An introduction to behaviours

The main purpose of this section is to characterise, quickly, the behaviours of high
order systems. Characterisation is done initially by considering which signals are
included in the behaviour, for example: steps, ramps, sinusoids and combinations of
these. For example:

Core characterisation points:

1. Constants are constant! [This also means they are convergent.]

2. Exponentials (with negative exponents) converge.
3. Exponentials (with positive exponents) diverge.
4. Sinusoids oscillate for ever and thus do not converge.
5. Ramps and parabola (not on plot) diverge.
6. Sinusoids and ramps multiplied by convergent exponentials, do converge.

Later, we extend this list to include core properties such as steady-state, convergence
and damping.
Characterisation Techniques: Often a signal or system will be captured through its
Laplace Transform representation, and thus one infers the underlying dynamics and
behaviours directly from the transform.
Background - assume students A key skill is to infer directly from the Laplace
are familiar with the look-up transform what a signal will do and its key
table for Laplace transforms. attributes. This done from the pole positions.
f(t) F(s) f(t) F(s)
1 1 sin wt w
s s  w2
2

t 1 coswt s
s2 s  w2
2

e  at 1
e  at sin wt w
sa ( s  a) 2  w 2

SIMPLE POLES
What is the link between the pole position 1

and the time domain behaviour? 0.9

0.8

 If the pole is in the left half plane, 0.7

1 1
1  e  3t   ; pole  3
then the exponential is convergent. 0.6
s s3
 If the pole is in the RHP, then the
0.5

0.4

exponential is divergent. 0.3

convergent
0.2

0.1

 at 1 1
 ; ebt 
0
e 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

sa s b 20

18

4 16

LHP RHP 1
e 
3t
14
; pole  3
2
12 s 3
10

0 6 divergent
4

2
-2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

-4
-4 -2 0 2 4

SUMMARY: Poles in LHP plane are denoted as stable and those in RHP as unstable
due to the associated behaviours being convergent and divergent respectively.
 LHP means left half plane, that is the real part is negative.
 RHP means right half plane, that is the real part is positive.

GENERALISING to other signals multiplied by an exponential

  A SIGNAL/SYSTEM IS STABLE IF IT IS CONVERGENT.
 A SIGNAL/SYSTEM IS UNSTABLE IF IT IS DIVERGENT.
sb
e bt cos wt  ; Convergent so stable {LHP poles at -b±𝑗𝑤}
( s  b) 2  w 2
1
te at  Divergent so unstable {RHP poles at a}
( s  a) 2
sa
e at sin wt  Divergent so unstable {RHP poles at a±𝑗𝑤}
( s  a) 2  w2
1
te bt 
( s  b) 2 Convergent so stable {LHP poles at –b}

Is there a link between the pole positions and whether the corresponding signal is
convergent?
1. It is clear that where the poles are in the LHP, the associated signal is
convergent, that is, has an exponential with a negative power.
2. Where the poles are in the RHP, the associated signal is divergent, that is, has
an exponential with a positive power.
3. Note that poles on the imaginary axis (excluding origin) imply behaviour that is
neither convergent or divergent in general (e.g. sinusoids oscillate).

4
1
e 2t  ; pole  2
s2
2

1
0 et  ; pole  1
s 1

-2

-4
-4 -2 0 2 4
s3
e 3t cos t  ;
( s  3) 2  12
poles  3  i
 Behaviour of a system or signal?
Often engineers will talk interchangeably about systems and signals – while this
could be considered sloppy, in practice it is permissible due to the explicit link
between the two. However, it is important that students understand the
distinction so they apply their learning and analysis appropriately.

WHAT IS A SIGNAL AND WHAT IS A SYSTEM? TRANSFER FUNCTIONS.

• Signals are functions of time such as u(t) and y(t). These have Laplace
transforms U(s) and Y(s). Examples could be:
𝑦𝑦(𝑡𝑡) = sin(2𝑡𝑡) , 𝑥𝑥(𝑡𝑡) = 𝑒𝑒 −𝑡𝑡 , 𝑧𝑧(𝑡𝑡) = 𝑒𝑒 −3𝑡𝑡 cos(0.4𝑡𝑡)
• A system consists of real components. Examples could be a car, a distillation
column, a human body and so forth.
• The behaviour of a system is characterised by its output (say y(t)) which is a
signal. This signal depends upon the input to the system (say u(t) – another
signal) as well as the underlying system design and parameter values.

