EXPONENTIAL CURVE FITTING TO AN OUTPUT SIGNAL

• Exponential curve fitting is a powerful tool in signals and

systems analysis for extracting meaningful information
from measurements or any available data.

n y[n]
• A measured signal may represent various phenomena such
0 y[0]
as decay, growth, or oscillation. The general form of an
1 y[1]
exponential function is:
y[n]=Aαn
N y[ N ]
y[n]
• The goal of exponential curve fitting is to find the
parameters A and α that best fit to the data. This is done by
solving a set of equations.

n
• In the simples case, if we have y[0] and y[1] then the
solutions of equations will yield: n
 y[1] 
y[n] y[n]  y[0]  
 y[0] 

• Once the exponential curve is fitted to the data, it can be

n
used for finding H(z) and the drawing a block diagram.
Question: An RC circuit is experimented by applying
a unit step signal. The measured values are as follows:
y[0]=18
y[1]=10
Find an exponential approximation for y[n].

Answer
Assumed approximation: y  n   A n
It should satisfy the follow,ng equations:
n  0  18  A 0 
 Solution: A=18 and   10 / 18
n  0  10  A 1 
y  n   18  (10 / 18) n  18  0.5556 n

y[n]  18  0.5556n
 LEAST SQUARES REGRESSION
It is used when the number of data points is greater
n y[n] yL [n]  ln( y[n]) y[n]  A n
than the order of the model.
0 y[0] yL [0]
yL [n]  ln A  n ln  It is used to find a line or a polynomial that best
1 y[1] yL [1]
ln A  a0 fits the observed data points by minimizing the
sum of the squared differences between the
ln   a1 observed and predicted values.
N y[ N ] yL [ N ]
yL [n]  a0  a1n
w[n]
yL [0]  a0 1 0 
yL [1]  a0  a1 1 1 
   yL [0] 
yL [2]  a0  2a1 1 2   a0   yL [1] 
     
yL [3]  a0  3a1 1 3   a1   
n
   
   L
y [ N ] 
yL [ N ]  a0  Na1 1 N 
a 
A  0   YL
 a1 
 a0  1 T
 a   ( A A) A YL
T

 1
1 1 1 1 1
A
0 1 2 3 N 
Question: An RC circuit is experimented by n y[n]
applying a unit step signal. The output is measured 0 25
at 4 different time instants as given in the table. Find 1 17
an exponential approximation for y[n]. 2 14
Answer:
3 8
1 0
1 1  4 6 
A  AT A   
1 2  6 14 
 
1 3
 0.7 0.3 yL [n]  3.235  0.3612n
( AT A) 1   
 0.3 0.2 
10.771
AT y   
14.350 
 a0  1 T  3.235 
a   ( AT
A) A y   0.3612 
 1  
y[n]  25.39  0.6968n
A  e  25.39 
a0

 y[ n ]  25.39  0.6968 n

  e a1  0.6968 n y[n] Appr.

0 25 25.39
1 17 17.69
2 14 12.33
3 8 8.592
Question: The impulse response of a system is recorded as given in n y[n]
the table. Find an exponential approximation for y[n]. 0 0.7
1 0.9
Answer: 2 1.8
3 3.0
1 0
1 1 
4 4.4
  5 10 
A  1 2   AT A   
  10 30 
1 3
1 4 
yL [n]  0.4349  0.4881n
 0.6 0.2 
( AT A) 1   
 0.2 0.1 
 2.706 
AT y   
10.293
 a0  1 T  0.4349 
a   ( AT
A) A y   0.4881 
 1  
y[n]  0.6473  1.6291n
A  e a0  0.6473 n y[n] Appr.
 y[ n ]  0.6473  1.6291n

  e a1  1.6291  0 0.7 0.6473

1 0.9 1.0546
2 1.8 1.7181
3 3.0 2.7990
4 4.4 4.5599
Computer code:
clear;clc;
yn=[0.7,0.9,1.8,3.0,4.4].';
yL=log(yn);

