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
|
<HTML>
<HEAD><TITLE>IB01PY - SLICOT Library Routine Documentation</TITLE>
</HEAD>
<BODY>
<H2><A Name="IB01PY">IB01PY</A></H2>
<H3>
Estimating system matrices B and D using a structure exploiting QR factorization
</H3>
<A HREF ="#Specification"><B>[Specification]</B></A>
<A HREF ="#Arguments"><B>[Arguments]</B></A>
<A HREF ="#Method"><B>[Method]</B></A>
<A HREF ="#References"><B>[References]</B></A>
<A HREF ="#Comments"><B>[Comments]</B></A>
<A HREF ="#Example"><B>[Example]</B></A>
<P>
<B><FONT SIZE="+1">Purpose</FONT></B>
<PRE>
1. To compute the triangular (QR) factor of the p-by-L*s
structured matrix Q,
[ Q_1s Q_1,s-1 Q_1,s-2 ... Q_12 Q_11 ]
[ 0 Q_1s Q_1,s-1 ... Q_13 Q_12 ]
Q = [ 0 0 Q_1s ... Q_14 Q_13 ],
[ : : : : : ]
[ 0 0 0 ... 0 Q_1s ]
and apply the transformations to the p-by-m matrix Kexpand,
[ K_1 ]
[ K_2 ]
Kexpand = [ K_3 ],
[ : ]
[ K_s ]
where, for MOESP approach (METH = 'M'), p = s*(L*s-n), and
Q_1i = u2(L*(i-1)+1:L*i,:)' is (Ls-n)-by-L, for i = 1:s,
u2 = Un(1:L*s,n+1:L*s), K_i = K(:,(i-1)*m+1:i*m) (i = 1:s)
is (Ls-n)-by-m, and for N4SID approach (METH = 'N'), p = s*(n+L),
and
[ -L_1|1 ] [ M_i-1 - L_1|i ]
Q_11 = [ ], Q_1i = [ ], i = 2:s,
[ I_L - L_2|1 ] [ -L_2|i ]
are (n+L)-by-L matrices, and
K_i = K(:,(i-1)*m+1:i*m), i = 1:s, is (n+L)-by-m.
The given matrices are:
For METH = 'M', u2 = Un(1:L*s,n+1:L*s),
K(1:Ls-n,1:m*s);
[ L_1|1 ... L_1|s ]
For METH = 'N', L = [ ], (n+L)-by-L*s,
[ L_2|1 ... L_2|s ]
M = [ M_1 ... M_s-1 ], n-by-L*(s-1), and
K, (n+L)-by-m*s.
Matrix M is the pseudoinverse of the matrix GaL,
built from the first n relevant singular
vectors, GaL = Un(1:L(s-1),1:n), and computed
by SLICOT Library routine IB01PD for METH = 'N'.
Matrix Q is triangularized (in R), exploiting its structure,
and the transformations are applied from the left to Kexpand.
2. To estimate the matrices B and D of a linear time-invariant
(LTI) state space model, using the factor R, transformed matrix
Kexpand, and the singular value decomposition information provided
by other routines.
IB01PY routine is intended for speed and efficient use of the
memory space. It is generally not recommended for METH = 'N', as
IB01PX routine can produce more accurate results.
</PRE>
<A name="Specification"><B><FONT SIZE="+1">Specification</FONT></B></A>
<PRE>
SUBROUTINE IB01PY( METH, JOB, NOBR, N, M, L, RANKR1, UL, LDUL,
$ R1, LDR1, TAU1, PGAL, LDPGAL, K, LDK, R, LDR,
$ H, LDH, B, LDB, D, LDD, TOL, IWORK, DWORK,
$ LDWORK, IWARN, INFO )
C .. Scalar Arguments ..
DOUBLE PRECISION TOL
INTEGER INFO, IWARN, L, LDB, LDD, LDH, LDK, LDPGAL,
$ LDR, LDR1, LDUL, LDWORK, M, N, NOBR, RANKR1
CHARACTER JOB, METH
C .. Array Arguments ..
