#313 Signal to noise ratio of a peak in the power spectrum

Enhancement
open
David Benn
Algorithms (37)
5
2013-03-16
2012-02-06
David Benn
No

Kevin Alton suggested that it would be useful to select a peak in a periodogram and determine signal to noise ratio.

Here's an exchange between us on the VStar team forum:

KA: Formal statistical significance of amplitude generated following DFT. It is critical to know what is true signal and what is noise following successive pre-whitening to reveal what can often be multiple periods for pulsating variables.

DB: The File -> Info menu item opens a dialog box that contains a variety of information including ANOVA, e.g.

Confidence Interval: Mean error bars denote 95% Confidence Interval (twice Standard Error)
anova: F-value: 1.705135 on 167 and 3693 degrees of freedom, p-value: < 0.000001
Is that what you are after?

Now, the only problem is that this ANOVA value currently corresponds to the series that is the current source of the mean data. ANOVA is recalculated when the raw mode's mean series is changed (due to a change of source series or bin size). See the Plot Control dialog in the View menu for how to specify mean series.

KA: There is a feature in Period04 which allows the user to select a peak in the power spectrum and determine the s/n ratio. This value can be used as a guide to establish whether a particular frequency is statistically meaningful. Therefore this capability would probably be best built into the Period Analysis window which pops up after each successive DCDFT has been completed.

Discussion