Same filters but using different methods give different results

Hi,

I am getting different results when using the [b,a] filter method and [z,p,k] method. My code goes like this:

for loop for cut-off frequecy (starts from 50 Hz and goes upto 5000 Hz with step size of 50 Hz) ---------------------

Read couple of csv files in a loop (1 to 7) -> Filtering out each of them using a Elliptical band/high pass 4th order filter for "x" cut-off frequency -> Calculating amplitude (which means adding the absolute values of this filtered data) of each file -> Then taking average of the amplitude of these 1-6 files -> dividing the amplitude of 7th file by the average amplitude of 1-6 files -> store this value -> loop starts again for cut-off frequency of "x+50" upto 5000 Hz

------------------------ end loop

Now, when I plot the graph of the values using [b,a] method and by using [z,p,k] method, I can see whole different value. Why would that be the case?

fl=0;
fh=9000;
fs=40000;
array_aW=zeros(100,1);

y1=readmatrix('1.csv');
y2=readmatrix('2.csv');
y3=readmatrix('3.csv');
y4=readmatrix('4.csv');
y5=readmatrix('5.csv');
y6=readmatrix('6.csv');
y7=readmatrix('7.csv');

for e=1:100
    fl=fl+50;

% [b,a]=ellip(2, 1, 40, [fl,fh]/(fs/2),'bandpass');
[z,p,k]=ellip(2, 1, 40, [fl,fh]/(fs/2),'bandpass');
[sos,g]=zp2sos(z,p,k);

% y_n1=filter(b,a,y1);
y_n1=filtfilt(sos,g,y1);
y_h1 = normalize(y_n1, 'range', [-1 1]);
amp1=sum(abs(y_h1));

% y_n3=filter(b,a,y2);
y_n2=filtfilt(sos,g,y2);
y_h2 = normalize(y_n2, 'range', [-1 1]);
amp2=sum(abs(y_h2));

y_n3=filtfilt(sos,g,y3);
%  y_n3=filter(b,a,y3);
y_h3 = normalize(y_n3, 'range', [-1 1]);
amp3=sum(abs(y_h3));

y_n4=filtfilt(sos,g,y4);
%  y_n4=filter(b,a,y4);
y_h4 = normalize(y_n4, 'range', [-1 1]);
amp4=sum(abs(y_h4));

y_n5=filtfilt(sos,g,y5);
%  y_n5=filter(b,a,y5);
y_h5 = normalize(y_n5, 'range', [-1 1]);
amp5=sum(abs(y_h5));

y_n6=filtfilt(sos,g,y6);
%  y_n6=filter(b,a,y6);
y_h6 = normalize(y_n6, 'range', [-1 1]);
amp6=sum(abs(y_h6));

%calculations
ampNoiseWeightedAvg=((0.033*amp1)+(0.033*amp2)+(0.033*amp3)+(0.05*amp4)+(0.05*amp5)+(0.8*amp6))/6;

%%%%%%%%%%%%7 file%%%%%%%%%%%%%
y_n7=filtfilt(sos,g,y7);
% y_n7=filter(b,a,y7);
y_h7 = normalize(y_n7, 'range', [-1 1]);
ampW=sum(abs(y_h7));

%%%%%%Figure of Merit%%%%%% 
   FigureOfMerit_ampW= ampW/ampNoiseWeightedAvg;
   array_aW(e,1)=FigureOfMerit_ampW;

end

Thanks,

Gagan

[Jan]

"My code goes like this" - do not paraphrase, what you code does, but post the code. If it contains a bug, the paraphrasation will conceal it.

[Giggs B.]

Got it @Jan. Thanks.

Answer 1

[Star Strider]

‘I am getting different results when using the [b,a] filter method and [z,p,k] method.’

I am not surprised. Filters using transfer function ‘[b,a]’ notation can be unstable. Ues the freqz function to explore the filter characteristics:

fl=0;
fh=9000;
fs=40000;

[b,a]=ellip(2, 1, 40, eps+[fl,fh]/(fs/2),'bandpass');

figure
freqz(b, a, 2^16, fs)

[z,p,k]=ellip(2, 1, 40, eps+[fl,fh]/(fs/2),'bandpass');
[sos,g]=zp2sos(z,p,k);

figure
freqz(sos, 2^16, fs)

The phase characteristics are significantly different.

This is the optimal approach for the best filter implementation:

[z,p,k]=ellip(2, 1, 40, [fl,fh]/(fs/2),'bandpass');

[sos,g]=zp2sos(z,p,k);

Note that ‘fl=0’ designs a lowpass filter. It would likely be more efficient to design the filter as a lowpass filter rather than a bandpass filter.

Also if ‘y1’ through ‘y7’ are equal-length column vectors (or can be made to be equal length column vectors), concatenate them in a matrix and call filtfilt once with your filter and the matrix. The ooutput will be a matrix of the filtered signals.

.

[Giggs B.]

@Star Strider, thank you, I understand now. Yes, all vectors are of same length. I will try your suggested way.

[Star Strider]

As always, my pleasure!