Semantic-aided Parallel Image Transmission Compatible with Practical System

TL;DR

The paper tackles the limited efficiency and robustness of traditional SSCC in image transmission by introducing ParaSC, a parallel-stream framework that integrates a DL-based semantic JSCC stream with a conventional SSCC stream. It develops a residual-conditioned rate adaptation and a dynamic PAGNet-driven fusion that allocates channel resources adaptively based on residual information and SNR, achieving improved performance across AWGN and Rayleigh channels. Theoretical analysis via a conditional ELBO formalizes the rate-distortion tradeoff, and extensive experiments demonstrate ParaSC’s superior reconstruction quality with lower semantic overhead and favorable complexity, indicating strong potential for real-world deployment. This approach provides a practical pathway to combine semantic communications with existing coding architectures, enhancing resilience to channel fluctuations while preserving compatibility with current systems.

Abstract

In this paper, we propose a novel semantic-aided image communication framework for supporting the compatibility with practical separation-based coding architectures. Particularly, the deep learning (DL)-based joint source-channel coding (JSCC) is integrated into the classical separate source-channel coding (SSCC) to transmit the images via the combination of semantic stream and image stream from DL networks and SSCC respectively, which we name as parallel-stream transmission. The positive coding gain stems from the sophisticated design of the JSCC encoder, which leverages the residual information neglected by the SSCC to enhance the learnable image features. Furthermore, a conditional rate adaptation mechanism is introduced to adjust the transmission rate of semantic stream according to residual, rendering the framework more flexible and efficient to bandwidth allocation. We also design a dynamic stream aggregation strategy at the receiver, which provides the composite framework with more robustness to signal-to-noise ratio (SNR) fluctuations in wireless systems compared to a single conventional codec. Finally, the proposed framework is verified to surpass the performance of both traditional and DL-based competitors in a large range of scenarios and meanwhile, maintains lightweight in terms of the transmission and computational complexity of semantic stream, which exhibits the potential to be applied in real systems.

Figures (14)

• Figure 1: An overview of the ParaSC framework. The blue line stands for the conventional branch while the orange line represents the semantic branch, with the transmitted streams denoted by dash-dotted lines. The grey dotted line shows the SNR feedback link. "CE&Mod", "Demod&CD", "RA Encoder" and "RA Decoder" represent the channel encoding plus modulation, demodulation plus channel decoding, rate-adaptive encoder, and rate-adaptive decoder, respectively.
• Figure 2: (a) The structure of the semantic encoder, where the left pathway processes image and the right pathway processes image residual. (b) Visualization of the residual attention mechanism. Compared to $\hat{\bm{y}}_j$, the high-frequency components are highlighted in $\bm{y}_{j+1}$.
• Figure 3: The architecture of the semantic decoder, which includes a novel PAGNet in the up-sampling block. The "DownConv" and "TransConv" denote the convolution layer and transpose convolution layer, respectively.
• Figure 4: Block diagram of PAGNet in the $\ell$-th up-sampler, where the left pathway deals with the up-sampled latents (starting with $\bm{v}_0=\bm{\hat{s}}$) and the right pathway processes the down-sampled latents (extracted from decoded image $\hat{\bm{x}}_c$).
• Figure 5: The conditional rate adaptation module in ParaSC, including a conditional entropy model, an RA encoder and an RA decoder. The RA encoder performs rate token encoding and rate transform successively The rate allocation is conditioned on the residual $\bm{x}_r$, which characterizes the compression loss of the conventional image stream.