Classification of Primary Users Using Deep Residual Learning

Abstract

The novel idea of cognitive radio (CR) for designing wireless communication devices was developed to mitigate the scarcity problems in the available spectrum. This problem came into light after the rapid development in the field of wireless communication devices. The CR promises to solve the scarcity problem by improving the utilization of the spectrum. This also solves the underutilization of the spectrum in some frequency bands. In general, CR has the ability of learning and adapting to their environment. CR allows secondary users (SU) to share the licensed spectrum of primary users (PU) provided that PU is not subject to interference. Spectrum sensing is a fundamental task of CR which helps it to obtain opportunistic spectrum access to its users. Various methods have been proposed to provide an efficient spectrum sensing. In order to improve the learning ability of CRs and to provide an efficient spectrum sensing method, machine learning algorithms can be applied. The objective is to analyze an efficiently performing machine learning algorithm that can be applied to the extracted data. The extracted dataset contains the PU’s transmission signal patterns as spectrograms, which can be used to obtain labeled classes, i.e., whether the PU is present or not. Following that, a classifier is developed from the analysis of the transmission patterns, which is trained and tested for sensing the PU transmission pattern in a dynamic environment.