Design and implementation of an adaptive LMS-based parallel system for noise cancellation

When a desired signal is encompassed by a noisy environment, active noise cancellation may be implemented to remove the background noise. The presented algorithm is based on the standard Least Mean Squares (LMS) algorithm developed by Bernard Widrow. Modifications to the LMS algorithm were made in order to optimize its performance in extracting a desired speech signal from a noisy environment. The system consists of two adaptive systems running in parallel, with one having a much higher convergence rate to provide rapid adaptation in a non-stationary environment. However, the output of the higher converging system results in distorted speech. Therefore, the second system, which runs at a lower convergence rate but regularly has its coefficients updated by the first system, provides the actual output of the desired signal. All of the algorithm development and simulation were initially performed in Matlab, and were then implemented on TMS320C6416 Digital Signal Processor (DSP) evaluation board to produce a real-time, noise-reduced speech signal.