Least-squares order statistic filters for signal restoration in dependent noise
L. Naaman, Alan Conrad Bovik · International Conference on Acoustics, Speech, and Signal Processing · 2002
C.G. Boncelet's algorithm (SIAM J. Sci. Stat. Comput., vol.8, p.868-76, Sept. 1987) is used to explore the OS filter design/analysis problem. In particular, the optimal filter for restoring nonrandom signals immersed in Markov noise, using the mean square error as an optimality criterion, is studied. The noise processes are modeled either as causal first-order autoregressive Gaussian or as first-order moving-average Gaussian. Various structural signal constraints are improved on the solution by stating them as local unbiasedness constraints.>