基于语义压缩的深度JSCC研究

In the future, the communication of agents needs to re-examine the purpose, way and process of communication. The communication between agents is to complete intelligent tasks, that is, pragmatic communication. Shannon\'s separation theorem has some limitations such as difficulty in extracting semantic information in pragmatically oriented semantic communication scenarios. Compared with the separation scheme, the joint source channel coding scheme can improve the overall performance of the system. In particular, the deep joint coding method is applied to the semantic communication coding terminal, which can efficiently extract the semantic features of the task. Aiming at the classification task that the receiver does not need to reconstruct the source, this paper improves the existing deep joint source channel coding technique and introduces a strategy mechanism in the semantic communication sender. Driven by classification tasks, the policy mechanism can learn the semantic features of source data, and the degree of semantic compression can be changed by adjusting the emphasis of the policy mechanism in the whole semantic coding network. Compared with the existing deep joint source channel coding schemes, the deep joint source channel coding scheme (JSCC) introduced in this paper greatly improves the classification performance under the condition of AWGN channel with the same feature compression ratio. Compared with the traditional JEPG encoding, the combined encoding scheme greatly improves the classification performance under the condition of low signal-to-noise ratio.