Pattern recognition in signal??

[Allstar_07]

I have a signal of 1's and 0's sampled at an even time spacing.  Is there any functions within LabVIEW that are designed or could be used to interrogate this signal and find any parts of it that repeat with a time offset.  I don't know what I'm looking for so can't just search the signal, I need something to look at all the samples and check to see whether there is a pattern (either a small sample period repeated later on or a pattern all the way through)  i.e. it needs to be intelligent enough to use different sample lengths and parts of the signal for comparison.

Many Thanks

[rpursley8]

The autocorrelation function should do that.  Peaks will occur where parts of your signal match.  The longer the matching segment is, the higher the peak.

[Allstar_07]

I've tried that on a sample signal but doesn't give the results I would like, are there any functions in any of the toolkits specifically for pattern recognition?

Thanks

[alipio]

Hello
As rpursley8 said, autocorrelation and crosscorrelation of part of your signal whith full one, is, probably, the best way to find parts that are repetitve and unknown " a priori".

It looks souspicious to me that the procedure failed in your signal
Please, take a look at attached jpg. You will need to scale data properly to achieve a good correlation results.
Hope it helps.
Alipio

[johnsold]

I agree with the posts above.

Are you sure that patterns really exist? The human eye/brain tries very hard to see patterns in things, whether any exist or not. If you know they exist, can you use any knowledge about the type of patterns to help in the recognition? If you post a small file with sample data and any informaton about the types of patterns you expect to find, perhaps someone can provide better suggestions.

Lynn

[Allstar_07]

I've attached a 'Sample' set of data.  I'm not sure if there are patterns so cannot really narrow the search down, I just hoped there was going to be either all the way through all parts of.  This sample may just be the first 3000 samples of a pattern of much greater length.  This is what i would like to look into.

I created some data and have a attached it - 'Constructed Data' i copied the first 15 ish samples on the end of the set, with randon values in the middle.  As you can see from the picture in the word document after plotting the auto-correlation it's not easy/impossible? to interpret that this is happening.  There is a big peak but I would like to be able to know the pattern etc etc.  I persume for this though you would need more advanced pattern finding software - something labVIEW doesn't have???

Thanks

[Allstar_07]

Didn't mean to do that!

Can you attach more than 1 file?!

[johnsold]

Here is a little VI I put together which will read your data files or simulate various signals. Your data looks more random than periodic at first glance.

Lynn