Frequency Selection Method Of HF Communication Based On Evolutionary Neural Network

A frequency selection method of HF communication based on evolutionary neural network is proposed in this paper.To improve the efficiency of frequency selection using the detecting data accumulated in HF communication system,we established a neural network prediction model of frequency quality.This model indirectly achieves the frequency selection task through predicting the communication quality of all potential frequencies in the preset frequency table of HF communication system and then choosing several frequencies with the highest estimated communication quality as detecting frequencies.Through analyzing the factors influencing the communication quality of short-wave ionosphere reflection channel,the input and output variables of neural network are determined.And through trial and error,the structure of neural network is determined.Since the BP algorithm is easily trapped in the local optima,the paper adopts an improved particle swarm optimization(PSO)to train the parameters of neural network,forming the prediction model of frequency quality based on evolutionary neural network.To enhance the performance of the training algorithm of neural network,a novel particle swarm optimization based on orthogonal information extraction is presented in the paper.This improved PSO periodically selects a particle in swarm to execute orthogonal experiment with the gene pool during the search process,extracts beneficial information within dimensions of the selected particle into the gene pool through significance test.We call this operation as orthogonal information extraction(OIE)operator.As the subsequent particles participating in OIE also obtain the favorable information in the gene pool from other particles,the orthogonal information feedback(OIF)is proceeding meanwhile along with the OIE.The OIE operator opens up a new channel of information exchange for swarm,reinforces the cooperation among particles,and elevates the search efficiency of the algorithm.Moreover,to strengthen the local search capability of particle swarm optimization,we embed the BP algorithm as a local search operator into PSO,and use the gene pool as the initial solution of the local search operator to accelerate its convergence speed.Finally,we designed four simulation experiments,including single link unidirectional communication frequency quality prediction simulation experiment,single link bidirectional communication frequency quality prediction simulation experiment,multi-link unidirectional communication frequency quality prediction simulation experiment and multi-link bidirectional communication frequency quality prediction simulation experiment,to verify the accuracy of the prediction model of frequency quality based on neural network proposed in the paper.The experimental results indicated that,for the communication between fixed stations,the model can fairly accurately predict their communication quality using certain frequency at some time in the future according to their historical SNR data,and for the communication between mobile stations,the model can reasonably predict the communication quality of a new link using the SNR data from the stations nearby.Therefore,the prediction model of frequency quality based on neural network established in this paper can predict the communication quality of certain frequency given the communication condition,and thus is qualified for the frequency selection task in HF communication.