Linear systems may be represented by transfer functions such as G(s) here.

The response of the system is denoted as Y(s)=G(s)U(s).

The Laplace transform of the system output Y(s) is the system transfer function G(s)
times the Laplace transform of the system input U(s).

NOTE the use of different words: (i) Laplace transform for signals and (ii) transfer
function for systems.
The transfer function will originate from an ODE or something similar, for example:
𝒅𝒅𝒅𝒅 𝟒𝟒
𝟐𝟐 + 𝟑𝟑𝟑𝟑 = 𝟒𝟒𝟒𝟒 → 𝒀𝒀(𝒔𝒔) = � � 𝑼𝑼(𝒔𝒔)
𝒅𝒅𝒅𝒅 𝟐𝟐𝟐𝟐 + 𝟑𝟑
KEY OBSERVATIONS:
1. The system behaviour is in effect contained in Y(s).
2. However Y(s) includes both G(s) and U(s) and thus has 2 behaviours:
(i) behaviours associated to the input signal u(t) and
(ii) behaviours associated to the system dynamics (from G(s) or the ODE
parameters).

It is normal to assume that the input U(s) is convergent, that is a step signal or
similar. Consequently any instability and the dominant behaviours come from G(s).

CONCLUSION: Engineers often talk interchangeably about the behaviour of the

system or the behaviour of output Y(s) as they reduce to the same analysis even
though one is a signal and one a system. For example:
• Instability refers to divergent system behaviour but critically can be assessed
based on ensuring either Y(s) or G(s) have only LHP poles (assuming that U(s) is
convergent).
• Any RHP pole implies divergence or instability.

The following notes focus on characterising the behaviours of Laplace transforms

(that is signals), but it is implicit that in fact we are talking about systems as the
signals are system outputs.
 Behaviours: speed of response
This note assumes that inferences about
output behaviour Y(s) can be made be solely
on an analysis of transfer function G(s).
Typically U(s) is assumed to be a step signal.

WHAT DO I MEAN BY FAST? HOW DO I QUANTIFY SPEED OF RESPONSE

Fast change of speed A fast turn could be One would expect 0-
would be anything less anything less than around a 60mph within about 4sec,
then about 20 seconds. minute. so fast means 2-3
seconds!

KEY OBSERVATIONS: Speed of response is a relative concept. You need to define

what you think is a reasonable expectation for a given context. Normal speed would
match open-loop poles, for example the time constant from an ODE model.

DEFINING SPEED OF RESPONSE USING FIRST ORDER MODELS

Consider the natural dynamics of a 1st order
model such as: 2

4 1
𝐺𝐺(𝑠𝑠) = ; 𝑇𝑇 = 0.5; 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 = − 1.5

𝑠𝑠 + 2 𝑇𝑇
A sketch of the step response Y(s)=G(s)[1/s] 1

includes the constant and exponential Time constant = 0.5

components of the behaviour. 0.5

Settling time is linked to the pole (or 0

equivalently time constant). 0 0.5 1 1.5 2 2.5 3
Time (sec)

SUMMARY: If the specified settling-time is around W seconds, then we would expect

that the time constant is about T=W/3 and the system pole is around pole=-3/W.
EXAMPLES of ‘ideal’ poles and settling times for different scenarios: T = -1/pole
Desired settling Required time constant (approx..) Required pole (sec or hr)
time
120 sec or 2min 40 sec or 2/3 min -0.025 or -1.5
0.01 sec 0.003sec -333
2hr or 7200 sec 2400 sec or (2/3) hr -0.0004 or -1.5

WHAT IS THE EXPECTED SPEED OF RESPONSE?

