Prediction based channel allocation performance for cognitive radio

Abstract

The interdependency, in a cognitive radio (CR) network, of spectrum sensing, occupancy modelling, channel switching and secondary user (SU) performance, is investigated. Achievable SU data throughput and primary user (PU) disruption rate have been examined for both theoretical test data as well as data obtained from real-world spectrum measurements done in Pretoria, South Africa. A channel switching simulator was developed to investigate SU performance, where a hidden Markov model (HMM) was employed to model and predict PU behaviour, from which proactive channel allocations could be made. Results show that CR performance may be improved if PU behaviour is accurately modelled, since accurate prediction allows the SU to make proactive channel switching decisions. It is further shown that a trade-off may exist between achievable SU throughput and average PU disruption rate. When using the prediction model, significant performance improvements, particularly under heavy traffic density conditions, of up to double the SU throughput and half the PU disruption rate were observed. Results obtained from a measurement campaign were comparable with those obtained from theoretical occupancy data, with an average similarity score of 95% for prediction accuracy, 90% for SU throughput and 70% for PU disruption rate.