filtfilt and filtfilthd give different results and behave differently under parallel for loop

These are actually two related questions.

First: I get two different results from filtfilt and filtfilthd. The latter gives the correct result, but the former gives an unstable result.

Second: when I try to use filtfilthd inside a parfor loop, I get an error message. Here is my code snippet and comments tell you what happens:

======

 clear all, close all, clc

% design bandpass filter 80 - 600 Hz for a signal of Fs=1500,
% also get the matrix and scale values to use in filtfilt

Fs=1500;
h1  = fdesign.bandpass(75, 80, 600, 610, 60, 1, 80, Fs);
Hd = design(h1, 'cheby2', 'MatchExactly',  'stopband');
sos=Hd.sosMatrix ;
scale=Hd.ScaleValues;

%input signal is random noise of 200 seconds , 
sig=rand(Fs*200,1);

% first use filtfilthd , this produces correct output, as seen in
% figure(1), which is bound between -1 and 1
output1=filtfilthd(Hd, sig);
figure(1), plot(output1), title('output1')

% First problem: I use the matrix and scales of the same filter in filtfilt. THis
% produces an unstable output, which shoots up to 10^32 . 

output2=filtfilt(sos, scale, sig);
figure(2), plot(output2,'r'), title('output2')

% open 4 labs, though this behavior happens with 2 or 12 labs which I tested.
desired_pool=4 ;
if ~matlabpool('size') %if no lab is open,
    matlabpool('OPEN', desired_pool) % open the desired # of labs
    sprintf('No Open Labs, Opened  %d pools. \n', desired_pool)
end

% Use filtfilt method (that gave unstable output) inside parfor. Parfor
% works , but not surprisignly produces the same unstable results again. 
%  which -all filtfilt shows the correct version of filtfilt is used i.e.:
% Lab 1:
%   /Applications/MATLAB_R2011b.app/toolbox/signal/signal/filtfilt.m

spmd
    which -all filtfilt,
    paroutput2=filtfilt(sos, scale, sig);
end
figure, plot(paroutput2{1}), title('parout2 (1)')

% Try to put the first version (filtfilthd) inside the parfor loop,
spmd
    which -all filtfilthd,
    paroutput1=filtfilthd(Hd, sig);
end

% I get this error message:
% Error using spmd_feval (line 8)
% Error detected on lab(s) 4
%
% Error in ParForError (line 31)
%     which -all filtfilthd,
%
% Caused by:
%
%     Coefficients must be nonempty.
%
%     Error stack:
%     schema>checkNonemptyVector at 54
%     dtfwnum_setnumerator.m at 11
%     setnumerator.m at 35
%     df2.m at 44
%     filtfilthd.m at 65

=============

As shown in the error message, it seems the coefficient are empty when filtfilthd is used inside parallel for loop. I made sure there are no other versions of filtfilt or filtfilthd on my machine (e.g Octave ones which behave erratic). My version is: MATLAB Version 7.13.0.564 (R2011b) Operating System: Mac OS X Version: 10.7.3 Build: 11D50b Java VM Version: Java 1.6.0_29-b11-402-11D50b with Apple Inc. Java HotSpot™ 64-Bit Server VM mixed mode

Answer 1

[Wayne King]

filtfilthd is not a MathWorks' function. Beginning in R2011a, the MathWorks' function filtfilt accepts filters in biquad filter object form. Since I see you are using R2011b why not just use filtfilt() with filter objects in SOS form?

[ArashMat Fazl]

Thanks for the reply.

I have used filtfilt with Hd.SOSmatrix and Hd.ScaleValues as shown above. Is that what you mean? Because filtfilt doesn't accept the filter object itself.

>>output1=filtfilt(Hd, sig);

Error using filtfilt (line 48)

Not enough input arguments.

Weird thing is that filtfilthd produces stable output, whereas filtfilt doesn't . Does filtfilt in the code above produce stable result on your machine?

Thanks,

Arash