Research on Narrowband Line Spectrum Noise Control Method Based on Nearest Neighbor Filter and BP Neural Network Feedback Mechanism

The filter-x least mean square (FxLMS) algorithm is widely used in active noise control (ANC) systems. However, because the a feedback based on minimization of error signal variance to update filter coefficients, it has certain delay, usually slow convergence speed, and system response time long easily affected by learning rate leading lack stability, which often fails achieve desired effect practice. In this paper, we propose an with nearest-neighbor trap structure neural network mechanism reduce coefficient FxLMS use realize parameter update, called NNR-BPFxLMS algorithm. schematic diagram given, performance analyzed. Under various conditions, shown simulation experiment that following three advantages: terms performance, higher reduction under same number sampling points, i.e., faster computer sound pipe experiment, for simple ideal line spectrum noise, compared speed improved more than 95% algorithm, real also 70%. insensitive step size changes. tracking its responds quickly sudden changes can cope complex requirements spectrum.