Research On Wideband Spectrum Detection Technology Based On Dynamic Neural Network

With the rapid development of 5G technology and next-generation wireless communication networks,wideband spectrum detection technology has been widely studied.Compared with narrowband spectrum detection,it can more flexibly implement cognitive users' access to spectrum holes and improve spectrum detection efficiency.Modulated Wideband Converter(MWC)can well solve the problem of high sampling rate required for wideband spectrum detection.Based on this,this paper aims at Sparse Bayesian Learning(SBL)based on simplified MWC.In the detection algorithm,the problem of over-reliance on the prior information of the signal and insufficient detection accuracy are proposed.A method of using a dynamic neural network for wideband spectrum detection is proposed.On the one hand,based on the dynamic neural network,based on the time correlation between the spectrum data,two network models suitable for spectrum detection are considered,namely Long Short Term Memory(LSTM)neural network and Elman(Elman)Neural Network.Construct new network structures for the two networks,train and learn to set the best detection parameters,analyze and compare their performance in broadband spectrum detection,and select the appropriate model.In the end,the dynamic LSTM neural network algorithm has the unique advantage of fully mining the signal itself.Under QPSK modulation,when the number of main user signals is 3 and the signal-to-noise ratio is 5dB,the detection probability is increased by 20% compared with the MWC-Elman algorithm.On the other hand,it is also the focus of this study.After determining the system model of the MWC-LSTM spectrum detection algorithm,the previous perceptual event and the current perceptual event are used as the input of the LSTM network model,and the characteristics of the network are used directly from the input compressed sampling value.Relevant information is extracted to estimate the support set.Next,we use one of the symmetric frequency bands to determine that the signal is completely detected.We analyze the MWC-OMP,MWC-SBL,and MWC-LSTM algorithms for single-signal and multi-signal spectrum detection at fixed positions.performance.The simulation results show that the proposed algorithm can effectively improve the detection probability compared with the other two algorithms without the sparse prior,and the false alarm probability is always kept in a lower range.Finally,the influence of the number of sampling channels on the three detection algorithms is studied,and it is found that the MWC-LSTM algorithm can achieve the same detection probability as other algorithms with fewer sampling channels.