Study Of Digital Filters Optimal Design

In the technical-application dominant of modern signal processing, communication engineering and electronic information technology, digital filters, as a kind of digital signal processing unit, plays a very significant role, and its application value is increasingly every days.Traditional digital filter design methods to develop a more mature, there are a large number of ready-made formulas and tables of parameters, and so is relatively simple to implement. However, the conventional digital filter design method is not suitable for high-end digital filter design, and not easy to accurately control the frequency of the passband and stopband boundaries, and thus between the performance indicators of the performance indicators will always be the ideal state error, with the performance requirements of modern high-tech products by leaps and bounds, more and more strict, making the traditional way to design filters up to less than the error range of error range requirements.Therefore, the theoretical knowledge of this paper, artificial neural network, the optimization design of the one-dimensional digital filters do a more in-depth study, so that approximate the ideal filter in the frequency domain.Artificial neural network algorithm has a significant advantage---not dependent on accurate model. To the above problems, this paper puts forward a kind of using cosine basis functions as a neuron feedforward neural network model to design digital filter. Initially, This article explains the superiority of the traditional neural network,lists several typical neurons excitation function to discuss. Then, this article explains the role of learning rate's adoption, introduces several kinds of typical learning rule, and the range of learning rate's adoption is discussed. Later, this article explains how to use cosine based neural network as algorithm model of optimization design of digital filter. Including set up the cosine base neural network model, demonstrate neural network system's stability and convergence, introduce train steps of the neural networking. Finally, this paper explains how to use MATLAB, to simulate the cosine base neural network model algorithm of digital filter design, and mainly introduces a FDAtool tools to get using neural network of initial values of the weighted coefficient method. From the MATLAB simulation result of the final, its performance index approach the perfect state, and global error index reached the e-015orders of magnitude. Not only will it analysis the longitudinal comparison with the traditional method of digital filter design's simulation results, but also its own weight coefficients for initial value's default is different,so it will produce different simulation results for the lateral comparative analysis. Final results show that the cosine base neural network model algorithm of digital filter design is a very excellent digital filter's design method.