How can I use a filter in a parfor with the more numerically stable z,p,k?

Hello,

I would like to use a filter in a parfor loop. Using [b,a] coefficients there is no problem. Because of a better stability I have to use the [z,p,k] coefficients. It perfectly works in a for loop. However, in a parfor loop the error massage "Not enough input arguments." appears. Does anyone know a workaround?

Here is a little example: Anything works but the last loop!

% Signal
x = sin(0:pi/1000:10*pi);

% Filter properties
N = 1;
fc = 1/(2*pi);
fs = 1000/pi;
Wn = fc/(fs/2);
ftype = 'low';

% Transfer Function design
[b,a] = butter(N,Wn,ftype); 

% Zero-Pole-Gain design
[z,p,k] = butter(N,Wn,ftype);
[sos,g] = zp2sos(z,p,k);
hd = dfilt.df2tsos(sos, g); 

%%Loops

for n=1:2
    y = filter(b,a,x);
end

parfor n=1:2
    y = filter(b,a,x);
end

for n=1:2
    y = filter(hd, x);
end

parfor n=1:2
    y = filter(hd, x);
end

[Jan]

Please post the complete error message.

[Gordon]

This is the complete error message ;)

Answer 1

[Aniruddha Katre]

Modifying the parfor loop as follows seems to resolve the error. The filter object was causing the error to occur for some reason, but creating the filter object on each worker separately fixes it.

parfor n=1:2
    hd = dfilt.df2tsos(sos, g); 
    y(n,:) = filter(hd, x);
end

[Edric Ellis]

There's an underlying problem sending objects of type dfilt.df2tsos into parfor loops. An additional workaround you can use to avoid re-creating the object on each iteration is:

hdc = parallel.pool.Constant(hd);
parfor ...
  ... = filter(hdc.Value, ...);
end

[Gordon]

Both perfectly works! Thank you very much for the very quick reply!