Finding Valley of Waveform (local minimum) without displacing wave

I am trying to find the peak and valley of a waveform. My data is attached in the matrix. It is a waveform repeated four times. I have written code based off of another thread. I am using findpeaks function, which worked well. I then used taking the max of the matrix and subtracting it to inverse the waveforms. However, it is displacing the waveform, which is a problem because I am trying to get the ratio of the peak to valley, so the displacing of the waveform is giving me inaccurate ratio. I will attach my code. Any ideas on how I might go about finding the valleys without displacing the waveform?

 if true
datain=example;
fs=4e4;
[pks,locs,w,p]=findpeaks(datain,fs,'NPeaks',4);
inverted=max(datain)-datain;
pksmean=mean(pks);
[valleys,locsvalley,wvalley,pvalley]=findpeaks(inverted,fs,'NPeaks',4,'WidthReference','halfheight');
valleysmean=mean(valleys);
peakvalleyratio=pksmean/valleysmean;
halfwidthaveragevalleys=mean(wvalley);

end

Answer 1

[Star Strider]

It’s easiest to use the subscripts rather than the times to get the valleys. Also, there appears to be a bug in findpeaks such that supplying the sampling frequency does not produce the correct times for the peak locations. An easy workaround is to provide a time vector instead and use that.

See if this does what you want:

D = load('Krispy Scripts example.mat');
datain=D.datain;
fs=4e4;
t = linspace(0, 1, length(datain))'/fs;                         % Time Vector
[pks,locs,w,p]=findpeaks(datain,t,'NPeaks',4);
inverted=max(datain)-datain;
pksmean=mean(pks);
[valleys,locsvalley,wvalley,pvalley]=findpeaks(inverted,'NPeaks',4,'WidthReference','halfheight');
valleysmean=mean(valleys);
peakvalleyratio=pksmean/valleysmean;
halfwidthaveragevalleys=mean(wvalley);

valleys_actual = datain(locsvalley);
valleys_time = t(locsvalley);

figure(1)
plot(t, datain)
hold on
plot(locs, pks, '^r', 'MarkerFaceColor','r')
plot(valleys_time, valleys_actual, 'vr', 'MarkerFaceColor','r')
hold off
grid

The Plot —

[Krispy Scripts]

Perfectly, thanks so much.

[Star Strider]

My pleasure.

[Krispy Scripts]

So is this time scale of 10e-3 not correct when the findpeaks function is run to generate its own plot?

[Star Strider]

When findpeaks generates its own plot (I don’t use that option so I don’t know), the times may display correctly.

I know it doesn’t return the correct time values as outputs with only the sampling frequency argument. It does return the correct times with a time vector, as my plot demonstrates.

----------

While I’m thinking about it, if you just invert (negate) the signal (rather than also offsetting it) and then use findpeaks, to get the original values of the valleys, just negate them:

... CODE ...
inverted = -datain;
[valleys,locsvalley,wvalley,pvalley]=findpeaks(inverted,t,'NPeaks',4,'WidthReference','halfheight');
... CODE ...
valleys_actual = -valleys;
valleys_time = locsvalley;
... CODE ...

You might have to tweak your findpeaks arguments with that approach to produce the results you want.

[Krispy Scripts]

My apologies, the time scale should read:

if true
   t=linspace(0,128, length(datain))'/fs;
end

for the data set. With that and fs=40,000 it puts it correctly into the right time domain. Thanks again for your help.

[Star Strider]

As always, my pleasure.

[Greg Dionne]

I think your observed issue with FINDPEAKS can be traced to how you set the input:

t = linspace(0, 1, length(datain))'/fs;

This should read instead:

 t = (0:length(datain)-1)'/fs;