A new adaptive filter structure: comparative study of NLMS, DST-LMS and DCT-LMS schemes applied to electromyographic signal modelling

The goal of this work is to present a transform domain algorithm called discrete sine transform with axis rotation LMS adaptive filter. Based on the least-mean-square (LMS) and discrete sine transform with axis rotation (DSTr) equations, a proposed algorithm is deduced. The authors compare its performance, through computer simulations, with normalized LMS (NLMS), discrete sine transform LMS (DST-LMS) and discrete cosine transform LMS (DCT-LMS) schemes. Another goal of this paper is the study of electromyographic (EMG) signal modeling using the DSTr-LMS algorithm. This is a type of signal that can be represented by an autoregressive model of the fourth order. Using an adaptive filter with adequate order as a predictor, its coefficients can be viewed as a representation of this signal. The learning curves of DSTr-LMS exhibit a better convergence rate and equivalent values of steady state mean-square error (MSE) if compared with learning curves of the other algorithms already mentioned. Therefore, the authors conclude that this set of coefficients is a good representation of this type of signal.