SDR Demonstration of Signal Classification in Real-Time Using Deep Learning
Christopher Gravelle, Ruolin Zhou · 2019
In this paper, we demonstrate a software defined radio (SDR) prototype with the capability of signal classification in real-time. Detection, classification, and characterization of wireless signals is a key step to efficiently utilize and effectively share spectrum for enabling SDR-based cognitive radio and intelligent radio. Convolutional neural network (CNN), a deep learning (DL) algorithm, is trained, tested, and used for wireless signal modulation classification. Two types of radio frequency (RF) front-ends, ADALM- PLUTO and Universal Software Radio Peripherals (USRP), are used to transmit and receive over-the-air signals in the demonstration. A classification accuracy of 95.5% is achieved using ADALM-PLUTO; and 96.25% is achieved using USRP N210 with SBX daughterboard.