Count Peaks in a Graph

Hello all,

I'm writing a code in matlab which will count the peaks, by (1) determining the intersection with a threshold and (2) a positive slope of the signal z(n)-z(n-1)>0

but the code isn't working. I'm stuck at counting the intersection points beteween the signal and the threshold.

P.S. I'm trying to avoid MATLAB's Specific functions, and to use basic code. because I'll convert this code to MathScript later on.

Thank you

Here's my code:

clc;
clear all;

rise=0;

%specs
f=1;
Amp1=1;
Amp2=0.5;
ts=1/8000;
T=6;
t=0:ts:T;

%signals
y=Amp1*sin(2*pi*f*t);
x=-Amp2*sin(2*pi*f*t);

%rectification and summation
y(y<0)=0;
x(x<0)=0;
z=x+y;

%thresh
thL=0.1*max(z);

for n=1:(length(z)-1)
    if (z == thL) % crossing thresh
        rise = rise + 1;
    end
end
disp(rise);


plot(z, 'b')
hold
line(xlim, [thL,thL], 'Color', 'r')
legend('Signal', 'Threshold');

Answer 1

[Cris LaPierre]

Is your signal always this clean? if so, this would work

for n=3:length(z)
    if z(n-1) - z(n-2) > 0 && z(n)-z(n-1) < 0 && z(n-1) >= thL
        rise = rise + 1;
    end
end

[Majd AG]

Thank you for your answer,

Well, this signal is just a simulation for the envelope of inhale and exhale sounds, so in reality it's different (check the attached picture).

But I didn't understand the algorithm you used, can you please explain it.

[Cris LaPierre]

It applies the definition of a peak.

• positive slope to the left
• negative slop to the right
• peak value is above the threshold

Note the index of my for loop - it starts at n=3.

Breath sounds are going to be super noisy (in a signal processing sense, not audibly), so I highly doubt this will work for you.

[Majd AG]

Makes sense.

I'll try to filter the signal as much as I can to make this code work.

Do you have any other suggestion for detection, given that I can't use MATLAB functions.

[Cris LaPierre]

You will definitely want to use a wider range than 3 points. It might also be a good idea to look at minimum width of the peak. If you want some ideas for things to consider, read the documentation on our findpeaks function.

Another idea might be to use the Find Local Extrema task in a live script. This will allow you to interactively explore peak detection of your signal. Once you have something reliable, you can conver the task to code. The code will be using MATLAB functions, but it might still provide you with insight on what things to consider to in your algorithm.

[Majd AG]

OK, sorry to bother you again

if I just want to plot the intersection point with the threshold and find it's index, what should i do?

the condition Z(n)==thL doesn't work.

[Cris LaPierre]

The chances of having a point in the blue signal exactly equal the value of the green line is pretty small. If you want the value where the two exactly equal each other, you are likely going to have to interpolate.

If you just want the point that is closest, you're going to have to invest some time coming up with a logic that is able to identfy that point. One thought I have - subtract the threshold value from the blue signal and take the absolute value. Then step through the absolute difference looking for the min value. Once the difference starts increasing again, you've moved past one crossing point and can start looking for the next one.