Adaptive filtering with finite wordlength constraints

Abstract

The paper describes the implementation and design of an adaptive digital filter. A brief introduction to the theory of adaptive filtering is first undertaken; this is followed by a description of the hardware used in the implementation. At this point, finite wordlength considerations are outlined. A consequence of these finite wordlength constraints – namely deadspace in the adaptive update equations – is investigated and a new method for reducing this deadspace, by use of an additive dither signal in the weight vector update algorithm, is proposed. Simulation results which confirm the utility of the method are included. Experimental results obtained from the hardware described are presented. A technique for determining required wordlength for digitally implemented adaptive filters is outlined. Finally, limitations of the design are discussed, along with possible applications.