Lecture 8
Grandfather of control and telecommunications
Two Flavours of Stability
Routh-Hurwitz Criterion
Motivation
Control systems must be designed to meet
performance criteria – e.g. step response
specs.
If the control system will be operated over a
long period of time, stability is usually also
necessary to be able to meet these criteria
We present two definitions of stability.
For LTI systems, both conditions turn out to
be very closely related
Stability w.r.t. Initial Conditions
𝐵𝐵(𝑠𝑠)
𝐺𝐺 𝑠𝑠 =
u(t)=0 𝐴𝐴(𝑠𝑠) Free response y(t) 0?
Arbitrary initial conditions
Say A(s), B(s) are given polynomials, e.g. from a diff. eq. model
𝑑𝑑𝑗𝑗 𝑦𝑦
If A(s) has order n, then initial conditions are 𝑗𝑗 (0− ), 𝑗𝑗 = 0, … , 𝑛𝑛 − 1.
𝑑𝑑𝑡𝑡
Definition
(input=zero)
Stable if output converges to zero for any
initial condition.
Unstable if output diverges for at least some
initial conditions
Marginally stable if output stays bounded
from all initial conditions and may converge to
zero for some initial conditions.
Visualisation
Equilibrium
Stable Marginally Stable Unstable
Free Response
Free response
𝑛𝑛 𝑑𝑑 𝑖𝑖 𝑦𝑦 𝑚𝑚 𝑑𝑑 𝑖𝑖 𝑢𝑢
∑𝑖𝑖=0 𝑎𝑎𝑖𝑖 𝑖𝑖 = ∑𝑖𝑖=0 𝑏𝑏𝑖𝑖 𝑖𝑖 = 0 , t≥0
𝑑𝑑𝑡𝑡 𝑑𝑑𝑡𝑡
𝑑𝑑 𝑗𝑗 𝑦𝑦
⟺ ∑𝑛𝑛𝑖𝑖=0 𝑎𝑎𝑖𝑖 𝑖𝑖
𝑠𝑠 𝑌𝑌 𝑠𝑠 − 𝑖𝑖−1
∑𝑗𝑗=0 𝑠𝑠 𝑗𝑗
0− =0
𝑑𝑑𝑡𝑡 𝑗𝑗
𝑑𝑑 𝑗𝑗 𝑦𝑦 − 𝑛𝑛
∑𝑛𝑛−1
𝑗𝑗=0 𝑠𝑠 𝑗𝑗
𝑗𝑗 0 ∑𝑖𝑖=𝑗𝑗+1 𝑎𝑎𝑖𝑖 � = poly(𝑛𝑛 − 1)
𝑑𝑑𝑡𝑡
⟺ 𝑌𝑌 𝑠𝑠 =
∑𝑛𝑛𝑖𝑖=0 𝑎𝑎𝑖𝑖 𝑠𝑠 𝑖𝑖 � = 𝐴𝐴(𝑠𝑠)
LTI Stability w.r.t. Initial Conditions
Stable:
All poles of A(s) strictly in the left-half
plane (LHP).
Marginally stable:
Non-repeated pole(s) on imaginary
axis, all other poles strictly in the LHP
Unstable:
Some pole(s) strictly in the right-half plane
OR repeated poles on the imaginary axis
You can understand this if you remember how terms in the partial fraction expansion
affect the impulse response.
Graphical Interpretation of Stability
Complex domain
(pole locations)
Brown colour represents a “forbidden (unstable) region”.
Bounded-Input Bounded-Output
(BIBO) Stability
System
Bounded input Bounded output?
u(t) y(t)
Zero initial conditions
Signals as functions of time
u(t) u(t)
1/ε
ε t
t
delta function or impulse δ(t) sinusoid
We measure the size of signal u using:
BIBO Stability Definition
(zero initial conditions)
Formally: For any 𝑢𝑢� > 0, there exists a finite
𝑦𝑦� > 0 such that for any input u with u(t) ≤ 𝑢𝑢� ,
𝑦𝑦 𝑡𝑡 ≤ 𝑦𝑦� , for all 𝑡𝑡 ≥ 0.
Informally: Bounded inputs yield bounded
outputs.
BIBO stability for LTI systems
BIBO stability holds if and only if (iff) the
impulse response g is absolutely integrable,
i.e.
∞
� 𝑔𝑔 𝑡𝑡 𝑑𝑑𝑑𝑑 < ∞
0
Proof of sufficiency ():
BIBO Stability
Proof of necessity (): By contradiction.
∞ 𝑇𝑇
Suppose ∫0 𝑔𝑔 𝑡𝑡 𝑑𝑑𝑑𝑑 = lim ∫0 𝑔𝑔 𝑡𝑡 𝑑𝑑𝑑𝑑 = ∞.
𝑇𝑇→∞
For an arbitrary T>0, apply an input
𝑢𝑢. sign 𝑔𝑔(𝑇𝑇 − 𝑡𝑡) , 0 ≤ 𝑡𝑡 ≤ 𝑇𝑇,
𝑢𝑢 𝑡𝑡 = �
0, else.
