> : 0:19 0:1444 > 8 9 L = : 2:21 . 1) If p is large then the weights for y1 and y2 will be those calculated above. K (1) = P (1j0)C T R2 + CP (1j0)C T .17). a combination of the measurements gives a better result than using only the best measurement. This gives 8 9 L = : 2:21 . i.1) + B(z )B (z.1 ) = 0 ) P (z ) = b1 zB (z ) = 2z (0:5z + 0:38) = z (z + 0:76) 1 Now we have to nd L such that 8 .1 ) = rP (z )P (z. . Thus. Assume that the a priori estimate of x is zero and the variance of x is p.54) we get P (1j0) = p and 8 9 .0:45 99 .0:76 0 9 ( . x ^(0j0) and P (0j0) = p From (11.40) A(z)A(z .L) = > : 1 0> where controllable canonical form have been used.20 8 1:45 .9 Vmin = 10 Using only y1 gives the variance 1 and only y2 the variance 9. In the example R1 = 0 the steady state gain will then be zero since the estimate will approach the true value of x. The minimum value of V is Problem 11.0:45 An alternative solution is to use (11.1 = 10pp+ 9 : 9 1 This gives 9p y (k) + p y (k) + 9 x x ^(kjk) = 10p +9 1 10p + 9 2 10p + 9 ^(kjk .

R0 = E x2 (0) = 1 :5P (k) K (h) = 10+ P (k) P (k + 1) = 0:25P (k) .L(k)x(k) = . First assume that = 0 and use linear quadratic design with Q1 = Q12 = 0 Q2 = % Q0 = 1 N =2 Theorem 11. x ^(k) with .21 a. Theorem 11. This gives k P (k) K (k) 0 1 0:25 1 0:125 0:056 100 . In this case = 1 and x(k) is reconstructed using a Kalman lter x ^(k + 1) = 0:5^ x(k) + u(k) + K (k) y (k) .Problem 11. 0:5P (k)K (k) with P (0) = R0 = 1. For di erent values of % we get % 1:0 0:1 0 L(0) 0:056 0:093 0:1 L(1) 0:250 0:455 0:5 b.5 gives R1 = R12 = 0 R2 = = 1 . S (N ) = S (2) = Q0 = 1 :52% S (1) = 0 %+1 ) ) :53 L(0) = % + 0 1 + 0:52 :5 L(1) = %0+ 1 The control law is u(k) = .1 gives the Riccati equation S (k) = 0:5 0:5 . L(k) S (k + 1) 0:5S (k) L(k) = % + S (k + 1) This gives .Ly (k).

1 8 1 9 > :1> p = 0:005 c.The control law is u(k) = .L(u)^ x(k).1 8 1 9 8 98 2+p p > > > 1 2 p :1 1 : : 1 > = 0:01 2+p p 2 2+ 2 2 p2 2 2 +1( 2 + 2 = 0:01 1 2 1 2 )p 2 2 2 =0 p2 . 1) + Ky (k) :1> x 819 8 9 8 9 8 2 09 > 1 :1 1 > : 0 2 +> :1>p:1 1 2 9. Problem 11. 1 8 98 2 +p p > > 1 : K=p 1 1 : p = 2 2 + p( 2 + 2) p 2 1 2 2+p 8 9 0:09458 8 91 2 = : 4 1 4 + 5 0:09458 = : 0:0846 0:0211 2 0:0052 + 0:01 2 1 2 + 1 2 r = 0:005 + 0:0052 + 0:01 4 5 = 0:09458 s 2 2 101 . x(k + 1) = x(k) + v (k) 819 y (k ) = > : 1 > x(k) + e(k) 8 2 09 R1 = 0:01 R2 = > : 01 2 > 2 819 x ^(k + 1jk) = 1 . (11. K > ^(kjk .22 a. 8 2 29 9 : 2 1 .47) ) p = p + 0:01 . 0:01p . 0:01 2 1 2 1+ 2 b. p2 .

