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
|
.TH abinv 1 "April 1993" "Scilab Group" "Scilab Function"
.so ../sci.an
.SH NAME
abinv - AB invariant subspace
.SH CALLING SEQUENCE
.nf
[X,dims,F,U,k,Z]=abinv(Sl,alfa,beta)
.fi
.SH PARAMETERS
.TP 10
sl
: \fVsyslin\fR list containing the matrices \fV[A,B,C,D]\fR.
.TP
alfa
: real number or vector (possibly complex, location of closed loop poles)
.TP
beta
: real number or vector (possibly complex, location of closed loop poles)
.TP
X
: orthogonal matrix of size nx (dim of state space).
.TP
dims
:
integer row vector \fVdims=[dimR,dimVg,dimV,noc,nos]\fR with \fVdimR<=dimVg<=dimV<=noc<=nos\fR
.TP
F
: real matrix (state feedback)
.TP
k
: integer (normal rank of \fVSl\fR)
.TP
Z
: non-singular linear system (\fVsyslin\fR list)
.SH DESCRIPTION
Output nulling subspace (maximal unobservable subspace) for
\fVSl\fR = linear system defined by a syslin list containing
the matrices [A,B,C,D] of \fVSl\fR.
The vector \fVdims=[dimR,dimVg,dimV,noc,nos]\fR gives the dimensions
of subspaces defined as columns od \fVX\fR.
The \fVdimV\fR first columns of \fVX\fR i.e \fVV=X(:,1:dimV)\fR,
span the AB-invariant subspace of \fVSl\fR i.e the unobservable subspace of
\fV(A+B*F,C+D*F)\fR. (\fVdimV=nx\fR iff C^(-1)(D)=X).
.LP
The \fVdimR\fR first columns of \fVX\fR i.e. \fVR=X(:,1:dimR)\fR spans
the controllable part of \fVSl\fR in \fVV\fR, \fV(dimR<=dimV)\fR. (\fVdimR=0\fR
for a left invertible system). \fVR\fR is the maximal controllability
subspace of \fVSl\fR in \fVkernel(C)\fR.
.LP
The \fVdimVg\fR first columns of \fVX\fR spans
\fVVg\fR=maximal AB-stabilizable subspace of \fVSl\fR. \fV(dimR<=dimVg<=dimV)\fR.
.LP
\fVF\fR is a decoupling feedback: for \fVX=[V,X2] (X2=X(:,dimV+1:nx))\fR one has
\fVX2'*(A+B*F)*V=0 and (C+D*F)*V=0\fR.
.LP
The zeros od \fVSl\fR are given by : \fVX0=X(:,dimR+1:dimV); spec(X0'*(A+B*F)*X0)\fR
i.e. there are \fVdimV-dimR\fR closed-loop fixed modes.
.LP
If the optional parameter \fValfa\fR is given as input,
the \fVdimR\fR controllable modes of \fV(A+BF)\fR in \fVV\fR
are set to \fValfa\fR (or to \fV[alfa(1), alfa(2), ...]\fR.
(\fValfa\fR can be a vector (real or complex pairs) or a (real) number).
Default value \fValfa=-1\fR.
.LP
If the optional real parameter \fVbeta\fR is given as input,
the \fVnoc-dimV\fR controllable modes of \fV(A+BF)\fR "outside" \fVV\fR
are set to \fVbeta\fR (or \fV[beta(1),beta(2),...]\fR). Default value \fVbeta=-1\fR.
.LP
In the \fVX,U\fR bases, the matrices
\fV[X'*(A+B*F)*X,X'*B*U;(C+D*F)*X,D*U]\fR
are displayed as follows:
.nf
[A11,*,*,*,*,*] [B11 * ]
[0,A22,*,*,*,*] [0 * ]
[0,0,A33,*,*,*] [0 * ]
[0,0,0,A44,*,*] [0 B42]
[0,0,0,0,A55,*] [0 0 ]
[0,0,0,0,0,A66] [0 0 ]
[0,0,0,*,*,*] [0 D2]
.fi
where the X-partitioning is defined by \fVdims\fR and
the U-partitioning is defined by \fVk\fR.
