0

I have a data file contains timestamps - data signals.

I was trying to detect the peak of the signals as well as the start and end of the peaks.

Referred to this post, I could detect the peak with no issue.

The questions is how to robustly detect the start / end of the peak signals?

How do I detect the start of continuous rising, or the end of continuous decreasing trend of the signals?

enter image description here

I am very new to digital signal processing. Any advise are welcome.

Xuhang
  • 445
  • 5
  • 20
  • 1
    Well, the main feature of your "false" start/end is that the entire thing seems to be above the range of your baseline measurements. So, surely you could try a simplistic approach like considering all points that lie between the regions where signal went above some number of standard deviations from "normal". – paddy Aug 31 '20 at 00:09
  • you should probably look into existing libraries – Cyril Jouve Aug 31 '20 at 00:11
  • Based on the [linked post](https://stackoverflow.com/questions/22583391/peak-signal-detection-in-realtime-timeseries-data) I guess the start/end points you want are the last-before-peaks and first-after-peaks points in the "noise range" between the two green lines in the *Result* graph in the accepted answer. – dxiv Aug 31 '20 at 00:24
  • 1
    A good first step to detecting the start/end of the peaks is to **define** the start/end of the peaks. Why do you consider the "false start/end" to be false? What objective criteria can you give for why the "false start" is false and the "true start" is the true? (Once you have an initial definition, draw some more pictures as a gut-check to make sure you have defined what you intended to define.) – JaMiT Aug 31 '20 at 01:15
  • @JaMiT . Thanks, it's good point to find the definition of start/end of the peaks. I followed the [matlab findpeaks] (https://www.mathworks.com/help/signal/ref/findpeaks.html) link for the definition. And the solution came along with the definition. – Xuhang Sep 11 '20 at 20:19

1 Answers1

0

I followed this matlab findpeaks link to define the start and end of peaks.

It describes the definition of prominence (thus the width) of peaks.

enter image description here

Xuhang
  • 445
  • 5
  • 20