Deep Learning Techniques for Cooperative Spectrum Sensing Under Generalized Fading Channels

Abstract

We consider the cooperative spectrum sensing problem in cognitive radios as a deep learning-based classification problem, under generalized fading scenarios. In particular, we carry out a performance comparison of well-known deep learning architectures such as deep neural networks, convolutional neural networks (CNN), long short term memory (LSTM) networks, CNN-LSTM networks and gated recurrent units (GRU). The features selected are maximum eigenvalue, energy statistic and maximum-minimum eigenvalue of the received sample correlation matrix. Through experimental studies, we show that GRU marginally outperforms other architectures, and usage of the maximum eigenvalue feature yields the best performance in terms of classification accuracy. Further, the variation in the accuracy performance of the GRU architecture with parameters such as the number of sensors, number of observations and fading parameters are discussed.