N=numel(yn);
A=zeros(N,2);
A(:,1)=1;
A(:,2)=[0:N-1];

x=(transpose(A)*A)^-1*transpose(A)*yL

M=exp(x(1));
alp=exp(x(2)); yL [n]  0.4349  0.4881n
n=[0:0.01:5];
ya=M*alp.^n
plot(n,ya,'r',[0:N-1],yn,'b*');
grid on;

y[n]  0.6473  1.6291n

n y[n] Appr.
0 0.7 0.6473
1 0.9 1.0546
2 1.8 1.7181
3 3.0 2.7990
4 4.4 4.5599
 PADE APPROXIMATION TO A TRANSFER FUNCTION
(an advanced method)

1 2 P a0  a1 z 1  a2 z 2  ...  aN z  N
H ( z )  h[0]  h[1]z  h[2]z  ...  h[ P]z 
1  b1 z 1  b2 z 2  ...  bM z  M

a0  a1 w  a2 w2  ...  aN w N
h[0]  h[1]w  h[2]w  ...  h[ P]w 
2 P

1  b1 w  b2 w2  ...  bM wM

• Pade uses rational functions to approximate a polynomial of H(z) via a ratio

of two polynomials.

• The order M and N depend on factors such as the desired accuracy and the
behavior of the original function. They should be such that M+N=P

• Calculation of an and bn is done by solving a system of equations derived

from matching the Taylor series expansions of the original function and the
rational function.

• Pade approximation has numerous applications in various fields such as

control theory, signal processing, computational physics and
electromagnetics.
 x[n] y[n]
Example: The transfer function of a system 
is given by 1 1 2 1
H ( z)  1  z  z  1  w  w2 δ[n-1]
2 2
Determine the unknown coefficents in the
following form of Pade approximation: δ[n-1]

1 a a w 0.5
1  w  w2  0 1
2 1  b1 w

Answer:
1 a a w
1  w  w2  0 1
2 1  b1 w
 1 2
1  w  w  1  b1 w   a0  a1 w
 2 
(1) w0   b1  1 w   b1  12  w2  a0  a1 w

b1  12  0  b1   12
b1  1  a1  a1  1
2

a0  1 y[n]
x[n]

1  12 w
f pade ( w)  δ[n-1]
1  12 w δ[n-1]

1  12 z 1 0.5 0.5
H pade ( z ) 
1  12 z 1
 Graph of H(eiω) versus ω

asd h[n]  H (ei )  H (ei )  H (ei )  20 log10 H (ei )

Useful identities
ei  cos   i sin 
ei  e  i
cos    1  2sin 2 ( / 2)  2 cos 2 ( / 2)  1
2

ei  e  i
sin    2sin( / 2) cos( / 2)
2i
20 log10 H (ei ) cos 2   12  12 cos 2
sin 2   12  12 cos 2
cos  cos   12 cos(   )  12 cos(   )
sin  sin   12 cos(   )  12 cos(   )
 sin  cos   12 sin(   )  12 sin(   )
p s
A cos   B sin   A2  B 2 cos   tan 1 (B /A) 
Example: Plot the graph of H(eiω) x[n] y[n]
versus ω. 

δ[n-1]

Answer:
H ( z )  1  z 1
H (ei )  1  e  i  e  i / 2 ei / 2  e  i / 2   2e  i / 2 cos( / 2)
H (ei )  2 cos( / 2)

H (ei )
2
2/ 2
1


0 π/2 
 1
Example: Plot the graph of H(eiω) δ[n+1]
versus ω. x[n] y[n]
2 
δ[n-1]

Answer: 1

H ( z )  z  2  z 1
H (ei )  ei  2  e  i  2  2 cos   4 cos 2 ( / 2)

H (ei )

4/ 2


0 π/2 
Example: Plot the graph of H(eiω) x[n] y[n]
versus ω. 
δ[n-1]

-1

Answer:
H ( z )  1  z 1
H (ei )  1  e  i  e  i / 2 ei / 2  e  i / 2   2ie  i / 2 sin( / 2)
H (ei )  2 sin( / 2)

H (ei )
2

2/ 2


0 π/2 
 -1
Example: Plot the graph of H(eiω) δ[n+1]
versus ω. x[n] y[n]
2 
δ[n-1]

Answer: -1

H ( z )   z  2  z 1
H (ei )  ei  2  e  i  2  2 cos   4sin 2 ( / 2)

H (ei )

4/ 2


0 π/2 
 DISRETE-TIME FILTERS
I) Low Pass Filter (LPF)

