Efficient Symmetric Algorithms for the Modified Covariance Method for Autoregressive Spectral Analysis

Two new algorithms are developed for the efficient computation of the parameters of an autoregressive (AR) pro- cess. The proposed algorithms are based on the simultaneous least squares (LS) minimization of the forward and backward error powers. This method is known as the forward-backward linear prediction (FBLP) or the modified covariance method and is appropriate for AR spectral analysis. One of the derived al- gorithms is for batch processing, and the other is for sequential processing. The developed algorithms are called symmetric be- cause they take advantage of the computationally attractive property of symmetry, which is imposed onto the FBLP prob- lem. Due to this fact, both schemes evolve around symmetric internal variables and offer substantial computational savings over previously derived algorithms.