Spectral estimation of short duration signals using time domain extrapolation

Time domain extrapolation is investigated to enhance the resolution of spectral estimators. When extrapolation is used the resolution of spectral peaks is enhanced, provided that the time series can be predicted. Backward and forward linear prediction is used to obtain the extrapolation. The prediction coefficients are estimated using the modified covariance methods or the modified forward-backward linear prediction technique. This allows a fairly robust time domain extrapolation that is less sensitive to model order matching than the standard parametric spectral estimators. Classical spectral analysis (periodogram) is performed on the extrapolated time series, providing improved resolution