DOUBLE PRECISION B(LDB, *), D(LDD, *), DWORK(*), H(LDH, *),
$ K(LDK, *), PGAL(LDPGAL, *), R(LDR, *),
$ R1(LDR1, *), TAU1(*), UL(LDUL, *)
INTEGER IWORK( * )
</PRE>
<A name="Arguments"><B><FONT SIZE="+1">Arguments</FONT></B></A>
<P>
<B>Mode Parameters</B>
<PRE>
METH CHARACTER*1
Specifies the subspace identification method to be used,
as follows:
= 'M': MOESP algorithm with past inputs and outputs;
= 'N': N4SID algorithm.
JOB CHARACTER*1
Specifies whether or not the matrices B and D should be
computed, as follows:
= 'B': compute the matrix B, but not the matrix D;
= 'D': compute both matrices B and D;
= 'N': do not compute the matrices B and D, but only the
R factor of Q and the transformed Kexpand.
</PRE>
<B>Input/Output Parameters</B>
<PRE>
NOBR (input) INTEGER
The number of block rows, s, in the input and output
Hankel matrices processed by other routines. NOBR > 1.
N (input) INTEGER
The order of the system. NOBR > N > 0.
M (input) INTEGER
The number of system inputs. M >= 0.
L (input) INTEGER
The number of system outputs. L > 0.
RANKR1 (input) INTEGER
The effective rank of the upper triangular matrix r1,
i.e., the triangular QR factor of the matrix GaL,
computed by SLICOT Library routine IB01PD. It is also
the effective rank of the matrix GaL. 0 <= RANKR1 <= N.
If JOB = 'N', or M = 0, or METH = 'N', this
parameter is not used.
UL (input/workspace) DOUBLE PRECISION array, dimension
( LDUL,L*NOBR )
On entry, if METH = 'M', the leading L*NOBR-by-L*NOBR
part of this array must contain the matrix Un of
relevant singular vectors. The first N columns of UN
need not be specified for this routine.
On entry, if METH = 'N', the leading (N+L)-by-L*NOBR
part of this array must contain the given matrix L.
On exit, the leading LDF-by-L*(NOBR-1) part of this array
is overwritten by the matrix F of the algorithm in [4],
where LDF = MAX( 1, L*NOBR-N-L ), if METH = 'M';
LDF = N, if METH = 'N'.
LDUL INTEGER
The leading dimension of the array UL.
LDUL >= L*NOBR, if METH = 'M';
LDUL >= N+L, if METH = 'N'.
R1 (input) DOUBLE PRECISION array, dimension ( LDR1,N )
If JOB <> 'N', M > 0, METH = 'M', and RANKR1 = N,
the leading L*(NOBR-1)-by-N part of this array must
contain details of the QR factorization of the matrix
GaL, as computed by SLICOT Library routine IB01PD.
Specifically, the leading N-by-N upper triangular part
must contain the upper triangular factor r1 of GaL,
and the lower L*(NOBR-1)-by-N trapezoidal part, together
with array TAU1, must contain the factored form of the
orthogonal matrix Q1 in the QR factorization of GaL.
If JOB = 'N', or M = 0, or METH = 'N', or METH = 'M'
and RANKR1 < N, this array is not referenced.
LDR1 INTEGER
The leading dimension of the array R1.
LDR1 >= L*(NOBR-1), if JOB <> 'N', M > 0, METH = 'M',
and RANKR1 = N;
LDR1 >= 1, otherwise.
TAU1 (input) DOUBLE PRECISION array, dimension ( N )
If JOB <> 'N', M > 0, METH = 'M', and RANKR1 = N,
this array must contain the scalar factors of the
elementary reflectors used in the QR factorization of the
matrix GaL, computed by SLICOT Library routine IB01PD.
If JOB = 'N', or M = 0, or METH = 'N', or METH = 'M'
and RANKR1 < N, this array is not referenced.
