Tex as A& M University

ECEN 314– Spring 2022

Signals and Systems

Module 6

Stability of LTI Systems

Learning Objectives

Be able to determine the stability of a system by observing

the locations of its poles in the s-plane.
Be able to use the Routh-Hurwitz procedure to determine
the stability of a system.
 Stability of LTI Systems
Previously we found that the output, 𝑦𝑦(𝑡𝑡), of an LTI system due to an input, 𝑥𝑥(𝑡𝑡) is
given by a convolution
∞ ∞
𝑦𝑦 𝑡𝑡 = 𝑥𝑥 𝑡𝑡 ∗ ℎ 𝑡𝑡 = � 𝑥𝑥 𝑣𝑣 ℎ 𝑡𝑡 − 𝑣𝑣 𝑑𝑑𝑑𝑑 = � ℎ 𝑣𝑣 𝑥𝑥 𝑡𝑡 − 𝑣𝑣 𝑑𝑑𝑣𝑣 .
−∞ −∞
In the next few slides, we seek to find under what conditions such a system would
exhibit stability.

Recall that a system is BIBO (bounded input, bounded output) stable if any input
which is always finite produces an output which is always finite.

The magnitude of the output is

∞
𝑦𝑦 𝑡𝑡 = � 𝑥𝑥 𝑣𝑣 ℎ 𝑡𝑡 − 𝑣𝑣 𝑑𝑑𝑑𝑑
−∞
∞ ∞
≤� 𝑥𝑥 𝑣𝑣 ℎ 𝑡𝑡 − 𝑣𝑣 𝑑𝑑𝑑𝑑 = � 𝑥𝑥 𝑣𝑣 ℎ 𝑡𝑡 − 𝑣𝑣 𝑑𝑑𝑑𝑑.
−∞ −∞
Now suppose the input is bounded so that 𝑥𝑥 𝑣𝑣 ≤ 𝑐𝑐1 for all 𝑡𝑡. Then
∞ ∞
𝑦𝑦 𝑡𝑡 ≤ 𝑐𝑐1 � ℎ 𝑡𝑡 − 𝑣𝑣 𝑑𝑑𝑑𝑑 = 𝑐𝑐1 � ℎ 𝑤𝑤 𝑑𝑑𝑤𝑤.
−∞ −∞
 Stability of LTI Systems
Consider an LTI system with a general transfer function of the form
(𝑠𝑠−𝑧𝑧1 )(𝑠𝑠−𝑧𝑧2 )⋯(𝑠𝑠−𝑧𝑧𝑚𝑚 )
𝐻𝐻 𝑠𝑠 = . (assume 𝑚𝑚 < 𝑛𝑛)
)(𝑠𝑠−𝑝𝑝 )⋯(𝑠𝑠−𝑝𝑝 )
(𝑠𝑠−𝑝𝑝1 2 𝑛𝑛

Furthermore, suppose all the poles are distinct (no repeated poles). Then using
partial fractions we can express the transfer function in the form
𝐴𝐴1 𝐴𝐴2 𝐴𝐴𝑛𝑛
𝐻𝐻 𝑠𝑠 = + +⋯ ,
𝑠𝑠 − 𝑝𝑝1 𝑠𝑠 − 𝑝𝑝2 𝑠𝑠 − 𝑝𝑝𝑛𝑛
where all of the constants 𝐴𝐴𝑘𝑘 are finite. The corresponding impulse response will be

ℎ 𝑡𝑡 = 𝐴𝐴1 𝑒𝑒 𝑝𝑝1𝑡𝑡 + 𝐴𝐴2 𝑒𝑒 𝑝𝑝2𝑡𝑡 + ⋯ + 𝐴𝐴𝑛𝑛 𝑒𝑒 𝑝𝑝𝑛𝑛𝑡𝑡 𝑢𝑢(𝑡𝑡).