Where there are more than one pole, a default is to assume the settling time is based
on the slowest pole. In practice, it may be faster but you cannot guarantee this.
5 Slowest pole is -2 0.8
G= or time constant
( s + 2)( s + 4)
0.5. 0.6

Expected settling is
0.4
1.5 sec.
0.2
This is a fair
approximation as 0
0 1 2 3
seen here.
5( s + 0.3) Slowest pole is -0.1 80
G= or time constant
( s + 0.2)( s + 0.1)
10. 60

Expected settling is 40

30 sec.
20

This is a fair 0
approximation is 0 10 20 30 40

seen here.
s + 0.2 Slowest pole is -0.1 1
G= or time constant
( s + 2)( s + 0.1) 0.8
10.
Expected settling is 0.6
30 sec.
0.4
This system has a
much faster initial 0.2
response, but the
0
expected settling 0 10 20 30 40

time is still a good

estimate.
SUMMARY:
1. Speed of response can sometimes be estimated from transfer function poles
(time constant is negative inverse of the pole and settling time is about 3-4
times the time constant).
2. However, if poles differ significantly in magnitude, only a full partial fraction
expansion will reveal the residue sizes and thus the contribution to behaviour
of each mode.
3. In the absence of more information, the slowest mode is assumed to dominate
the settling time as the fast modes will converge much faster, so eventually
only the slow mode will remain.
4. The further a pole moves into the LHP, the faster the implied convergence. For
example:
• A pole at -1 has a settling time of 3-4 sec.
• A pole at -5 has a settling time of 0.6-0.8 sec.

LHP RHP

X X X X
Converge Converge Diverge Diverge
fast slow slow fast

REMARK: In line with the analysis of 1st order systems, it would not be uncommon
to use the words ‘settling time’ to define the time by which the response reaches
to within 2-5% of steady-state. This is 3-4 times the dominant time constant.
 Behaviour: links to pole positions
This note assumes that inferences about
output behaviour Y(s) can be made be solely
on an analysis of transfer function G(s).
Typically U(s) is assumed to be a step signal.
We know that speed of response/convergence is linked to real poles. This summary
shows how the response shape in general is linked to all the pole positions
including those which are complex.

WHAT IS THE EXPECTED SHAPE OF THE RESPONSE?

There is an explicit link between pole positions and the corresponding time domain
signal. Students should understand this link so they can infer behaviour just from a
transfer function.
3( s + 1) Expected settling
G=
3

( s + 1) 2 + 6 2 is 3 sec. (dotted
red line based on
2

⇒ e −t (3 cos 6t ) e-t component). 1

POLES = -1±6j Significant

0
Imaginary part oscillation in
bigger than real transients. -1

part. Frequency of -2

Real part of pole is - oscillation is 6

1 or time constant 1. rad/s - taken from -3
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

imag(pole).
3( s + 1) Expected settling
G=
3

( s + 1) 2 + 12 is 3 sec (based on
real part of pole).
2

⇒ e −t (3 cos t ) Oscillation is 1

POLES = -1±j much slower than 0

above (just 1
Imaginary part same rad/s) and thus is -1

as real part. Real less noticeable as -2

part of pole is -1 or decay largely

time constant 1. complete within a -3
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

single period.
SUMMARY: Complex poles imply oscillation – this is obvious from the table of
Laplace transforms as a pole at -a+jw implies a signal of the form e-atsin(wt). The
decay is governed by the real part of the pole and the oscillation speed by the
imaginary part.

WHAT CONSTITUTES ACCEPTABLE OR GOOD BEHAVIOUR?

1. Oscillation is undesirable in general, so the imaginary part should be small
relative to the real part of the pole; this ensures the decay dominates the
oscillation.
2. The real part should be well into the LHP to ensure fast convergence.
3. Anything on the RHP implies divergence/instability and hence is unacceptable.
4. Anything close to he imaginary axis implies the oscillation dominates decay and
is unacceptable.
An indication of the likely interpretation of different pole positions is give here. The
shaded blue area is one where the poles may or may not be acceptable depending
upon residue sizes and context.

Too oscillatory RHP

Not allowed
Debatable
LHP
OK
Real
Too slow axis
Good/fast Good/ok

Imaginary axis

The scale, that is what is slow and fast, is defined by the context and thus requires
the engineer to make a decision.

QUESTIONS: Using the above figure and observations, what do you expect from the
following systems? Check your answers by doing some simulations in MATLAB noting
that an expectation is not a guarantee.
10 Poles are at -2±j2√2 so imaginary part is larger.
G= 2 Expect the oscillation to dominate decay and therefore
s + 4 s + 12
expect it to be unacceptable. However, dominance is
slight and therefore this may be borderline acceptable.
Expected settling time will be around 1.5 sec (time
constant is about 0.5).
 3 Poles are at -2±j√2 so imaginary part is larger.
G= Expect the decay to dominate oscillation and therefore
s 2 + 4s + 6
expect acceptable responses.
Expected settling time will be around 1.5 sec (time
constant is about 0.5).
3 Poles are at -2±j√11,-0.1 so imaginary part is larger.
G= 2 Expect the oscillation to dominate decay in transients
( s + 4 s + 15)( s + 0.1)
and therefore unacceptable. However, expected
overall settling time will be around 30 sec (time
constant is about 10 for slowest mode) and oscillatory
mode should decay in about 1.5 sec. so oscillations
may have a irrelevant impact overall.

