Auction-Based RIS Allocation With DRL: Controlling the Cost-Performance Trade-Off

Researchers Martin Mark Zan and Stefan Schwarz have published a paper proposing an auction-based system for dynamically allocating Reconfigurable Intelligent Surfaces (RISs) in next-generation wireless networks. In their model, RIS units—smart surfaces that can reflect and manipulate radio waves to improve signal quality—are deployed at cell edges and operated by an independent entity. Multiple base stations then compete to lease these shared RISs using a simultaneously ascending auction format, where each station bids based on the estimated utility of acquiring additional surfaces.

To optimize this complex, multi-agent bidding process, the researchers integrated Deep Reinforcement Learning (DRL). Each base station employs a DRL agent that learns to maximize network performance (spectral efficiency) while strictly adhering to predefined budget constraints. Their simulations in clustered cell-edge environments show that this RL-based bidding significantly outperforms traditional heuristic strategies. A key innovation is a tunable parameter that controls the bidding aggressiveness of the RL agents, allowing network operators to flexibly manage the trade-off between improved performance and infrastructure cost. This work highlights the potential of combining economic mechanisms like auctions with adaptive AI to manage scarce, shared physical resources in future 6G systems.