1:2 + 1:2 0:4) 1:2 = 4:28 This implies that the prediction variance is almost the same as the variance of y when m 3. 1:2q + 0:4)(q2 + f1q + f2) + g0q + g1 which gives F (q) = q2 .0:2q + 0:1 The predictor is then given by 0:2q 2 + 0:1q y (k) y ^(k + 1jk) = q. 0:14 G(q) = .17) and the variance of the prediction error is given by (12. For m = 1 we get q2 .4 we can compute the variance of y to var(y ) = 2 2 :4)+2(.19). 1:2q + 0:4 + g0 q + g1 which gives G(q) = g0q + g1 = . 1:4q + 0:5 and the variance of the prediction error Ey ~2(k + 1jk) = 2 = 4 For m = 2 q(q 2 .0:088q + 0:056 2 2 ) = 4:24 Ey ~ (k + 3jk) = 2 (1 + f12 + f2 Using Theorem 10. 1:4q + 0:5) = (q2 .1 The m-step ahead predictor is given by where G is obtained from the identity (12.Solutions to Chapter 12 Problem 12.0:4(1+0:4)) = 22 (1+1:4 +0:5(1)(1+0 . 1:2q + 0:4)(q + f1 ) + g0 q + g1 This gives F (q) = q . 1:4q + 0:5 = q2 .1:4 0:5) 1:2+2 0:5(1:22. 0:2 G(q) = .1:4. 0:2q . 102 (q ) y (k) y ^(k + mjk) = qG C (q) . 2 . 1:4q + 0:5) = (q 2 . 0:42)(1 + 0:4) + (.0:14q + 0:08 and Ey ~2(k + 2jk) = 2(1 + f12 ) = 4:16 For m = 3 we get q 2(q 2 .

fm. The C -polynomial has a zero outside the unit circle. a) . 0:9)(q + f1) + g0 This gives F (q ) = q + 1:1 G(g ) = 0:99 103 . a f2 = (.2 The identity (12. The two-step-ahead predictor is obtained from q(q + 0:2) = (q . a2 + a2(m.2 + ::: + fm.1 + g0 Ey ~(k + mjk) = 2(1 + (c .1 ) + g0 This gives f1 = c .1) 1 . a)2 + a2 (c .a)2(c . 0:9y(k .17) is qm.1 = (.2 + fm.1 + f1q m. Example 12.2(c . . .a) q +(c c and the variance of the prediction error is The solution is c = a + f1 0 = af1 + f2 .a)(c .1 0 = afm. 0 = afm.a)m. a)2 + ! 2(m. . 1) = 5(e(k) + 0:2e(k . 1)) b. a 2 2 = 1 + (c .1 shows how the C -polynomial should be changed.3 a. a) The m-step ahead predictor is m. a)q y (k) y ^(k + mjk) = (. a) g0 = (. It is replaced by C (z ) = 5z + 1 = 5(z + 0:2) The equivalent process is thus y (k) .Problem 12. a) f3 = (.a)m. .1(c .1 .1(q + c) = (q + a)(q m.2) (c . a)2) Problem 12. a) 1 .

zd is the deterministic part of z . y (k). 1) y ^(k + mjk) = qG 0 1 C (q) which gives the predicted values and their standard deviation .5 330 4 330 .This predictor is and :99q y (k) y ^(k + 2jk) = q0+ 0:2 Ey ~2(k + 2jk) = 25(1 + 1:12) = 55:25 Problem 12. . . k z(k) y(k) zd (k) 1 320 10 310 2 320 0 320 3 325 .10 340 5 350 0 350 6 370 10 360 7 375 5 370 The prediction of the demand for August through November is z ^(8j7) = zd (8) + y ^(8j7) . m 1 2 3 4 104 y ^(7 + mj7) 4:5 3:7 3:0 2:5 zd (7 + m) 380 390 400 410 z ^(7 + mj7) 384:5 393:7 403:0 402:6 5 6:1 6:8 7:2 . 3 and 4 step ahead predictors of y.17) and give m 1 2 3 4 The prediction is F (q) 1 q + 0:7 q2 + 0:7q + 0:59 q3 + 0:7q2 + 0:59q + 0:48 G(g) 0:7q + 0:1 0:59q + 0:07 0:48q + 0:06 0:40q + 0:05 (q ) y (k) = g y (k) + g y (k . z ^(11j7) = zd (11) + y ^(11j7) We thus need the 1.4 Using the data given up to time k = 7 it is possible to calculate y (k) and zd (k) = z (k) . Those are given by solving the identity (12. 2.

