Counting the peaks

Projects Programming
1

Hi everyone, I'm back with my timegrapher project using ESP32. For my project I'm using a mechanical watch with 21600 tics per hour. On the first graph we can clearly see 6 tics in a second and in a hour we'll clearly get those 21600 tics

vibration graph
vibration graph1598×618 9.22 KB

My goal is to count those tics, as we can see, those tics are not only one line, there are like 6 "blocks" with different peaks. My question is to know if it is possible to count the peak of those tics or is it possible to have a graph only with peaks, just like the black line on the graph

wanted graph
wanted graph1598×618 50 KB

2

Surely you have the underlying numbers for the graph. What do they show for the various peaks? Once you know the average number, you can test for a number greater than the average. When true, that is the beginning of a peak. When the numbers are back to the average, the peak has ended and you can add one to a counter.

3

Hi Paul, I've tried your solution, but I've a problem with it.
I've set the average to 20 base on the tic with the lowest peak, and in one tic we've for example this [19,12,15,21,0,10,5,22], the counter count 2 times. There are tics which cross the average many times and the counter count it all those times

4

You could get something like the black line in your second figure by keeping a running average of the absolute value of the samples. A running average length on the order of 0.1 second / sample_period is what you've drawn.

5

An Exponentially Weighted Moving Average (EWMA Moving average - Wikipedia ) is simpler to code than a rolling window, and takes less memory:

     const float alpha = 0.2;  // adjust between 0,1.  Higher numbers are more responsive, lower numbers are more smooth.

     float ewma =0;
     ...
     ewma = alpha * newdata + (1-alpha)* ewma;

     ....
6

Or the more efficient formulation

ewma += alpha * (newdata - ewma) ;

Here the newdata would either be the absolute value of the signal or its square.

Another approach is to gate the tick-detection in time, so after detecting a spike you skip 0.1s worth of samples before enabling re-triggering.

1 Like
7

This topic was automatically closed 180 days after the last reply. New replies are no longer allowed.