IN 227 Control Systems Design

Lecture 3

Instructor: G R Jayanth
Department of Instrumentation and Applied Physics
Ph: 22933197
E-mail: [email protected]
 Issues with time domain analysis of linear system
Background
• The time domain analysis of linear time-invariant systems, described in Lecture-2, is essentially complete. The
equations provide explicit response even to the most general case, i.e., for arbitrary order ‘n’, and arbitrary initial
conditions and inputs.
• We noted in lecture#2 that an intermediate step necessary to obtain the response of a linear system to any general
input is to first obtain its response g(t) to an impulse input. Generally, in practice, the dynamic equation of the
system is not available, and for complex engineering systems, first-principles based modeling is impractical. So, one
prefers to characterize the system experimentally. However, an impulse input, being “infinitely large” in magnitude
(for however short a length of time) always poses the threat of damaging the system under study. While some
systems can be subjected to such inputs without irreversibly damaging them, it is an impractical proposition to
“strike” any/every system with a hammer or apply a voltage spike and still expect the tested physical system to be
intact. There appears to be no obvious way around this problem if we stick only to time-domain approach.
Secondly, the convolution integral, which needs to be solved to obtain the response, is not an intuitive step: it is
hard to predict the response of the system given the specified input.
• The root cause for this problem is that we chose to express an arbitrary input as a sum of impulses and thus, the
response as a sum of impulse responses. In this lecture, we shall first investigate how an input u(t) can be
represented by signals that are “safer” than an impulse. Then, we shall try to use this representation and try to
obtain the impulse response without actually applying an impulse.
 The Fourier series representation of a signal

• In our attempt to express an arbitrary input in terms of “standard inputs” other than an impulse, we shall
start first with arbitrary periodic inputs u(t) of period T, viz., u(t+T)=u(t).
• It is well known that a periodic signal u(t) can be represented as a sum of sinusoids of period T and its
higher harmonics, i.e.,
 2   2   2   2 
u (t ) = a0 + a11 cos  t  + a12 sin  t  + ... + ak1 cos  k t  + ak 2 sin  k t  + ...
 T   T   T   T 
T 2 T 2
 2   2 
where ak1 = 2
T  u ( t ) cos  k t  dt
 T 
and ak 2 = 2
T  u ( t ) sin  k t  dt
 T 
k = 0,1, 2...
−T 2 −T 2

• Thus, it is seen that sine and cosine signals of frequency 1/T and its higher harmonics can form an
alternate set of “standard inputs” other than impulses with which to represent any arbitrary periodic input
of frequency 1/T. The quantity of each signal is determined by the Fourier series coefficients ak, bk.
• For the sake of convenience, the Fourier series above shall be written in terms of ω0 =2π/T as
 a − ja12   a11 + ja12 
u (t ) = a0 +  11  ( cos 0 t + j sin 0t ) +   ( cos 0t − j sin 0t ) + ...
 2   2 
k =
=  ak e jk0t
k =−
T 2
where a = a − ja = 1
k k1 k2
T  u ( t ) e− jk0t dt
−T 2
 u(t)
1

The Fourier series representation of a signal

-T/2 0 T/2 T t

−1, −T  t  0 -1

• Example: A square wave given by u ( t ) =  2
can be represented by means of Fourier series
 1, 0  t  T
 2

 0, k  even

with coefficients ak =  2 . The magnitude of these coefficients is sketched below:
, k  odd
 j k


ak
-5 -3 -1 0 1 3 5 k

• It is seen that infinitely many complex exponentials exp(ikωt) are needed to accurately reconstruct a
square wave. It is also worth pointing that complex exponentials of higher frequencies are required
to construct sharper features, such as discontinuities, while those of lesser frequencies are required
to construct the smoother features. Incase we choose to truncate the expansion and attempt to
construct the square wave with finite number of terms, the sharp features cannot be reconstructed
very well (Fig. on the right).

u'(t)
1

-T/2 0 T/2 T t
-1
 The Fourier Transform
• Suppose now that the period of the signal T→ infinity, or more practically, the signal does not repeat
during the period of our observation/experiment. Then the signal u(t) can be written as

u(t)
   1 T /2 
u (t ) = im
T →
 ak e jk0t
= im
T →
 
T u (t )e − jk0t dt  e jk0t
 -T/2 0 T/2 t
k =− k =−  −T /2 
• writing 1 =  the above equation can be rewritten as
T 2
  T /2      d
d  jt
u (t ) = Lt
 →0
  
