Combining repetitive curves into one average curve

So I have this curve that I obtained/created from mulitple discrete time signals. I was wondering if it was possible to combine these curves into one average curve? I need to do some calculations based on amplitude, peak width, etc. and it will be easier if the curves were averaged. The figure only shows 4 waves but theres approximately 1000 in the full dataset.

[dpb]

Attach a sample dataset as a .mat file if want somebody to try something specific.

Those look clean- and repetitive-enough from the thumbnail to try something as simple as findpeaks on the signal and the negative of the signal (to find minima) and then just iterate through the sections.

Undoubtedly they'll be off by a few points in length, so to combine you'll undoubtedly want/need to do an interpolation to one fixed (average?) number of points.

[Dhakshenan Pushparajan]

I have used findpeaks to find the max and min already. Thank you for your suggestion however!

As this is just sample data, it was relatively easy to do the calculations I need (time to peak, time bw systolic arch and diastolic arch, width at different %of the amplitude)

I'm more concerned about when I repeat the process with live data. The data would be more errenous even after filtering for noise. So averaging the curves would be the best thing to do then to limit any abnormalities. Do you have any suggestions on what I should do?

Thanks in advance

[dpb]

Same as I outlined above -- once you find the breakpoints, iterate through those pairs and then interpolate them over a common range to average.

[Dhakshenan Pushparajan]

Thank you! I will try that ASAP

Answer 1

[Star Strider]

One option is to do something similar to the approach in How do I find and plot the average of action potentials from a trace?

[Dhakshenan Pushparajan]

Thank you so much! Would that approach be possible for two iterations of the waveform as well?

[Star Strider]

Probably. You will need to try it to see if it does. For something like an aortic pressure contour (or similarly a fingertip photoplythysmography signal), it would be straightforward for a regular heart rate. For a variable heart rate, it could be more of a challenge.

[Dhakshenan Pushparajan]

Hey sorry to bother you. I have been trying to figure it out for the past couple of days now and I can't seem to figure it out. I've attached the sample data I'm using. Could it possible that because it is a 1D sample, I'm running into issues??

Thanks in advance

[Star Strider]

Try this:

y = load('clean sample.txt');
t = linspace(0, numel(y), numel(y))/1000;                           % Assume 1kHz Sampling Frequency (Supply Correct Time Vector)
Ts = mean(diff(t));
Fs = 1/Ts;

FTy = fft(y-mean(y))/numel(y);                                      % Calculate & Plot Fourier Transform
Fv = linspace(0, 1, numel(t)/2+1)*Fs/2;
Iv = 1:numel(Fv);

figure
plot(Fv, abs(FTy(Iv))*2)
grid
xlim([0 30])
xlabel('Frequency')
ylabel('Amplitude')

yf = highpass(y, 0.001*Fs, Fs);                                     % Eliminate Baseline Variation

[pks,plocs] = findpeaks(yf, 'MinPeakProminence',0.5);               % Peaks & Location Indices
[vls,vlocs] = findpeaks(-yf, 'MinPeakProminence',0.5);              % Valleys & Location Indices

figure
plot(t, yf)                                                         % Detail Of Original Waveform
hold on
plot(t(plocs),pks,'^r') 
hold off
xlim([0 2.5])
xlabel('Time (s)')
ylabel('Amplitude (Unit)')

Vidx = mean(diff(vlocs));
for k = 1:numel(pks)
    if (plocs(k)+320) <= numel(t)
        idxrng = max(1,plocs(k)-55):min(numel(t),plocs(k)+320);
        ensb(k,1:numel(idxrng)) = yf(idxrng);
    end
end

figure
plot((1:numel(idxrng)), ensb)
grid
xlabel('Time (s)')
ylabel('Amplitude (Unit)')

ensb_mean = mean(ensb);
ensb_std = std(ensb);
tci = tinv(0.975,size(ensb,1)-1);                                   % 95% Critical t-Statistic
err = tci*ensb_std/sqrt(size(ensb,1));                              % 95% Confidence Intervals

figure
plot(t(1:numel(ensb_mean)), ensb_mean, '-b')
hold on
plot(t(1:numel(ensb_mean)), ensb_mean+err, '--r')
plot(t(1:numel(ensb_mean)), ensb_mean-err, '--r')
hold off
grid
xlabel('Time (s)')
ylabel('Amplitude (Unit)')
legend('Ensemble Average', '±95% Confidence Intervals')

producing:

.

[Dhakshenan Pushparajan]

Thank you so much!! This is perfect

If you don't mind me asking, how did you get so good at MATLAB. I'm just starting out and want to improve

[Star Strider]

As always, my pleasure!

I have been using MATLAB since the early 1990s, and have been involved in Answers for about the last 7 or 8 years. In addition to writing code for my own applications, working with Answers problems has been a benefit to me, because it allowed me to experiment and learn.