1
$\begingroup$

Please help understanding the DSP usage of Autocorrelation & Cross-correlation

It seems this is strongly linked to calculating phase offsets, frequency offset for carrier recovery, symbol timing errors, recovery algorithms, what else?

What can I get from Cross-correlation? Started list below, please help

  • Biased version can give difference in constant phase offset between two signals, useful to rotate one set of symbols back into place.
  • Biased version can give The difference in power between the two which is useful if I want to get them to the same amplitude.

**What can I get from Auto-correlation? Started list below, please help **

  • I think that the magnitude or absolute of unbiased and unnormalised with no frequency offset or error can give me the bandwidth or Symbol rate of the signal on the Y axis where there is the maximum spike.
  • The frequency offset can be obtained from the imaginary part of the autocorrelation, how this is implemented to correct the offset in carrier recovery is still alluding me.
$\endgroup$
3
  • 1
    $\begingroup$ For the autocorrelation, this post gives some answers. $\endgroup$
    – Peter K.
    May 30 at 16:24
  • 1
    $\begingroup$ As you learn the details, keep the big picture in mind: correlation is a measure of the similarity between two signals. $\endgroup$
    – MBaz
    May 30 at 16:30
  • 1
    $\begingroup$ I'd recommend moving your lists to answers, and making each answer a community wiki. $\endgroup$
    – Peter K.
    May 31 at 14:58

2 Answers 2

3
$\begingroup$

In audio, I have used autocorrelation, or something like it, to accurately determine the period and fundamental frequency of a quasi-periodic audio signal; a musical note. Pitch detection

I have used cross-correlation to measure the difference, in time, that a single acoustic signal impinges upon two different microphones. From that inter-aural distance, we can compute an angular position of that single sound source is relative to the line connecting the two microphones. Localization

$\endgroup$
2
$\begingroup$

In addition to the uses robert bristow-johnson (very well) provided, one can use cross-correlation to detect a (known) signal in noise. For more information on this use you can search for "matched-filter" approaches in Detection Theory. A personal preference of a related textbook is "Fundamentals of Statistical Signal Processing: Detection Theory" by Steven M. Kay, but of course you can search in your preferred sources.

$\endgroup$
2
  • 1
    $\begingroup$ Oh yeah. Matched filters. Also for binary or M-ary communication (which is really your first example). You have a constellation of signals all associated with a message containing some bits (the simplest is, of course, binary which is a constellation of two given signals, one for a 0 and another for 1. But you could have a constellation of 16 signals in a 4 $\times$ 4 grid or something like that. I think we call those "symbols" or similar. Given a noisy received signal you cross-correlate each symbol in the constellation and see which one the received signal resonates best with. $\endgroup$ Jun 26 at 18:09
  • $\begingroup$ Exactly the examples I had in mind, thanks for the extra info :). $\endgroup$
    – ZaellixA
    Jun 27 at 9:04

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.