Improved LMS estimation via structural detection

摘要

In various adaptive estimation applications, the unknown channel is characterized by an impulse response which consists of extended regions of negligible response or 'inactivity'. We consider the LMS estimation of a channel that may be well approximated by an FIR (filter) model with only a few nonzero tap coefficients within a given delay horizon or tap length n. When the number of nonzero tap coefficients m is small compared to the delay horizon n, the performance of the LMS estimator is greatly enhanced when this specific structure is exploited. We propose a consistent algorithm that performs identification of nonzero taps only.