deg B = 2 The Diophantine equation is qd. 0:7(2 + (k . 1) = u + u ~(k .27) (q ) y (k) = . 2) + (k) . Introduce e(k) = 2 + (k) u(k) = u + u ~(k) where (k) is zero mean white noise.1 C (q ) = A(q ) F (q ) + G(q ) q(q3 + 0:8q2 + 0:25q ) = (q3 . 0:2 and :1q y (k) .1 + f1 0:25 = 0:5 . q2 + 0:5q B = q + 0:5 C = q3 + 0:8q 2 + 0:25q It is easily seen that C is a stable polynomial. 1)) =u ~(k . 2) + 2 + (k) . 0:6 u(k) = q 0 .g. 0:7 (k . by the stability triangle. This is a necessary condition for the minimum variance design method. e.5 A = q3 . The process is now y(k) .0:6 and the problem is reduced to the standard problem. B(G q )F (q) (q + 0:5)(q + 1:8) The loss function is Ey 2 = 0:52(1 + 1:82) = 1:06 The noise sequence has a non zero mean value. 0:5y (k . q2 + 0:5q )(q + f1 ) + (g0q2 + g1q + g2) Identifying coe cients of powers of q gives 0:8 = .Problem 12.0:9 g2 = 0 The minimum variance regulator is obtained from (12.0:25 + f1 = 1:55 g1 = .1:55q 2 + 0:9q y (k) u(k) = . f1 + g0 0 = 0:5f1 + g1 0 = g2 Solving these equations f1 = 1:8 g0 = .6 . 1) + u + 0:6 Choose u = . The pole excess is d = deg A . The identity gives :1q y (k) u ~ = q0 . 0:2 105 The polynomials are Problem 12.

1 q +10:6 e(k) + 1 q u(k) or (q 2 + 0:1q .9 a.2 and Fig. Problem 12. This implies that u(k .3 shows the output and the control signal for the same noise realization when the minimum variance controller is used with d = 1 and d = 2.1 shows the output of the open loop system for one realization of the noise sequence. 12.8 a. The identity gives F (q) = q + c . qa+ (c .2) 2 1 (! ) = 2 1 + a2 + 2a cos ! This implies that = 1 and a = 0:6 gives the desired spectral density.7 a. 2) cannot be used to decrease the variance of y (k). The expression above gives the optimal controller u(k) = 0 if a = 0. Assume that 1 H (z) = z + a Sending white noise e(k) through this lter gives the spectral density (see Theorem 10. a) b. B(G q)F (q) y (k) Problem 12. Fig.Problem 12. 12. F (q) = q + 3:2 G(q) = 5:64q 2 . The process is now described by y(k) = 1 . The identity gives for d = 1 and for d = 2 F (q ) = 1 G(q) = 3:2q + 0:2 and the minimum variance in the two cases are d=1: Ey 2 = 1 d=2: Ey 2 = 1 + 3:22 = 11:24 b. c)q y (k) u(k) = . 2:24q The minimum variance controller is (q ) u(k) = . a G(q) = a(a . 0:3) y (k) = (q + 0:6) u(k) + q e(k) 106 . The output is drifting since the A-polynomial contains an integrator. c)q and the minimum variance controller is (a . Fig. 12. The process is then a moving average process of rst order. 01:5q.

Ky (k) (q 2 + 0:1q . Use the controller This gives u(k) = .(q + 0:6) K y (k) + q e(k) y(k) = q2 + (0:1 + K )qq + (0:6K . 100 Time 150 200 Output 0 −10 0 50 100 150 200 20 Input 0 −20 0 50 100 Time 150 200 The output and the control signal when d = 1 and the minimum variance controller is used for the process in Problem 12. Figure 12. 0:3) y (k) = .200 Output 100 0 0 50 Figure 12.0:5 < K < 1:5 107 .1 10 The output of the open loop system in Problem 12.8. 0:3) e(k) The system is stable if .2 b.8.

