Low Complexity Adaptive Noise Cancellation in Audio Systems using LMS Algorithm

Abstract

Noise reduction is crucial to improving real-time acoustic speech communication quality and intelligibility. We present a low-complexity adaptive noise cancellation system based on the Least Mean Squares (LMS) algorithm in a Digital Signal Processing (DSP) environment. The main objective of this research is to reduce the computational complexity of the algorithm and provide a computationally efficient noise cancellation system for implementation in resource-constrained and real-time systems. The noise cancellation system is based on an initial input representing the desired speech signal corrupted by noise, and a reference input representing highly-correlated noise. The LMS adapts the filter coefficients to minimise the MSE between the wanted audio signal and the filtered signal. Particular attention is given to the choice of step size for convergence rate and stability. The noise reduction allows the signal-to-noise ratio (SNR) to be significantly improved, and the convergence rate is also fast compared to conventional filters. The results demonstrate that the proposed method offers significant noise reduction, improvement in SNR and rapid convergence compared to other traditional filters. It can reduce noise under various conditions and so it is effective for use in speech enhancement, hearing aids and mobile telephony. The adaptive noise cancellation system using the LMS algorithm provides a simple, fast, effective, scalable and real-time approach to audio signal enhancement with reduced computational load, which can be used in an embedded digital signal processing (DSP) system.