A Distributed Arithmetic Based Approach for the Realization of the Signed-Regressor LMS Adaptive Filter

This paper presents a distributed arithmetic (DA) based approach for the implementation of signedregressor LMS adaptive filter. DA, although is an efficient technique for the implementation of fixed coefficient filters, the adaptive filter implementation using DA is not a straight-forward task as the partialproducts of the filter weights have to be updated in every iteration. This is achieved by storing the partialproducts of the signum values of the input samples in a look-up-table (LUT). It has been shown that this LUT can be updated to accommodate the partial-products of newest set of samples in an efficient way using the circular- shifting of its address bits. Results indicate that the proposed filter can give better throughputs compared to multiply-and-accumulate (MAC) based implementation and can be effective when implementing large filters. With proper choice of system parameters, the proposed architecture for a 32- tap filter consumes around 87% less number of adder units while providing similar throughput performance compared to most recent existing DA based architecture.