A simple method for power spectral estimation using subband decomposition

A method for estimating the power spectral density (PSD) of a given input sequence is proposed. The method is derived as a simple extension of the subband decomposition fast Fourier transform algorithm. A simple tree-structured subband decomposition of the input sequence is applied using only additions and subtractions. A minor addition to the algorithm allows at each decomposition stage the computation of an estimate of the autocorrelation coefficients of the respective input sequence. The overall PSD estimate is then obtained by properly combining the outputs of the separate stages. The method is best suited for estimating the frequency band containing most of the signal power and can be used by itself as an estimate and/or as an integral part of data-driven algorithms to compute a limited number of transform points, to initialize transform domain adaptive algorithms.<<ETX>>