A Deep Joint Source Channel Coding Scheme with Adaptive HARQ for Image Transmission

To address the insufficient image restoration quality of Deep Joint Source-Channel Coding (DJSCC) systems under low signal-to-noise ratio (SNR) conditions, this paper proposes an enhanced system called DJSCC-H, which integrates a Hybrid Automatic Repeat Request (HARQ) mechanism. By designing a HARQ neural network module (HARQnn) that leverages the Log-Likelihood Ratio (LLR) of transmitted data, channel SNR, and a custom retransmission penalty factor, the paper enables the module to adaptively generate retransmission probabilities and optimize the weighted merging strategy for multi-round transmission data. As a result, this module effectively improves transmission quality under low SNR conditions. Experimental results show that in AWGN channels, when the SNR is 0-5 dB, the DJSCC-H system achieves maximum performance improvements of 18.62% in Peak Signal-to-Noise Ratio (PSNR) and 7.79% in Structural Similarity Index (SSIM) for image reconstruction compared to the baseline DJSCC system, verifying the effectiveness of deep integration of the retransmission mechanism with DJSCC. This paper provides an enhanced scheme for image transmission under low-reliability channels.