A new architecture for implementing pipelined FIR ADF based on classification of coefficients

In this paper, we propose a new method for implementing pipelined finite-impulse response (FIR) adaptive digital filter (ADF), with an aim of reducing the maximum delay of the filtering portion of conventional delayed least mean square (DLMS) pipelined ADF. We achieve a filtering section with a maximum delay of one by simplifying a pre-upsampled and a post-downsampled FIR filter using the concept of classification of coefficients. This reduction is independent of the order of the filter, which is an advantage when the order of the filter is very large, and as a result the method can also be applied to infinite impulse response (IIR) filters. Furthermore, when the proposed method is compared with the transpose ADF, which has a filtering section with zero delay, it is realized that it significantly reduces the maximum delay associated with updating the coefficients of FIR ADF. The effect of this is that, the proposed method exhibits a higher convergence speed in comparison to the transpose FIR ADF.