1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
|
// Scilab ( http://www.scilab.org/ ) - This file is part of Scilab
// Copyright (C) INRIA -
//
// This file must be used under the terms of the CeCILL.
// This source file is licensed as described in the file COPYING, which
// you should have received as part of this distribution. The terms
// are also available at
// http://www.cecill.info/licences/Licence_CeCILL_V2.1-en.txt
function [x,frequ]=linfn(G,PREC,RELTOL,options);
//[x,frequ]=linfn(G,PREC,RELTOL,options);
// Computes the Linf (or Hinf) norm of a transfer function
// -1
// G(s) = D + C (sI - A) B
//
// This norm is well-defined as soon as the realization
// (A,B,C) has no imaginary eval which is both controllable and observable.
//
// The algorithm follows the paper by G. Robel (AC-34 pp. 882-884, 1989).
// The case D=0 is not treated separately due to superior accuracy of
// the general method when (A,B,C) is nearly nonminimal.
//
// In the general case (A neither stable nor antistable), no upper bound is
// prespecified. If by contrast A is stable or antistable, lower
// and upper bounds are computed using the associated Lyapunov
// solutions (see Glover).
// On input:
// ---------
// * G is a syslin list
// * PREC is the desired relative accuracy on the norm
// * RELTOL: relative threshold to decide when an eigenvalue can be
// considered on the imaginary axis.
// * available options are
// - 'trace': traces each bisection step, i.e., displays the lower
// and upper bounds and the current test point.
// - 'cond': estimates a confidence index on the computed value
// and issues a warning if computations are
// ill-conditioned
//
// On output:
// ---------
// * x is the computed norm.
// * freq: list of the frequencies for which ||G|| is attained, i.e.,
// such that ||G (j om)|| = ||G||. If -1 is in the list, the norm
// is attained at infinity. If -2 is in the list, G is all-pass in
// some direction so that ||G (j omega)|| = ||G|| for all
// frequencies omega.
//!
//
// Called macros:
// -------------
// heval_test, cond_test, list_set
//
// History:
// -------
// author: P. Gahinet, INRIA
// last modification: Oct 3nd, 1991
//****************************************************************************
//constants
//*********
INIT_LOW=1.0e-4; INIT_UPP=1.0e5; INFTY=10e10;
frequ=[];
//user interface. The default values are:
// PREC=1.0e-3; RELTOL=1.0e-10; options='nul';
//************************************************
[lhs,rhs]=argn(0);
select rhs,
case 0 then
error(msprintf(gettext("%s: Wrong number of input arguments: At least %d expected.\n"),"linfn",1))
case 1 then
PREC=1.0e-3; RELTOL=1.0e-10; options="nul";
case 2 then
RELTOL=1.0e-10;
if type(PREC)==10 then
iopt=2
options=PREC; PREC=1.0e-3;
else
options="nul";
end,
case 3 then
if type(RELTOL)==10 then
iopt=3
options=RELTOL; RELTOL=1.0e-10;
else
options="nul";
end,
end
if typeof(G)<>"state-space" then
error(msprintf(gettext("%s: Wrong type for input argument #%d: Linear state space expected.\n"),"linfn",1))
end
if G.dt<>"c" then
error(msprintf(gettext("%s: Wrong type for argument #%d: In continuous time expected.\n"),"linfn",1))
end
if type(options)<>10|and(options<>["t","nul"]) then
error(msprintf(gettext("%s: Wrong value for input argument #%d: Must be in the set {%s}.\n"),"linfn",iopt,"""t"",""nul"""))
end
if type(PREC)<>1|size(PREC,"*")<>1|~isreal(PREC)|PREC<=0 then