Some step response plots to consider alongside the expected answers above.

1.4

1.2

0.8

0.6

0.4 1st
2nd
3rd
0.2

0
0 2 4 6 8 10
 Behaviours: signal steady-state values & FVT
The focus of this summary is to consider what type of signal has a finite
asymptotic value, that is: y ∞ = lim t →∞ y (t ) and moreover y∞ is bounded.
Secondly, can we classify those asymptotic values in a meaningful way.

WHICH OF THESE SIGNALS HAVE A BOUNDED AYSMPTOTIC VALUE?

1.5
0.04 t 0.04 t
e , e sin 2t , 0.2t , 1

divergent 0.5

e −0.2t , e −0.2t sin 2t , 0.2te −0.2t

0

-0.5

oscillatory -1

convergent
sin 2t -1.5
0 1 2 3 4 5 6 7 8 9 10

1. A pure sinusoid does not have an asymptotic value ALTHOUGH IT IS

BOUNDED.
2. An exponential with a positive exponent does not have an asymptotic value
(divergent).
3. Positive powers of time (t,t2,...) do not have asymptotic values (divergent).
ONLY constants and signals multiplied by an exponential with a negative
exponent converge to a fixed value. HEREAFTER CONSIDER ONLY CONVERGENT
SIGNALS.

What is the asymptotic value for the following convergent signals?

e-4t0 e-0.2t0 e-0.05t sin2t0 3+2t e-0.001t3+0 -2+2t e-tcos5t-2+0

MOST OF these signals have an asymptotic value of ZERO – only the constants
remain non-zero!

SUMMARY: The only signal with a non-zero asymptotic value is a constant!

HEAVISIDE STEP FUNCTION – this is a 1.2

convenient mathematical mechanism 1

for representing both: 0.8

1. A step input 0.6

2. A constant in future time.

0.4

0.2

H (t ) = 0 t < 0 1 0
L[ H (t )] =
H (t ) = 1 t ≥ 0 s
-0.2
-1 -0.5 0 0.5 1 1.5 2 2.5 3

INTERIM SUMMARY: A convergent signal will have a non-zero asymptotic value (or
steady-state) if and only if it contains a step function, that is, the partial fraction
expansion includes a term (A/s).
• By inspection, the asymptotic value will be A – assuming all other
components converge to zero.
• The steady-state can be determined by doing the partial fraction expansion
and finding A.

FINAL VALUE THEOREM (simple shortcut to find A)

WARNING: The FVT can only be applied if the signal is convergent. If applied to the
Laplace Transform of a divergent signal, the answer will be meaningless.
Students may note this is the same as
()
lim f t = lim sF s ( ( )) using the cover-up rule to find the
residue for the term (A/s) in the partial
t →∞ s →0
fraction expansion.

EXAMPLES of applying the FVT showing consistency between time domain and
result
 1   0 
 x(t ) = e ; X (s ) = lim t →∞ x(t ) = 0; lim s →0 sX (s ) = = 0
− at

 s + a  a 
 b   b 
−bt
 y (t ) = 1 − e ; Y ( s ) =  lim t →∞ y (t ) = 1; lim s →0 sY (s ) = = 1
 s ( s + b)   b 
 2   0 
−t
 z (t ) = e sin 2t ; Z ( s ) =  lim t →∞ z (t ) = 0; lim s →0 sZ (s ) = = 0

2
( s + 1) + 4   5 
1 0
�𝑤𝑤(𝑡𝑡) = 𝑡𝑡𝑒𝑒 𝑎𝑎𝑎𝑎 ; 𝑊𝑊(𝑠𝑠) = � � lim 𝑤𝑤(𝑡𝑡) = ∞; lim 𝑠𝑠𝑠𝑠(𝑠𝑠) = 2 = 0�
(𝑠𝑠 − 𝑎𝑎)2 𝑡𝑡→∞ 𝑠𝑠→0 𝑎𝑎
Inconsistent because signal not
convergent so FVT not valid!

