Applying Periodic Retraining to Survival Analysis-Based Dynamic Spectrum Access Algorithms

In situations of spectrum scarcity, tactical networks must use available spectrum efficiently while still maintaining hierarchical access and minimal interference between users. We present simulation results for survival analysis-based dynamic spectrum access algorithms that were previously described in literature. We demonstrate the efficacy of using much shorter training sequences to build the non-parametric estimate of the cumulative hazard function, which is then used to predict remaining idle time. The algorithms are tested on simulated spectrum occupancy data that features time-varying mean occupied and vacant period lengths. We also introduce periodic retraining in order to adapt to changing channel conditions. Our results clearly demonstrate the benefits of periodically rebuilding the estimate of the cumulative hazard function, which requires data that is already gathered in the course of normal operation.