error(msprintf(gettext("%s: Wrong value for input argument #%d: Must be a positive scalar.\n"),"linfn",2))
end
if type(RELTOL)<>1|size(RELTOL,"*")<>1|~isreal(RELTOL)|RELTOL<=0 then
error(msprintf(gettext("%s: Wrong value for input argument #%d: Must be a positive scalar.\n"),"linfn",3))
end
//recover realization
//*******************
[a,b,c,d]=abcd(G);
//SCALING
//******* ||B||.||C||
// Scale A,B,C so that ||A||=||B||=||C||=1. With scale:= -----------,
// ||A||
// and DD:=D/scale, AA:=A/||A||, BB:=B/||B||, CC:=C/||C||, we have
// -1
// || G || = scale * || DD + CC (sI - AA) BB ||
//
// From now on, the parameters A,B,C,D are scaled
noa=norm(a,"inf"); nob=norm(b,"inf"); noc=norm(c,"inf"); nobc=nob*noc;
if nobc==0, x=norm(d); return, end
scale=nobc/noa; a=a/noa; b=b/nob; c=c/noc; d=d/scale; nd=norm(d);
//test the spectrum of A
//**********************
s=real(spec(a));
if min(abs(s)) < RELTOL then
mprintf(gettext("%s: WARNING: the A matrix has eigenvalues near the imaginary axis.\n"),"linfn");
end
//Search window initialization
//****************************
// Initialize the search window [lower,upper] where `lower' and `upper'
// are lower and upper bounds on the Linf norm of G. When no such
// bounds are available, the window is arbitrarily set to [INIT_LOW,INIT_UPP]
// and the variables LOW and UPP keep record of these initial values so that
// the window can be extended if necessary.
if max(s)*min(s) > 0 then
// A is stable or antistable: use associated Lyapunov equations
// to derive lower and upper bounds.
p=lyap(a',-b*b',"c");
q=lyap(a,-c'*c,"c");
s=sqrt(abs(spec(p*q)));
lower=max(nd,max(s)); LOW=0;
upper=nd+2*sum(diag(s)); UPP=100*upper;
else
if nd==0 then
lower=INIT_LOW; LOW=INIT_LOW;
else
lower=nd; LOW=0;
end
upper=INIT_UPP; UPP=INIT_UPP;
end
//form the constant parts of the pencil (E,F) (see G. Robel).
//***********************************************************
[na,na]=size(a); twona=2*na;
[p,m]=size(d);
nf=twona+min(m,p); //size of e and f
// to ensure that D'*D is of size min(m,p), replace (a,b,c,d) by
// (a',c',b',d') if m>p
if m>p then
a=a'; d=d'; aux=b; b=c'; c=aux';
end
e=eye(2*na,2*na); e(nf,nf)=0;
nul=0; nul(na,na)=0;
f=[a,nul;-c'*c,-a']; f(nf,nf)=0;
dd=d'*d; Cd=c'*d;
//----------------------
// BISECTION STARTS
//----------------------
while %t,
ga=sqrt(lower*upper); //test point gamma = log middle of [lower,upper]
if part(options,1)=="t" then
write(%io(2),[scale*lower,scale*ga,scale*upper],..
"(''lower,current,upper = '',3e20.10)");
end
bga=b/ga; Cdga=Cd/ga;
f(1:na,twona+1:nf)=-bga;
f(na+1:twona,twona+1:nf)=Cdga;
f(twona+1:nf,1:nf)=[Cdga',bga',eye(dd)-dd/(ga**2)];
// Test for generalized eigenvalues on the imaginary axis
// ------------------------------------------------------
[dist,frequ]=heval_test(e,f,RELTOL,"test");
if dist < RELTOL then
lower=ga; LOW=0;
// eigenvalue on the imaginary axis: gamma < ||G||
else
upper=ga; UPP=100*upper;
// gamma > ||G||
end
// Search window management:
//--------------------------
// If the gamma-iteration runs into one of the initial arbitrary bounds
// LOW or UPP, extend the search window to allow for continuation
if ga<10*LOW then
lower=LOW/10; LOW=lower;
end
// expand search window toward gamma<<1
if ga>UPP/10 then
upper=UPP*10; UPP=upper;
end
// expand search window toward gamma>>1
// Termination tests
//------------------
if lower > INFTY then
mprintf(gettext("%s: Controllable & observable mode(s) of A near the imaginary axis"),"linfn");
x=scale*lower;
return;
else
