Sensitivity Study And Application Of Complex Variabal Weight Function Neural Network

The monograph New Theory and Methods of Neural Network presents a new type of artificial neural network and training algorithm-spline weight function neural network learning algorithm. Right after training, the weight of neural network is a function of input samples.This new concept overcomes the defect of traditional neural network that the training weight is difficult to reflect the information of training samples, also avoids many other shortcomings like local minima and slow convergence. Complex variabal weight function neural network is the implementation of the weight functions neural networks in complex field, so it has all the advantages of weight functions neural networks. Domestic and foreign scholars have done many researches about the sensitivity of neural network. After calculating the weight function of neural network, we can analyze the output error and the degree of deviation from the taget samples by the trends of sensitivity.In this paper, based on the sensitivity analysis of neural network, analyzes the sensitivity of complex variabal weight function neural network, and calculates the sensitivity formula. In the theory part, first introduces the model of complex variabal weight function neural network and the methods for determining the weight function; then introduces the related concepts of neural network sensitivity, including its definition, major applications and analysis methods.Then does the study of sensitivity of complex variabal weight function neural network, using the recursive method to derive the sensitivity formula. Mean while simulation experiments about the complex variabal weight function neural network and its sensitivity have been done by Matlab. The validity and correctness of theoretical analysis can be determined by the simulation results.Using the senitivity of complex variabal weight function neural network, this paper designs a kind of spectrum shaper in the field of cognitive radio. The design model is given combines complex variabal weight function neural network. And then by calculating the sensitivity of the system to avoid the frequency band, achieves a reliable sharing of spectrum. Both the theoretical analysis and simulation showes that the spectrum shaper has good results in spectrum efficiency and spectrum avoiding performance.