266K + 1 = 0 which has the roots K . 0:25)q + 0:5 e(k) 108 . Only the root K = 0:004 gives a stable closed loop system. 1.4 gives an expression for the variance of a second order process. The minimum variance is Ey 2 = 1 since d = 1. c.10 Output 0 −10 0 50 100 150 200 20 Input 0 −20 0 50 100 Time 150 200 Figure 12. :7 + 0:6K I2 (K ) = (1:3 .839 and 0. (0:1 + K )2) dI2 dK = 0 This gives the third order equation 72K 3 + 12K 2 .Ky (k) we get the closed loop system q 2 + 0:5q y (k) = q2 + (K .2:009. The minimum value of I2 is obtained from Theorem 10. With the proportional controller u(k) = . Problem 12. For K = 1 we get I2 = 3:87. 0:6K )((00 :7 + 0:6K )2 .10 a. The value of the variance is I2 (0:004) = 1:12 d.2 but when d = 2.3 Same as Fig.004. 12.

0:25 0:5 > . 2) . From Example 3.) and K = 2:125 give an unstable closed loop system and the calculation of I2 is no longer valid. 1:5q 2 + 0:7q B(q) = q . K = 0:5(1:75 . 1:5y(k .1:25 < K < 1:75 Both K = 3:25 (the second root in a. K + 0:25 or . 0:25) 1:5 . 0:5u(k .1 + K . 0:5 Note that the process zero (0. deg Bm deg A . 3) + v (k) gives A(q ) = q 3 . 4:25K + 3:25 = 0 with the solutions K = 1 and K = 3:25. deg B + .2 we nd that the closed loop system is stable if 0:5 > . 2) = u(k .Using the results in Theorem 10. deg B = 2 deg Ao 2 deg A . 1 109 . Problem 12. 0:25) I2 (K ) = (1 . This means that the process zero can be cancelled without problem. 0:25)2(1 . K )(1:25 + K ) Taking the derivative of I2 and putting the derivative equal to zero leads to the equation K 2 .11 y (k) . (K .5) is stable and well damped. b.4 gives B0 = 1 + 0:52 = 1:25 B1 = 2(1 0:5 + 0:5 0) = 1 B2 = 0 e1 = 1:5 and 1:25 1:5 . K = 1 gives the variance 4 I2 (1) = 3 This is minimal variance for the present control law. 1) + 0:7y (k . With minimum variance control we would get E y 2 = 1.1 . The degree conditions gives deg Am . 0:5) 2:125 . (K . deg Am .

v (k) = 0 Deadbeat Control Am(q) = q 2 B + (q) = q . S (q) y (k) u(k) = R (q ) c R(q) where R(q) = R1(q)B + (q) = (q + 1:5)(q . 1) Minimum variance The polynomial C is given by C (q) = q3 . (q) = 1 0 (q ) = 1 Bm Ao (q) = q2 The polynomials R1 and S are obtained from the Diophantine equation A(z )R1(z ) + B . 1:5z 2 + 0:7z )(z + r1) + s0 z 2 + s1 z + s2 = z 4 with solution This gives the regulator r1 = 1:5 s0 = 1:5r1 . 0:5 B .0:7r1 = . 1:05q u(k) = . B(G q)F (q) y (k) where the polynomials F and G are obtained from q d. 0:5) b. 0:7 = 1:55 s1 = .1:05 s2 = 0 T (q) u (k) .a. 1:55q2 . 0:2q 2) = (q 3 . 0:2e(k . 0:5) T (q) = Bm Ao = q 2 Assuming that uc (k) = 0 gives S (q ) y(k) = . R (q ) (q + 1:5)(q . 1 the equation becomes (z 3 . (z)S (z) = Am(z)Ao(z) Recalling the condition deg S = deg A . 1 = 1 deg G = 2 q(q3 .1 C (q) = A(q )F (q) + G(q) which in this case is deg F = d . 1:5q2 + 0:7q)(q + f1 ) + g0 q 2 + g1q + g2 110 . 0:2q2 The minimum variance control law can be written as (q ) u(k) = . v (k) = e(k) .