2   u (t )e − jk (  ) t
dt  e

jk (  ) t
= 2
e

 u (t )e − j t
 
dt  =
2
e jtU ( j )
k =−  −T /2  −  −  −

• Thus an aperiodic signal u(t) can be represented as a sum of complex exponentials exp(ikωt) each of
infinitesimally small magnitude dω. The “quantity” of each such signal necessary to construct u(t) is given

by
 u(t )e
− jt
U ( j ) = dt
−


 u(t )e
− jt
U ( j ) = dt
• The term −
, which replaces the terms ak in the Fourier series representation of a signal,
is defined as the Fourier Transform of u(t). 

 U ( j)e d
− 1 jt
• The inverse Fourier transform is given by F (U ( j ) ) = u(t ) =
1
2
−

• It is worth noting at this point that we have succeeded in representing an arbitrary input u(t) in terms of
“standard inputs” exp(ikωt). These terms, unlike impulses, are finite in magnitude and thus, do not
potentially damage the system.
• U(jω) is also called the spectrum or the “frequency content” of the signal u(t).
 The Fourier Transform

• Fourier transforms of some common inputs:


Impulse:
  (t )e
− jt − jt
F ( (t ) ) = dt = e =1
t =0
−


Decreasing exponential: F e ( − at
) 
1(t ) = e−at e− jt dt =
1
a + j
0

1, t  0 1
Heaviside step 1(t): 1(t ) =  F (1(t )) = +  ( )
0, t  0 j

1
Sinusoid u (t ) = sin 0t U ( j ) =  ( − 0 ) −  ( + 0 )
2j
 Properties of the Fourier Transform

• Some important properties of Fourier transforms:

1  
• Time scaling: If U(jω) is the Fourier transform of u(t) then k U  j k  is the Fourier Transform of u (kt )
 − jt 

1  j 

1
Proof: U ( j ) =  u(t )e
− jt
dt . Define t  = kt . Then F[u (kt )] = u (t )e k dt  = U 
−
k k  k 
−

• This result indicates that if the signal is “speeded up” in the time domain by choosing k>1, then the
resulting signal requires more of the higher frequency complex exponentials to construct it, i.e., the
spectrum of the signal gets expanded. Likewise, if the signal is slowed down by choosing k<1, the resulting
waveform requires less of the higher frequency complex exponentials to construct it but more of the lower
frequency complex exponentials. Thus, the spectrum of the signal gets compressed.
Nominal signal Slow signal Fast signal
u(kt)

u(kt)
u(t)

0 t 0 t t

U(ω)
U(ω)

U(ω)

Fourier
Spectra

ω 0 ω 0 ω
0
 Properties of the Fourier Transform
t
• Fourier transform of a convolution x(t ) =  u( ) g (t −  )d
0
 

  u( ) g (t −  ) d e
− jt
F  x(t ) = X ( j ) = dt
− −
 

 u( )d  g (t −  ) e
− jt
= dt
− −
 

 u( )e  g (t1) e
− j − jt1
• Define t1 = t −  . Then X ( j ) = d dt1 = U ( j )G( j )
− −

• The expression above has special significance: It indicates that the Fourier transform of the impulse
response G(jω) can be obtained as G(jω)=X(jω)/U(jω), i.e., apply any input u(t), record the corresponding
response x(t) and take the ratio of their respective Fourier transforms.

• The time domain impulse response is obtained as g (t ) = 1 G( j )e jt d
2 
−

• It is worth noting at this point that the mathematical tool of Fourier transforms enables us to obtain the
all-important impulse response without actually applying an impulse input to the system, but indeed, by
applying any desired input. With the impulse response at hand, it is possible to evaluate the response to
any other input.
• It is desirable to apply a broad-spectrum input so that U ( j )  0 over the bandwidth of the LTI system.
 The Laplace Transform

