Optimizing Spectrum Use in Wireless Networks by Learning Agents

The paper examines several distributed mechanisms that can be used in wireless networks consisting of base stations transmitting onto a population of receivers. The overall goal of the algorithms is to optimize a global measure: the sum of capacities of the channels formed between transmitters and receivers in the presence of interference. We introduce a mathematical model of the operation of an OFDM-based wireless network and on this premise we pose the problem of interference in a game theoretic setting. We propose a simple but expressive way of modeling “smart devices” as learning and executing agents and introduce three types of such agents. This part of the work is carried out in the scope of the multi-armed bandit framework. One of the three algorithms we propose is well known and widely studied and two others are new variants of known algorithms seemingly not yet studied. By numerical simulations we show that these mechanisms improve network performance in the considered model. We offer some basic heuristic explanations of this improvement and identify future work.