Deep Learning Based Turbo Codes And Modulation

With the rapid development of information technology,communication services tend to be diversified,and higher requirements are put forward for the reliability of information transmission in communication systems.As a basic component of communication system,channel coding enables information to resist channel interference in the process of transmission.With the development of artificial intelligence,Deep Learning(DL)has been increasingly used in the field of communication to solve practical problems.In recent years,some scholars have proposed to apply DL to end-to-end channel compilation code systems and have shown its good performance.Existing methods of DL-based channel coding and decoding train network models under a specific signal-to-noise ratio(SNR),however,when deploying real communication systems,it is not guaranteed that the channel conditions are consistent with the training,resulting in the need to store a large number of models for different SNRs.In addition,conventional wireless communication system models are generally designed and optimized based on modules.However,theoretically,the approach of optimizing submodules individually in a communication system does not guarantee that the overall obtained performance is optimal.In this paper,we focus on DL-based channel compilation code techniques,and DL-based joint channel coding and modulation,and make improvements to communication system channel coding and decoding methods and modulation and demodulation scheme in terms of reliability transmission and complexity,and the specific work is summarized as follows.(1)The encoding and decoding algorithm of Turbo code,digital modulation theory and the basic principle of DL are studied in detail.This thesis introduces the structure of Turbo code encoder,including the design of interleaver and eliminator,and its decoding process.For the joint coded modulation theory,the principle of correlated joint coded modulation scheme is described.The structure and design of neural network in DL are described in detail.(2)The autoencoder based encoding and decoding scheme of Turbo code is studied.Firstly,the communication system model and neural network structure under the autoencoder based Turbo coding and decoding scheme are discussed.Relevant simulation experiments compare the network structure performance under different channel SNR.The performance of DL-based turbo coding and decoding scheme under different code lengths is compared.(3)An autoencoder based channel encoding and decoding scheme for Turbo code with adaptive signal-to-noise ratio is proposed.Firstly,the autoencoder based Turbo coding and decoding scheme is analyzed.By introducing attention mechanism,an autoencoder based SNR adaptive Turbo channel coding and decoding scheme is proposed.The simulation results show that by introducing attention mechanism into channel coding and decoding,the proposed adaptive channel SNR structure can generate matched coding codewords according to different channel conditions,effectively deal with the changes of channel conditions and greatly reduce the storage overhead of neural network parameters at the device end.(4)A DL-based joint channel coding and modulation scheme is proposed.The performance of the DL-based joint channel coding modulation scheme based on convolutional neural network and fully connected neural network is explored.When the channel is a non-AWGN channel,the DL-based joint channel coding and modulation scheme is compared with the conventional channel coding and modulation scheme.The simulation results show that the proposed DL-based joint channel coding and modulation scheme outperforms the conventional channel coding and modulation scheme under additive Tdistributed noise channels.