• Drawbacks of the Fourier Transform: While the Fourier Transform is an ingenious route to obtain the
impulse response, it possesses a couple of important problems. The Fourier transforms of periodic signals
contain delta-functions, which, by their very nature, are inconvenient to deal with. Further, for signals that
increase with time, such as a ramp, increasing exponential etc…, the Fourier transform does not exist since
the Fourier integral is not finite. Since such signals are routinely encountered in LTI system responses, they
are of importance to us and the inability to use Fourier transforms to analyze them is a big handicap.
• In order to get the Fourier integral to be finite, such increasing signals can be weighted down by
decreasing exponentials exp(-σt) with an appropriately high rate of decrease. In such a case the signal is
related to its Fourier transform as
  

e e 
− t 1 jt 1 jt
e u (t ) = d U ( j ) = d  e− t u (t )e − jt dt
2 2
− − 0
   

e  e 
1  t jt − t − jt 1 ( + j )t
• Rearranging this we get u (t ) = e d e e u (t )dt = d  e−( + j )t u (t )dt
2 2
− 0 − 0

• Defining s =  + j and noting that d = ds / j , u(t) can be written in terms of the complex exponential
exp(-st) at  + j   + j

  
1 − s 1
u(t ) = e ds u( )e d =
st
est dsU ( s)
2 j 2 j
 − j 0  − j

• As with the Fourier transform, the coefficient U (s) =  u(t )e− st dt represents the frequency spectrum of the
signal for complex frequencies s =  + j 0

• U(s) is called the Laplace Transform of u(t)

 Laplace Transform
Laplace Transform 
• Definition: The laplace transform F(s) of f(t) is defined as F ( s) =  f (t )e − st dt where s is a general complex number
given by s=σ+jω 0

• Related note: Mathematicians and engineers have invented several transforms to help simplify solving linear
differential equations. A few

others include
• Mellin Transform: F ( z ) =  t z −1 f (t )dt
0

1
• Sumudu Transform: S (u ) =  f (t )e−t / u dt
0
u

• Properties
(1) Note that the Laplace transform is a complex-valued function of a complex number. Thus, its value is determined
by two variables-σ and ω. Thus, it can be thought of as describing a surface in the complex plane.
(2) Note also that a Laplace transform operation is a linear operation: the relationship between the time domain
signal f(t) and its modified version F(s) satisfies superposition and scaling.
(3) Existence: The Laplace transform exists for functions that are piecewise continuous (defined in the previous
lecture) and is of exponential order.
A function is said to be of “exponential order” if there exists a constants K , c and t1 such that f (t )  kect t  t1
Examples: t , eat , a0 + a1t + a2t 2 + ..
(4) Uniqueness: Every function u(t) for which the Laplace transform exists possess a unique Laplace transform.
 Laplace Transforms of physical systems
• The Laplace transform ‘transports’ us from time-domain into a new
domain, viz., the s-plane. To obtain a working knowledge of the
transformed systems, it is imperative that we establish connections
between linear systems in the time-domain and their transformed
counterparts.
• Region of convergence: It is worthwhile to start familiarizing ourselves with
the s-plane by noting that the Laplace transform of a function f(t) does not Region of
exist for all ‘s’. Since f is of exponential order, we have, f  et . Thus, convergence
   
. t 1 t 1
Im(s)
   f e dt +  f e dt   f e dt +  e e dt
− st − st − st − st − st t − st
L( f ) = fe dt  f e dt =
0− 0− 0− t1 0− 0−

• Thus, we see that the necessary condition for L(f) to exist is that Re(s)>ρ.
This region of the s-plane is called the region of convergence.
ρ Re(s)
 Laplace transforms of some common signals
• A dirac delta function (Note that the integral starts from 0-)

L− () =  (t )e − st dt = e − s (0) = 1
0−
0 t  0
• Heaviside Step: A Heaviside step 1(t) is defined as 1(t ) = 
1 t  0

1
L(1(t )) =  1(t )e − st dt =
0
s

• Exponential Function:

1
L(e at ) =  e at e − st dt =
0
s−a
jt
e − e− jt e jt + e− jt
• A sinusoid: sin t = cos t =
2j 2
 s
L(sin t ) = 2 L(cos t ) = 2
s + 2
s + 2
• Power of t (f(t)=tn):
n!
L(t n ) =
s n+1
Note that the above expression can be obtained from the Laplace transform of an exponential by expanding eat
on the LHS and 1/(s-a) on the RHS and subsequently comparing the coefficients of an on the LHS and the RHS
 References
(1) Alan Oppenheim, Alan Willsky, Hamid Nawab, Signals and Systems, Pearson (2002)
(2) K Ogata, Modern Control Engineering, PHI (2010)