H (ei )
1
i
H (e )   h[n]e
n 
 in

 2
1 1
 
i in
h[n]  H ( e ) e d   H (ei )ein d 
2 
2 0
 c 0 c  C is called the cut-off frequency

Steps to be followed in synthesis of  c

1 1 c Area
 H (e )e d  2  1e d  
i
systems h[0]  0 0

2  c
2
 c c
Step 1: Evaluate the integrals to find 1 1 ein sin(n )
 H (e )e d  2  1e d  2 in   n c
i in in
h[n] 
h[n] 2 c  c

Step 2: Find H(z)

H (ei )   h[2]ei 2  h[1]ei  h[0]e0  h[1]e i  h[2]e i 2 
Step 3: Determine the difference  c
  for n=0
equation h[n]  
 sin(nc ) for n  0
  n
Step 4: Implement the system by
Area
drawing a block digram Property: h[0] 
2
Example: Determine h[n] of a LPF by calculating for -4<n<4

Answer:
 /2
1 2 / 2 1 x h[n]
n  0  h[0] 
2 

 /2
1e0 d  
2

2 -3 -0.1061
 /2 -2 0
1 sin(n / 2)
n  0  h[n] 
2 
 /2
1ein d  
n
-1 0.3183
0 0.5
 1 1 0.3183
 2 for n=0
h[n]   2 0
 sin(n / 2) for n  0 3 -0.1061
 n

H ( z )  -0.1061z 3  0.3183z  0.5  0.3183z 1 -0.1061z 3

 Example: Draw the block diagram of the given system.
H ( z )  -0.1061z 3  0.3183z  0.5  0.3183z 1 -0.1061z 3

Answer:
-0.1061x[n  3]  0.3183x[n  1]  0.5 x[n]  0.3183x[n  1]-0.1061x[n  3]  y[n]

-0.1061

δ[n+2]

0.3183

δ[n+1]

x[n] y[n]
0.5 
δ[n-1]

0.3183

δ[n-2]

-0.1061

H (ei )
18
Example: Find h[0] of the given filter in the graph.
Answer: h[0]  1 The area  1 18 1 2   2    4.5
  
π/6
2 2 2  6
0 π/3
-π/3 -π/6
3
 II) High Pass Filter (HPF)

H (ei )
1 
H (e )  i
 h[n]e
n 
 in

 2
1 1
  H (e
i in i
h[n]  H ( e ) e d   )ein d 
 c 0 c  2 
2 0

 
1 2 c
 H (e )e d  2  1e d  1  
i
Steps to be followed in h[0]  0 0

2 
synthesis of systems c

 
1 2 1 sin(n )  sin(c n)  sin( nc )
 H (ei )ein d    1e
in
h[n]  d  2 
Step 1: Evaluate the integrals 2 
2 c
2 n n
to find h[n]
H (ei )   h[2]ei 2  h[1]ei  h[0]e0  h[1]e i  h[2]e i 2 
Step 2: Find H(z)  c
 1  for n=0
h[n]  

Step 3: Determine the   sin(nc ) for n  0
  n
difference equation

Step 4: Implement the system

by drawing a block digram
Property: H HPF  1  H LPF
 H (ei )
18 Example: Find h[0] of the given filter in the graph.

 -π/3 -π/6 0 π/6 π/3  

Answer:
1
h[0]  The area
2
1 1    1 1
 2  18  18   18     4.5
2  2 6 6 12 6 
Example: Determine h[n] of a HPF by calculating for -4<n<4

Answer:

1 1
 1e d  2
n  0  h[0]  2 0 x h[n]
2 /2 -3 0.1061

1 sin(n / 2) -2 0
 1e
in
n  0  h[n]  2 d  
2 /2
n -1 -0.3183
0 0.5
 1
 for n=0 1 -0.3183
2
h[n]   2 0
  sin(n / 2) for n  0 3 0.1061
 n

H ( z )  0.1061z 3  0.3183z  0.5  0.3183z 1 +0.1061z 3

Example: Draw the block diagram of the given system.
H ( z )  0.1061z 3 -0.3183z  0.5-0.3183z 1 +0.1061z 3

Answer:
0.1061x[n  3]-0.3183x[n  1]  0.5 x[n]-0.3183x[n  1]+0.1061x[n  3]  y[n]

0.1061

δ[n+2]

-0.3183

δ[n+1]

x[n] y[n]
0.5 
δ[n-1]

