An Improved Least Mean Square Algorithm for Adaptive Filter in Active Noise Control Application

Abstract

System identification process is used to model the transfer function of the auxiliary path in Active Noise Control (ANC) system, in which its accuracy and the speed of adaptation will determine the performance and stability of the system particularly for broadband noise attenuation. The process of identification is considered successful when the error signal, the difference between the input signal of a microphone and the output signal of an adaptive filter is minimized and converged at steady state. The method of Least Mean Square (LMS) is used as an adaptive algorithm in ANC application due to its simplicity and robustness in implementation. This paper presents an improved LMS algorithm to address the convergence performance of the error signal in a system identification process for ANC headset, in which repeated updates on filter weight are carried out within every sampled audio data. The proposed work uses field programmable gate arrays to realize real-time hardware implementation of LMS adaptive filter with the repeated updates of filter weight at 48 kHz data sampling rate. Results from the simulations have predicted error convergence for several selections of learning constant μ, while the hardware implementation further verified the results from simulation with more stringent selection of learning constant due to the time-varying environment.