example8_4.m

用dsp解压mp3程序的算法

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% example8_4.m - This program compare the linear convolution in time 
%                domain and fast convolution in frequency domain
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

close all, clear all;

x = randn(1,1024);
h = hanning(64)';

total_length = length(x)+length(h)-1; 

% Linear convolution in time domain

t0 = clock;
y_t = conv(x,h);
t_t = etime(clock,t0)

% Linear convolution in time domain

t1 = clock;
X = fft(x,total_length);
H = fft(h,total_length);
Y = X.*H;
y_f = ifft(Y,total_length);
t_f=etime(clock,t1)

% Plot the results

figure;
subplot(211), stem(y_t); title('Time domain linear convolution');
subplot(212), stem(real(y_f));title('Frequency domain fast convolution');

fprintf('Time taken in time domain %d \n',t_t);
fprintf('Time taken in freq domain %d \n',t_f);

% This portion computes the number of Nh that is required for efficient
% FFT-based linear convolution.
% Assumption Nh < Nx

Nx = 128;               % number of input samples,Nx(fixed)
point = 0;
for Nh = 1:1:Nx         % number of coefficient samples,Nh (increasing)
N = Nh+Nx-1;
lfft = 6*N*log2(N)+4*N; % computation load using FFT-based linear convolution
lc = Nx*Nh;             % computation load using linear convolution
if lc>lfft              % Only when the linear conv load > FFT load
    point = [point Nh]; % note the value of point(2) which indicates 
end                     % the minimum value of Nh that provides efficient
end                     % FFT-based linear convolution