Vector Decimation Harmonic Mean-Based Algorithm for Online Acoustic Feedback Active Noise Control

This research addresses the acoustic feedback and stability performance of Active Noise Control (ANC) systems. The Filtered Cross Least Mean Square (FxLMS) algorithm is one of ANC’s most successful adaptive algorithms for reducing undesired noise. The ANC filters have their convergence speeds reduced by significant acoustic feedback, making them unstable in the worst-case scenario. It is possible to eliminate acoustic feedback using a fixed feedback neutralization filter that can be acquired offline. However, the feedback path could change over time, so we might have to make modifications often while the ANC system is online. This paper proposes two enhancements to the FxLMS algorithm to address the ANC’s performance difficulties. To address acoustic feedback problems, the first alteration improves the ANC duct’s feedback path modeling by incorporating the online Feedback Naturalization (FN) concept with an Active function normalized LMS algorithm (AF-NLMS). The second modification, the vector-decimation-based method, is proposed in this paper as a way to improve stability, convergence speed, and robustness. It is achieved by decimating the filter weight vector into small vectors and then updating each vector using the harmonic mean error of the VSS FxLMS. To ensure the enhanced effectiveness of the proposed solution, the AF-FxLMS, online-AFFNFxLMS, and HM-AFFNFxLMS algorithms are used to update the filter weights as a demonstration. The proposed method has superior noise reduction effectiveness compared to other algorithms, which indiscriminately attenuate the control filter across all frequencies. The validation of the proposed approach was achieved via simulations and experimental tests in a PVC duct ANC system, showing that robust stability can be achieved while keeping relatively good performance.