1 .1)F (q)e(k) = F (q .0:91 g2 = 0 The minimum variance controller is thus given by :25q2 . The system can then be written as 1 u(k) + C1 e (k) y(k) = B A1 A1 1 (12.4 C (q )R1(q )e(k) = C (q . 0:3q . 1:5f1 + g0 = 0 0:7f1 + g1 = 0 g2 = 0 with solution f1 = 1:3 g0 = 1:25 g1 = . The variance of the output is then simply calculated as E (y 2) = (1 + 1:32 + 0:32) 2 = 2:78 In the minimum variance case the output is 2 y (k) = q. Fig. 12.1)e(k) which also is an MA process.12 Introduce the polynomials A1(q ) = A(q )D(q) B1(q ) = B (q)D(q ) C1(q) = A(q)C (q) and the noise e1 = e. The output is in the deadbeat case given by CR1 y(k) = ARCR + BS e(k) = Am Ao e(k) = q .1 )R1(q . 0:5)(q + 1:3) c.1 )e(k) = (1 . 0:2q .1)e(k) = (1 + 1:3q . (q1. 1:5 = . The output variance is E (y2) = (1 + 1:32) 2 = 2:69 2 d.(d.2)e(k) which is a moving average (MA) process. Problem 12.5 shows the output and the sum of the square of the output for the regulators in a and b.0:2 0:7 .1 )(1 + 1:5q .which yields the equations f1 .1) 111 . 0:91q y(k) u(k) = .1)e(k) = (1 + 1:3q.

1) where F and G satisfy the equation qd.1 )D (q . G1(q ) = A(q )G(q ).1 )A (q . The deadbeat controller is used in the upper diagram and the minimum variance controller in the middle diagram.11. G (q . A1 = AD and C1 = AC will also bo monic. deg B1 = deg A .1 ) y (k) u(k) = . 1 and deg G = deg D . 100 Time 150 200 Since A. C and D are monic. The minimum-variance controller for the rewritten system (12. i.1 )F (q. deg B = d. B G (q )F (q ) y (k) 1 1 (12. We see that if A = D the controller reduces to the normal minimum variance controller. B(G q)D(q)F (q) B (q .3) is equivalent to qd. 112 . The accumulated loss functions are shown in the lower diagram.2) (12.1 A(q)C (q) = A(q )D(q)F1(q ) + G1 (q) which implies that A must divide G1.10 Output 0 −10 10 0 50 100 150 200 Output 0 −10 500 0 50 100 150 200 Loss 0 0 50 Figure 12.1 C1 (q) = A1 (q)F1(q) + G1(q) Equation (12.3) where qd1. Putting F = F1 gives the minimum-variance control law (q )A(q ) y (k) = . 1.e. Let d1 = deg A1 .4 The output and the sum of the square of the output for the two regulators in Problem 12.1) is then calculated as 1(q ) u(k) = .1C (q) = D(q)F (q) + G(q) with deg F = d . deeadbeat (dashed) and minimum variance (solid).

c(qbq b b In this case ) g0 = c Problem 12. 0:7q .1) e(k .Problem 12. 0:56q .1 )(1 . 0:5q .1 e(k) y1(k) = 1 . 1) u(k) = . 0:5q . a. 2) + C e(k .12 gives q + c = q 1 + g0 The minimum variance controller is thus + a) y (k) = . The minimum variance controller for y1 is then u(k) = .1) = 0:94 .1 y1(k) 1 B u(k .1 )(1 . 2 2:78 c.1 1 . 1).1) = 1 + q .15 We have 1 .1 ) + q . 01 u ( k . 0:7q .1 1 u ( k . 1 q y2 (k) = 1 + q.13 A(q ) = q + a B (q ) = b C (q ) = q + c D (q ) = q d=1 The identity in Problem 12.2 (g0 + g1q .1 ) This gives F G (q .1 )(1 . 2) + = (1 + q . ac y (k .1) (1 + q .1)(1 + f1 q . Assume that y1 can be measured. 0:7q .1 = (1 + q .0:2y1(k) b. The minimum variance controller for y2 when y2 is measurable is obtained by using the identity 1 . c y (k) .1 . 1) To normalize the notations it is convenient to introduce a new noise "(k) = e(k . The variances of y1 and y2 are 2=1 Ey1 2= 1 Ey2 1.1 113 . 0:5q .1 (q . 1) + :7q. 0:7q. 1) =A A 1 A 1 .

