Hi Bob, In PSD estimates, you can take this to be 1.
The use of the logarithm in PSD calculations is different than the use of the logarithm in measurements like sound pressure level where the reference level is critical.
The reason you typically take the log in PSD estimation is for variance stabilization. The variance in a PSD estimate like the periodogram is frequency-dependent. Taking the log() removes that frequency dependence and makes the PSD estimate much easier to interpret.
Have you looked at aryule() in the Signal Processing Toolbox?
Basically the Yule-Walker spectral estimators:
- use aryule() to obtain the AR coefficients
- return the error variance along with the AR coefficients from aryule
- use freqz() to estimate the frequency response
- use e*abs(h).^2 as the basis for PSD estimate where e is the prediction error from aryule() and h is the frequency response from freqz()
- scale appropriately with the sampling frequency (if in Hz) or (2*pi) if in radians/sample.
Best Answer