Derivatives and smoothing

Share icon

Hi everybody,

I am perforning derivative analysis to reveal overlapping bands.....Genuinely, i am fitting the second or fourth derivative to get the number and positions of my overlapping bands and then i return to the original spectrum where i perform constraint fitting.....so i keep my positions fixed (i know them from the fitting the derivative spectra) and then i let the software to vary the other parameters (height and width) untill it gives the best fitting result. Does this approach sound reasonable? However, the higher the derivative, the more pronounced the noise.....The S/N ratio is sufficient to do this but sometimes some smoothing is necessary. I try to apply smoothing (when needed) step by step and i always compare with the original spectrum to make sure that i do not distort any features. Any suggestions? As i said sometimes it is just impossible to further improve the S/N ratio experimentally no matter how many accumulations i obtain. Thank you

Regards
Wow. Fourth derivative, eh? That's a lot of derivatives. I think Igor's automatic peak detector used in the Multipeak Fit package makes do with just the third derivative. But it also has some very complicated heuristic to decide just how much noise there is, and therefor how much to smooth. It often gets it right, but often over-smooths a bit. To see it, load Analysis->Packages->Multipeak Fitting->MultipeakFit 2.

We do the fitting to the raw data, so some error in the location, width and height of peaks is tolerated; the fit polishes the result to the optimum (or it fails :).

John Weeks
WaveMetrics, Inc.
support@wavemetrics.com

March 4, 2015 at 04:47 pm - Permalink