Allowing for the possibility that some or all of the poles may be complex, write 𝑝𝑝𝑘𝑘 =
𝜎𝜎𝑘𝑘 + 𝑗𝑗𝜔𝜔𝑘𝑘 . Then
𝑛𝑛

ℎ(𝑡𝑡) ≤ � 𝐴𝐴𝑘𝑘 exp 𝜎𝜎𝑘𝑘 𝑡𝑡 + 𝑗𝑗𝜔𝜔𝑘𝑘 𝑡𝑡 𝑢𝑢(𝑡𝑡)

𝑘𝑘=1
𝑛𝑛 𝑛𝑛

= � 𝐴𝐴𝑘𝑘 exp 𝜎𝜎𝑘𝑘 𝑡𝑡 exp 𝑗𝑗𝜔𝜔𝑘𝑘 𝑡𝑡 𝑢𝑢 𝑡𝑡 = � 𝐴𝐴𝑘𝑘 exp 𝜎𝜎𝑘𝑘 𝑡𝑡 𝑢𝑢 𝑡𝑡

𝑘𝑘=1 𝑘𝑘=1
 Stability of LTI Systems
∞ ∞ 𝑛𝑛

𝑦𝑦 𝑡𝑡 ≤ 𝑐𝑐1 � ℎ 𝑡𝑡 𝑑𝑑𝑡𝑡 ≤ 𝑐𝑐1 � � 𝐴𝐴𝑘𝑘 exp 𝜎𝜎𝑘𝑘 𝑡𝑡 𝑢𝑢 𝑡𝑡 𝑑𝑑𝑑𝑑

−∞ −∞ 𝑘𝑘=1
𝑛𝑛 ∞ 𝑛𝑛 ∞
1
= 𝑐𝑐1 � 𝐴𝐴𝑘𝑘 � exp 𝜎𝜎𝑘𝑘 𝑡𝑡 𝑑𝑑𝑑𝑑 = 𝑐𝑐1 � 𝐴𝐴𝑘𝑘 exp 𝜎𝜎𝑘𝑘 𝑡𝑡 � .
0 𝜎𝜎𝑘𝑘 0
𝑘𝑘=1 𝑘𝑘=1

The term exp 𝜎𝜎𝑘𝑘 𝑡𝑡 evaluated at 𝑡𝑡 = ∞ will be finite only if 𝜎𝜎𝑘𝑘 < 0. In which case we
have
𝑛𝑛
1
𝑦𝑦 𝑡𝑡 ≤ 𝑐𝑐1 � 𝐴𝐴𝑘𝑘 − .
𝜎𝜎𝑘𝑘
𝑘𝑘=1
 We have shown that in this case, the output is bounded. Note: A similar line of
reasoning can be followed if some or all of the poles are repeated.

We have proven the following important result:

Theorem: An LTI system described by an ODE is BIBO

stable if all of the poles of the transfer function are in the
left half of the complex s-plane (real parts are negative).
 Examples
Example 1: Example 2:
𝑦𝑦 ′′ 𝑡𝑡 − 𝑦𝑦 𝑡𝑡 = 𝑥𝑥(𝑡𝑡) 𝑦𝑦 ′′ 𝑡𝑡 + 𝑦𝑦 𝑡𝑡 = 𝑥𝑥(𝑡𝑡)
𝑠𝑠 2 + 1 𝑌𝑌 𝑠𝑠 = 𝑋𝑋(𝑠𝑠)
𝑠𝑠 2 − 1 𝑌𝑌 𝑠𝑠 = 𝑋𝑋(𝑠𝑠) 𝑌𝑌(𝑠𝑠) 1
𝑌𝑌(𝑠𝑠) 1 𝐻𝐻 𝑠𝑠 = =
𝐻𝐻 𝑠𝑠 = = 𝑋𝑋(𝑠𝑠) 𝑠𝑠 2 + 1
𝑋𝑋(𝑠𝑠) 𝑠𝑠 2 − 1 1
1 =
= (𝑠𝑠 + 𝑗𝑗)(𝑠𝑠 − 𝑗𝑗)
(𝑠𝑠 + 1)(𝑠𝑠 − 1)
Both poles are on the imaginary axis.
One pole in the RHP, one pole in the We call this a “marginally stable”
LHP the system is not stable. system (marginally stable = not
stable).
𝜔𝜔 𝜔𝜔

*
𝜎𝜎 𝜎𝜎
* *
*
 Determining Stability of High
Order Systems
For higher order systems, finding the roots of the polynomials may be difficult. We
can always resort to numerical computations, but there are some results that can
determine stability without having to explicitly calculate the roots of the
characteristic polynomial.

