1
$\begingroup$

I’m trying to analyze a repeated data stream from a sensor. The data probably has events at the 100us scale, it occurs at a fairly random interval, and the signals are extremely similar. The problem is that I don’t have any sensors that have sub-millisecond refresh rates. The best I can do with what I have is about 1ms sensor reads.

So . . .

The idea is that if I sample a large number of event the chances are that the sampling ticks are falling on different sections of the signal. If this is true then a series of 10-20 events should give me the data that I need to reconstruct the signal. The signals repeat in a fairly random fashion and are very similar in each dataset so the chances are that each sampling run is gathering “pieces” of the total dataset.

Is there a name for this sort of procedure so that I can start to research how to do it? Are there algorithms or toolsets in the open source world to accomplish this task?

Right now I’m literally graphing the data on graph paper and overlaying the image. This is less than ideal . . .

Improve this question
$\endgroup$
3
  • $\begingroup$ Do you know the time at which each event starts, and the time at which each sensor reading is taken? I suspect that a sensor reading at time T is actually an average of the signal over some window of times, say $[T-\delta,T]$, which would mess up your idea. Do you know that the shape of the signal/waveform is identical for all events? What is the duration of each event, and how many events occur per second? $\endgroup$
    – D.W. ♦
    Commented Apr 14, 2021 at 7:03
  • $\begingroup$ The entire event is about 100ms in duration. Each repeat is nearly identical. The actual time of the the events occurrence is somewhat random so I can't predict exactly when they will occur - but about every 4-5 seconds. $\endgroup$
    – Richard Wayne Barlow
    Commented Apr 14, 2021 at 15:19
  • $\begingroup$ That super-resolution page is interesting. Something like that might be useful. I wish they referenced some analytics tools that I could dig into there! $\endgroup$
    – Richard Wayne Barlow
    Commented Apr 14, 2021 at 15:24

2 Answers 2

Reset to default
1
$\begingroup$

If the events occur at random, unknown times, all you can get are random samples of the signal, and by collecting enough of them, the statistical distribution of the signal values.

If the signal is known to be monotonic, say increasing, with a neglectable falloff time, you can retrieve it by inverting the cumulative density function. Otherwise, the reconstruction does not seem possible.

Improve this answer
$\endgroup$
0
$\begingroup$

It's not the same, but it reminds me of super-resolution.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.