REPEAT WARNING: FVT can only be applied if the signal is convergent. If applied to
the Laplace Transform of a divergent signal, the answer will be meaningless.
EXAMPLES of applying the FVT and links to partial fraction expansion
Suggest you validate these with MATLAB:
e.g. y=impulse(tf([1 6],[1 4 3]),1000);y(end)

5 A 2 B + Cs D  0 
G(s) = = + 2 + lim s →0 sG (s ) = = 0
2
( s + 1)( s + s + 4)( s + 2) s + 1 s + s + 4 s + 2  8 
No factor (1/s) so expect zero.
5 A 2 B + Cs D  5 
H ( s) = = + + lim s →0 sH (s ) = = A
s ( s 2 + s + 4)( s + 2) s s 2 + s + 4 s + 2  8 
Same result as cover up rule!
s+6  6
lim s →0 sL(s ) = 
L( s ) =  14 
s ( s + 2)( s 2 + 3s + 7)
Same result as cover up rule!
s+6 
lim s →0 sM (s ) =
0 
= 0
M ( s) =  − 56 
( s − 4)( s + 2)( s 2 + 3s + 7)
INCORRECT because signal not
convergent so FVT not valid!
s+6 
lim s →0 sN (s ) =
0 
= 0
M (s) =  26 
( s + 4)( s 2 + 7)
INCORRECT because signal not
convergent – it is sinusoidal!

REMINDER: If a transform does not include a single factor ‘s’ in the denominator, the
final value will be zero or infinite (except for pure sinusoids). FVT only applies to
convergent signals.
Behaviours: system steady-state values & FVT
In practice we are interested in the values
of the output of a system G(s). This note
focuses on the asymptotic output or
steady-state. We can deduce this using the final
value theorem.
𝐥𝐥𝐥𝐥𝐥𝐥 𝒚𝒚(𝒕𝒕) = 𝐥𝐥𝐥𝐥𝐥𝐥 𝒔𝒔𝒔𝒔(𝒔𝒔)
𝒕𝒕→∞ 𝒔𝒔→𝟎𝟎

WARNING: FVT can only be applied if the signal is convergent so we need to be

sure that the system output is convergent. Always check this first.

STEADY-STATE OUTPUT OF A SYSTEM

It is normal to assume U(s) is a constant The output can be computed as:
in a scenario where we are trying to find 𝐴𝐴
𝑌𝑌(𝑠𝑠) = 𝐺𝐺(𝑠𝑠)𝑈𝑈(𝑠𝑠) = 𝐺𝐺(𝑠𝑠)
the steady-state output. [If the input 𝑠𝑠
was not constant, neither would the
output be!] The output is convergent, if and only if
Assume that U(s)=A/s. G(s) has LHP poles. Check this first.
Assuming Y(s) represents a convergent signal, use the FVT to find the asymptotic
value, that is:
𝑨𝑨
𝐥𝐥𝐥𝐥𝐥𝐥 𝒚𝒚(𝒕𝒕) = 𝐥𝐥𝐥𝐥𝐥𝐥 𝒔𝒔𝒔𝒔(𝒔𝒔) = 𝐥𝐥𝐥𝐥𝐥𝐥 𝒔𝒔𝒔𝒔(𝒔𝒔) = 𝐥𝐥𝐥𝐥𝐥𝐥 𝑮𝑮(𝒔𝒔)𝑨𝑨 = 𝑮𝑮(𝟎𝟎)𝑨𝑨
𝒕𝒕→∞ 𝒔𝒔→𝟎𝟎 𝒔𝒔→𝟎𝟎 𝒔𝒔 𝒔𝒔→𝟎𝟎

