in reply to
Finding local maxima/minima in noisy, pediodic data
Thank you all for your comments.
I ended up doing two smoothing passes before I attempt to find any extrema: one is smoothed twice by a running average with a window of 1.1*$period (which I find with a simplified version of the autocorrelation func.), while the second is smoothed with a running median algorithm. The first (almost) completely removes the pulses, leaving only the "carrier wave" in place, while the second preserves the shape of the pulses but removes the spiky noise from them.
This gives me a robust crossover detection: if the running median is higher than the large scale smoothed wave, then I have to look for a maximum, else a minimum.
This algorithm is somewhat complex, however, it is (or seems to be) robust enough.
Ultimately, this algorithm will be the core of a Gtk2 GUI application that lets the user (my boss) visualize the data, find the peaks and mark/move/delete/add/export them at will. However, I'll write this later.
As for Perl not being the best tool for this particular problem, or a particular class of DSP-related problems -- well, I hesitate to say it out loud on this hallowed ground, but yes, quite likely that is true. There maybe other languages, platforms or tools which may be more efficient or easier to use for this problem than Perl; however, Perl is something I'm comfortable with. I happen to like Perl. :)
There is also the factor that I need my application to be portable, specifically, I develop it on linux but it'll have to run on windows. It is my experience that this works with Perl; I wouldn't be sure about other languages (for example, I never wrote anything for Windows in C).
Additionally, Perl is free, which is important to me for both practical and ethical standpoints.