and the controller 0:56q u(k) = . . 0:941. It is now only necessary to predict one step ahead.1 1 y1 (k) + B = "(k + 2) + AGA A1A C C .1 y (k) u(k) = . + q . B u(k . 1) + B u(k) 1 y2 (k + 2) = "(k + 2) + AGA C 1 C A1A 1 .1 )"(k) = "(k . . G1 q u(k) 1 1 G1 y (k) + B u(k) = "(k + 2) + C A 1 1 The variance is thus minimized when 1 y (k) u(k) = .1 y2(k) = (1 + q. 2) and y1 (k) = (1 + q. B u(k . For the speci c system we get 0:56q .1 1 114 . 0:8q. 1) + e(k . 11.1)(1 + q .1 )e(k + 1) The variances of the two signals are 2 = 1 + ( + 1)2 + Ey1 2 = 1+1 = 2 Ey2 2 = 1:68 d. AGB 1 1 which is an admissible control law. 1) "(k + 1) = C 1 C A y (k) .1G1 This gives y2(k + 2) = ACA e(k + 2) + ABA u(k) 1 1 G 1 e(k + 1) + B u(k) = e(k + 2) + A A AA 1 1 But This gives A y (k) .1 y2 (k) This gives . Introduce the identity C = A A1 + q. In this case both y1 and y2 are measurable and y1 (k) will contain more recent information about the noise process than y2 (k).

1 )y2(k + 1) = (1 + q . 1) and y1(k) = (1 + q.1)"(k + 1) = (1 + q . Assume that the system has a stable inverse.With this control law we get y2 (k) = "(k) = e(k . d) .d G v (k) 1 (G y (k) + B F u(k) + D F v (k)) = F e(k + d) + C The minimum variance controller is thus u(k) = . q .17) can be used here also and gives B u(k) + D v(k) y(k + d) = C e ( k + d ) + A A A B G = F e(k + d) + A e(k) + A u(k) + D A v (k ) A y (k) . D v (k . Problem 12. BGF y (k) . q. Also assume that deg A = deg B = deg D i. d) = F e(k + d) + G A C C C D v (k) u ( k ) + +B A A G y(k) + B C .1)e(k) + D (q .e.d G u(k) = F e(k + d) + C AC D + A C C . d) + C (q .16 The same arguments can be used in this case as when deriving the normal minimum variance controller. d) The identity (12. D B v (k) 115 .1)v (k .1)e(k) The variances of the two output signals are 2= 1+ Ey1 2 = 1:64 and 2=1 Ey2 We can thus conclude that the extra measurement will greatly improve the control of the output of the system.1)u(k .1 )y(k) = B (q . that the process can be written as A (q . B u(k .

q s +r . 1 p1 = . 0:9)(z + r1) + zs0 The solution is given by p1 .1 ) = A(z)A(z.0:9 r(1 + p2 1) = 1 + 1:81 which has two solutions. 0:9) + 1 1 .0:9r1 4p s0 = 0:4 + 9 1 1 5p r1 = 9 1 which results in the controller 116 0 q y (k) u(k) = .5.1 . 0:5 = . 0:9 B(q) = q C (q ) = q . 0:5) = (z .1 r(1 + p2 1 ) + rp1 z + rp1 z = 1:81 + 1 .1 .17 A(q)y(k) = B (q)u(k) + C (q )e(k) A(q) = q . Let P (z) be the closed loop system characteristic equation (12. 0:5 LQG-Control: Minimize E (y2 + u2).1 + p1) = (z .45) rP (z)P (z.0:5p1 = .Problem 12.1) r(z + p1)(z. R (q ) PC = AR + BS or This gives (z + p1 )(z . 0:9 z . See computational procedure in Section 12. one of which gives a stable P (z ) 1:81 + ( 1 + 1:81 )2 . The Diophantine equation to be solved is S (q ) y (k) u(k) = .0:9 + r1 + s0 . 0:9 z This gives the system of equations rp1 = . 1 + 1:8 1:8 Determine the LQG-regulator by means of pole placement: r Am = P (z ) = z + p1 Ao = C (z ) = z . 0:9)(z.1 ) P (z) contains stable zeros of the right hand expression (Lemma 12. 0:5 Control law: where S (0) = 0.1) + B (z )B(z.

