Intelligent Spectrum Sensing and Access Using Deep Learning for Cognitive Communication Systems

Wireless communications are expanding rapidly; as a result, at times, the radio frequency spectrum will be overcrowded (some bands used heavily) while other bands remain underutilized. An investigation into an intelligent spectrum sensing and access framework utilizing deep learning for cognitive communication systems has been conducted. The proposed system utilizes a convolutional neural network to identify whether a given spectrum band is occupied or vacant under various levels of interference. The CNN was trained and tested on synthetic wireless signal data generated within a simulated environment. Results from this research demonstrated the use of the proposed CNN resulted in higher detection probabilities and lower false alarm rates than those produced by energy detection methods. When utilized in conjunction with intelligent decision making processes related to access, the CNN also improved spectrum usage efficiency. Although the signals used to generate the training data were simulated, results show that deep learning can enhance cognitive communication systems efficiency.