arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.14018eess.SP
收藏

扩展的通用联合源-信道编码用于数字语义通信:提升信道适应性

Extended Universal Joint Source-Channel Coding for Digital Semantic Communications: Improving Channel-Adaptability

Seoul National University(首尔大学)

Eunsoo Kim, Yoon Huh, Wan Choi

AI总结:

本文提出扩展通用联合源-信道编码框架,通过动态码本生成和超网络归一化提升信道适应性,实现高效语义通信。

摘要:

近年来,基于深度学习(DL)的联合源-信道编码(JSCC)的进展使在动态无线环境中实现高效的语义通信成为可能。在这些方法中,基于向量量化(VQ)的JSCC有效地将高维语义特征向量映射到紧凑的码字索引以进行数字调制。然而,现有方法,包括通用JSCC(uJSCC),依赖于固定、调制特定的编码器、解码器和码本,限制了对细粒度SNR变化的适应性。我们提出了一种扩展的通用JSCC(euJSCC)框架,该框架在单一模型内实现了SNR和调制自适应的传输。euJSCC采用基于超网络的归一化层进行细粒度特征向量归一化,并采用动态码本生成(DCG)网络,根据块级SNR优化调制特定的基码本。为处理由多个相干块组成的块衰落信道,采用了内-外编码器-解码器架构,其中外编码器和解码器捕捉长期信道统计信息,而内编码器和解码器优化特征向量以与块级码本对齐。采用两阶段训练策略,即先在AWGN信道上预训练,再在块衰落信道上微调,以确保稳定的收敛。在图像传输实验中,euJSCC在块衰落和AWGN信道上均优于最先进的信道自适应数字JSCC方案。

英文摘要:

Deep learning (DL)-based joint source--channel coding (JSCC), particularly vector quantization (VQ)-based JSCC, enables efficient semantic communication by mapping high-dimensional feature vectors into compact codeword indices for digital modulation. However, existing methods, including universal JSCC (uJSCC), rely on fixed, modulation-specific encoders, decoders, and codebooks, limiting adaptation to fine-grained channel variations. We propose an extended universal JSCC (euJSCC) framework that enables adaptive transmission across both modulation orders and fine-grained channel conditions within a single model using hypernetwork-based normalization and a dynamic codebook generation (DCG) network that refines modulation-specific base codebooks according to channel quality indicator (CQI) feedback. To support time-varying channels with periodic CQI feedback, euJSCC employs an inner--outer encoder--decoder architecture, where the outer modules capture long-term channel statistics and the inner modules refine feature vectors to align with block-wise codebooks. A two-phase training strategy first pretrains the outer modules and DCG network over additive white Gaussian noise (AWGN) channels to establish stable feature--codebook alignment and then fine-tunes the full model over time-varying channels. Experiments demonstrate that euJSCC consistently outperforms state-of-the-art channel-adaptive digital JSCC schemes in image transmission, while additional evaluations under practical channel and feedback conditions and on a downstream task further validate its effectiveness.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