Multiple parallel branch with folding architecture for multichannel filtered-x least mean square algorithm
Dong-Yuan Shi, Jianjun He, Chuang Shi, Tatsuya Murao, Woon-Seng Gan
Abstract
Multichannel active noise control (MCANC) systems are commonly used in acoustic noise or vibration control, such as large-dimension ventilation ducts, open windows and mechanical structures. However, its computational load far exceeds the capabilities of digital signal processors (DSPs) and microcontrollers. Even the field programmable gate array (FPGA) cannot straightforwardly cope with the exponential increase in the computation load of MCANC systems. A novel architecture, called the multiple parallel branch with folding, is proposed for the J × J × M (J reference microphones, J secondary sources and Merror microphones) MCANC implementation with the floating-point arithmetic. This architecture addresses the tradeoff between throughput and hardware resource consumption by using parallel execution and folding. The proposed architecture is validated in an experimental setup that carries out a 4 × 4 × 4 multichannel filtered-x least mean square (FxLMS) algorithm achieving the sampling rate and throughput of 24 KHz and 18.4 Mbps, respectively.
BibTeX
@inproceedings{icassp2017_multipleparallel,
title = {Multiple parallel branch with folding architecture for multichannel filtered-x least mean square algorithm},
author = {Dong-Yuan Shi and Jianjun He and Chuang Shi and Tatsuya Murao and Woon-Seng Gan},
booktitle = {ICASSP 2017},
year = {2017}
}