Note, it is often convenient to use the shorthand G(0) in general. Exceptions are
where G(0) is not defined:
• If G(s) includes integrators, lim 𝐺𝐺(𝑠𝑠) = ∞. Hence, one cannot determine the
𝑠𝑠→0
asymptotic output using the formulae above.
SUMMARY
If all of the poles of G(s) are strictly in the LHP, that is there are no RHP factors and
no poles on the imaginary axis, then the asymptotic system output for a step input
of magnitude A is given by:
𝐥𝐥𝐥𝐥𝐥𝐥 𝒚𝒚(𝒕𝒕) = 𝑮𝑮(𝟎𝟎)𝑨𝑨
𝒕𝒕→∞
SYSTEM STEADY-STATE GAIN
Steady-state gain
is the asymptotic limt →∞ y (t ) lim s →0 sY ( s ) sG ( s )U ( s )
ratio of the system K = = = lim s →0
output to input; limt →∞ u (t ) lim s →0 sU ( s ) sU ( s )
again this assumes
that the input is
constant
K = lim s →0 G ( s )
(asymptotically).
NOTE: We do not need to define U(s) explicitly here and need
only assume it is convergent.
𝐥𝐥𝐥𝐥𝐥𝐥 𝑮𝑮(𝒔𝒔) = 𝑮𝑮(𝟎𝟎) 𝒐𝒐𝒐𝒐 ∞
𝒔𝒔→𝟎𝟎
Shortcut
The infinity arises if and only if there are poles on the origin.
REMARK: Steady-state gain assumes that the output is convergent and thus is well
defined irrespective of the poles of G(s).
Once students have studied feedback, they will realise that it is quite normal for a
system with RHP poles to have a convergent output, BUT, this requires very careful
selection of the signal U(s) (in effect U(s) cancels any RHP factors).

STEADY-STATE GAIN OR STEADY-STATE OUTPUT?

It is easy to mix up these two definitions, especially as they often give the same
answer when the input magnitude is unity, so we need to clarify the differences.
Always defined and meaningful irrespective
Steady-state gain: Given as G(0) of poles, although could be infinite if G(s)
has poles on the origin.
A is the input magnitude and needed here
Steady-state output: Given as G(0)A but not used in steady-state gain.
Only defined if G(s) has strictly LHP poles.
EXAMPLES of computations: For the system transfer functions given below,
determine both the steady-state gain SSG and also the steady-state output SSY for
a unit step response. [Thus means that U(s)=1/s and thus A=1]
𝒔𝒔 + 𝟏𝟏 SSG = G(0)=1/24, SSY = G(0) = 1/24
𝑮𝑮(𝒔𝒔) = System is stable so both give the same
(𝒔𝒔 + 𝟔𝟔)(𝒔𝒔 + 𝟒𝟒)
answer.
𝒔𝒔𝟐𝟐 + 𝟒𝟒 SSG = F(0)= 4/12, SSY = F(0)= 4/12
𝑭𝑭(𝒔𝒔) = 𝟐𝟐
(𝒔𝒔 + 𝒔𝒔 + 𝟐𝟐)(𝒔𝒔 + 𝟔𝟔) System is stable so both give the same
answer.
𝒔𝒔 + 𝟎𝟎. 𝟒𝟒 SSG = H(0) = -0.4/0.12, SSY = ∞
𝑯𝑯(𝒔𝒔) =
(𝒔𝒔 − 𝟎𝟎. 𝟐𝟐)(𝒔𝒔 + 𝟎𝟎. 𝟔𝟔) System is unstable (RHP pole) so SSY is not
defined (y(t) is divergent).
(𝒔𝒔 − 𝟐𝟐)(𝒔𝒔 + 𝟐𝟐) SSG = V(0) = 4/4, SSY = ∞
𝑽𝑽(𝒔𝒔) = System is unstable (RHP pole) so SSY is not
(𝒔𝒔 + 𝟏𝟏)(𝒔𝒔 − 𝟒𝟒) defined (y(t) is divergent).
𝒔𝒔 + 𝟏𝟏 SSG = W(0) = ∞ as there is an integrator
𝑾𝑾(𝒔𝒔) = SSY not defined as there is an integrator so
𝒔𝒔(𝒔𝒔 + 𝟏𝟏𝟏𝟏) y(t) is divergent.
𝒔𝒔 + 𝟒𝟒 SSG = G(0)=4/18, SSY = G(0) = 4/18
𝒁𝒁(𝒔𝒔) = System is stable so both give the same
(𝒔𝒔 + 𝟔𝟔)(𝒔𝒔 + 𝟏𝟏)(𝒔𝒔 + 𝟑𝟑)
answer.
 Behaviours: system stability
In practice we are interested in the values
of the output of a system G(s). This note
focuses on the convergence of the
output.

