Large Language Models as Bidding Agents in Repeated HetNet Auction

A team of researchers including Ismail Lotfi, Ali Ghrayeb, and Merouane Debbah has published a groundbreaking paper, 'Large Language Models as Bidding Agents in Repeated HetNet Auction,' accepted at WCNC 2026. The work investigates a major shift in wireless network resource allocation, moving from traditional centralized optimization to a distributed auction-based framework. In this new model, each base station runs its own multi-channel auction, and user equipments (UEs) must strategically decide both which base station to associate with and how much to bid for spectrum, all under budget constraints and through repeated interactions. This transforms the problem from a one-shot optimization into a complex, long-term economic game.

The core innovation is using LLMs as reasoning agents for the UEs. Unlike classical algorithms that use fixed, myopic policies, the LLM agents can analyze historical auction outcomes, anticipate competitor behavior, and adapt their bidding strategies across multiple episodes. Simulation results showed these AI-powered agents consistently outperformed benchmark algorithms, securing higher channel access frequency while using their budgets more efficiently. This research paves the way for deploying lightweight, edge-capable LLMs to create intelligent, decentralized markets for spectrum and other resources in future 6G and beyond networks, where dynamic and efficient allocation is critical.