Adversarial bandit approach for RIS-aided OFDM communication.

Area of Science:

• Wireless Communication
• Signal Processing
• Artificial Intelligence

Background:

• Sixth-generation (6G) wireless systems require advanced technologies like reconfigurable intelligent surfaces (RISs) for diverse service management.
• Current RIS implementations face real-time operation challenges, particularly high-dimensional, fully passive RISs needing extensive channel estimation overhead.
• Semi-passive RIS offers a potential solution by minimizing active elements and reducing pilot requirements.

Purpose of the Study:

• To investigate the feasibility and performance of semi-passive RIS in 6G wireless systems.
• To develop low-training overhead and energy-efficient schemes for RIS-based beamforming.
• To compare proposed methods against existing deep learning-based approaches.

Main Methods:

• Exploration of semi-passive RIS with minimal active elements, requiring only two pilots per channel coherence time.
• Development and application of two adversarial bandit-based schemes for reflection beamforming.
• Analysis of deep learning (DL) model quality using state-of-the-art prediction trends.

Main Results:

• Proposed adversarial bandit schemes demonstrate significant performance gains compared to DL-based methods.
• The semi-passive RIS approach drastically reduces channel estimation overhead.
• The developed DL models show promise for practical RIS deployment.

Conclusions:

• Semi-passive RIS, coupled with adversarial bandit algorithms, presents a viable and efficient solution for 6G wireless systems.
• The proposed schemes offer substantial improvements in training overhead and energy efficiency for RIS beamforming.
• This research paves the way for more practical and scalable RIS implementations in future wireless networks.