Arbitrarily high sampling rate adaptive filters

Abstract

Adaptive filters have an inherent feedback from error signal back to the adaptation of the coefficients. This represents a problem in high sampling rate realizations. We demonstrate a realization of adaptive filters for which there is no theoretical limit on sampling rate in a given speed of hardware, at the expense of additional hardware and latency. Our realization does not change the input-output characteristics aside from finite precision effects, and hence does not degrade the filter tracking capability. The basis for our realization is the adaptive lattice filter, which uses only local feedback in adapting reflection coefficients at each stage, and for which the recursive portion of the adaptation algorithms in each stage is linear. Our realization is based on these two properties and the basic technique of look-ahead computation. Several forms of our realization applying to different recursive least-squares and stochastic-gradient adaptation algorithms are described.