Sparse Vector Coding Based Robust Semantic Communication for Dynamic Environment

Semantic communications have attracted great attention for their capacity to reduce transmitted data while maintaining task performance. However, existing semantic communication models primarily rely on end-to-end deep neural network architecture, posing adaptability and compatibility challenges in practical applications with dynamic environment. Inspired by sparse vector coding (SVC), this paper introduces an SVC-based robust semantic communication (SVC-SC) scheme. In this scheme, SVC is integrated for transmitting discrete semantic features derived from feature extraction and vector quantization. SVC is robust to channel changes, and its parameters can be flexibly adjusted based on channel conditions, which avoids the issue of insufficient adaptability caused by the fixed parameters in end-to-end model fitting. Simulation results demonstrate that the proposed SVC-SC scheme is adaptable to various signal-noise ratios (SNRs), and capable of achieving highly reliable semantic transmission as well as dynamic control of transmission rate.