gpuArray/filter doesn't support multi-channel lag terms?

It seems that the gpuArray implementation of the filter function doesn't support lag terms with multiple input channels.

For example, the cpu implementation works just fine:

B = 0.5;
A = [1, -0.5];
X = (randn(100,4));
Zi = X(1,:);
Y = filter(B, A , X , Zi);

However, Once I try the gpu implementation,

B = 0.5;
A = [1, -0.5];
X = gpuArray(randn(100,4));
Zi = X(1,:);
Y = filter(B, A , X , Zi);

I get the following error:

Error using gpuArray/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.

Alternatively, the gpu approach does work with multiple channels if I insert [] for the lag term.

Any ideas?

Answer 1

[Joss Knight]

This is a bug that was fixed in R2016b, so you must have an earlier version of MATLAB. If you can't upgrade, you can avoid it by specifying more than 2 filter taps, or using the same initial condition for all inputs.

I believe you can get an identical answer by adding some additional zeros:

Y = filter(B, [A, 0] , X , [Zi; zeros(1,size(Zi,2))]);

This should avoid the problem.