Selective Fixed-filter Active Noise Control with Compensation Filter Based on Modified Error Filtered-x Discrete Cosine Transform Least Mean Square Algorithm

抄録

This paper presents a selective fixed-filter active noise control (ANC) system incorporating a compensation filter, based on a modified error filtered-x discrete cosine transform least mean square (Fx-DCT-LMS) algorithm. In the proposed framework, a compensation filter is cascaded with a fixed noise control filter, which is selected using a convolutional neural network (CNN). Adaptive signal processing is then applied to compensate for the discrepancy between the fixed filter and the optimal solution. The sliding discrete cosine transform (SDCT) enables filter selection using spectrograms generated from arbitrary-length sample windows, which serve as input to the CNN. Notably, since the proposed method utilizes the DCT to compute the spectrogram, it significantly reduces computational complexity compared to conventional DFT-based spectrogram generation. Moreover, by employing the DCT-LMS algorithm, the compensation filter achieves faster convergence than the conventional filtered-x normalized LMS (Fx-NLMS) algorithm. Simulation results using real-world impulse responses demonstrate that the proposed system can achieve up to 10 dB of noise reduction under varying noise conditions.