Abstract:In decentralized active noise control (ANC) systems, crosstalk between multichannel secondary sources and error microphones significantly degrades control accuracy. Moreover, prefiltering reference signals in filtered-x (Fx) type algorithms may further introduce modeling errors. A theoretical analysis of the Fx-based decentralized control algorithm was performed, which reveals how prefiltering and crosstalk affect the control performance. Then, a hybrid method combining fixed-value neural networks and adaptive strategies was proposed for efficient decentralized ANC. The adaptive filter models the primary path of its own channel online using the least mean square (LMS) algorithm while the neural network (named DecNet) is u…
Abstract:In decentralized active noise control (ANC) systems, crosstalk between multichannel secondary sources and error microphones significantly degrades control accuracy. Moreover, prefiltering reference signals in filtered-x (Fx) type algorithms may further introduce modeling errors. A theoretical analysis of the Fx-based decentralized control algorithm was performed, which reveals how prefiltering and crosstalk affect the control performance. Then, a hybrid method combining fixed-value neural networks and adaptive strategies was proposed for efficient decentralized ANC. The adaptive filter models the primary path of its own channel online using the least mean square (LMS) algorithm while the neural network (named DecNet) is used for secondary paths inverting and decoupling. The hybrid DecNet-LMS algorithm was implemented in the time domain to guarantee causality and avoid latency. Simulation results with measured acoustic paths show that the proposed method outperforms the existing ANC algorithms using either traditional adaptive filters or neural network-based fixed-coefficient methods under different acoustic conditions.
| Subjects: | Systems and Control (eess.SY); Signal Processing (eess.SP) |
| Cite as: | arXiv:2511.03162 [eess.SY] |
| (or arXiv:2511.03162v1 [eess.SY] for this version) | |
| https://doi.org/10.48550/arXiv.2511.03162 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yijing Chu [view email] [v1] Wed, 5 Nov 2025 04:15:23 UTC (567 KB)