MULTICHANNEL FILTERED-X LMS FOR ACTIVE NOISE CONTROL ON MULTICORE SYSTEMS

摘要

In recent years, the level of development of the CPU (Central Processing Unit) hardware has reached near saturation. Thus to overcome the computational limitations of such systems, current technology is betting on multicore central processing units or strategies based on general purpose graphical processing units (GP-GPU). These systems allow us to execute shared processes between different cores if these processes are suitably parallelized. Although the main theory of multichannel active noise control has been already developed, implementations of these systems have not reached the general market mainly due to the high computational cost of such systems, which grows proportionally with the number of channels. Therefore, when using control algorithms based on sample by sample processing (as the multiple channel filtered-x LMS), either the number of channels or the highest working frequency is limited by the hardware system. However, the use of multicore platforms allows implementing these systems without increasing multichannel runtime, and therefore increasing the number of control channels without having to decrease the working frequency of the system. This article describes how to implement the modified filtered-x LMS algorithm for multichannel active noise control on a multicore system. It highlights its benefits and challenges, and explains how the computational gain has been achieved.