A high performance spectral estimation method: The AMW algorithm

In this paper, we propose a new spectral estimation method, the AMW algorithm, to avoid the line-splitting and biased frequency estimates, which usually happen in the Burg´s MEM spectral estimates in some cases of high signal-to-noise ratio and short-data records. The L1-norm of the observed data vector has been adopted as the weight of the prediction error. The reflection coefficient of the lattice prediction filter has been developed, and the AMW algorithm follows with adoption of Levinson recursion. The performance of the proposed method are studied by computer simulation in detail and compared with the other modified MEM algorithms, such as data-adaptive weithted (DAW) algorithm and optimum tapped weighted (OTW) algorithm. The stabitity of the algorithm Is also rejustified rigorously by using Jury´s stabitity test rule in the appendix.