Predictive Multi-user Dynamic Spectrum Allocation Using Hidden Semi-Markov Model

Abstract

For wideband Cognitive Radio (CR) devices, the availability of primary user (PU) activity prediction model can save sensing energy and time, track spectrum variations, reduce switching overhead, improve spectrum usage and lower interference to PU. Using Universal Software Radio Peripheral (USRP) generated data, this paper has verified the superiority of Hidden Semi-Markov model (HSMM) over the conventional HMM in learning and predicting temporal correlations in PU spectral activity under non-geometric distribution of state durations. Numerical simulations further show that in a coordinated multi-SU/PU scenario, compared to the traditional random Carrier Sense Multiple Access (CSMA), the proposed HSMM-dynamic spectrum allocation (DSA) protocol significantly reduces aggregate interference to PU. Depending on the number of SU, there exists a trade-off between “interference-free coexistence with PU” and “self-existence of a large number of SU”. This problem is addressed further by statistically modeling probability of free channels and using it to optimize the number of SU that can be allocated. The proposed scheme outperforms the conventional sensing based dynamic spectrum allocation (DSA) in terms of reduced PU interference, lower spectrum handoff requirements and higher spectrum utilization efficiency. The predictive optimized HSMM-DSA can be used as an additional layer of intelligence for any CR device for efficient spectrum sharing.