Random sampling for effective spectrum sensing in cognitive radio time slotted environment

Abstract

In the context of wireless communication, Cognitive Radio has been proposed to ensure an optimal use of the spectrum. It divides the communication transceivers into two categories. Primary User has the priority to use the spectrum band while secondary user is an opportunistic user that can transmit on that band whenever it is available. In order to determine whether a concerned band is occupied or vacant, spectrum sensing functionality is needed together with an appropriate algorithm able to analyze the produced data in order to understand whether a primary user is transmitting or not. In this paper, we contribute by investigating a narrowband spectrum sensing based on two proposed parameters obtained randomly sampling the signal and a Multilayer Perceptron neural network (MLP) classifier. We show that the use of random sampling is more effective than the use of classical uniform sampling. We analyzed the activity status in the GSM channel by using a synthetically generated signal: the signal has been sampled using both uniform and random sampling, than it has been reconstructed and finally classified with the aim of a neural network based classifier. The performances in terms of Receiver Operating Characteristic curves for different Signal to Noise Ratio are presented.