Deep Learning Based Performance of Cooperative Sensing in Cognitive Radio Network

Abstract

In cooperative spectrum sensing in cognitive radio network for the detection of primary user (PU), the detection in classical methods solely depend on signal power and threshold. The selection of threshold is important issue which defines the level of accuracy of detection of PU. This paper focuses on machine learning based prediction of presence of PU based on recorded data training which also shows solution for the problem of various signal strength confusing issues. The model is tested using support vector machine (SVM) based linear binary classifier for combinations of recorded signal strengths from simulated experimental data. The deep learning based method is also tested using recurrent neural network configured using long short term memory (LSTM) and gated recurrent unit (GRU) layers in the model. The performance is compared for the accuracy of PU detection and deep learning approach shows better performance.