Energy-Constrained Distributed MAC in CR-IoT Networks: A Budgeted Multi-Player Multi-Armed Bandit Approach

The multi-armed bandit (MAB) framework plays a key role in designing medium access control (MAC) strategies for wireless networks where multiple users must select channels without prior knowledge of the statistical characteristics on time-varying wireless links. In this work, we focus on energy-efficient channel access for distributed underlay Cognitive Radio Internet of Things (CR-IoT) networks, where each user has the learning capability and operates with a fixed energy budget. In this network, users do not have prior knowledge of channel quality statistics and cannot communicate with each other. All users operate in a time-slotted manner, where in each slot, they allocate channels and transmit information over their allocated channels. We formulate the distributed MAC problem within the budgeted multi-player multi-armed bandit framework, where channels are modeled as arms and users as players. To maximize the expected traffic volume over the network’s lifetime, we propose a novel distributed online learning MAC strategy that estimates throughput-to-energy ratios for channels and performs channel allocation using a modified Bertsekas auction method. Our theoretical analysis demonstrates that the proposed strategy converges to the optimal MAC performance with logarithmic regret, accounting for users’ energy constraints. To evaluate the effectiveness, we compare the proposed algorithm with existing methods under the 5G urban micro-street canyon channel model, such as the classical UCB algorithm and the random algorithm. Simulation results demonstrate that the proposed strategy achieves a 70% improvement in energy efficiency compared to the random algorithm. This work offers a tailored solution for a variety of practical CR-IoT networks, including underlay low-power wireless LANs, massive IoT D2D networks, and ad hoc sensor networks.