DIGITAL FILTER DESIGN

Filters are the fundamental unit in the processing from all the analog and digital data
from all the type of measurement and sensing devices.They are used to filter out or sample out
the sensed analog values so that noise occurred due to external environment causing the
uneven rises in the amplitude could be reduced.There are widely two types of filter (i.e) analog
and digital.In case of the analog filters they are made with the help of the schematic that
consists of passive elements like inductor and capacitor and active elements like resistor.
Generally low pass filter,band pass filter, band reject filter and high pass filter are its different
variants.But due to presence of the inductor of larger sizes these element could take larger
space which would not be used in case of the size as the main constraint in the design.So
digital filters are only way to make it fit into the available size.

Digital Filters consists of adder , multiplier and memory unit that performs the
mathematical operations on a sampled digitized signal to reduce / enhance certain features of
the processed signal.These filters are linear in phase (i.e) they exhibit the linear behavior in its
equivalent frequency spectrum.Digital Filters are easy to simulate and design due to its
computation limitation under the sampling period. But it additionally requires the components
like ADC, DAC and DSP processors.But compared to the analog filters these digital filters
exhibit the high accuracy.The main applications of this type are in Audio Processing, Image
Processing, Modems and Equalizers.

Digital Filters are broadly classified into two types: Infinite Impulse Response(IIR) filter
and Finite Impulse Response Filter(FIR). IIR filter as the name suggests the length of its
impulse response is infinite . The output depends on the present input as well as their past
inputs and outputs.It consists of fewer coefficients so comparatively it offers lower computational
complexity . But the two major drawbacks for this type of design is its stability and phase
response. Since it consists of the feedback path improper filter coefficient calculation will lead to
wrong results. The frequency response of this type of filter is non linear in nature so the group
delay may increase / decrease which becomes difficult to predict the time taken for its entire
calculation.So they are mainly used in the resource constrained applications like real time
Embedded Systems, Wireless Communication and Biomedical Instrumentation .The schematic
of IIR filter is shown below. It describes the present output y[n] depends on the present input
x[n] , past inputs (x[n-1],x[n-2]) and past outputs(y[n-1],y[n-2]).
 x[n] -> Present Input
x[n-1] , x[n-2] -> Past Inputs
bo,b1,b2 -> Filter Coefficients
z -1 -> Memory Elements to store the past
inputs and outputs
y [n] -> Preset Output
y[n-1] , y[n-2] -> Past Outputs

Finite Impulse Response (FIR) Filters has the impulse response of finite in
length. The output mainly depends on the present input and the past inputs not on the present
outputs.The impulse response of an N th order discrete time FIR filter takes precisely N+1
samples before it then settles it to zero. Since the feedback is absent this type of filter are
stable.FIR filters exhibits linear behavior in the frequency spectrum so that it produces the
constant delay for all the frequencies , this makes it useful in various applications like audio
equalization and image processing .It consists of larger number of filter coefficients compared to
the IIR filter so it results in higher computational capability and memory requirements.These
filters are more efficient , lower complexity and they are free of limit cycle oscillations.But te
memory requirement and execution time are relatively more high.These filters exhibits the sharp
cutoff in the frequency spectrum. FIR filters are mainly used in digital communication
applications at the intermediate frequencies of the receiver. The schematic of the FIR Filter is
shown below.It describes that the output signal y[n] depends on only the present input x[n] and
past inputs (x[n-1],x[n-2]).

x[n] -> Present Input

x[n-1] , x[n-2] -> Past Inputs
bo,b1,b2 -> Filter Coefficients
z -1 -> Memory Elements to store the past
inputs
y [n] -> Preset Output
STEPS INVOLVED IN THE IMPLEMENTATION

1) Choose an ideal / desired frequency

response Hd(w).

2)Take the Inverse Fourier Transform of Hd(w)

to get hd(n).

3)Sample Hd(w) at finite number of points to get

H(k)

4)If hd(n) is determined then convert to finite

duration h(n)

5)Take the z transform of h(n) to get H(z) (i.e)

transfer function of the digital filer

6) Choose the suitable structure/method and

realize the filter

7)Calculate the Finite word length effects in the

obtained response

8)Implement the Realized filter structure using

the Hardware Description Languages like
Verilog HDL.

