Browse Theses

This dissertation deals, in general, with adaptive filtering and presents several least mean-square (LMS)-type algorithms that could be used in many adaptive filtering applications like system identification, noise cancellation, channel equalization, and signal prediction. The aim is to improve the performance of these applications by minimizing the mean-square error, increasing the rate of convergence, and improving the tracking capability of the proposed adaptive algorithms. In addition, the research work in this dissertation introduces an adaptive noise canceller for solving the signal leakage or crosstalk problem.

Two proposed LMS-type algorithms, based mainly on error normalization, are analyzed and simulated in system identification using different cases of correlated and uncorrelated input data in stationary and nonstationary environments. Simulation results show significant improvements of the proposed algorithms over other algorithms in achieving higher rates of convergence with the same value of steady-state mean-square error. Two more algorithms are analyzed and simulated in adaptive noise cancellation using real speech signals. The first one is a time-varying step-size algorithm that employs both data and error normalization in its update recursion. The second algorithm uses a step-size that varies in time between two hard limits based on a predetermined nonlinear decreasing function of signal to noise ratio (SNR) estimated at every iteration. Simulation results of these algorithms show their superior performance over other algorithms in minimizing signal distortion and reverberation in the recovered speech.

Finally, a new adaptive noise canceller is proposed to improve the system performance in the presence of crosstalk or signal leakage in the reference input. The proposed adaptive noise canceller consists of three microphones and two adaptive filters that automatically adjust their coefficients through LMS-type algorithms to produce an output that is as close as possible to the original speech signal. Simulation results in both stationary and nonstationary noise environments show significant improvements of this proposed adaptive noise canceller over the conventional one.