1 B (q ) = h C (q ) = q + c The minimum variance regulator is given by the Diophantine PC = AR + BS where The solution is P (q ) = qd.0:70 1:197 V ar u 0:135 0:0658 0:0148 Problem 12. S y (k) + C e(k) u ( k ) + y(k) = B A A A R A or CR e(k) = R e(k) = q + r1 e(k) y(k) = ARCR e ( k ) = + BS PC P q + p1 Theorem 10. S y (k) = . S e(k) = .16 we get a polynomial description of the process A(q )y (k) = B (q )u(k) + C (q )e(k) where A(q) = q . s0 q e(k) u(k) = . p2 The input u is 1 In the following table the calculations are summarized for = 0:1 1 and 10.For the closed loop system we get C e(k) = B .4 gives s2 0 V ar u = 1 . S R e(k) = .1 B(q) = h r0 = h s0 = c + 1 and the minimum variance regulator is 1 y (k) u(k) = .0:077 1:0012 1 .4 gives 2 1 . p2 1 p1 V ar y 0:1 .0:36 1:030 10 . c + h 117 . R R P P q + p1 Theorem 10. 2r1p1 V ar y = 1 + r 1 .24 From Example 12.

1 q . v (kh) y(kh) = x(kh) + (kh) To obtain the LQ-controller we have to sample the loss function.16) x(kh + h) = x(kh) + hu(kh) + v (kh + h) .L = 1 . h h p3 = 3 . From (11.1(. .In LQ . 481) PC = AR + BS S (0) = 0 (q + p1 )(q + c) = (q . 1:4z + 0:65 B (z ) = z .)L L = (Q2 + .p1c 1 (p + c + 1 + p c) s0 = h 1 1 and thus 1 (p + c + 1 + p c)q h u(k) = . p c 1 y (k) 1 s ds = h 2 Z0 h 3 Q2 = s2 ds = h 3 0 Zh 1 ds = h 2 Problem 12.29 The process is described by A(z ) = z 2 . 0:2 C (z ) = z 2 + 0:4z 118 .p1 p p 1 3 + .T S + Q12 ) and the solution is ( S = ph 12 p 1 3+p3 L= h 2+ 3 The closed loop system has a pole in . design we use a state-space description of the process (see Example 12.8) we get Q1 = Q12 = Zh 0 The Riccati equation is S = T S + Q1 . 2 2+ 3 and P (z ) = z + p1 To get the regulator we solve the Diophantine (see p.T S . 1)(q + r1 ) + hs0 q which has the solution r0 = .).T S .(11.p1 = . LT (Q2 + .6) .

Minimum variance control var(y ) = 1 Dead-beat control var(y ) = (1 + 0:42) 2 = 4:64 2=4 which gives Problem 12. To get the minimum variance controller we use the identity A(z ) F (z ) + G(z) = z d. 1:4 q . 0:65 y u = .a G y = ay u = . BF The closed loop system becomes y (k) = e(k) + ce(k . 0:65 q . 1:8q . 0:2 b.30 a. 0:2 c. 0:65 y u = .1 C (z) z 2 . Assume that C = 0 and use the identity C = AF + G This gives and the controller F =1 G = . 1) and the variance var(y ) = 1 + c2 119 .a. 1:4z + 0:65 + g0z + g1 = z 2 + 0:4z This gives The control law is G(z ) = 1:8z . BF q . The dead-beat controller is obtained from the identity A(z)F (z ) + G(z ) = z 2 G(z) = 1:4z . 0:65 G y = .