-0.3183

δ[n-2]

0.1061
 III) Band Pass Filter (BPF)

H (ei )

1 H (e )  i
 h[n]e
n 
 in

 2
1 1
 
i in
h[n]  H ( e ) e d   H (ei )ein d 
 C 2 C1 0 C1 C 2 
2 2 0


Steps to be followed in 1 2(C 2  C1 ) C 2  C1
synthesis of systems
h[0] 
2 

H (ei )d  
2



1 sin(nC 2 )  sin(nC1 )
Step 1: Evaluate the integrals h[n] 
2 

H (ei )d  
n
to find h[n]

Step 2: Find H(z) H (ei )   h[2]ei 2  h[1]ei  h[0]e 0  h[1]e i  h[2]e i 2 
 C 2  C1
Step 3: Determine the  for n=0
h[n]  

difference equation
 sin(nC 2 )  sin(nC1 ) for n  0
 n
Step 4: Implement the system
by drawing a block digram Area
Property: h[0] 
2
%% Symbolic integration
% Integrals may be evaluated
% by using int command
clear;clc;
syms w H n;
H=3*exp(-2*i*w)+4*cos(3*w); fft
int(H*exp(i*w*n),w,-pi,pi)

Time DTFT
%% Fast Fourier Transform command
signal signal
% The function H in the graph can be
% transformed into h[n] directly
% by using fft
ifft
clear;clc;
H=[3,5,1].';
h=ifft(H)
H=fft(h)

%%Numerical integration
% Trapezoidal rule may be used for this.
clear;clc;
a=-pi; Trapezoidal rule:
b=+pi; ba
K=20; h
h=(b-a)/K; K
w=a+[0:K]*h; k  a  kh
n=3; f ( )  H (ei )ei n
f=(3*exp(-2*i*w)+4*cos(3*w)).*exp(i*w*n);
b K 1
hn=h/2*f(1)+h/2*f(K+1)+h*sum(f(2:K-1)); h h
hn  f ( )d   f (0 )  f (K )  h f (k )
a
2 2 k 1
 Usage of FFT in the filter design

Example: Using π/3 as the cut-off frequency, design an LPF.

clear;clc;format short;
DTFT=@(h)ifftshift(fft(ifftshift(h)));
iDTFT=@(H)fftshift(ifft(fftshift(H)));
N=20;
wc=pi/3;

n=[-N:N];w=n*pi/N;

He=heaviside(w+wc)-heaviside(w-wc);
h=iDTFT(He);h(abs(n)>7)=0;
Ha=DTFT(h);
subplot(2,1,1);
plot(w,abs(He),'r',w,abs(Ha),'k','linewidth',2);
xlim([-1 1]*pi);ylim([0 1]*1.2);
grid on;title('H(e^{iw})');

subplot(2,1,2);
bar(n,real(h),'b','barwidth',0.2);
xlim([-1 1]*N/2);ylim([-.1 0.4]);
grid on;title('h[n]');
 III) Band Stop Filter (BSF)

H (ei )

1
i
H (e )   h[n]e
n 
 in

 2
1 1
 
i in
h[n]  H ( e ) e d   H (ei )ein d 
 C 2 C1 0 C1 C 2 
2 2 0


1 C1   C 2
Steps to be followed in h[0] 
2 

H (ei )d  



synthesis of systems
C 1 
1 1
 
in
h[n]  2e d   2ein d 
Step 1: Evaluate the integrals 2 0
2 C 2
to find h[n]

Step 2: Find H(z)

Step 3: Determine the

difference equation

Step 4: Implement the system

by drawing a block digram
 ANALOG LOW PASS FILTERS (LPF)

1
H ( ) 
1  j L / R

1 1
H ( )  H ( ) 
1  j RC 1  j RC

| |
 ANALOG HIGH PASS FILTERS (HPF)

1 1
H ( )  H ( ) 
1  j / ( RC ) 1  jR / ( L)

| |
 ANALOG BAND PASS FILTERS (BPF)

1
1 H ( ) 
H ( )   1 
j 1  1  jR C 
1   L    L 
R C 
 ANALOG BAND STOP FILTERS (BSF)

1
H ( )  1
 1  H ( ) 
1  jR /  L  j  1 
 C  1 / C 
R   L 