First consider a system with several real poles in the LHP. Then the characteristic
polynomial will be of the form
𝑏𝑏 𝑠𝑠 = 𝑠𝑠 + 𝑝𝑝1 𝑠𝑠 + 𝑝𝑝2 ⋯ 𝑠𝑠 + 𝑝𝑝𝑛𝑛 ,
where the 𝑝𝑝𝑘𝑘 are now all positive numbers (poles at 𝑠𝑠 = −𝑝𝑝𝑘𝑘 ). When we multiply
this out, the resulting polynomial will have all positive coefficients.

Similarly with a pair of complex conjugate poles in the LHP (𝑠𝑠 = −𝜎𝜎𝑘𝑘 ± 𝑗𝑗𝜔𝜔𝑘𝑘 , with
𝜎𝜎𝑘𝑘 > 0) the polynomial will look like
𝑠𝑠 + 𝜎𝜎𝑘𝑘 + 𝑗𝑗𝜔𝜔𝑘𝑘 𝑠𝑠 + 𝜎𝜎𝑘𝑘 − 𝑗𝑗𝜔𝜔𝑘𝑘 = 𝑠𝑠 2 + 2𝜎𝜎𝑘𝑘 + 𝜎𝜎𝑘𝑘2 + 𝜔𝜔𝑘𝑘2 .
This also has all positive coefficients.

A stable system will have a characteristic polynomial that has either all positive or all
negative coefficients. If the coefficients alternate signs it is necessarily unstable. If
the coefficients do not alternate signs, it could still be unstable.
 Routh-Hurwitz Criterion
There is a procedure that goes by the name of Routh-Hurwitz that can be used to
determine if a system is stable or not. It involves the creation of the RH table which
can be generated following a fairly simple pattern:

Step 1: Populate the first two rows of the RH table with the coefficients of the
characteristic polynomial starting with the highest order coefficient and alternately
using the first two rows.
A C
Step 2: Starting in Row 3, form each new entry (𝐸𝐸) according to B D
𝐵𝐵𝐵𝐵 − 𝐴𝐴𝐴𝐴
𝐸𝐸 = , E
𝐵𝐵
where 𝐴𝐴 and 𝐵𝐵 are the leftmost entries in the two rows above 𝐸𝐸 and 𝐶𝐶 and 𝐷𝐷 are
the entries from the two rows above 𝐸𝐸 and one column to the right of 𝐸𝐸. Continue
until all entries are 0.

Step 3: The number of roots in the RHP will be equal to the number of sign
changes in the entries in the first column.  The system will be stable if all the
entries in the first column are of the same sign.
 Example

𝑏𝑏 𝑠𝑠 = 𝑠𝑠 4 + 4𝑠𝑠 3 + 𝑠𝑠 2 + 14𝑠𝑠 + 40

1�4−1�14
R3, C1: 𝐸𝐸 = = −2.5
4
Col. 1 Col. 2 Col. 3
Row 1 1 1 40 4�40−1�0
R3, C2: 𝐸𝐸 = = 40
4
Row 2 4 14 0

Row 3 −2.5 40 0
R4, C1: 𝐸𝐸 =
(−2.5)�14−4�40
= 78
−2.5
Row 4 78 0 0

Row 5 40 0 0
R5, C1: 𝐸𝐸 =
78�40−(−2.5)�0
= 40
78

Going down Column 1, we encounter two sign changes in the entries 4 → −2.5 and
− 2.5 → 78. Therefore there should be two roots in the RHP.

