Realization of Delayed Least Mean Square Adaptive Algorithm using Verilog HDL for EEG Signals

Abstract

An efficient architecture for the implementation of delayed least mean square (DLMS) adaptive filter is presented in this paper. It is shown that the proposed architectures reduces the register complexity and also supports the faster convergence. Compared to transpose form, the direct form LMS adaptive filter has fast convergence but both has most similar critical path. Further it is shown that in most of the practical cases, very small adaptation delay is sufficient enough to implement a direct-form LMS adaptive filter where in normal cases a very high sampling rate is required and also it shows that no pipelining approach is necessary. From the above discussed estimations three different architectures of LMS adaptive filter has been designed. They are, first design comprise of zero delays i.e., with no adaptation delays, second design comprises of only single delay i.e., with only one adaptation delay, and lastly the third design comprises of two adaptation delays. Among all the three designs zero adaptation delay structure gives efficient performance comparatively. Design with zero adaptation delay involves the minimum energy per sample (EPS) and also minimum area compared to other two designs. The aim of this thesis is to design an efficient filter structures to create a system-on-chip (SoC) solution by using an optimized code for solving various adaptive filtering problems in the system. In this thesis our main focus is on interference cancellation in electroencephalogram (EEG) applications by using the proposed filter structures. Modern field programmable gate arrays (FPGAs) have the resources that are required to design an effective adaptive filtering structures. The designs are evaluated in terms of design time, area and delays.