how to implement a fast filter algorithm

I have the following filter, but it runs very slow. How can I make it faster?

if true
  % code
end
function [yfilt] = HPfilter(y, fs)
N    = 5001;     % Filter order (time delay will be (N-1)/2)
F6dB = 1500;     % 6-dB Frequency
h = fdesign.highpass('n,fc', N, F6dB, fs);
Hd = design(h, 'window');
yfilt1 = filter(Hd,y);
y2 = fliplr(y);
yfilt2 = filter(Hd,y2);
yfilt2 = fliplr(yfilt2);
offset = yfilt2(N) - yfilt1(N);
yfilt1 = yfilt1 + offset;
yfilt = [yfilt2(1:(N-1)) yfilt1(N:end)];

I run the filter twice from both sides to elminiate the effect of the impulse response.

The sampling rate of the data (fs) is 5.12 MHz. The cutoff for the high pass is 1.5 kHz. I would reduce the order number if I could, but can't accept the degradation of performance. Some of the data files have 500,000 points.

Is there any way to remove the low frequency stuff of my data in a faster way?

Thanks for your help in advance.

Answer 1

[Wayne King]

I would try using filtfilt.m instead of filter() followed by your other code to do zerophase filtering.

Replace:

 yfilt1 = filter(Hd,y);

with

 yfilt = filtfilt(Hd.Numerator,1,y);

Or

 yfilt = filtfilt(Hd.sosMatrix,Hd.ScaleValues,y);

and then eliminate the rest.

Also, why do you need to do the filter specification and design inside the function? That is expensive. I would recommend doing that outside the function in which case you just pass filtfilt() the filter design object if you follow my recommendation.

[Kilian]

Thanks for your advice. However, filtfilt doesn't work on a filter object. I get an error "not enough input arguments'.

y = filtfilt(b,a,x) y = filtfilt(SOS,G,x)

How can I obtain b and a or SOS and G from the filter object?

[Wayne King]

@Jerome, Sorry!!! I've edited the post to correct that.

Answer 2

[Jan]

Matlab's filter implementation has a potential for speed improvements. See Fex: FilterM. Unfortunately the version posted there does not exploit the simpler processing for the FIR type filter. It is updated soon.

[Royi Avital]

Is it still faster than the latest MATLAB versions?