Averaging analysis of partial-update adaptive filters

A simple and attractive way to reduce the hardware complexity and power consumption of an adaptive filter is to implement it using a partial-update technique whereby only a subset of the adaptive filter coefficients are updated at each iteration. For certain cyclostationary or periodic input signals partial-update techniques become susceptible to divergence problems unlike their full-update counterparts. In this paper we present an averaging analysis for the class of partial-update stochastic gradient algorithms and characterize the conditions that lead to unstable behaviour. The conclusions of the paper are supported by extensive simulation examples.