power spectral density plots

Hi all,

How do we interpret a power spectral density plot. I have used modified covariance and burg methods and plotted using MATLAB. What does the peaks at some frequencies suggest? and what is the difference between positive and negative power in the graph? how do we know that which frequencies variations are strong and at which frequencies variations are weak.. can anyone help me please..

Thanks

1 Answer

The plot is not of power. The units of the $y$ axis are dB/Hz. The dB part means that a decibel scale is used, which means it is a scale relative to a reference value.

All that negative values on a decibel scale mean is that the actual power is less than the reference power (because the ratio is less than one).

The higher the decibel value, the more powerful the signal.

Your signal has a major peak at about 4Hz, and lesser peaks at 28Hz and 39Hz. There might be something happening at about 12Hz, but it's hard to say without more information (or data).

• Thanks for your response Peter. This is a PSD estimate of half second EEG signal. I am trying to see If PSD can be used as a feature for identifying individuals. So according to this plot EEG signal has maximum strength/power at 4Hz freq and then 12 Hz and lesser strength around 28 Hz and 40 Hz. So power is decreasing with increasing frequency right? Does it also mean that signal variations are strong at frequencies where we see the peaks i.e. 4 Hz, 28 Hz and 40 Hz? Thanks for your help. – Hemang Shrivastava Sep 8 '15 at 2:31
• Yes, that's correct: less power at higher frequencies (except for the bumps mentioned). "Strong" might be too strong a word. It means there are periodicities / frequencies / correlations that are higher at those frequencies than the others. – Peter K. Sep 8 '15 at 11:29