Active Noise Control System Based on Even Mirror Fourier Filter Using an Efficient Adaptive Algorithm

The EMF (Even mirror Fourier) expansion functions are the effective solution for compensating for nonlinear distortion due to their universal approximation ability. However, since the ANC system contains nonlinearity that comes from many different causes, the expansion function needs to use a large enough memory length and order. This can make the EMF-based ANC system heavy. The paper proposes an efficient ANC system based on the EMF filter and set-membership learning strategy with an estimated error bound. For this purpose, we exploit the data-dependent learning algorithm to reduce the cost of updating the weights in the EMF filter. The filtering weights that contribute little to the model's estimation process will be ignored thanks to the data-selective update strategy. Additionally, to avoid performance degradation when using the set-membership algorithm in the ANC system, an estimated error bound (EEB) has been proposed to replace the predefined error bound. The EEB is designed to adapt to nonlinear variations, thus the proposed ANC system will significantly improve the noise reduction quality. Compared with other nonlinear controllers based on the Volterra or Trigonometric expansion functions, the proposed controller is effective in both the computational complexity and denoising performance.