if upper < 1.0e-10 then
x=scale*upper;
mprintf(gettext("%s: All modes of A are nearly nonminimal so that || G || is almost 0.\n"),"linfn");
return;
else
if 1-lower/upper < PREC,
ga=sqrt(lower*upper);
x=scale*ga;
// Compute all the frequencies achieving ||G||
if lower<>0 then
bga=b/lower; Cdga=Cd/lower;
f(1:na,twona+1:nf)=-bga;
f(na+1:twona,twona+1:nf)=Cdga;
f(twona+1:nf,1:nf)=[Cdga',bga',eye(dd)-dd/(lower**2)];
[dist,frequ]=heval_test(e,f,RELTOL,"freq");
end
if frequ==[] then
mprintf(gettext("%s: The computed value of || G || may be inaccurate.\n"),"linfn");
end
// evaluate the condition of the eigenproblem of (e,f) near || G ||
if part(options,1)=="c" then
gt=1.1*ga;
f=[a,nul,-b/gt;cc,at,Cd/gt;dc/gt,bt/gt,eye(dd)-dd/(gt**2)]
co=cond_test(e,f,frequ,RELTOL);
if co < RELTOL then
mprintf(gettext("%s: The computed value of || G || may be inaccurate.\n"),"linfn");
end
end
//-----------
return;
end,
end,
end
end//end while
endfunction
function [dist,frequ]=heval_test(e,f,TOL,option);
//[dist,frequ]=heval_test(e,f,TOL,option);
// This procedure estimates the distance of the generalized spectrum
// of the pencil f - lambda e to the imaginary axis. Here e is always
// of the form diag(I_(nf-nz),0_nz). The distance is 0 whenever there are
// (nearly) infinite eigenvalues or eigenvalues of the form 0/0.
//
// The eigenvalues are computed via a generalized Schur decomposition
// of f - lambda e . Let (a(i),b(i)) : i=1..nf be the output of gspec.
// Three cases must be distinguished:
// * both a(i) and b(i) are << 1 -> singularity of the pencil
// * b(i)<<1 and a(i) close to 1 -> infinite eigenvalue
// * both a(i) and b(i) are close to 1 -> finite eigenvalue.
//
// Let nz denote the rank deficiency of e which is also the size of D'*D.
// For gamma > ||G||, the generalized spectrum of (e,f) consists of exactly
// nz infinite modes and nf-nz finite ones. By contrast, there may be
// additional singularities or infinite modes for ||D|| <= gamma <= ||G||,
// depending on whether ||D|| = ||G|| or ||D|| < ||G||.
//
// If ||D|| < ||G||, there are still exactly nz infinite modes for gamma in
// [ ||D|| , ||G|| ]. At gamma=||G||, some finite mode(s) hit the imaginary
// axis and their imaginary part omega is such that ||G(j omega)|| = ||G||.
//
// If ||D|| == ||G|| now, we always have ||G (infinity)|| = ||G||
// and if moreover some pair (a(i),b(i)) is nearly (0,0), then
// || G (j omega) || = || G || for all omega's (direction along which G is
// all-pass). Note that G is all-pass iff there are nz pairs
// (a(i),b(i)) nearly equal to (0,0). Finally, finite modes which hit
// the imaginary axis still yield frequencies for which ||G|| is attained.
//
// Two options are available in this function:
// * When option='test', the function counts the number of finite modes.
// If less than nf-nz (nf=order of f), it concludes gamma <= ||G|| and returns
// dist=0. Otherwise, it estimates the distance of the finite spectrum
// min | Re(l_i) |
// to the imag. axis computed as --------------- where the l_i's denote
// max | l_i |
// the pencil finite eigenvalues.
// * With the option 'freq' (used for gamma = ||G||), it furthermore returns
// all frequencies for which ||G|| is attained. Infinite frequencies are
// denoted by -1 and if ||G(j omega)|| == ||G|| for all omega's, frequ=[-2];
//
//
// Input:
// * (e,f): pencil
// * TOL: relative tolerance on the size of eigenvalue real parts.
// * option: 'test' or 'freq'.
