No abstract

The enormous success of wireless technology has recently led to an explosive demand for, and hence a shortage of, bandwidth resources. This expected shortage problem is reported to be primarily due to the inefficient, static nature of current spectrum allotment methods. As an initial step towards solving this shortage problem, FCC 1 opens up for the so-called opportunistic spectrum access (OSA), which allows unlicensed users to exploit unused licensed spectrum, but in a manner that limits interference to licensed users. Fortunately, technological advances enabled cognitive radios, which are viewed as intelligent communication systems that can learn from their surrounding environment by themselves, and adapt their internal operating parameters in real-time also by themselves to improve spectrum efficiency. Cognitive radios have recently been recognized as the key enabling technology for realizing OSA. In this work, we propose a machine learning-based scheme that will exploit the cognitive radios' capabilities to enable effective OSA, thus improving the efficiency of spectrum utilization. Our proposed learning technique does not require prior knowledge of the environment's characteristics and dynamics, yet can still achieve high performances by learning from interaction with the environment.

No abstract

The expected shortage in spectrum supply is well understood to be primarily due to the inefficient, static nature of current spectrum allocation policies. In order to address this problem, Federal Communications Commission promotes the so called opportunistic spectrum access (OSA) to be applied on cognitive radio networks (CRNs). In short, the idea behind OSA is allowing unlicensed users to use unused licensed spectra as long as they do not cause interference to licensed users. In this paper, we present and evaluate learning schemes that allow unlicensed users to locate and use spectrum opportunities effectively, thus improving efficiency of CRNs. We separately consider two models: single and multiple unlicensed user(s). For the latter model, we present two schemes: noncooperative and cooperative Q-learning. All proposed schemes do not require prior knowledge or prediction models of the environment's dynamics and behaviors, yet can still achieve high performance by learning from interaction with the environment. Using simulations, we show that the proposed schemes achieve good performances in terms of throughput and fairness.