.LP
\fVA11\fR is \fV(dimR x dimR)\fR and has its eigenvalues set to \fValfa(i)'s\fR.
The pair \fV(A11,B11)\fR is controllable and \fVB11\fR has \fVnu-k\fR columns.
\fVA22\fR is a stable \fV(dimVg-dimR x dimVg-dimR)\fR matrix.
\fVA33\fR is an unstable \fV(dimV-dimVg x dimV-dimVg)\fR matrix (see \fVst_ility\fR).
.LP
\fVA44\fR is \fV(noc-dimV x noc-dimV)\fR and has its eigenvalues set to \fVbeta(i)'s\fR.
The pair \fV(A44,B42)\fR is controllable.
\fVA55\fR is a stable \fV(nos-noc x nos-noc)\fR matrix.
\fVA66\fR is an unstable \fV(nx-nos x nx-nos)\fR matrix (see \fVst_ility\fR).
.LP
\fVZ\fR is a column compression of \fVSl\fR and \fVk\fR is
the normal rank of \fVSl\fR i.e \fVSl*Z\fR is a column-compressed linear
system. \fVk\fR is the column dimensions of \fVB42,B52,B62\fR and \fVD2\fR.
\fV[B42;B52;B62;D2]\fR is full column rank and has rank \fVk\fR.
.LP
This function can be used for the (exact) disturbance decoupling problem.
.nf
DDPS:
Find u=Fx+Rd which reject Q*d and stabilizes the plant:
xdot=Ax+Bu+Qd
y=Cx+Du
DDPS has a solution iff Im(Q) is included in Vg + Im(B).
Let G=(X(:,dimVg+1:nx))'= left annihilator of Vg i.e. G*Vg=0;
B2=G*B; Q2=G*Q; DDPS solvable iff B2(:,1:k)*R1 + Q2 =0 has a solution.
R=[R1;*] is the solution (with F=output of abinv).
Im(Q2) is in Im(B2) means row-compression of B2=>row-compression of Q2
Then C*[(sI-A-B*F)^(-1)+D]*(Q+B*R) =0 (<=>G*(Q+B*R)=0)
.fi
.SH EXAMPLE
.nf
nu=3;ny=4;nx=7;
nrt=2;ngt=3;ng0=3;nvt=5;rk=2;
flag=list('on',nrt,ngt,ng0,nvt,rk);
Sl=ssrand(ny,nu,nx,flag);alfa=-1;beta=-2;
[X,dims,F,U,k,Z]=abinv(Sl,alfa,beta);
[A,B,C,D]=abcd(Sl);dimV=dims(3);dimR=dims(1);
V=X(:,1:dimV);X2=X(:,dimV+1:nx);
X2'*(A+B*F)*V
(C+D*F)*V
X0=X(:,dimR+1:dimV); spec(X0'*(A+B*F)*X0)
trzeros(Sl)
spec(A+B*F) //nr=2 evals at -1 and noc-dimV=2 evals at -2.
clean(ss2tf(Sl*Z))
A=diag(1:6);A(2,2)=-7;A(5,5)=-9;B=[1,2;0,3;0,4;0,5;0,0;0,0];C=[zeros(3,3),eye(3,3)];
sl=syslin('c',A,B,C);sl=ss2ss(sl,rand(6,6));
[X,dims,F,U,k,Z]=abinv(sl,alfa,beta);
[A,B,C,D]=abcd(sl);clean(X'*(A+B*F)*X)
clean(X'*B*U)
.fi
.SH AUTHOR
F.D.
.SH SEE ALSO
cainv, st_ility, ssrand, ss2ss
|