0:8z ) = (z 2 . 0:8q "(k) y(k) = q2 . 0:8) The new process is where " has the standard deviation 1:25. 1:25) = 1:252(z . Assume that C (z ) = z + c ^.1 C (z ) = A(z )F (z) + G(z ) z (z2 .31 The C polynomial has a pole outside the unit circle. 0:8)(z .b.1 . 0:8 y (k) b. a The closed loop system is now q + c e(k) y (k ) = q +c ^ which has the variance. The minimum variance controller for this model is given by F =1 G=c ^. a. see Theorem 6. 0:09 (q ) 0:03(q .4 . 1:1q + 0:3 z m. Replace C by a new polynomial obtained from spectral factorization C (z )C (z.) c 2 ^ 2 It is better to use a guessed value c ^ if 2c 0<c ^< 1+ c2 Problem 12. 1:1z + 0:3)(z + f1) + g0z + g1 which gives and F (z) = z + 0:3 G(z ) = 0:03z . 1:25)(z .1 ) = (z . 2cc ^ var(y ) = (1 +1c. 3) y ^(k + 2jk) = qG C (q) y (k) = q . The prediction error variance is Ey ~ = 1:252(1 + 0:32) = 1:70 120 . The 2-step ahead predictor is given by q2 .1 .

1:7z2 + 0:8z . 0:9) =(q 3 . Factor B (z ) as B(z) = B+ (z ) B .3. so B + ( z ) = . 1:7f2 . 0:1) = (z3 .32 a. (z) = .0:1f1 + 0:8f2 .2 B.(z) with B .0:1f2 .1:7f0 + f1 0:09 = 0:8f0 . (q ) =A(q)F (q ) + B .e. (z)G(q) q 3(q . 1:7q 2 + 0:8q . Since B (z ) has a root outside the unit circle we use the procedure in Theorem 12. 1:18q + 0:16 y (k) u(k) = . BF 2(q .. 2g1 + 1:8g2 0 = . 1:7f1 + f2 + 1:8g0 0 = .0:9q + 1)(g0q 2 + g1 q + g2 ) This gives the system of equations 1 = f0 . This gives the identity zC (z ) = A(z)F (z) + G(z) where deg F = 1 and deg G = 2. The output variance when using the minimum variance controller in a. is given by 2 = 1 + 1:62 = 3:56 var(y ) = 1 + f1 c.1 = .1C (q)B . z3(z . 1:18z + 0:16 and the controller is G y (k) = .1:862 g2 = 0:272 121 . 0:1)(z + f1 ) + g0z 2 + g1z + g2 The solution is F (z) = z + 1:6 G(z) = 1:92z2 . i. d = 2.0:1f0 + 0:8f1 . 2g0 + 1:8g1 0 = . 0:9)(q + 1:6) b. 0:1)(q . 0:1)(f0q2 + f1q + f2 ) + (. 2g2 which has the solution f0 = 1:000 f1 = 0:700 f2 = 2:721 g0 = 2:490 g1 = .Problem 12. (q ) monic.0:9q + 1 The Diophantine to solve is qd. 1:92q2 .

4 gives the output variance V ar y = 94:7 or Ay(k) = Bu(k) + Ce(k) = B . 1 B .1:7 + r1 + 0:9s0 = . 1:7r1 + 0:9s1 + s0 = 0 0:7r1 + s1 = 0 r1 = 0:526 s0 = 0:526 s1 = . 1)(q . .9 p. q d. BG + F y (k) + Ce(k) . B +G (q )F (q ) q2 + 0:7q + 2:721 The closed loop system is CF e(k) = F e(k) y (k) = AF CF e ( k ) = . 0:7) AR + BS = qd. 0:136 y (k) u(k) = . 0:7) (q . 1)(q .Equation (12.0:368 CR = (1 + 0:526q.33 Determine the controller from :9q + 1 u(k) + q (q . 0:7) e(k) y (k) = (q . 2 0:7q + 2:721 e(k) =q + q(q .1 C (q . = .01)( q .1)e(k) y (k) = ARCR e ( k ) = + BS qC var y (k) = (1 + 0:5262) 2 = 1:27 2 Compare to Example 12. 0:7) This gives the system of equations . 0:7)(q + r1) + (0:9q + 1)(s0q + s1 ) = q 2(q . 468 2 2 var y (k) = 20 19 = 1:053 122 with the solution .0:7 0:7 .1:245q 2 + 0:931q . GB F y (k) + Ce(k) Problem 12. 1 + GB q CB . d . 0:9) Theorem 10.31) gives (q ) y (k) = .