DESIGN TECHNIQUES FOR FIR FILTERS

The well known methods of designing FIR filters are as follows

1. Fourier Series Method

2. Frequency Sampling Method
3. Window Method
FOURIER SERIES METHOD

In this method desired frequency response converted to the Fourier Series by replacing
w by 2*pi*f*T. The Fourier Coefficients are calculated by taking the inverse fourier transform of
Hd(w) which is the desired impulse response of the filter hd(n). The finite impulse response
could be obtained by truncating the infinite impulse response hd(n) to N samples.

But the major drawback in this method of design is that the truncation operation of the
Fourier Series of the impulse response causes oscillation in the pass band and the
stopband.This effect is known as Gibbs phenomena.

FREQUENCY SAMPLING METHOD

In this method the ideal frequency response is sampled at sufficient number of points
(i.e) N points.These samples are the DFT coefficients of the impulse response of the filter which
is obtained by performing Inverse Discrete Fourier Transform Operation.The final expression is
multiplied by the exponential factor for getting the filter coefficients of symmetry in nature. With
the respect to consideration of the origin point these are subdivided into Type I and Type II
designs.

This technique can be used for any given magnitude response. This method useful for
design of the non prototype filters where the desired magnitude response can take any irregular
shape. Even Though this type of design is simple and direct in designing the results are not
optimal because the response generally deviates from what is desired between the samples.If
the desired frequency response is undersampled then the resulting impulse response will be
time aliased for some extent.

WINDOW METHOD

In this method the desired frequency response of the filter H d(w)is taken Inverse Discrete
Fourier Transform to obtain its impulse response hd(n).As per the selected window sequence
w(n) then by multiplying it with the impulse response results in the infinite duration then convert
it to a finite duration impulse response .This method eliminates the ringing effect at the
bandedge and results in lower sidelobes at the expense of an increase in the width of the
transition band of the filter.

As the length of the window increases the main lobe width is reduced which reduces the
width of the transition band but it also introduces more ripple in the frequency response.Since
the result comprises of the convolution operation the resultant can’t be optimal . The width of the
transition bands depends on the width of the main lobe of the frequency response of the window
function.The discontinuities in H(w) become transition bands between values on either side of
the discontinuity.
 Some of the commonly used windows are

1)Rectangular Window

The window function for an N point rectangular window is given by

Width of the main Lobe: 4*⊼/N

As the window is made longer main lobe becomes narrower and higher and the side
lobes become more concatenated around w=0 but the amplitude of side lobes is unaffected.
The increase in the length does not reduce the amplitude of the ripples but increases the
frequency.
Due to convolution of the desired response ans the window response side lobes gives
rise to ripples in both the pass band and stop band. The amplitude of ripples is dictated by the
amplitude of the side lobes.This occurrence of the side lobes at the pass band and stop band is
is known a Gibb’s phenomena. This effect can be reduced by less abrupt truncation of the filter
coefficients.

2)Hanning Window

The window function for an n point Hanning Window is given by

Width of the main Lobe: 8*⊼/N

This kind of filter results in the smaller ripples in both the pass band and stop band.So
the width of the main lobe is maximum and the magnitude of the side lobe is reduces so it
exhibits lower Gibbs phenomena effect.

3)Hamming Window
 The window function for an n point Hamming Window is given by

Width of the main Lobe: 8*⊼/N

In this kind of filter the magnitude of the first side lobe is decreased by 10 db relative to
the hanning window.Hamming window generates lesser oscillations in the side lobes than the
hanning window for the same lobe width.

4)Blacman Window

The window function for an N point Blackman Window is given by

Width of the main Lobe: 12*⊼/N - highest among all the windows
The main lobe width can be reduced by increasing the value of N.Blackman window
exhibits the higher magnitudes of stop band attenuation and peak of the first side lobe.

5) Kaiser Window

The window function for an N point Kaiser Window is given by

 Kaiser window exhibits minimum stop band attenuation and pass band ripple. Since it is
represented as the function in the time domain its Fourier Transform has maximum energy in
the main lobe. By varying the parameter ɑ , the desired side lobe level and main lobe peak can
be achieved.The main lobe width can also be varied with respect to length N. So Kaiser window
remains superior in comparison to all other windows.

FINITE WORD LENGTH EFFECTS

TABULATION ILLUSTRATING THE EFFICIENCY IN DIGITAL FILTERING

TIMING DIAGRAM OF VARIOUS WINDOWS

FLOATING POINT UNIT IMPLEMENTATION

PERFORMANCE ANALYSIS OF THE DESIGN