An Adversarial Learning Enabled Hybrid Automatic Repeat Request Mechanism for Satellite Remote Sensing Image Transmission

Semantic communication leverages inherent meanings within diverse data contents, increasingly integrated into knowledge-based image communication frameworks, which attracts current research that predominantly targets computational methods while often neglecting transmission mechanisms. Specifically, the issue of quantization noise in semantic compression and transmission remains inadequately addressed by existing approaches, significantly degrading the recovery of transmitted data with essential integrity and utility. In this paper, we propose an adversarial learning-based quantization selective hybrid automatic repeat request (ALQS-HARQ) semantic transmission mechanism for image data in the remote-sensing satellite network, which could resist quantization noise by dynamically configuring the amounts of quantized bits in the encoder of semantic data. In particular, we develop a quantized bit-merging-enabled retransmission scheme with an intelligent decision-maker and a standardized packet header by proposing a novel metric to evaluate the completion degree of semantic recovery. Experimental results show that the proposed mechanism exhibits superior performance in terms of task success rate with less overhead, highlighting its excellent intelligence capability compared to the typical image data retransmission mechanism.