Implementation of CNN-Inception Deep Learning for Cognitive Radio Based on Modulation Classifications

Abstract

With the growing demand for new wireless services and applications, as well as the growing number of wireless users, the available spectrum is becoming increasingly scarce. As a result, the Federal Communications Commission (FCC) explored new ways to manage radio frequency resources. Cognitive radio technology is an innovative radio design philosophy that aims to increase the maximum exploitation of the physical spectrum by harnessing unused and underutilized spectrum in dynamic environments. Modulation recognition is the most important part of the spectrum sensing for cognitive radio. In this paper, we proposed and implement a developed Convolutional Neural Network technique called CNN-inception for modulation classification, the introduced model is a new module inspired by the idea of a naive version of the initiation module. We examine it with two types of inputs first one is the complex (I/Q) time-domain, the second one is the FFT of the signals as features to train the neurons. We test the proposed model by using two popular datasets MIGOU-MOD dataset and the RadioML2016.10a dataset, the results show that we were able to achieve maximum accuracy of 93.22% which is very competitive and better than many other proposed techniques