An Algorithm for Periodic Noise Reduction Based on ICA

The traditional independent component analysis （ICA） relies on strong assumptions. Among them is the requirement that the number of sensors is more than or equal to that of sources. Thereby, the applications of ICA are limited by the requirement for the number of sources, which might be impractical for noise reduction in many cases. In this paper, a novel method for constructing a mixed signal is proposed based on the holistic periodicity of periodic noises. Much better signal-to-noise ratio （SNR） could be obtained by using ICA with the constructed mixed signal. The algorithm presented proves to be fast, computationally simple and efficient. Simulation and experimental results demonstrate the feasibility and good performance of the proposed algorithm.