ECG spectrum wrong?

Hi all,

I have a question about an EKG spectrum.

Someone gave me this ECG spectrum for an exercise.

I'm not sure but the spectrum looks weird.

You clearly see a 50 Hz tone and its harmonics.

The whole thing looks like a modulation.

I would say there was no anti aliasing filter used.

Or does an ECG spectrum with 50 Hz + harmonics look like that?

Thanks for the help.

Best Regards.

Answer 1

[Daniel M]

I'd say it looks fairly normal once you notch filter the line noise. But there is other preprocessing to do, like a low pass filter. You would probably get a lot out of this article <https://www.ncbi.nlm.nih.gov/m/pubmed/20851409/>

[Star Strider]

+1 for the PubMed citation!

Answer 2

[Daniel M]

I'm not sure how you're doing your notch filters, but try this:

Fs = 625; % or whatever your sampling frequency is.
Fn = Fs/2; % Nyquist frequency
numHarmonics = 4; % Let's do 50, 100, 150, 200, 250
lineFreq = 50; % Hz
for fq = ((0:numHarmonics)+1) * lineFreq
    Fl = fq + [-1, 1]; % notch around Fl. Could try [-2, 2] if too tight
    [z,p,k] = butter(6, Fl/Fn, 'stop');
    sos = zp2sos(z,p,k);
    data = filtfilt(sos, 1, data); % assumes data is [time x ... dimensions]
    % overwrites data, and filters sequentially for each notch
end

[Stephan Lallinger]

Thank you for your code.

I tested several preprocessing steps last week.

Now iam using an IIR highpass and FIR lowpass (0.05 Hz - 150 Hz).

For power line noise i tried a comb notch filter.

I will test your code also, thank you for that!

Answer 3

[Stephan Lallinger]

Thank you for your help.

Now i'am using an IIR HPF and a FIR LPF.

Iam not happy with the notch filters. Now i try to implement an adaptive filter for line noise.

Thank you!

Answer 4

[Stephan Lallinger]

Hi,

sorry again i got a question again about preprocessing.

For simplicity, an ECG with a noise at only 50 Hz.

In my picture you can see that the Notch filter distorts the ECG in the QRS complex.

I use a 150 Hz low pass filter, with a lower cut-off frequency I get the error off, but I want to stay at 150 Hz.

Here's the notch I use:

%%noise to filter
f_noise = 50;

%%normalize frequency
f_notch = f_noise/fs;

%%place zeros
N =[exp(j*2*pi*f_notch);exp(-j*2*pi*f_notch)];

%%place poles with same angular as zeros
%%distance to unit circle
P = 0.5*N;
%get coefficients
b = poly(N); a = poly(P);

Is there a technique to reduce this afterswing, except for a lower cutoff frequency of the low-pass filter?

I can imagine it is due to the phase response and the associated group delay.

Best regards.

[Daniel M]

You may want to post this as a new question so more people can see it. But I have a few comments.

For your bandpass filter, try this

data = detrend(data); % remove the mean
[B, A] = butter(4, [0.5/(fs/2) 150/(fs/2)]);
filt_data = filtfilt(B,A,data);

As for notch filters, I either use the code I posted above, or the function ft_preproc_dftfilter() in the Fieldtrip toolbox. (It uses a similar approach to yours, which I'm not familiar with, so your issue could be something small.)

But your issue might be aliasing. What is your sampling frequency? If it is under 300, that could cause an issue with filter at 150 Hz.

[Stephan Lallinger]

It was not that smart to post a new question here. I will open a new one.

I really appreciate your effort.

I tried your notch filter today and it doesn't overshoot, but it does ringing in the signal.

I looked at the impulse response and it oscillates very long.

I sample at 500 Hz, there should be no aliasing effects.

I think I need to find a trade off. I just don't have any experience in signal processing, so it's hard to figure out where to start.

[Daniel M]

Ok good idea to open a new question. I will look out for it, and see if others can give advice too. In the meantime, think you could post your data here? I'd like to tinker with it a bit.

[Stephan Lallinger]

Of couse i can it is excel. I hope this is fine.

fs = 500 Hz;

for reading i use:

ecg_signal = xlsread( 'mySigXx.xlsx');

First signal contains 50 Hz + harmonics.

Second signal is with the ringing effect when notch is used.

EDIT: for second signal you need to add 50 Hz noise.

I do it like this

fs = 500;
Ts = 1/fs;
t=0:1:length(ecg_signal)-1; %time vector depending on ecg signal
t = t/fs; %normalize time vector

ecg_signal = ecg_signal' + 0.5*sin(2*pi*50*t); % add Noise