PGAL (input) DOUBLE PRECISION array, dimension
( LDPGAL,L*(NOBR-1) )
If METH = 'N', or JOB <> 'N', M > 0, METH = 'M' and
RANKR1 < N, the leading N-by-L*(NOBR-1) part of this
array must contain the pseudoinverse of the matrix GaL,
as computed by SLICOT Library routine IB01PD.
If METH = 'M' and JOB = 'N', or M = 0, or
RANKR1 = N, this array is not referenced.
LDPGAL INTEGER
The leading dimension of the array PGAL.
LDPGAL >= N, if METH = 'N', or JOB <> 'N', M > 0,
and METH = 'M' and RANKR1 < N;
LDPGAL >= 1, otherwise.
K (input/output) DOUBLE PRECISION array, dimension
( LDK,M*NOBR )
On entry, the leading (p/s)-by-M*NOBR part of this array
must contain the given matrix K defined above.
On exit, the leading (p/s)-by-M*NOBR part of this array
contains the transformed matrix K.
LDK INTEGER
The leading dimension of the array K. LDK >= p/s.
R (output) DOUBLE PRECISION array, dimension ( LDR,L*NOBR )
If JOB = 'N', or M = 0, or Q has full rank, the
leading L*NOBR-by-L*NOBR upper triangular part of this
array contains the R factor of the QR factorization of
the matrix Q.
If JOB <> 'N', M > 0, and Q has not a full rank, the
leading L*NOBR-by-L*NOBR upper trapezoidal part of this
array contains details of the complete orhogonal
factorization of the matrix Q, as constructed by SLICOT
Library routines MB03OD and MB02QY.
LDR INTEGER
The leading dimension of the array R. LDR >= L*NOBR.
H (output) DOUBLE PRECISION array, dimension ( LDH,M )
If JOB = 'N' or M = 0, the leading L*NOBR-by-M part
of this array contains the updated part of the matrix
Kexpand corresponding to the upper triangular factor R
in the QR factorization of the matrix Q.
If JOB <> 'N', M > 0, and METH = 'N' or METH = 'M'
and RANKR1 < N, the leading L*NOBR-by-M part of this
array contains the minimum norm least squares solution of
the linear system Q*X = Kexpand, from which the matrices
B and D are found. The first NOBR-1 row blocks of X
appear in the reverse order in H.
If JOB <> 'N', M > 0, METH = 'M' and RANKR1 = N, the
leading L*(NOBR-1)-by-M part of this array contains the
matrix product Q1'*X, and the subarray
L*(NOBR-1)+1:L*NOBR-by-M contains the corresponding
submatrix of X, with X defined in the phrase above.
LDH INTEGER
The leading dimension of the array H. LDH >= L*NOBR.
B (output) DOUBLE PRECISION array, dimension ( LDB,M )
If M > 0, JOB = 'B' or 'D' and INFO = 0, the leading
N-by-M part of this array contains the system input
matrix.
If M = 0 or JOB = 'N', this array is not referenced.
LDB INTEGER
The leading dimension of the array B.
LDB >= N, if M > 0 and JOB = 'B' or 'D';
LDB >= 1, if M = 0 or JOB = 'N'.
D (output) DOUBLE PRECISION array, dimension ( LDD,M )
If M > 0, JOB = 'D' and INFO = 0, the leading
L-by-M part of this array contains the system input-output
matrix.
If M = 0 or JOB = 'B' or 'N', this array is not
referenced.
LDD INTEGER
The leading dimension of the array D.
LDD >= L, if M > 0 and JOB = 'D';
LDD >= 1, if M = 0 or JOB = 'B' or 'N'.
</PRE>
<B>Tolerances</B>
<PRE>
TOL DOUBLE PRECISION
The tolerance to be used for estimating the rank of
matrices. If the user sets TOL > 0, then the given value
of TOL is used as a lower bound for the reciprocal
condition number; an m-by-n matrix whose estimated
condition number is less than 1/TOL is considered to
be of full rank. If the user sets TOL <= 0, then an
implicitly computed, default tolerance, defined by
TOLDEF = m*n*EPS, is used instead, where EPS is the
relative machine precision (see LAPACK Library routine
DLAMCH).
