Sensitivity Study, Algorithm Implementation And Application Of The Second Type B-spline Weight Function Neural Network

Sensitivity refers to the influence from the disturbance of input. The study of the sensitivity ofneural network is to construct stronger neural network. After the training of weight function neuralnetwork, the weight is about the input sample, which simplified structure of the neural networkstructure, and solve the defect of traditional neural network algorithm for many years.This paper quoted the topology structure and training algorithm of the second type weightfunction neural network from document [1]. According to the formula and property of B-splinefunction, the paper use the B-spline function to approximate the theory weight function. The modelerror formula and approximation noise error formula of the second type B-spline weight functionneural network was deduced. Then according to the statistical definition of sensitivity, then deducethe formula of the second B-spline weight function neural network.In this paper the sensitivity formula of the second B-spline weight function neural network isdeduced, and analyze the effect of the outside interference noise and sample noise for the network.This paper use the Matlab to simulate the formula of sensitivity, and it is validated that the heproposed formula of sensitivity is correct, and it also proved the second type B-spline weightfunction neural network has good approximation ability and generalization ability.This paper uses the sensitivity of the second type B-spline weight function neural network torecognize the modulation signal. According to the theory of the digital modulation signal, thepaper extracted five instantaneous parameters of transient signals. It is proved from the simulationresults that the second type B-spline weight function neural network classifier has simplestructure and high recognition rate. The recognition rate will be higher after adding the sensitivityanalysis.