Stability
Stability
Introduction:
Unstable systems: The signal levels of the output or some internal states may become
too high and cause signal distortion or even damage to the system.
General observations:
o The solution of the continuous-time state-space formulation depends on the
matrix function:
n 1
At
e m (t ) Am
m0
1
0 (t ) 1 1 12 1n 1 1 t e
(t ) n 1 t
1
2
1 2 2 2 2 e
. . . . . .
. . . . . .
n 1 (t ) 1 2 n 1 en t
n n n
o From the above results we can conclude that the system becomes unstable
when one or more eigenvalues have a non-negative real part.
o The solution of the discrete-time state-space formulation depends on the
matrix function:
n 1
m (k ) A m
Ak
e
m 0
1
0 (k ) 1 1 12 1n 1 1k
(k ) 2 n 1 k
1 1 2 2 2 2
. . . . . .
. . . . . .
n 1 (k ) 1 2 n 1 k
n n n n
o From the above results we can conclude that the system becomes unstable
when the magnitude of one or more eigenvalues is larger than or equal 1.
Sources of instability:
o Input: BIBO (assuming zero state).
o Internal states: Asymptotic and marginal (assuming zero input)
Zero-state (Input-output) stability: BIBO
Continuous-time systems:
o BIBO: Every bounded input results a bounded output. The system must be
initially relaxed (zero state).
g (t ) dt M
0
where M is a constant and g (t ) is the impulse response of the system.
o Filter response of a BIBO system, as t :
For u(t) = a, y (t ) a G ( s ) s 0
For u(t) = sin(0 t), y (t ) G ( j 0 ) sin( 0t G ( j 0 )
o A proper rational G(s) is stable iff every pole of G(s) has a negative real part.
o The transfer matrix is given by:
adj ( H ) C adj ( s I A) B
G ( s ) C ( s I A) 1 B D Since H 1 G (s) D
|H | | sI A |
As a result, the poles of G(s) can be calculated from the eigenvalues of A.
Every pole of G(s) is an eigenvalue of A.
Because of possible cancellations, every eigenvalue of A is not
necessarily a pole of G(s).
o If every eigenvalue has a negative real part, then the system is BIBO stable.
o Even if some eigenvalues have a zero or positive real part, the system may
still be BIBO stable:
The matrix C or B is zero, which eliminates all s-terms in G(s).
(Example 5.2, text p. 126)
Pole-zero cancellations: A zero can cancel a pole at the same location.
If this happens to unstable poles, then the system is theoretically BIBO
stable even some eigenvalues have a zero or positive real part.
Since pole-zero cancellation requires the locations of the pole and the
zero involved to match exactly, a condition that is difficult to meet and
even more difficult to maintain, one should pay close attention to the
stability of such systems and consider the practical issues involved in
pole-zero cancellation.
Discrete-time systems:
o BIBO: Every bounded input results a bounded output. The system must be
initially relaxed (zero state).
g (k ) M
k 0
where M is a constant and g (t ) is the impulse response of the system.
o Filter response of a BIBO system, as k :
For u(k) = a, y ( k ) a G ( z ) z 1
j 0 j 0
For u(k) = sin(0 k), y (k ) G (e ) sin( 0 k G (e )
o A proper rational G(z) is stable iff the magnitude of every pole of G(z) < 1.
o The transfer matrix is given by:
adj ( H ) C adj ( z I A) B
G ( z ) C ( z I A) 1 B D Since H 1 G( z ) D
|H | | zI A|
As a result, the poles of G(z) can be calculated from the eigenvalues of A.
Every pole of G(z) is an eigenvalue of A.
Because of possible cancellations, every eigenvalue of A is not
necessarily a pole of G(z).
o If every eigenvalue has a magnitude < 1, then the system is BIBO stable.
o Even if some eigenvalues have a magnitude > 1, it may still be BIBO stable:
The matrix C or B is zero, which eliminates all z-terms in G(z).
Pole-zero cancellations: If unstable poles are cancelled by zeros, then
the system is theoretically BIBO stable even some |i| > 1.
Since pole-zero cancellation requires exact match of pole/zero
locations, a condition that is difficult to meet and even more difficult
to maintain, the stability of such systems must be watched closely.
Zero-input (internal) stability: Marginal and asymptotic
Basic information:
o Minimal polynomial () (p. 62): It is a monic polynomial. In general, () is
difficult to compute. For Jordan form matrices (upper triangular block):
( ) ( 1 ) n ( 2 ) n ...( i ) n where ni is called the index of i, which
1 2 i
o If all eigenvalues of A have negative real parts, then the unique solution of the
Lyapunov equation:
T A t
T
N e A t dt
A M + M A = -N is M e
0
M
2.2 1.6
1.6 4.8
Leading principal minors: 2.2 > 0,
M 8
The system is asymptotically stable.
o If all |s| < 1, then the unique solution of the Lyapunov equation:
AT
m
T M N Am
MA MA=N is
m0