//
// Output
// * dist: distance of the spectrum to the imaginary axis as defined above.
// * frequ: list of frequencies for which ||G|| is attained.
//
//!
//balancing
frequ=[]; evals=[];
[f,xx]=balanc(f);
[nf,nf]=size(f);
nz=nf-sum(diag(e)); //rank deficiency of e
//Generalized Schur decomposition of the pencil (f,e)
//---------------------------------------------------
[a,b]=spec(f,e);
if option=="test" then
//***********************************
//Simple test and computation of dist
//***********************************
// Check that there are exactly nz infinite modes of (e,f) and compute dist
nai=0; //nai: number of infinite or (0,0) modes (b(i) << 1)
for i=1:nf,
bi=abs(b(i));
if bi < 100*TOL then
nai=nai+1;
else
evals=[evals,a(i)/bi];
end
end
if nai>nz then
dist=0;
else
dist=min(abs(real(evals)))/max(abs(evals));
end
else
//option = 'freq'
//*************************************************
//Compute the frequency for which ||G|| is attained
//*************************************************
// Here gamma is appx equal to ||G||. Distinguish two cases:
// ||D|| < ||G|| and ||D|| = ||G||.
if min(svd(f(nf-nz+1:nf,nf-nz+1:nf))) < TOL then
//-----------------------------------------------
// f(nf-nz+1:nf,nf-nz+1:nf)= I - (D'*D)/||G||**2 -> case ||D||=||G||
noa=max(abs(a)); na=0; //na -> # pairs (0,0)
frequ=[-1]; //||G|| is always attained for s=infinity in this case
for i=1:nf,
bi=b(i);
if abs(bi) < 100*TOL then
if abs(a(i)) < 100*TOL*noa, na=na+1; end
else
evals=[evals,a(i)/bi];
end
end
if na>0 then // G is all-pass along some direction
frequ=[-2];
if na>=nz, mprintf(_("G is all-pass")); end
else
if evals<>[] then
maxabs=max(abs(evals));
for i=1:max(size(evals)),
if abs(real(evals(i))) <= TOL*maxabs,
frequ=[frequ,abs(imag(evals(i)))];
end
end,
end,
end
else
// case ||D|| < ||G||
//------------------------
for i=1:nf,
bi=b(i);
if abs(bi) > 100*TOL, evals=[evals,a(i)/bi]; end
end
maxabs=max(abs(evals));
for i=1:max(size(evals)),
if abs(real(evals(i))) <= TOL*maxabs then
frequ=[frequ,abs(imag(evals(i)))];
end
end
end //endif
frequ=list_set(frequ,1.0e-5); //eliminate redundancy in frequ
end //endif
endfunction
function [c]=cond_test(e,f,frequ,TOL);
//[c]=cond_test(e,f,frequ,TOL);
// This procedure returns a confidence index for the computed gamma = ||G||
// at which some generalized eigenvalue(s) of (e,f) meets the imaginary
// axis. Specifically, it considers gamma := 1.1*computed value of ||G||
// and computes how close (e,f) is to have imaginary eigenvalues
// for this gamma. If very close, this indicates (e,f) has generalized
// eigenvalues near the imaginary axis for all gamma's in an interval
// around ||G|| whence the computed value is likely to be inaccurate.
//!
[nf,nf]=size(f);
c=1;
for i=1:max(size(frequ)),
s=svd(f-%i*frequ(i)*e);
c=min(c,s(nf)/s(1));
if c < TOL then return; end
end
endfunction
function [l]=list_set(l,TOL);
//[l]=list_set(l,TOL);
// eliminates redundant elements in a list. Two entries are considered
// identical when their difference is smaller then TOL (in relative terms)
//!
nl=max(size(l));
i=1;
while i < nl,
entry=l(i); TOLabs=TOL*entry;
j=i+1;
while j <= nl,
if abs(l(j)-entry) <= TOLabs then
l=[l(1:j-1),l(j+1:nl)];
nl=nl-1;
else
j=j+1;
end
end
i=i+1;
end
endfunction
|