Вы находитесь на странице: 1из 2

Digital FIR Filter Design Using Evolutionary

Optimization

Satabdi Chatterjee
ECE Department
Durgapur Institute of Advanced Technology and Management,Durgapur
Satabdi.ece@rahul.ac.in

Abstract— Electric filters are used in practically all circuits which maintains a population of structures that evolves according to rules of
require the separation of signals according to their frequencies. selection, recombination and survival referred to as genetic operators.
Applications include noise rejection and signal separation in A shared environment determines the fitness or performance of each
industrial and measurement circuits, feedback of phase and amplitude individual in the population.
control in servo loops, smoothing of digitally generated analog (D-A)
Methodology:
signals, audio signal sampling and sound enhancement , channel
separation and signal enhancement in communication circuits.
DEWM is applied in order to obtain the actual filter response as
Such filters can be built from passive RLC components,
possible to the ideal response. The designed FIR filter with h(n)
electromechanical devices, crystals, or with resistors, capacitors and
individuals is even symmetric and of even order. The frequency
op-amps (active filters). Active filters are applicable over a wide
response of the FIR filter can be written as in equation
frequency range, are inexpensive and offer high input impedance, low N
output impedance, gain and a wide variety of responses. H (wk ) = �h( n)e - jwn
The frequency domain behavior of a filter is described n=0
mathematically in terms of its transfer function or network function. 2p k
This is the ratio of the Laplace transform of its output and input Where wk = ; H (wk ) is the Fourier transform complex vector.
N
signals. The voltage transfer function H(s) of a filter can therefore be This is the FIR filter frequency response. The frequency in [0, p ]
written as: interval is sampled with N points.
Vout ( s) H d (w ) = [ H d (w1 ), H d (w2 ), H d (w3 ),....H d (w N )]T
H ( s) =
Vin ( s ) H i (w ) = [ H i (w1 ), H i (w 2 ), H i (w3 ),....H i (w N )]T
Filters can be design by optimization algorithms. Optimization is the Where H d represents the approximate magnitude response of the
process of making something better. Optimization consists in trying designed filter; H i represents the magnitude response of the ideal
variations on an initial concept and using the information gained to filter and for low pass and high pass it is given as in H i (w ) = 1
improve on the idea. Optimization is the process of adjusting the
for 0≤ w ≤ wc
inputs to or characteristics of a device, mathematical process, or
experiment to find the desirable result. The input consists of = 0 otherwise.
variables; the process or function is known as the cost function, H i (w ) = 0 for 0 ≤ w ≤ wc
objective function or fitness function and the output is the cost or = 1 otherwise.
fitness. If the process is an experiment, then the variables are physical Where wc is the cut off frequency of either low pass or high pass
inputs to the experiment. Optimization refers to the selection of a best fitter.
element from some set of available values. An optimization problem The error function or fitness function used in this method is
consists of maximizing or minimizing a real function by K

