Research On Communication Signal Modulation Identification Algorithm And FPGA Implementation In Multi-path Rayleigh Channel

The development of modern society is fast,and the development of science and technology is changing with each passing day.The development of science and technology must be accompanied by the exchange of information.The exchange of information requires the support of communication technology,so the rapid development of communication technology supports the development of other industries.Modulation recognition of communication signals have many achievements,but most methods can only detect specific way of modulation,which has great limitations.At the same time,the feature extraction of most modulation recognition methods of communication signal uses a traditional pattern of recognition method,which requires manual extraction of the characteristics of the communication signal and a lot of work.This pa per starts with feature-based extraction modulation recognition algorithm based on decision theory and deep learning network;studies the mechanism and implementation method of communication signal modulation recognition algorithm;and implements the algorithm under FPGA platform structure.In this paper,firstly,15 kinds of communication signal modulation methods involved are analyzed and explained,such as ASK signal,PSK signal,FSK signal,OFDM signal,QAM signal and MSK signal.The analysis of these 15 kinds of characteristics of communication signal modulation is also included.Afterwards,through the research of feature-based extraction modulation recognition algorithm based on decision theory and deep learning network,successfully using the feature-based extraction recognition algorithm to recognize the communication signal's modulation at higher signal-to-noise ratio,the experiment shows that the algorithm has achieved good results in the environment with high signal-to-noise ratio.However,in the environment with low signal-tonoise ratio,the performance of the feature-based modulation recognition algorithm based on decision theory is drastically reduced.Therefore,this paper proposes the convolutional neural network structure under two different application scenarios.The experimental results show that the deeper structure of the network layer has lower SNR sensitivity,but the calculation volume is relatively large,and the networks with a relatively small number of structures have the smaller calculation.What's more,the signal can be recognized precisely when the SNR is greater than 5 d B.Compared with other algorithms of the same kind of channel,other algorithms have lower recognition rate than this algorithm when the channel influence is lower than this algorithm.Finally,the Open CL platform is used to implement the nine-layers convolutional neural network.Through the convolution pooling module,the global average pooling module,and the fully-connected module of the designation of Open CL kernel,the algorithm running speed is accelerated by parallelizing its structure.Comparing the migration results of FPGA algorithm and the simulation results of Python program algorithm,the error of the algorithm implementation in the two architectures is 0.FPGA quantization error can not affect the classification result.The algorithm implementation result is within the FPGA quantization error,so FPGA algorithm is transplanted with zero error.