Enhanced Deep Learning Architectures for Spectrum Sensing in Cellular Networks

Abstract

The expansion of 5G technologies and the Internet of Things (IoT) increases the demand for spectrum efficiency. In future smart city and Industrial IoT (IIoT) applications, the number of wireless users and IoT devices will be excessive. The effect will be spectrum congestion. Moreover, the existing wireless technology has security flaws and inadequate service quality. Cognitive Radio (CR) technology intends to enhance the functioning of the existing system and meet the growing bandwidth needs of users. Spectrum awareness with identification of various signal patterns, is crucial in a cellular system environment. In this work, two deep neural network architectures are presented to distinguish 5G NR (new Rradio) signals from Long-Term Evolution (LTE) signals. This paper presents AlexNet and SqueezeNet architectures for the classification of NR signal with LTE signal. The analysis is conducted by training the classifiers with three distinct optimizers, including RMSprop (root mean squared propagation), ADAM (adaptive moment estimation) and SGDM (stochastic gradient descent with momentum), In addition, performance study is conducted at three distinct training frequencies to assess the classifiers’ superiority.