In this study, we design and implement two algorithms for dynamic spectrum access (DSA) that are based on survival analysis. They use a non-parametric estimate of the cumulative hazard function to predict the remaining idle time available for secondary transmission subject to the constraint of a preset probability of successful completion. In addition to theoretical performance analysis of the algorithms, we evaluate them using data collected from a Long Term Evolution (LTE) band to model primary user activity to demonstrate their effectiveness in real-world scenarios, even at fine time scales. The algorithms are run in different configurations, i.e., they are trained and run on a few combinations of data sets. Our results show that as long as the cumulative hazard functions are fairly similar across datasets, the algorithms can be trained on one day's dataset and run on that of another day's without any significant degradation of performance. The algorithms achieve fairly high white space utilization and have a measured probability of interference that always stays below the preset threshold.
Citation: IEEE Transactions on Cognitive Communications and Networking
Pub Type: Journals
Dynamic spectrum access, spectrum sharing, survival analysis, hazard function