0
$\begingroup$

As a research to answer my question, I've been reading several posts on this stack with a common thread of "comparing signals". The methods recurrently suggested were to use DTW, correlation or MSE etc.

The two signals I'm trying to compare are generated by piezoelectric single-axis accelerometer and they are periodic.

Out of 4 variables playing a role in signal amplitudes, two of the variables are different(A,B resp) and the rest are same(C,D resp).

My aim is to measure quantitatively how different the two signals are on the basis of variable B. At the moment, a qualitative way to go about this, according to my supervisors, is to look which among the two signals have the higher peaks.

I want to get a quantitative measure to compare the two signal's peaks or maybe the whole dataset which consists of 20,000 datapoints for a period of 10 sec. Are peaks/amplitudes the right way to go about this?

$\endgroup$
2
  • $\begingroup$ Please don't deliberately be vague. You're the one asking the question, because you want to acquire expertise that you don't yet have. But: to know what can be omitted from a question to get a useful answer for your use case, you'd need exactly that expertise. You intentionally being vague just tells everyone that they can't know whether they're actually helping you with your problem, or wasting their time, because chances are very high that an answer will get a comment like "thank you, but that doesn't address my unspecific problem". $\endgroup$ – mmmm Apr 17 at 8:39
  • $\begingroup$ @mmmm regarding your concern you can simply flag the question as "opinion based" :) $\endgroup$ – MimSaad Apr 17 at 18:26
0
$\begingroup$

I suggest using a correlation metric as a comparison. Remove any mean in the signals and after subtracting the mean, compare the signals using the Pearson Correlation Coefficient to have a single data point that scales between +1 and -1 with regards to their direct correlation assuming any time offset factors have been removed, and even better as a first evaluation use the cross-correlation function which would provide the above (without normalization) for every offset in time, in case their are delays between the results.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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