ANALYSIS OF FINITE WORD-LENGTH EFFECTS

ANALYSIS OF FINITE WORD-LENGTH EFFECTS

ANALYSIS OF FINITE WORD-LENGTH EFFECTS

The MATLAB function a2dR given below can be used to generate the decimal equivalent of the binary representation in sign-magnitude form of a decimal number with a specified number of bits for the fractional part obtained by rounding.

function beq = a2dR(d,n)

% BEQ = A2DR(D, N) generates the decimal

% equivalent beq of the binary representation

% of a decimal number D with N bits for the

% magnitude part obtained by rounding

%

m = 1; d1 = abs(d);

while fix(d1) > 0

d1 = abs(d)/(10^m);

m = m+1;

end

beq = 0;d1 = d1 + 2^(-n-1);

for k = 1:n

beq = fix(d1*2)/(2^k) + beq;

d1 = (d1*2) - fix(d1*2);

end

beq = sign(d).*beq.*10^(m-1);

To compare the performance of the direct-form realization of an IIR transfer

function with that of a cascade realization with both realized with quantized coefficients.

Program P9 2 evaluates the effect of multiplier coefficient quantizations on the cascade

form realization

% Program

% Coefficient Quantization Effects on Cascade

% Realization of an IIR Transfer Function

clf;

[z,p,k] = ellip(6,0.05,60,0.4);

[b,a] = zp2tf(z,p,k);

[g,w] = gain(b,a);

sos = zp2sos(z,p,k);

sosq = a2dT(sos,6);

R1 = sosq(1,:);R2 = sosq(2,:);R3 = sosq(3,:);

b1 = conv(R1(1:3),R2(1:3));bq = conv(R3(1:3),b1);

a1 = conv(R1(4:6),R2(4:6));aq = conv(R3(4:6),a1);

[gq,w] = gain(bq,aq);

plot(w/pi,g,’b’, w/pi,gq,’r--’);

axis([0 1 -80 1]);grid

xlabel(’\omega /\pi’);ylabel(’Gain, dB’);

title(’original - solid line; quantized - dashed line’);

pause

zplane(b,a);

hold on;

pzplot(bq,aq);

title(’Original pole-zero locations: x, o; New pole-zero locations: +, *’)

The above two programs can be modified easily to investigate the multiplier coefficient

effects on an FIR transfer function, as illustrated by Program P9 3 given below for the

direct-form realization.

% Program

% Coefficient Quantization Effects on Direct Form

% Realization of an FIR Transfer Function

%

clf;

f = [0 0.4 0.45 1];

m = [1 1 0 0];

b = firpm(19, f, m);

[g,w] = gain(b,1);

bq = a2dT(b,5);

[gq,w] = gain(bq, 1);

plot(w/pi,g,’b-’,w/pi,gq,’r--’);

\axis([0 1 -60 10]);grid

xlabel(’\omega /\pi’); ylabel(’Gain, dB’);

legend(’original’, ’quantized’);

pause zplane(b);

hold on

pzplot(bq);

hold off

title(’Original pole-zero locations: x, o; New pole-zero locations: +, *’)

to investigate the propagation of input quantization-noise to the output of a causal, stable LTI digital filter, the function noisepwr1 given below can be employed

function nvar = noisepwr1(num,den)

% Computes the output noise variance due

% to input quantization of a digital filter

% based on a partial-fraction approach

%

% num and den are the numerator and denominator

% polynomial coefficients of the IIR transfer function

%

[r,p,K] = residue(num,den);

R = size(r,1);

R2 = size(K,1);

if R2 > 1

disp(’Cannot continue...’);

return;

end

if R2 == 1

nvar = K^2;

else

nvar = 0;

end

% Compute roundoff noise variance

for k = 1:R,

for m = 1:R,

integral = r(k)*conj(r(m))/(1-p(k)*conj(p(m)));

nvar = nvar + integral;

end

end

disp(’Output Noise Variance = ’);disp(real(nvar))}