filtfilt.m fails for input array x with 3 or more dimensions

Dear all,

filtfilt.m does not work for ndim-arrays anymore, e.g.:

 x = randn(1000,3,2);   % i.e., random signal with 3x2 channels
 [b,a] = butter(2,0.1); % 2nd order Butterworth lowpass filter

In older matlab versions, e.g., R2007b, I could say

 fx = filtfilt(b,a,x);

and got the filtered signal fx with the same size as x. In R2011a the same command

 fx = filtfilt(b,a,x);

returns the following error:

??? Error using ==> filter

Initial conditions must be a vector of length max(length(a),length(b))-1, or an array with the leading dimension of size max(length(a),length(b))-1 and with remaining dimensions matching those of x.

Error in ==> filtfilt>ffMultiChan at 296

 yc3 = filter(b(:,ii),a(:,ii),xt,zo);

Error in ==> filtfilt at 78

 y = ffMultiChan(b,a,x,zi,nfact,L);

My impression is that the dimensionality of the initial and final conditions, Zi and Zf is not correctly passed to filter.m.

Of course, I can circumvent the error by filtering data with maximal two dimensions, e.g., by reshaping before and after filtering or as follows:

 fx = zeros(size(x));   % allocation
 fx(:,:,1) = filtfilt(b,a,x(:,:,1));
 fx(:,:,2) = filtfilt(b,a,x(:,:,2));

However, I just would like that filtfilt.m can handle n-dimensional input x as in earlier versions.

Thanks in advance for any help,

   Helmut

Answer 1

[Jan]

Does it help to specify the dimension to operate on?

You could try to use Fex: FiltFiltM, which is much faster than Matlab's implementation, while it replies identical values. FILTFILT wastes a lot of time with expanding the signal by initial and final phases, while FiltFiltM calculates the edges separately. In addition FILTFILT reverses the expanded signal twice, while FiltFiltM uses a fast C-mex to emulate FILTER with allowing to filter in teh reverse direction. In sonsequence FiltFiltM is 5 (short signal) to 20 (large signal) times faster.

Answer 2

[Nitesh Chowdhury]

hi,

I'm new to using MATLAB and i'm using 2014a version, trying to filter a sensor signal which has a lot of noise. The data is in 1*16060double matrix. I have attempted multiple different variations of the command filtfilt but I keep getting errors. Also what should be the values for 'a' and 'b'?

I have already used transpose, converted into a 2*16060double matrix with time in the new row but nothing works. Here are the errors that I have got.

>

> filtfilt(1,10,ax)
Index exceeds matrix dimensions.

Error in filtfilt>ffMultiChan (line 320)
    [~,zo] = filter(b(:,ii),a(:,ii),xt,zi(:,ii)*xt(1,:)); % outer product

Error in filtfilt (line 106)
    y = ffMultiChan(b,a,x,zi,nfact,L);

>> filtfilt(1000,10000,ax)
Index exceeds matrix dimensions.

Error in filtfilt>ffMultiChan (line 320)
    [~,zo] = filter(b(:,ii),a(:,ii),xt,zi(:,ii)*xt(1,:)); % outer product

Error in filtfilt (line 106)
    y = ffMultiChan(b,a,x,zi,nfact,L);

>> filtfilt(17000,17000,ax)
Index exceeds matrix dimensions.

Error in filtfilt>ffMultiChan (line 320)
    [~,zo] = filter(b(:,ii),a(:,ii),xt,zi(:,ii)*xt(1,:)); % outer product

Error in filtfilt (line 106)
    y = ffMultiChan(b,a,x,zi,nfact,L);

>> filtfilt(17000,17000,ax)
Attempted to access xt(1); index out of bounds because numel(xt)=0.

Error in filtfilt>ffOneChan (line 260)
    [~,zo] = filter(b(:,ii),a(:,ii), xt, zi(:,ii)*xt(1)); % yc1 not needed

Error in filtfilt (line 103)
        y = ffOneChan(b,a,x,zi,nfact,L);

[Jan]

Please post a new question in a new thread and not as an answer of another question. Thanks.

Do you get the same error for this code:

ax = rand(1, 16060);
y  = filtfilt(1,10,ax);

?