Subband decomposition using multichannel AR spectral estimation

Subband decomposition has been shown to be a useful tool for spectral estimation, in particular when parametric methods have to be considered. Indeed, the loss of observed samples due to decimation can be compensated by the use of a suitable model, if available. This paper studies a subband multichannel autoregressive spectral estimation (SMASE) method. The proposed method decomposes the observed signal through an appropriate filter bank and processes the decimated signals by means of a multichannel autoregressive (AR) model. This model takes advantage of known correlations between different subband signals. This a priori knowledge allows to improve spectral estimation performance. Simulation results illustrate the interest of the proposed methodology for signals with continuous spectra and for sinusoids.