Research On Cognitive Radio Spectrum Prediction Based On Dynamic Fuzzy Neural Network

Cognitive radio is a feasible method to alleviate the shortage of radio spectrum resources and make full use of free spectrum to improve spectrum utilization.Spectrum prediction is a technique for analyzing the using rules of historical frequency bands to predict the future free frequency.The cognitive users can access the channel effectively with the spectrum sensing technology in real time.Spectrum prediction technology can reduce the interference caused by unauthorized users on the authorized users,avoid unnecessary energy loss,improve the accuracy and reliability of access,so as to achieve a more reasonable and effective use of spectrum.So far the main spectrum prediction methods are based on regression analysis,based on Markov model and machine learning,etc.In this paper,firstly,several different methods of cognitive radio spectrum prediction technology are studied.Secondly,the basic principles of neural network and fuzzy system are explained and the necessity of the combination of them are analyzed.Then,the internal structure and principle of the dynamic fuzzy neural network are researched and some examples are taken to illustrate its characteristics.According to the principle of queuing theory,a model for the channel state of cognitive radio system are established and the channel state data are generated as the spectral data for the spectrum prediction.Studying on the internal structure of neural network and analysis its algorithm process,a spectral prediction model based on dynamic fuzzy neural network are established,in the process of learning and training the network structure are adjusted and optimized adaptively based on data sequence.The MATLAB simulation is used to complete the prediction and comparison of the dynamic fuzzy neural network system,which illustrated the feasibility and superiority of the method on this basis.Then the prediction model based on different signal-to-noise ratio are analyzed and the stability and accuracy of the dynamic fuzzy neural network prediction model are compared.For multi-channel spectrum prediction,how the time domain correlation between multiple channels make influence on prediction performance are explained.