Channel Coding Recognition Of Non-cooperative Communication Based On Deep Learning

Source coding and channel coding are required before transmitting signals in a complete digital communication system.The purpose of source coding is to reduce redundancy,while channel coding is to improve data transmission efficiency and reduce bit error rate,so that the receiver can better identify and correct errors in the data transmission process.In the field of non-collaborative communication,in order to correctly obtain the information in the intercepted signals,the non-collaborative end must perform channel decoding after demodulation of the received signal.By researching on channel coding recognition,the channel decoding can be carried out simply according to channel coding recognition and parameter estimation of the intercepted signals.Therefore,this thesis conducts in-depth research on channel coding recognition and parameter estimation,unknown channel coding discrimination and channel coding recognition without demodulation,through proposing feature extraction algorithms to improve the accuracy of channel coding recognition,and using recognition probability distribution to design algorithms to achieve unknown channel coding discrimination and channel coding parameter estimation,and channel coding recognition without demodulation is realized by combining artificial features and raw signal samples.The main work is as follows:1.Four new feature extraction algorithms aiming at channel codes including autocorrelative code weight,advanced run-length,GFFT square Euclidean distance and FFT square Euclidean distance are proposed to form multi-dimensional features with four existing feature extraction algorithms.Under the condition that the signal-to-noise ratio is greater than-3dB,the recognition accuracy of 11 different types of channel codes can reach more than 95.1%;2.An algorithm of unknown channel coding discrimination is proposed,and the accuracy of discriminating unknown channel cods can reach 85% under the condition of signal-to-noise ratio ranging from-5dB to 4dB,and a parameter estimation algorithm of RS code is extended,which realizes that the accuracy of parameter estimation of RS code can reach 99.9% under the condition of signal to noise ratio ranging from-5dB to 4dB;3.A channel coding recognition algorithm without demodulation is proposed.16 artificial features extracted from time domain,transformation domain and high-order cumulant are used to combine with signal samples to form feature vectors,and the recognition accuracy of 22 types of measured signals can reach more than 95.6% when the signal-to-noise ratio is greater than 12 dB.