Recent advances in deep learning have led to increased interest in solving high-efficiency end-to-end transmission problems using methods that employ the nonlinear property of neural networks. These methods, we call semantic coding, extract semantic features of the source signal across space and time, and design source-channel coding methods to transmit these features over wireless channels. Rapid progress has led to numerous research papers, but a consolidation of the discovered knowledge has not yet emerged. In this article, we gather ideas to categorize the expansive aspects on semantic coding as two paradigms, i.e., explicit and implicit semantic coding. We first focus on those two paradigms of semantic coding by identifying their common and different components in building semantic communication systems. We then focus on the applications of semantic coding to different transmission tasks. Our article highlights the improved quality, flexibility, and capability brought by semantic coded transmission. Finally, we point out future directions.

翻译：深度学习的最新进展使得利用神经网络非线性特性解决高效端到端传输问题的方法日益受到关注。这些方法被称为语义编码，它们提取源信号在时空维度中的语义特征，并设计源-信道编码方法以通过无线信道传输这些特征。该领域的快速发展催生了大量研究论文，但尚未形成系统性的知识整合。本文通过概念分类，将语义编码的广泛维度归纳为两种范式，即显式语义编码与隐式语义编码。我们首先聚焦这两种范式，通过识别其在构建语义通信系统中的共性组件与差异性组件展开分析，随后探讨语义编码在不同传输任务中的应用。本文重点阐释了语义编码传输所带来的质量提升、灵活性与能力增强，最后展望了未来研究方向。