Adaptive equalization using multirate filtering techniques

The authors present a computationally efficient blind equalization architecture that is based on multirate and subband adaptive filtering techniques. The received signal is first partitioned by a filter bank into a set of narrowband frequency bins that are approximately nonoverlapping. The adaptive processing is then performed at a lower sampling rate, thus achieving a considerable savings in complexity. Furthermore, the signal power in each frequency bin is normalized to obtain a more uniform convergence rate across the adaptive filter coefficients. The overall convergence rate of the filter is thus dramatically improved compared to that of conventional time- and frequency-domain implementations. Computer simulations of this filter bank equalizer in a multipath environment are presented to illustrate its transient and steady-state convergence properties