CONTEXT: In general terms the output of a system is something useful, perhaps a

temperature, speed, concentration, quality of a product. Hence, typically one
expects this output to be bounded and converge to a finite asymptotic value.

STABILITY: Without wanted to get tied up in the multiple possible definitions of

‘stability’, for an introductory course it is sufficient to understand that stability
means that: for a convergent input signal u(t), the output signal y(t) is also
convergent, that is:
𝐥𝐥𝐥𝐥𝐥𝐥 𝒖𝒖(𝒕𝒕) = 𝒇𝒇𝒇𝒇𝒇𝒇𝒇𝒇𝒇𝒇𝒇𝒇 → 𝐥𝐥𝐥𝐥𝐥𝐥 𝒚𝒚(𝒕𝒕) = 𝒇𝒇𝒇𝒇𝒇𝒇𝒇𝒇𝒇𝒇𝒇𝒇
𝒕𝒕→∞ 𝒕𝒕→∞

SUMMARY: Stability and convergence can be treated as equivalent concepts for

many scenarios. If an output signal converges, the underlying system is stable.

What Laplace transforms result in convergent signals to zero?

e-4t0 e-0.2t0 e-5t sin2t0 2t e-0.1t0 e-tcos5t 0
1 1 2 2 (𝑠𝑠 + 1)
𝑠𝑠 + 4 𝑠𝑠 + 0.2 (𝑠𝑠 + 5)2 + 22 (𝑠𝑠 + 0.1)2 (𝑠𝑠 + 1)2 + 52
Pole at -4 Pole at -0.2 Poles at -5 pm 2j Poles at -0.1 Poles at -1 pm 5j
What do these signals have in common? The real part of the poles is strictly less
than zero.

What Laplace transforms result in convergent signals to non-zero?

A A t ∞ 2+ e-5t  2 3+te-2t 3 2t+e-t ∞
𝐴𝐴 1 𝑠𝑠 + 7 2 1 2 1
+ +
𝑠𝑠 𝑠𝑠 2 𝑠𝑠(𝑠𝑠 + 5) 𝑠𝑠 (𝑠𝑠 + 2)2 (𝑠𝑠)2 𝑠𝑠 + 1
Pole at 0 2 poles at 0 Poles at 0, -5 Poles at 0, -2, -2 Poles at -1, 0 ,0
As long as the other poles have negative real parts, a single pole at the origin
results in a non-zero steady-state. More than one pole at the origin results in
divergence.
 What Laplace transforms result in divergent signals?
e4t ∞ e0.2t ∞ e5t sin2t ∞ 2t e0.1t ∞ etcos5t ∞
1 1 2 2 (𝑠𝑠 − 1)
𝑠𝑠 − 4 𝑠𝑠 − 0.2 (𝑠𝑠 − 5)2 + 22 (𝑠𝑠 − 0.1)2 (𝑠𝑠 − 1)2 + 52
Pole at 4 Pole at 0.2 Poles at 5 pm 2j Poles at 0.1 Poles at 1 pm 5j
If the real part of any pole is strictly greater than zero, the underlying signal is
divergent.

SUMMARY: If the output signal of a system is divergent, assuming a convergent

input, we talk about the system being unstable. Readers will note that this statement
is equivalent to simply checking the position of the poles:
• All poles must have real parts strictly less than zero for the system to be stable.
• If any poles have positive real parts, the system is unstable.

EXAMPLES: Which of the following transfer functions represent open-loop stable

systems.

