Research On Joint Source-Channel Wireless Image Communication Technology Based On Multi-Level Semantic Information

With the advent of the 5G era,the surge in the amount of data transmitted makes the communication technology at the signal level approach the Shannon limit.How to effectively solve the challenges brought by the era of massive data to communication technology is a hot topic of current research.End-to-end communication system based on deep learning design provides a new solution.Therefore,this paper focuses on the end-to-end communication system,and proposes an end-to-end communication system with higher transmission quality,transmission efficiency and stability of the shortcomings of the existing design.The specific codec and channel design schemes are as following:The design scheme of codec technology under the channel is determined.In view of the unstable transmission performance of the existing wireless communication system of Joint Source-Channel Coding(JSCC)without pilot,the global information and location information of images cannot be effectively captured.A joint source channel wireless image transmission system based on multilevel semantic information is proposed.The transmitter of the system is composed of two networks: the feature extraction network is used to extract the high-level semantic features of the image,compress the information of image transmission,and improve the bandwidth utilization;Feature retention network is used to preserve low-level semantic features and image details to improve communication quality.The receiving end is also composed of two networks.The received high-level semantic features and low-level semantic features after the feature enhancement network are fused in the same dimension,and then the feature recovery network is used for image dimension recovery,and the image location information is effectively used for image reconstruction to improve transmission stability.Design scheme of communication system under unknown channel.Aiming at the problem that channel modeling of some end-to-end communication system is limited to determining the noise model and unable to model the unknown channel,a Multi-Level-D JSCC image transmission system which can adapt to different noise environments or has automatic learning ability is proposed.It consists of two parts: coding method based on multi-level semantic information and DDPM-C channel model method.DDPM-C model models unknown channels into two main stages: diffusion process and reverse diffusion process.In the diffusion stage,Gaussian noise is gradually applied to the image features after passing through the channel to destroy the original features and make them eventually destroyed into Gaussian noise.In the reverse diffusion stage,an inverse model is learned,Gaussian noise is taken as the input,and image features similar to the training data are restored through multiple reverse diffusion steps,so as to achieve the effect of simulating the channel.This method of image feature generation is to restore image feature step by step rather than directly map the relationship between Gaussian noise and image feature.The design scheme of codec technology under the channel is determined.In view of the unstable transmission performance of the existing wireless communication system of Joint Source-Channel Coding(JSCC)without pilot,the global information and location information of images cannot be effectively captured.A joint source channel wireless image transmission system based on multilevel semantic information is proposed.The transmitter of the system is composed of two networks: the feature extraction network is used to extract the high-level semantic features of the image,compress the information of image transmission,and improve the bandwidth utilization;Feature retention network is used to preserve low-level semantic features and image details to improve communication quality.The receiving end is also composed of two networks.The received high-level semantic features and low-level semantic features after the feature enhancement network are fused in the same dimension,and then the feature recovery network is used for image dimension recovery,and the image location information is effectively used for image reconstruction to improve transmission stability.Design scheme of communication system under unknown channel.Aiming at the problem that channel modeling of some end-to-end communication system is limited to determining the noise model and unable to model the unknown channel,a Multi-Level-D JSCC image transmission system which can adapt to different noise environments or has automatic learning ability is proposed.It consists of two parts: coding method based on multi-level semantic information and DDPM-C channel model method.DDPM-C model models unknown channels into two main stages: diffusion process and reverse diffusion process.In the diffusion stage,Gaussian noise is gradually applied to the image features after passing through the channel to destroy the original features and make them eventually destroyed into Gaussian noise.In the reverse diffusion stage,an inverse model is learned,Gaussian noise is taken as the input,and image features similar to the training data are restored through multiple reverse diffusion steps,so as to achieve the effect of simulating the channel.This method of image feature generation is to restore image feature step by step rather than directly map the relationship between Gaussian noise and image feature.In this paper,an image communication system is built and the CIFAR-10 dataset is used for training under Gaussian channel and Rayleigh fading channel.The experimental results show that the Multi-Level JSCC algorithm proposed in this paper can effectively transmit and recover image information.Furthermore,the newly proposed Multi-Level JSCC-D algorithm achieves higher peak signal-to-noise ratio than the Multi-Level JSCC algorithm under the same simulation conditions.