Convolutional Neural Network for Cooperative Spectrum Sensing with Spatio-Temporal Dataset

Spectrum sensing can be considered as a classification problem in machine learning. Adaptability is one of the key features one looks for in the classification model and neural networks aid in achieving it. In this paper, we consider the spatio-temporal system model of cooperative spectrum sensing scenario for which a convolutional neural network (CNN) is trained to classify the sensed signal from primary user as available/busy. We have synthesized the necessary dataset for different scenario depending on how the primary users are placed with respect to a set of secondary users and also for different noise floor regimes. Performance is evaluated in terms of the accuracy and test loss for different scenario. The CNN built is very minimalistic in architecture when the scenario is simple and is slightly modified with the scenario or noise-floor. The accuracy of classification is highest when the noise floor is -124dBm and the primary user is surrounded by a good number of sensing secondary users.