Capon´s time-frequency representation with nonstationary AR autocorrelation

A novel approach for the spectral analysis of nonstationary signals is presented. For this purpose, Capon´s time frequency representation (CTFR) is employed. It is shown that CTFR is an upper bound on the range of nonunique solutions for power estimation of a complex sinusoid contaminated with unknown noise. A new local autocorrelation function using a nonstationary auto-regressive (NAR) model is defined and used in CTFR. This method efficiently models the autocorrelations of NAR processes. Synthetic signals are generated in order to illustrate the superiority of CTFR with the NAR model in comparison to other methods.