Wireless Synonymous Image Transmission with Joint Spatial-Power Domain Adaptation

This paper proposes an end-to-end learnable wireless image transmission scheme, with joint spatial-power domain adaptation through spatial synonymity and power allocation. In this scheme, the transmitter employs a spatial synonym-based clustering algorithm to adaptively extract vital semantic information from the latent features of the image for wireless transmission. Afterwards, an adaptive power allocation strategy is adopted to reduce the impact of channel noise. At the receiver, after a corresponding signal detection, a synonym decoder with a prediction model is used to achieve accurate image reconstruction from partial information. Extensive experiments demonstrate that the proposed method can achieve superior bandwidth utilization and image reconstruction quality in Rayleigh fading channels, outperforming existing Deep Joint Source Channel Coding (DJSCC) schemes.