Semantic Radio Access Networks: Architecture, State-of-the-Art, and Future Directions

Rui Meng, Zixuan Huang, Jingshu Yan, Mengying Sun, Yiming Liu, Chenyuan Feng, Xiaodong Xu, Zhidi Zhang, Song Gao, Ping Zhang, Tony Q. S. QuekarXiv 2025 · DOI · arXiv

TL;DR

This paper provides a comprehensive blueprint for Semantic Radio Access Networks (SemRAN), a transformative 6G architecture that shifts the focus of wireless communication from transmitting "syntactic bits" to delivering "semantic meaning." By integrating Artificial Intelligence (AI) across all network layers, SemRAN aims to shatter the efficiency bottlenecks of 5G, offering massive gains in spectrum efficiency, interference resistance, and user experience.

What Problem Does This Paper Address?

Traditional Radio Access Networks (RANs) are reaching a point of diminishing returns. To meet the extreme demands of 6G, we’ve historically relied on "brute force" methods: adding more antennas, expanding spectrum, or cranking up power efficiency. However, these evolutionary paths are hitting physical and economic limits.

The core issue is that current networks are "meaning-blind." They treat every bit as equally important, whether it’s a critical autonomous braking command or background noise in a video stream. This bit-level delivery (syntactic communication) is inefficient in complex, resource-constrained environments. The authors argue for a paradigm shift toward Semantic Communication (SemCom), where the network understands the context and intent of data, allowing it to transmit only the essential "meaning" required to complete a task.

Methods

The paper systematically outlines a four-layer architecture for SemRAN and reviews the cutting-edge techniques within each:

  1. Physical Layer (PHY): Moves beyond traditional modulation to Joint Source-Channel Coding (JSCC) and generative AI models. These techniques extract compact semantic features from raw data, making transmissions much more robust to noise and fading.
  2. Data Link Layer (DLL): Introduces Semantic HARQ (Hybrid Automatic Repeat Request), which prioritizes the retransmission of "meaningful" features rather than just corrupted bits, and semantic-aware resource scheduling.
  3. Network Layer: Employs Knowledge Bases (KBs)—essentially the "memory" of the network—to provide context that helps the receiver reconstruct information from minimal data. It also introduces the Age of Semantic Information (AoSI) to measure the "freshness" of meaning rather than just packet latency.
  4. Security Plane: Addresses new vulnerabilities unique to AI-native networks, such as model poisoning, gradient leakage, and adversarial attacks on semantic encoders.

Datasets

As a comprehensive survey and architectural proposal, the paper does not introduce a new experimental dataset. Instead, it references a wide array of data modalities used in the cited literature, including:

  • Text: Natural language processing datasets for semantic text transmission.
  • Images/Video: High-resolution datasets used to evaluate generative reconstruction and JSCC.
  • CSI (Channel State Information): Real-world and simulated wireless channel data for semantic-aware beamforming and feedback.

Key Insights and Ideas

  • Breaking the Shannon Limit: By moving from bit-accuracy to task-accuracy, SemRAN can effectively operate "beyond" traditional Shannon limits, achieving successful communication even when the signal-to-noise ratio (SNR) is extremely low.
  • The "Intellicise" Network: The authors use the term "intellicise" (intelligent and concise) to describe the goal of SemRAN—using AI to transmit the absolute minimum amount of data necessary to convey a specific meaning.
  • Knowledge as a Resource: In SemRAN, a Knowledge Base (KB) is as vital as bandwidth or power. A shared "background" between transmitter and receiver allows for extreme data compression.
  • Meaning-Centric Metrics: Traditional metrics like Bit Error Rate (BER) are replaced or augmented by semantic metrics like Semantic Distortion and AoSI, which better reflect the actual utility of the communication to the end-user or task.

Key Results

While the paper is a survey, it highlights several landmark quantitative results from the field:

  • Bandwidth Efficiency: One cited study (Gao et al.) demonstrated that semantic video transmission could reduce bandwidth consumption by 36.7% compared to traditional methods while maintaining the same perceived quality.
  • Overhead Reduction: Semantic-based image HARQ schemes have shown the ability to reduce transmission overhead by over 60% while improving image perception metrics (LPIPS) by 20%.
  • Robustness: Semantic coding schemes (like DeepJSCC) consistently outperform traditional separation-based coding in "low-SNR regimes," avoiding the "cliff effect" where signal quality drops off catastrophically.

Limitations and Future Work

The authors identify several "bottleneck" areas that require urgent research:

  • Theoretical Framework: We still lack a mathematical "Semantic Information Theory" as robust as Shannon’s original work to guide system design.
  • Model Management: How do we coordinate thousands of different AI models across a global network? Issues of model selection, switching, and cross-vendor compatibility remain unsolved.
  • Hybrid Coexistence: For the foreseeable future, semantic "meaning" traffic will have to share the airwaves with traditional "bit" traffic. Designing a network that handles both efficiently is a major challenge.
  • Security and Privacy: Semantic models are "black boxes" that can be exploited. Protecting the privacy of the data used to train these models and ensuring they aren't "poisoned" by attackers is critical for deployment.

Summary updated

Abstract

Radio Access Network (RAN) is a bridge between user devices and the core network in mobile communication systems, responsible for the transmission and reception of wireless signals and air interface management. In recent years, Semantic Communication (SemCom) has represented a transformative communication paradigm that prioritizes meaning-level transmission over conventional bit-level delivery, thus providing improved spectrum efficiency, anti-interference ability in complex environments, flexible resource allocation, and enhanced user experience for RAN. However, there is still a lack of comprehensive reviews on the integration of SemCom into RAN. Motivated by this, we systematically explore recent advancements in Semantic RAN (SemRAN). We begin by introducing the fundamentals of RAN and SemCom, identifying the limitations of conventional RAN, and outlining the overall architecture of SemRAN. Subsequently, we review representative techniques of SemRAN across physical layer, data link layer, network layer, and security plane. Furthermore, we envision future services and applications enabled by SemRAN, alongside its current standardization progress. Finally, we conclude by identifying critical research challenges and outlining forward-looking directions to guide subsequent investigations in this burgeoning field.
BibTeX
@misc{meng2025semantic,
  author = {Rui Meng and Zixuan Huang and Jingshu Yan and Mengying Sun and Yiming Liu and Chenyuan Feng and Xiaodong Xu and Zhidi Zhang and Song Gao and Ping Zhang and Tony Q. S. Quek},
  title = {{Semantic Radio Access Networks: Architecture, State-of-the-Art, and Future Directions}},
  year = {2025},
  eprint = {2512.20917},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2512.20917},
}
Figure