Channel Usage Patterns and Their Impact on the Effectiveness of Machine Learning for Dynamic Channel Selection

Abstract

The diverse behavior of different primary users (PU) in various spectrum bands impacts a cognitive radio’s ability to exploit spectrum holes. This chapter summarizes the results of our previous studies on the impact of the complexity of primary users’ behavior on the performance of learning algorithms applied to dynamic channel selection. In particular, we characterize the observable spectrum utilization with respect to the duty cycle of the channels and to the complexity of the primary user’s activity. We use the term complexity to refer to the unpredictability associated with the primary user’s wireless resource usage, which we quantitatively characterize using Lempel-Ziv complexity. We evaluate the effectiveness of two learning-based dynamic channel selection algorithms by testing them with real spectrum occupancy data collected in the GSM, ISM, and DECT bands. Our results show that learning performance is highly correlated with the level of PU activity, estimated by the duty cycle, and the amount of structure in the use of spectrum, estimated by the Lempel-Ziv complexity.