Abstract

Today with the growing demand for more data transmissions and increased network capacities, cognitive radio technology is ever more relevant. Traditional static spectrum allocation is no longer a feasible option. Through dynamic spectrum access, cognitive devices are able to tap into unused licensed spectrum bands. Thus, improving the spectrum utilization efficiency and fueling spectrum scarce applications. Cognitive Radio (CR) networks consist of smart radio devices that have the ability to sense and adapt to the rapidly changing radio environment. A cognitive device goes through a process of intelligent decision-making, which intrinsically shapes them into smart devices.

Motivated by the superior performance of machine learning in various research paradigms, a cooperative Secondary Network (SN) is proposed that operates under a hybrid underlay-interweave access model. By taking advantage of both access models, the SN maximizes its throughput. A detection problem is formulated for each access model and Machine Learning (ML) techniques are applied to the SN, namely Gaussian Mixture Model (GMM), Support Vector Machine (SVM), and Naive Bayes' (NB) to classify the state of the channel. The multi-class SVM (MSVM) algorithm is reformulated and used to further classify the state of each primary user in the network. The performance of the hybrid network is evaluated based on the Receiver Operating Characteristics (ROC) and classification accuracy. In addition, we show that the accuracy of the MSVM is improved through the cooperation of the secondary users. Our results show that the proposed ML-based hybrid model is robust to low Signal-to-Noise Ratio (SNR) environments, and yields an improved performance compared with traditional cooperative sensing techniques. Moreover, we show that the Gaussian SVM surpasses other proposed learning algorithms achieving as high as an 80% detection rate with as low as 10% false alarm.

Energy detection-based spectrum sensing, relies on measuring the energy level in the spectrum, and accordingly deciding the current occupancy state of the channel. Therefore, CR devices are required to determine the corresponding channel state given a measured energy level. CR networks that use supervised learning techniques to perform the sensing task require data examples of energy levels and the corresponding channel state for training purposes , i.e., labeled data. Having readily available labeled data is a complex task for CR networks, since it requires cooperation from both primary and secondary users. Such cooperation violates the ground rules for the interweave and underlay CR access models. Tackling the problem of labeled data scarcity in practical CR applications, we propose a two-stage learning framework for cooperative spectrum sensing. The algorithm combines the superior performance of the SVM algorithm and low cost training data of the GMM. Thus, rendering the two-stage learning framework suitable for practical CR applications. Finally, a system model is proposed and the performance of the system is evaluated based on the ROC for its upper and lower performance bounds. Additionally, our results show that the two-stage learning attains a higher detection performance compared with using the GMM algorithm.