In fact, the roots of this polynomial are 𝑟𝑟1 = −1, 𝑟𝑟1 = −2, 𝑟𝑟1 = 1 + 𝑗𝑗, 𝑟𝑟1 = 1 − 𝑗𝑗.
 Example

Use Routh-Hurwitz to determine the number of zeros in the RHP of the polynomial
𝑏𝑏 𝑠𝑠 = 𝑠𝑠 4 + 10𝑠𝑠 3 + 35𝑠𝑠 2 + 50𝑠𝑠 + 24
 Feedback Control Systems

desired actual
speed error speed
+
+
_ Actuator Plant
Controller

measured speed

Sensor

Each block in this control system can be modelled as a system with a transfer
function. To simplify the system, we will combine the Controller/Actuator/Plant into
a single equivalent system
𝑥𝑥(𝑡𝑡) + 𝐻𝐻(𝑠𝑠) 𝑦𝑦(𝑡𝑡)
+
_
𝑌𝑌 𝑠𝑠 = 𝑋𝑋 𝑠𝑠 − 𝑌𝑌 𝑠𝑠 𝐺𝐺 𝑠𝑠 𝐻𝐻 𝑠𝑠
𝑌𝑌(𝑠𝑠) 𝐺𝐺 𝑠𝑠 𝐻𝐻(𝑠𝑠)
=
𝐺𝐺(𝑠𝑠) 𝑋𝑋(𝑠𝑠) 1 + 𝐺𝐺 𝑠𝑠 𝐻𝐻(𝑠𝑠)
 Feedback Control Systems
Suppose in our generic feedback control system, 𝐺𝐺 𝑠𝑠 = 1  the speedometer
perfectly and instantaneously measures the actual speed.

𝐾𝐾
Furthermore, suppose 𝐻𝐻 𝑠𝑠 = . For what values of 𝐾𝐾 will the
𝑠𝑠 4 +10𝑠𝑠 3 +35𝑠𝑠 2 +50𝑠𝑠+24
feedback control system exhibit a stable behavior?

In this case, the overall system has a transfer function given by

𝑌𝑌(𝑠𝑠) 𝐺𝐺 𝑠𝑠 𝐻𝐻(𝑠𝑠)
=
𝑋𝑋(𝑠𝑠) 1 + 𝐺𝐺 𝑠𝑠 𝐻𝐻(𝑠𝑠)
𝐾𝐾
𝑠𝑠 4 + 10𝑠𝑠 3 + 35𝑠𝑠 2 + 50𝑠𝑠 + 24
=
𝐾𝐾
1+ 4
𝑠𝑠 + 10𝑠𝑠 3 + 35𝑠𝑠 2 + 50𝑠𝑠 + 24
𝐾𝐾
= 4 .
𝑠𝑠 + 10𝑠𝑠 3 + 35𝑠𝑠 2 + 50𝑠𝑠 + 24 + 𝐾𝐾

From the previous example, we know that 𝑠𝑠 4 + 10𝑠𝑠 3 + 35𝑠𝑠 2 + 50𝑠𝑠 + 24 has all of its
roots in the LHP. We can use Routh-Hurwitz to help us determine which values of 𝐾𝐾
will lead to stability.
Feedback Control Systems

𝑏𝑏 𝑠𝑠 = 𝑠𝑠 4 + 10𝑠𝑠 3 + 35𝑠𝑠 2 + 50𝑠𝑠 + 24 + 𝐾𝐾 Since the first 3 terms in the first

column are all positive, the
remaining terms must also be
Col. 1 Col. 2 Col. 3 positive for the system to be
Row 1 1 35 24 + 𝐾𝐾 stable.
𝐾𝐾
Row 2 10 50 0 42 − > 0 → 𝐾𝐾 < 126
3
Row 3 30 24 + 𝐾𝐾 0 24 + 𝐾𝐾 > 0 → 𝐾𝐾 > −24
Row 4 𝐾𝐾 0 0
42 − The system will be stable for any
3
𝐾𝐾 in the range
Row 5 24 + 𝐾𝐾 0 0
−24 < 𝐾𝐾 < 126
Row 6 0 0 0