𝑇𝑇
⇒ 𝑦𝑦 𝑇𝑇 = � 𝑔𝑔 𝑡𝑡 𝑢𝑢 𝑇𝑇 − 𝑡𝑡 𝑑𝑑𝑑𝑑
0
𝑇𝑇
= ∫0 𝑔𝑔(𝑡𝑡)𝑢𝑢. sign 𝑔𝑔(𝑡𝑡) 𝑑𝑑𝑑𝑑
𝑇𝑇
= 𝑢𝑢. ∫0 |𝑔𝑔 𝑡𝑡 |𝑑𝑑𝑑𝑑 ⟶ ∞. QED
BIBO for LTI Systems with
Rational Transfer Functions
BIBO stability iff
Transfer function is proper (why?), and
All poles lie strictly in the LHP (why?).
If no unstable pole-zero cancellations in B(s)/A(s),
BIBO stability Stability w.r.t. initial conditions
(but the reverse implication is not true)
If no unstable pole-zero cancellations, and B(s)/A(s) is
proper
BIBO stability Stability w.r.t. initial conditions
Why pay heed to a pole if cancelled by
a zero anyway?
Pole-zero cancellation said to occur if there is
a zero 𝛽𝛽𝑘𝑘 which equals a pole 𝛼𝛼𝑗𝑗 .
In this case, the factor (𝑠𝑠 − 𝛽𝛽𝑘𝑘 ) in the
numerator B(s) formally cancels out the factor
(𝑠𝑠 − 𝛼𝛼𝑗𝑗 ) in the denominator A(s), and both
factors disappear from the transfer function
G(s) = B(s)/A(s).
So why do we care…?
Why Cancelled Poles Still Count
A cancelled pole 𝛼𝛼𝑗𝑗 is no problem if initial
conditions are zero. But in practice, initial
conditions may not be exactly zero.
Recall y(t) = free response (due to nonzero initial
conditions) + forced response (due to nonzero
input).
Cancelled pole 𝛼𝛼𝑗𝑗 appears in the free response!
Could affect transient performance if too slow.
If 𝛼𝛼𝑗𝑗 unstable Trouble as t∞
Example of How Pole-Zero
Cancellations Arise
Consider a DE system
𝑦𝑦̈ − 𝑦𝑦 = 𝑢𝑢̇ − 𝑢𝑢, where 𝑢𝑢 𝑡𝑡 = 0, 𝑡𝑡 < 0.
2
̇ − )+𝑦𝑦(0− ))= 𝑠𝑠 − 1 𝑈𝑈 𝑠𝑠
(𝑠𝑠 −1)𝑌𝑌(𝑠𝑠) − (𝑠𝑠 𝑦𝑦(0
1 ̇ − )+𝑦𝑦(0− )
𝑠𝑠𝑦𝑦(0
𝑌𝑌 𝑠𝑠 = 𝑈𝑈 𝑠𝑠 +
𝑠𝑠+1 (𝑠𝑠+1)(𝑠𝑠−1)
System is BIBO stable, but not stable w.r.t.
initial conditions
Example of System That’s Stable w.r.t.
Initial Cond’s, but isn’t BIBO stable
Determining stability for LTI systems
Consider a transfer function
Poles are the solutions of either equation:
The second polynomial is “monic” and often used in tests.
Radical Facts
A quadratic polynomial
Roots of cubic and quartic polynomials can be found in
terms of 3rd and 4th roots (radicals). But the formulas are
impossible to remember
Galois (1811-32): Polynomial equations of order 5 or
higher cannot generally be expressed in terms of
radicals!
So how to manually check stability for systems of order>2 ?
A necessary condition for stability
If a monic polynomial
has all roots with negative real parts, then all
its coefficients are positive; that is
If a coefficient is negative or zero, then some
roots have positive or zero real part.
This is also sufficient condition for first and
second order polynomials.
Proof:
Recall that an arbitrary monic polynomial in s
with real coefficients can always be written as
a product of monic linear and quadratic terms
with real coefficients:
Both linear and quadratic terms have all
roots with strictly negative real parts iff all
their coefficients are positive.
Test:
If some coefficient in the characteristic
polynomial is negative or zero, then there are
some poles with positive/zero real part.
If all coefficients in the characteristic
polynomial are all positive, then use Routh-
Hurwitz criterion to check whether there are
any poles with positive/zero real part.
Examples
Consider
This polynomial has all positive coefficients but two roots on the imaginary axis:
Routh-Hurwitz test
For polynomial with positive coefficients:
we form the following (Routh) array:
Example of computing entries in the third row
Example of computing entries in the third row
Routh-Hurwitz test
The polynomial with positive coefficients
has all roots with strictly negative real parts if
and only if all elements in the first column of
the Routh array do not change their sign.
No need to solve for the roots!
Example
For a first order polynomial
Routh array is
Example
For a quadratic polynomial
Routh array is
Example
For third order polynomials
Routh array is
Example
For fourth order polynomials we have:
Example
Consider a feedback system with
The characteristic polynomial is
Find all values of K for which the system is
stable.
Example
We use formulas for 3rd order polynomials:
The system is stable if
Example
Consider the same system.
For which values of K would all roots of the
characteristic polynomial have real parts
strictly smaller than -1?
We modify the polynomial by introducing
Example
Routh array is
Hence, we have the conditions:
or
Exercise
Form the Routh array for the following
polynomial:
