Scribd
Upload
6 views

LPF FP

This document describes designing a fixed point FIR filter. It generates a noisy signal by adding noise to a recorded voice signal. It then generates filter coefficients for a low pass filter and implements the filter in fixed point form for different word lengths. It calculates the error between the fixed point and floating point representations and plots the filtered signals and errors, showing that the error decreases as word length increases.

Uploaded by

Sohini Roy
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
6 views

LPF FP

This document describes designing a fixed point FIR filter. It generates a noisy signal by adding noise to a recorded voice signal. It then generates filter coefficients for a low pass filter and implements the filter in fixed point form for different word lengths. It calculates the error between the fixed point and floating point representations and plots the filtered signals and errors, showing that the error decreases as word length increases.

Uploaded by

Sohini Roy
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 17

PROGRAMMABLE AND EMBEDDED SYSTEMS ASSIGNMENT PART 2

FIR FIXED POINT FILTER DESIGN

SOHINI ROY , 18IE33007

warning('off','all')

GENERATION OF A NOISY SIGNAL FROM A RECORDED VOICE SIGNAL

[V,Fs] = audioread('D:\LABS\PES\Comb_Filter_Design\Voice.m4a');
%sound(V,Fs);
[N,Fs] = audioread('D:\LABS\PES\Comb_Filter_Design\Noise6.m4a');
%sound(N,Fs);
Fs = 10000;
audiowrite('Voice_write.wav',V,Fs)
audiowrite('Noise_write.wav',N,Fs)

[V,Fs] = audioread('D:\LABS\PES\Comb_Filter_Design\Voice_write.wav');
[N,Fs] = audioread('D:\LABS\PES\Comb_Filter_Design\Noise_write.wav');

x = randn(10000,1); %% Random Gaussian Distributed Noise Signal Generation


h = ones(20,1)/20;
y= filter(h,1,x);

f = 50;
Ts=1e-04;
w = 0.01*sin(2*pi*f*Ts*[0:(length(y)-1)]'); %Pure Sinusoidal signal
yn=y+w;

V1 = 10*V(50000:54000);
N1 = N(50000:54000)+yn(1000:5000);
X = V1+N1;
plot(X)
title ("Noisy Signal")

1
GENERATING THE FILTER COEFFICIENTS

%%filterDesigner (Generates LPF coefficients with fc=1KHz and Fs = 10KHz)


% 31 ORDER LOW PASS FILTER
h1=load('Low_Pass_Filter.mat')

h1 = struct with fields:


Hlp: [-0.0057 0.0061 0.0173 0.0233 0.0206 0.0087 -0.0096 -0.0283 -0.0398 -0.0373 -0.0169 0.0207 0.0698 0.1209 0.

h2 = struct2cell( h1 );
h = cell2mat( h2 );

% For a Word Length of 32


W = 8

W = 8

%Determing the length of the fraction


hmax = max(abs(h))

hmax = 0.1871

b = log((2^(W-1)-1)/hmax)

2
b = 6.5204

%Getting the fraction length


b = 6

b = 6

%Getting the Fixed point Coefficients


ht = h/(2^b)

ht = 1×32
-0.0001 0.0001 0.0003 0.0004 0.0003 0.0001 -0.0002 -0.0004

%Getting the Filtered Waveform in Fixed Point Form


hf = sfi(h,W,b);
Xf = sfi(X,W,b);
yfixed = conv(Xf,hf);
yfloat = conv(X,h);

eh = 100*(h-double(hf))./h;
ex = 100*(X-double(Xf))./X;
ey = 100*(yfloat-double(yfixed))./yfloat;

bar(eh)
title("Error in Filter Coefficients values")

3
bar(ex)
title("Error in Filter Input values")

bar(ey)
title("Error in Filter Output values")

4
plot(yfloat) % Filtered signal using the floating point representation
title ("Filtered Signal in Floating Point")

5
plot(yfixed) % Filtered signal using the fixed point representation
title ("Filtered Signal in Fixed Point")

6
Here we shall see the results of filtering with fixed point coefficients and the comparison of plots for different
word lengths and the corresponsing fraction length hence obtained.

% For a Word Length of 8


W = 16

W = 16

%Determing the length of the fraction


hmax = max(abs(h))

hmax = 0.1871

b = log((2^(W-1)-1)/hmax)

b = 12.0734

%Getting the fraction length


b = 12

b = 12

%Getting the Fixed point Coefficients


ht = h/(2^b)

ht = 1×32
10-4 ×
-0.0139 0.0149 0.0423 0.0568 0.0503 0.0211 -0.0235 -0.0691

%Getting the Filtered Waveform in Fixed Point Form


hf = sfi(h,W,b);
Xf = sfi(X,W,b);
yfixed = conv(Xf,hf);
yfloat = conv(X,h);

eh = 100*(h-double(hf))./h;
ex = 100*(X-double(Xf))./X;
ey = 100*(yfloat-double(yfixed))./yfloat;

bar(eh)
title("Error in Filter Coefficients values")

7
bar(ex)
title("Error in Filter Input values")

8
bar(ey)
title("Error in Filter Output values")

9
plot(yfloat) % Filtered signal using the floating point representation
title ("Filtered Signal in Floating Point")

10
plot(yfixed) % Filtered signal using the fixed point representation
title ("Filtered Signal in Fixed Point")

11
% For a Word Length of 32
W = 32

W = 32

%Determing the length of the fraction


hmax = max(abs(h))

hmax = 0.1871

b = log((2^(W-1)-1)/hmax)

b = 23.1638

%Getting the fraction length


b = 23

b = 23

%Getting the Fixed point Coefficients


ht = h/(2^b)

ht = 1×32
10-7 ×
-0.0068 0.0073 0.0206 0.0277 0.0245 0.0103 -0.0115 -0.0337

%Getting the Filtered Waveform in Fixed Point Form


hf = sfi(h,W,b);
Xf = sfi(X,W,b);
yfixed = conv(Xf,hf);
yfloat = conv(X,h);

eh = 100*(h-double(hf))./h;
ex = 100*(X-double(Xf))./X;
ey = 100*(yfloat-double(yfixed))./yfloat;

bar(eh)
title("Error in Filter Coefficients values")

12
bar(ex)
title("Error in Filter Input values")

13
bar(ey)
title("Error in Filter Output values")

14
plot(yfloat) % Filtered signal using the floating point representation
title ("Filtered Signal in Floating Point")

15
plot(yfixed) % Filtered signal using the fixed point representation
title ("Filtered Signal in Fixed Point")

16
The error converges to a very small value for Word length 32 , Thus as the word length increases error reduces
and filtered signal becomes more accurate.

17

You might also like