Research About The Predict Techniques Of Frequency Hopping Time Series

Prediction is an important technology in the domain of signal processing. In theelectronic warfare of military confrontation with frequency hopping communication, theprediction of frequency hopping sequences is of great significance. The frequencyhopping sequence is a special kind of nonlinear sequence, with some pseudo-randomcharacteristics, and can not effectively predicted by using traditional linear predictmethods. To carry out effective forecast, construct the suitable nonlinear predictor isneeded.In this paper, we research and realized the nonlinear predictors, which based onBernstein polynomial, support vector machine (SVM) and Radial Basis Function (RBF)neural network, and research the application of the nonlinear predictor in the actualfrequency hopping communication.The main research work of this paper is shown as following:1. According to the issues that the predict errors of time series rapidly accumulatedin multi-step forecast which affects the predict accuracy, we proposed an algorithmbased on Bernstein polynomial approximation and local modeling theorem with variableframe length and interpolation points.2. By means of optimize methods to solve the problem of machine learning, weanalyzed the applications of support vector machine in classification, and extend itacross regression to realize the prediction which based on support vector machine,following the idea of local modeling theorem.3. According to the issues that the predict performances limited for the singleforecast method, a hybrid model forecast algorithm based on Radial Basis Functionneural network was proposed to model time series, which combined the approximationprinciple of RBF neural network with the characteristics of the sequences to bepredicted.4. According to the issues that noise affect the predict accuracy in the actualfrequency hopping communications system, we analyzed the robustness of the predictalgorithm based on RBF, and proposed the PCA method to extract the main elements toreduce the effects of noise on the forecast results.5. According to the issues of poor forecast performances in the multi-stepprediction of frequency hopping time series, a scheme to improve the long-term forecastperformances was proposed, which based on the D-S evidence theory to make a fusionfor forecast results with variety of algorithms.