A new adaptive eigendecomposition algorithm based on a first-order perturbation criterion

A new adaptive eigendecomposition algorithm that can be used for on-line high-resolution spectral/spatial analysis is presented. The formulation of the algorithm is based on the interpretation of the correction term in the recursive update of the data covariance matrix estimate as a perturbation term, with the forgetting factor playing the role of a perturbation parameter. Following this interpretation, a first-order perturbation analysis is made to obtain a new recursion expressing the eigenstructure estimate of the true data covariance matrix at time k, in terms of the eigenstructure estimate at time k-1. The resulting algorithm can be realized by means of M linear combiners with nonlinear weight-vector adaption equations, where M is the signal-subspace dimensionality. Comparative simulation results for narrowband array data indicate very good performance of the proposed algorithm