Block Time and Frequency Domain Modified Covariance Algorithms for Spectral Analysis

Block modified covariance algorithms are proposed for autoregressive (AR) parametric spectral estimation. First, we develop the block modified covariance algorithm (BMCA) which can be implemented either in the time or in the frequency domain—with the latter being more efficient in highorder cases. A block algorithm is also developed for the energy weighted combined forward and backward prediction. This algorithm is called energy weighted BMCA (EWBMCA) and its performance is analogous to that of the weighted covariance method proposed by Nikias and Scott. Time-varying convergence factors, designed to minimize the error energy from one iteration to the next, are given for both algorithms. In addition, three updating schemes are presented, namely block-by-block, sample-by-sample, and sample-by-sample with time-scale separation. The performance of the proposed algorithms is examined with stationary and nonstationary narrowband and broadband processes, and also with sinusoids in noise. Lastly, we discuss the computational complexity of the proposed algorithms and we give performance comparisons to existing modified covariance algorithms.