This parameter is not used if M = 0 or JOB = 'N'.
</PRE>
<B>Workspace</B>
<PRE>
IWORK INTEGER array, dimension ( LIWORK )
where LIWORK >= 0, if JOB = 'N', or M = 0;
LIWORK >= L*NOBR, if JOB <> 'N', and M > 0.
DWORK DOUBLE PRECISION array, dimension ( LDWORK )
On exit, if INFO = 0, DWORK(1) returns the optimal value
of LDWORK, and, if JOB <> 'N', and M > 0, DWORK(2)
contains the reciprocal condition number of the triangular
factor of the matrix R.
On exit, if INFO = -28, DWORK(1) returns the minimum
value of LDWORK.
LDWORK INTEGER
The length of the array DWORK.
LDWORK >= MAX( 2*L, L*NOBR, L+M*NOBR ),
if JOB = 'N', or M = 0;
LDWORK >= MAX( L+M*NOBR, L*NOBR + MAX( 3*L*NOBR+1, M ) ),
if JOB <> 'N', and M > 0.
For good performance, LDWORK should be larger.
</PRE>
<B>Warning Indicator</B>
<PRE>
IWARN INTEGER
= 0: no warning;
= 4: the least squares problem to be solved has a
rank-deficient coefficient matrix.
</PRE>
<B>Error Indicator</B>
<PRE>
INFO INTEGER
= 0: successful exit;
< 0: if INFO = -i, the i-th argument had an illegal
value;
= 3: a singular upper triangular matrix was found.
</PRE>
<A name="Method"><B><FONT SIZE="+1">Method</FONT></B></A>
<PRE>
The QR factorization is computed exploiting the structure,
as described in [4].
The matrices B and D are then obtained by solving certain
linear systems in a least squares sense.
</PRE>
<A name="References"><B><FONT SIZE="+1">References</FONT></B></A>
<PRE>
[1] Verhaegen M., and Dewilde, P.
Subspace Model Identification. Part 1: The output-error
state-space model identification class of algorithms.
Int. J. Control, 56, pp. 1187-1210, 1992.
[2] Van Overschee, P., and De Moor, B.
N4SID: Two Subspace Algorithms for the Identification
of Combined Deterministic-Stochastic Systems.
Automatica, Vol.30, No.1, pp. 75-93, 1994.
[3] Van Overschee, P.
Subspace Identification : Theory - Implementation -
Applications.
Ph. D. Thesis, Department of Electrical Engineering,
Katholieke Universiteit Leuven, Belgium, Feb. 1995.
[4] Sima, V.
Subspace-based Algorithms for Multivariable System
Identification.
Studies in Informatics and Control, 5, pp. 335-344, 1996.
</PRE>
<A name="Numerical Aspects"><B><FONT SIZE="+1">Numerical Aspects</FONT></B></A>
<PRE>
The implemented method for computing the triangular factor and
updating Kexpand is numerically stable.
</PRE>
<A name="Comments"><B><FONT SIZE="+1">Further Comments</FONT></B></A>
<PRE>
The computed matrices B and D are not the least squares solutions
delivered by either MOESP or N4SID algorithms, except for the
special case n = s - 1, L = 1. However, the computed B and D are
frequently good enough estimates, especially for METH = 'M'.
Better estimates could be obtained by calling SLICOT Library
routine IB01PX, but it is less efficient, and requires much more
workspace.
</PRE>
<A name="Example"><B><FONT SIZE="+1">Example</FONT></B></A>
<P>
<B>Program Text</B>
<PRE>
None
</PRE>
<B>Program Data</B>
<PRE>
None
</PRE>
<B>Program Results</B>
<PRE>
None
</PRE>
<HR>
<A HREF=support.html><B>Return to Supporting Routines index</B></A></BODY>
</HTML>
|
