Abstract
In the area of radio communications, spectrum allotment is a crucial task. Modulation identification plays a major role or character in the wireless communication system. Automatic Modulation Classification (AMC) is a crucial segment in Cognitive Radio (CR) to identify the neighborhood dischargers to prevent the radio interference and to increase the spectrum performance. The goal of the AMC is to recognize the modulation types (classes) of received signals in CR without the prior information of channel and signal. The approach is to use the Cycle-Consistent Adversarial Networks (CycleGAN) for data transformation from the lower SNR signal to the higher SNR value and observe the experimental result. In this project the dataset used is RADIOML 2018.01A (new) dataset which is from 2018 and presently it is accessible from DeepSig website. The data set consist of twenty-four analog and radio modulation type signals (classes) each of which contains twenty-six variety of SNR values ranges from -20dB to 30dB. This project observed that accuracy to predict the signal modulation increases when lower SNR signal values is transformed to any higher SNR signal value using CycleGAN.