A-weighting filter to lookup table

hello,

i'm working at an embedded SPL meter, it has to filter out the audio using A-weighting.

I've build a script to test A-weighting filters, but here comes the problem.

My idea is to build a lookup table, that contains the weight value for a specific frequency, then use this table to weighting the signal using fft.

I'm investigating three filters, one build upong the fdesign function, one two using filter coefficients based functions.

I've two question and one problem:

1) is my method correct?

2) what filter table i've to use? if i test it seems that the two coef based filters differs from an offset.

download sample code here: https://www.dropbox.com/s/62ueiv3rmc8l3tc/test.zip

thank you!

[Daniel Shub]

It is not clear what you are doing, what you are getting, and what you expect. Without this information, it is hard to answer your question.

Answer 1

[Wayne King]

How about using

fdesign.audioweighting

if you have the DSP System Toolbox? That will give you A-weighting.

Answer 2

[Al Hasan]

Hey Davide,

Was your method correct ?

I have attached an image is that A weighted filtering

By any chance do you know what they are doing to the above image to get the below image in my attachment, I have data to make the above image dunno how to make the below one from the above

Thanks, Hasan

Answer 3

[Davide]

no-one reply me, so i'm not whure my method worked. one of my questions, is about the use of fdesign.audioweighting, i've found other implementation of the A-weighting, that differs from DSP matlab one. i'm asking in my question why it differs. i mean A-weighting should always the same.

[Wayne King]

You should provide more detail about your above statement.

Answer 4

[Davide]

i'm using this script

%clear matlab workspace
clear;

%build the filter using fdesign
Fsf = 22050; %frequency used for filter creation
HawfA = fdesign.audioweighting('WT,Class','A',1,Fsf);
Afilter = design(HawfA);
%plot filter 1
%get filter freq resp
[AH, AW] = freqz(Afilter);
%plot filter
figure(1)
semilogx(AW*1000,20*log10(abs(AH)));
xlabel('Frequency [Hz]'); ylabel('Gain [dB]');
title(['A-weighting, Fs = ',int2str(Fsf),' Hz']);
legend('Filter');
grid on

%gaintable size
gaintablesize = 32;

%build filter lookup table
Flookup = 22050; %frequency used for lookuptable creation
WindowSize = gaintablesize*2; %samples for lookuptable creation / window length
filter_gaintable = freqz(Afilter, WindowSize, 'whole'); %built for the sized window

%build test signal
Fs = 22050; %sampling frequency of signal
T = 1/Fs; %sample time
L = WindowSize*20; %length of signal (20 * window)
t = (0:L-1)*T; %time vector
y = 0.7*sin(2*pi*50*t) + sin(2*pi*2000*t) + 2*randn(size(t)); %sum of a 50Hz and a 2000 Hz sinusoid plus noise
signal = y';

%estimate signal level (within each windowed segment).
windowIntervals = [1:WindowSize:(length(signal)-WindowSize)];
windowTime = t(windowIntervals+round((WindowSize-1)/2));
%will contain the full signal
signalT = [];
filteredsignalT = [];
filteredsignalfreqrespT = [];
%will contain the dBA computations
dBA1 = zeros(length(windowIntervals),1);
dBA2 = zeros(length(windowIntervals),1);
dBA3 = zeros(length(windowIntervals),1);
for i = [1:length(windowIntervals)]
  %take a part of the signal
    signalP = signal(windowIntervals(i)-1+[1:WindowSize]);

      %filter signal part by filter function
      filteredsignalP = filter(Afilter, signalP);

      %filter signal part by frequency response 
      NFFTP = length(signalP);
      fftsignalP = fft(signalP);
      fftfilteredsignalP = fftsignalP .* filter_gaintable;
      filteredsignalfreqrespP = ifft(fftfilteredsignalP, NFFTP);

      %built the full signal
      signalT = cat(1, signalT, signalP);
      filteredsignalT = cat(1, filteredsignalT, filteredsignalP);
      filteredsignalfreqrespT = cat(1, filteredsignalfreqrespT, filteredsignalfreqrespP);

      %calcuatate rms of the signal part
      rmssignalP = sqrt(mean(signalP.^2));
      rmsfilteredsignalP = sqrt(mean(filteredsignalP.^2));
      rmsfilteredsignalfreqrespP = sqrt(mean(filteredsignalfreqrespP.^2));

      % Estimate dBA value using Parseval's relation.
      dBA1(i) = 20*log10(rmssignalP);
      dBA2(i) = 20*log10(rmsfilteredsignalP);
      dBA3(i) = 20*log10(rmsfilteredsignalfreqrespP);
  end

% %plot full signal, and filtered signal
% figure(2)
% plot([1:length(signalT)], signalT, '-', [1:length(filteredsignalT)], filteredsignalT, '-', [1:length(filteredsignalfreqrespT)], filteredsignalfreqrespT);
% title(['compare signal (1 window), to filtered signal']);
% legend('signal', 'filtered (using filter function)', 'filtered (using response table)');
% grid on

%plot out results
figure(3)
plot([1:length(dBA1)],dBA1,'-',[1:length(dBA2)],dBA2,'-',[1:length(dBA3)],dBA3);
legend('dBA signal','dBA filtered (using filter function)', 'dBA filtered (using weight table)',3);
grid on

%save gaintable to xls
C(:,1) = filter_gaintable(1:length(filter_gaintable)/2); %retain frequencies below Nyquist rate
M = xlswrite('aweightingtable.xls', C);

[Daniel Shub]

Okay, and what happens when you run the code? What are you expecting to happen? What would you like to happen? Ideally you should strip the code down to be as minimal as possible. Right now it looks like it should plot 3 figures. If those figures are not related to the problem, don't post that part of the code. You right to an xls file, presumably this isn't the problem. The simpiler the code, the more easily we can help you.