The Application And Research Of Cosine Base Neural Network In Optimization Design Of FIR Digital Filter

With the rapid development of electronic information technology and digital signal processing technology, the function of digital filter is also increasing day by day,all sectors to the digital filter request is higher and higher.Although the design method of traditional digital filter is mature, the realization is simple, but because of the low accuracy of wave filtering, high time complexity, huge structure, inflexible to use, and the wave filtering efficiency of digital filter is not ideal. Especially for the specific field of wave filtering applications, the traditional digital filters cannot reach the satisfactory result, and it directly affects the sustainable development of this field. So people are committed to research various new digital filters, and also strive to improve the function of the digital filters and gradually expand the application field. Modern digital filter can be said to be a process of improvement and optimization of traditional digital filter, over the years, many experts and scholars at home and abroad have done a lot of research work on the design and optimization of the digital filter, constantly put forward some new optimization design method, and obtain a satisfactory effect. There are a wide variety of digital filters, according to the filter implementation of the network structure, digital filters can be divided into infinite impulse response digital filter and finite impulse response filter. Compared with IIR digital filter, FIR digital filter is easy to implement and the system is absolute stability, so it is widely used. This paper outlined the characteristics of phase and amplitude function of FIR digital filter, described the design method of traditional digital filter, analyzed the deficiency and pointed out the necessity of using the neural network to optimize the design of the FIR digital filter.This paper introduced two typical feed-forward neural networks,gave the models and algorithm steps of this two kinds of neural networks, and analyzed the difference between the way of approachingand training algorithm.To improve the order of the digital filter,heighten the effect of the wave filtering, making the filter has a greater stopband attenuation in the same condition,The neural network algorithm can be combined with the optimization design of FIR digital filter. Because of the amplitude frequency characteristic function of FIR digital filter is finite length Fourier series, we can design FIR digital filter by using cosine base neural network model.In this paper, an optimal design algorithm of FIR digital filter based on cosine base neural network model is presented,analyzed its disadvantages and improved the algorithm,that is to adjust the weights of the neural network training steps. Training the weights by the method of recursive least square to achieve the purpose of optimization filter. Then, the FIR digital filter is further optimized on the basis of the improved algorithm.At the same time, it is optimized to transition zone sampling point of FIR digital filter,and got a better transition zone sampling value through a lot of simulation experiments.Through the MATLAB experiment simulation test, the analysis of experimental results shows,the improved algorithm is better,stopband attenuation of the filter is larger and the performance is better.