The Research Of Adaptive Noise Cancellation Based On Neural Network

Noise cancellation is a very important component of the field of signal processing. The neutral network for adaptive noise in the technology to eliminate background noise is effective since neutral network solve the relayed nonlinear noise which traditional adaptive noise canceller can not solve.In this paper, the primary study is as follows:(1) first of all, the paper introduces the principle of adaptive noise cancellation. And then Least Mean Square. Least Squares and natural network algorithm are researched deeply. It has conducted a comparative analysis of them and pointed out their advantages and disadvantages. Also, some parameters related to the system of adaptive noise cancellation are analyzed. (2)It has focused the research on the problems of back-propagation (BP) structure, algorithms principle and then has analytically discussed the problems of BP Neural Network. In the view of BP flaws, the paper has proposed an improved algorithm-LM algorithm principle. (3)Based on MATLAB platform, the paper has designed a neural network model for noise cancellation. It has focused on the determination of model parameters. Exactly, the input layer neurons nodes are determined by the input samples, output nodes determined by output samples. As to the determination of hidden layer, firstly, by the empirical formula hidden layer nodes has taken an initial value and then the actual learning of the network to adjust to the value. The network training algorithm has used an improved L-M algorithm. Finally, improved algorithm has compared with the standard BP algorithm, the improved convergence of the algorithm faster and better and accuracy. (4)Based on the models of natural network adaptive noise cancellation and LMS adaptive noise cancellation, the paper has studied the noise canceling capability from two noise non-linear correlation and linear correlation. Also it has comparatively analyzed their performance including de-noising ability and convergence speed. The effects of offset noise:we has calculated the signal to noise ratio of both systems and analyzed the two systems of de-noising in time domain and frequency domain map. In the convergence speed:it has compared the error convergence curves of the two systems. The results have proved that an Adaptive Noise Cancellation Controller based on Neural Network has good performance both two noises related linear and two noise nonlinear, while a Adaptive Noise Cancellation Controller based on LMS has efficient in the noise cancellation only when two noise related linear. (5)At last, the paper has analyzed the data gained in case study, and the results show the system can comparatively eliminate noise.