AI-Assisted MAC for Reconfigurable Intelligent Surface-Aided Wireless Networks: Challenges and Opportunities

Xuelin Cao, Bo Yang, Chongwen Huang, Chau Yuen, Marco Di Renzo, Zhu Han, Dusit Niyato, H. Vincent Poor, Lajos HanzoIEEE Communications Magazine 2021 · DOI · OpenReview

TL;DR

This paper tackles the nascent challenge of designing Medium Access Control (MAC) protocols for Reconfigurable Intelligent Surface (RIS)-aided wireless networks. It proposes and analyzes three AI-assisted MAC architectures (centralized, distributed, and hybrid) to optimize performance in various RIS-enabled multi-user scenarios, paving the way for more efficient and intelligent wireless communication systems.

What Problem Does This Paper Address?

Reconfigurable Intelligent Surfaces (RISs) are a promising technology for enhancing wireless communication by intelligently manipulating the radio environment. While their benefits at the physical layer (e.g., improved coverage, data rates, and energy efficiency) are well-established, the crucial Medium Access Control (MAC) layer for managing multiple users accessing RIS-enabled channels is still in its early stages. Existing research has primarily focused on single-user or downlink scenarios, leaving a gap in understanding and developing effective MAC strategies for multi-user uplink communications. This paper aims to bridge this gap by exploring various RIS-aided multi-user scenarios and proposing AI-driven MAC architectures to address the complexities of resource allocation, interference management, and efficient access in these novel networks.

Methods

The authors propose three AI-assisted MAC architectures:

  • Centralized AI-Assisted MAC: The Base Station (BS) acts as a central controller, coordinating RIS configurations and resource allocation for all users. This approach leverages deep learning for efficient computation of RIS phase shifts and resource management, aiming to satisfy Quality of Service (QoS) demands.
  • Distributed AI-Assisted MAC: Each user independently manages its access and computes RIS configurations using reinforcement learning (RL). This decentralized approach is designed for low-latency requirements and can handle unpredictable user demands, with RIS controllers acting passively.
  • Hybrid AI-Assisted MAC: This framework combines elements of both centralized and distributed approaches, offering flexibility by allowing users to switch between scheduled and competing transmission modes within a single frame. Different cases of hybrid designs are presented, integrating scheduled and competing transmissions, or handling requests and scheduled transmissions.

The paper also categorizes existing RIS-aided MAC protocols based on their scenarios, objectives, and key features, providing a comprehensive overview of the state-of-the-art.

Datasets

Not discussed in the paper. The paper focuses on proposing and evaluating theoretical frameworks and simulation-based performance, rather than using pre-existing datasets.

Key Insights and Ideas

  • The MAC Layer is the Next Frontier for RIS: While physical layer benefits of RIS are clear, unlocking their full potential hinges on intelligent MAC layer design.
  • AI is Crucial for RIS MAC Complexity: The dynamic and complex nature of RIS-aided networks necessitates AI (deep learning and reinforcement learning) to efficiently manage resources, estimate channels, and configure RISs in real-time.
  • Centralized vs. Distributed Trade-offs: Centralized AI-MAC excels at meeting strict QoS, while distributed AI-MAC is better suited for low-latency and unpredictable environments. Hybrid approaches offer a balance.
  • RIS Configuration as a Classification/Regression Problem: Depending on whether RIS phase shifts are discrete or continuous, AI can be applied as a classification or regression problem, respectively.
  • Scalability Challenges: As the number of RISs increases, the computational complexity for centralized approaches grows, making distributed or hybrid solutions more favorable.

Key Results

The paper presents simulation results demonstrating the performance of the proposed AI-assisted MAC frameworks.

  • System Throughput: AI-assisted MAC schemes consistently outperform their non-AI counterparts, with AI reducing computing time. Centralized MAC achieves higher throughput with an increasing number of transmitters, while distributed MAC performs best at low transmitter counts. Hybrid MAC (Case 1) falls in between.
  • Energy Efficiency (EE): EE generally decreases with an increasing number of RISs due to higher computational complexity. For a small number of RISs (1-2), centralized AI-MAC shows better EE. However, as the number of RISs grows (4-16), distributed AI-MAC gains an edge due to lower overhead compared to centralized approaches.

Limitations and Future Work

  • Limited Datasets: The paper acknowledges that collecting sufficient labeled data for training AI models in RIS-aided networks is a significant challenge. Future work could explore semi-supervised, transfer, or autonomous learning to overcome this.
  • Privacy and Security: Centralized learning may involve privacy-sensitive data. Distributed learning, particularly federated learning, is suggested as a promising direction for enhancing privacy and security.
  • Spectrum Sensing: Accurate spectrum sensing is critical for distributed AI-MAC to avoid collisions. Conventional methods struggle with hidden/exposed terminal problems, suggesting a need for intelligent spectrum learning with RIS cooperation.
  • Advanced AI Models: The authors call for a community effort to develop advanced AI computing models specifically for MAC design in RIS-aided networks.
  • mmWave and THz Communications: Extending AI-assisted MAC solutions to mmWave and THz bands is highlighted as an important future research area.
  • Privacy-Preserving and Smart Sensing: Developing privacy-preserving AI-assisted MAC and incorporating smart sensing capabilities are identified as exciting and challenging future directions.

Summary updated

Abstract

Recently, significant research attention has been devoted to the study of reconfigurable intelligent surfaces (RISs), which are capable of reconfiguring the wireless propagation environment by exploiting the unique properties of metamaterials-based integrated large arrays of inexpensive antennas. Existing research demonstrates that RISs significantly improve the physical layer performance, including the wireless coverage, achievable data rate and energy efficiency. However, the medium access control (MAC) of multiple users accessing an RIS-enabled channel is still in its infancy, while many open issues remain to be addressed. In this article, we present four typical RIS-aided multi-user scenarios with special emphasis on the MAC schemes. We then propose and elaborate upon centralized, distributed and hybrid artificial-intelligence (AI)-assisted MAC architectures in RIS-aided multi-user communications systems. Finally, we discuss some challenges, perspectives and potential applications of RISs as they are related to MAC design.
BibTeX
@article{cao2021ai,
  author = {Xuelin Cao and Bo Yang and Chongwen Huang and Chau Yuen and Marco Di Renzo and Zhu Han and Dusit Niyato and H. Vincent Poor and Lajos Hanzo},
  title = {{AI-Assisted MAC for Reconfigurable Intelligent Surface-Aided Wireless Networks: Challenges and Opportunities}},
  journal = {IEEE Communications Magazine},
  year = {2021},
  doi = {10.1109/mcom.001.2001146},
}
Figure