Simulating Spectrum Auctions: A Reinforcement Learning Approach

Spectrum allocation is a complicated and multifaceted matter that necessitates effective methods for balancing economic efficiency, equitable distribution, and income generation. This study uses reinforcement learning approaches to replicate the Simultaneous Multiple Round Auction (SMRA) format, a widely used format for spectrum auctions. Q-Iearning and Deep Q-Iearning (DQN) are used to simulate the bidding behavior, allowing artificial agent bidders to bid based on previous actions in previous rounds of the auction and competitor behavior. The findings showcase that reinforcement learning can improve the understanding of strategic decision-making in auctions, reducing exposure risk and assuring more efficient spectrum allocations. Compared to traditional simulation methods like using random generators for the bids of the agents, the simulation of reinforcement learning-driven auctions improves the understanding of how an auction can be designed to endure both economic efficiency and allocation efficiency. This study contributes to the understanding of SMRA auction intricacies, with an emphasis on how reinforcement learning artificial bidder agents can optimize bidding strategies over time in a way that mimics actual bidders. Results mark significant progress in understanding how bidders act in simultaneous multiple-round auctions.