Re: How do I find peaks in noisy data?

[TedPride]

Not that hard. If you want an absolute cutoff (500) rather than percentage (6% of max), just change the lines assigning values to $cutoff.

use strict;
use warnings;

my @n = (101, 203, 321, 45, 67, 156, 203, 502, 899, 2003, 5007, 8020, 
+7301, 5030, 3045, 1243, 567, 321, 234, 45, 123, 453);
my $cutoff = .06;

my $max = 0;
for (@n) {
    $max = $_ if $_ > $max;
}
$cutoff = $max * $cutoff;

for (@n) {
    $_ = 0 if $_ < $cutoff;
    print $_ ? '-' x ($_ * 50 / $max) : '', "\n";
}

Returns:

---
-----
------------
-------------------------------
--------------------------------------------------
---------------------------------------------
-------------------------------
------------------
-------
---

EDIT: Hmm, you're right, the background noise could theoretically vary, as well as the height of the peaks. I guess a more complicated algorithm is needed. Perhaps the cutoff level could be set to some percentage of the closest peak, then any islands remaining would be removed if they had a bad width-height aggregate (this would be the complicated part). The idea is to set the cutoff low enough so you don't lose much of the edges of the islands, but also to remove any islands which are just background noise.

Then any parts which have been removed can be analyzed to see if there's a regular pattern. If there is, that pattern is extrapolated and removed from the remaining islands.

Altogether too much work, imho :) I'm sure there is a utility written in C or C++ somewhere that does this already, probably faster and better than anything we can write on the spur of the moment. Unless what you're trying to do is something simple like automatically trimming sound files, I'd personally just get a third-party utility and write a wrapper for it.