systematically choosing input values from within an allowed set and {�[|| H d (e jwi ) | - | H i (e jwi ) ||]}2
computing the value of the function. Error = i =1
N
The error fitness function to be minimized using the evolutionary
algorithms like DE and DEWM. Each algorithm tries to minimize the
Keywords—MATLAB Software, Optimization. error and thus optimizes the filter response.
Introduction : Evolutionary optimization forms a set of solution
in that they are involved by biological evolution such as Basic Wavelet Theory:
reproduction, mutation, recombination, natural selection and survival Some seismic signals can be modeled by combining translational and
of the fittest. Candidate solutions to the optimization problem play dilations of an oscillatory function with a finite duration called a
the role of individuals in a population, and the cost function wavelet. There are two types of wavelet transform i.e. continuous
determines the environment within which the solutions lie. Evolution wavelet transform and discrete wavelet transform. The continuous
of the population then takes place after the repeated application of wavelet transform Wa ,b ( x ) of function f ( x) with respect to a
above operators. There are two main forces that form the basis of mother wavelety ( x ) �L2 (�) . The wavelet transform function
Wa ,b ( x ) can be written as
evolutionarysystems: recombination and mutation. Evolutionary
+�
Algorithm is an algorithm which incorporates aspects of natural 1
Wa ,b ( x ) = �f ( x )y *
a,b ( x ) dx
selection or survival of the fittest. An evolutionary algorithm Cy -�
1 x-b
) ; x ��, a, b ��, a>0.
x can be randomly generated within the interval [-2.5Xa, 2.5Xa].
Where y a , b ( x) = y(
a a The value of dilation parameter a is set to vary with the value of k/K.
Here (*) denotes the complex conjugate; where k is the current iteration number and K is maximum iteration
a denotes the dilation(scale) parameter; number. In order to perform a local search when k value is large, the
b denotes the translation (shift) parameter. value of a increases as k/K increases to reduce the significance of the
It is noted that a parameter controls the spread of the wavelet and b mutation. Hence a monotonic increasing function governing a and
determines its control position. A set of basis function y a ,b ( x ) is k/K may be written in the following equation
xw m
derived from scaling and shifting the mother wavelet. The basis a = e - ln( g1)(1- k / K ) + ln( g1)
function of the transform is known as daughter wavelet. The mother Where xw m is the shape parameter of monotonic increasing function,
wavelet has to satisfy the following condition g1 is the upper limit of parameter a. the value of a is between 1 to
_
+�
| y (w ) |2 1000. The magnitude of mutation operator s
Cy = 2p � dw < �
-�
w
Where y (w ) is the fourier transform of y (w ) and is given by
-
Decreases as a increases towards g1 with increasing iteration cycle.
following equation Changing the parameter x m will change the characteristics of
+�
- 1 monotonic increasing function.
y (w ) = y ( x)e

- jw x
dx
2p -�
y ( x) is confined to a finite domain and is bounded.
Association of wavelet mutation: It is proposed that every element
of the particle of the population will mutate. Among the population a
randomly selected i th particle and its j th element (within the limit Paper reference:
[ S j ,max , S j ,min ] ) at the k th iteration ( Si(,kj ) ; vector Si( k ) same as
1.G.Zaho, X.Peng , “Design of FIR filters with differential Evolution”, 8 th
- -
International Conference on Electronic Measurements and Instruments, 2007,
X i , g / updated v i , g will undergo mutation as given in the
ICEMI’07, pp 748-751.
(k ) (k ) - -
following Si , j ; vector Si same as X i , g / updated vi, g will
2.Karaboga N, Cettinkayal B.: “Design of digital FIR filters using differential
undergo mutation as given in the following equation Evolution algorithm,” Circuits Systems Signal Processing, 2006,25,(5), pp
649-660.

�Si(,kj ) + s ( S j ,max - Sik, j )if s > 0;
Si(,kj ) = � ( k ) 3.K.S. Reddy, M.S. Bharath, S.K.Sahoo, S.Sinha, J.P.Reddy, “ Design of
�Si , j + s ( Sik, j - S j ,min )if s � 0; Low Power, High Performance FIR filter using Modified Differential
1 x Evolution Algorithm”, International Symposium on Electronic System
Where s = y a ,0 ( x ) = y( )
a a Design,(ISED) 2011, pp62-66.
A Moore wavelet is defined in the following equation
x 2
x2
-( ) 4.S.Chattopadhyay, S.K.Sanayal, A.Chandra, “ A novel self adaptive
Thus s = 1 e x
a
-
y ( x) = e 2
cos(5 x) 2
cos(5( )) differential evolution algorithm for efficient design of multiplier less low pass
a a FIR filter”, International Conference on Sustainable Energy and Intelligent
Systems (SEISCON) 2011, pp 733-738.

5.G.Liu, Y. Li , G.He, “ Design of digital FIR filters using differential


evolution algorithm based on reserved genes” , IEEE Conference on
Evolutionary Computation(CEC), 2010, pp-1-7.

6.Sangeeta Mondal, Rajib Kar, Durbadal Mondal, Shakti Prasad Ghosal,


“Differential Evolution With Wavelet Mutation in Digital FIR Filter Design”.
Graphical presentation of Morelet Wavelet function

From this figure it is clear that as the dilation (scale) parameter


increases the amplitude y a ,0 ( x ) will be scale down.
In order to enhance the searching performance in fine tuning stage,
this property will be utilized in mutation operation. In the interval [-
2.5, 2.5] the mother wavelet contains the maximum energy of 99%.

Вам также может понравиться