𝒔𝒔 + 𝟏𝟏 Both poles are in the LHP – system is stable

𝑮𝑮(𝒔𝒔) =
(𝒔𝒔 + 𝟔𝟔)(𝒔𝒔 + 𝟒𝟒)
𝒔𝒔𝟐𝟐 + 𝟒𝟒 All poles are in the LHP – system is stable
𝑭𝑭(𝒔𝒔) = 𝟐𝟐
(𝒔𝒔 + 𝒔𝒔 + 𝟐𝟐)(𝒔𝒔 + 𝟏𝟏)
𝒔𝒔 + 𝟎𝟎. 𝟒𝟒 One pole is in RHP, system is unstable.
𝑯𝑯(𝒔𝒔) =
(𝒔𝒔 − 𝟎𝟎. 𝟐𝟐)(𝒔𝒔 + 𝟎𝟎. 𝟔𝟔)
(𝒔𝒔 − 𝟐𝟐)(𝒔𝒔 + 𝟐𝟐) One pole is in RHP, system is unstable.
𝑽𝑽(𝒔𝒔) =
(𝒔𝒔 + 𝟓𝟓)(𝒔𝒔 + 𝟏𝟏)(𝒔𝒔 − 𝟒𝟒)
𝒔𝒔 + 𝟏𝟏 One pole is on the origin (an integrator) and
𝑾𝑾(𝒔𝒔) = the other pole is in the LHP. Would be
𝒔𝒔(𝒔𝒔 + 𝟏𝟏𝟏𝟏) considered borderline stable, although some
industries would classify this unstable.
𝒔𝒔 + 𝟒𝟒 Double integrator so unstable.
𝒁𝒁(𝒔𝒔) =
(𝒔𝒔 + 𝟔𝟔)(𝒔𝒔 + 𝟏𝟏)𝒔𝒔𝟐𝟐
 Behaviours: summary
In practice we are interested in the
behaviour of the output of a system
G(s).

CONTEXT: In general terms it is useful to be able to

compare and contrast the behaviour of different systems. In order to do this we
need a language to undertake the comparison. We do this with a number of
quantitative measures.

Quantitative measures of behaviour.

1. Stability – does the output converge when the input is convergent?
2. Speed of convergence – how long does it take for the output to approach
steady-state (say to within 2% or 5%). Maybe called settling time.
3. Oscillation and overshoot – does the response oscillate and if so, how quickly
does the oscillation decay (usally expressed in terms of damping ratios).
4. Steady-state gain – what steady-state value is achieved for the output and
how does this compare to the steady-state input?

Readers will note that all the resources on behaviours have being explaining and
defining these measures.

REMARK: Sometimes readers will also be interested in rise-time, that is how

quickly a response accelerates towards its steady-state. However, in general terms
it is not possible to infer this directly from a Laplace transform or its poles and
some numerical computations are required.

Characterise the behaviours of systems with the following transfer functions

𝑠𝑠 + 1 • 2 LHP poles so stable
( )
𝐺𝐺 𝑠𝑠 = • Steady-state gain = 1/24
(𝑠𝑠 + 6)(𝑠𝑠 + 4)
• Slowest pole at -4 so expected settling
time is about 0.75-1 sec. (3-4 time
constants).
• Over-damped as only real poles.
 𝑠𝑠 2 + 4 • 3 LHP poles so stable
𝐹𝐹 (𝑠𝑠) = • Steady-state gain = 4/12
(𝑠𝑠 + 2)(𝑠𝑠 + 6)(𝑠𝑠 + 1)
• Slowest pole at -1 (T=1) so expected
settling time is about 3-4 sec.
• Over-damped as only real poles.
𝑠𝑠 + 0.4 • 1 RHP poles so unstable
𝐻𝐻 (𝑠𝑠) = • Steady-state gain = -2/0.6
(𝑠𝑠 − 0.2)(𝑠𝑠 + 0.6)
• Divergent, so settling time NA.
• Divergent so damping NA.
(𝑠𝑠 − 2)(𝑠𝑠 + 2) • 1 RHP poles so unstable
𝑉𝑉 (𝑠𝑠) = • Steady-state gain = 1/3
(𝑠𝑠 + 1)(𝑠𝑠 − 4)(𝑠𝑠 + 3) • Divergent, so settling time NA.
• Divergent so damping NA.
𝑠𝑠 + 1 • 1 LHP pole and 1 integrator so
𝑊𝑊 (𝑠𝑠) = borderline stable.
𝑠𝑠(𝑠𝑠 + 10) • Steady-state gain = infinite.
• Slowest pole at -10 (T=0.1) so expected
settling time is about 0.3-0.4 sec.
• Over-damped.
𝑠𝑠 + 4 • 3 LHP poles so stable
𝑍𝑍(𝑠𝑠) = • Steady-state gain = 4/6
(𝑠𝑠 + 6)(𝑠𝑠 2 + 0.1𝑠𝑠 + 1)
• Slowest ‘real part of pole’ at -0.05
(T=20) so expected settling time is
about 60-80 sec.
• Under-damped due to 2 complex poles.
Imag. part far bigger than real part so
expect significant oscillation.