Deep Learning based CP-OFDM Signal Classification with Data Augmentation

Nowadays, most Radio Access Technologies (RATs) employ one or other variants of Orthogonal Frequency Division Multiplexing (OFDM) modulation. This includes 4G, 5G, WiFi, Bluetooth, DVB-T and some Unmanned Aerial Vehicles (UAV) data links. For this reason, a classifier that differentiates among multicarrier communication signals must be an important component of any spectral monitoring solution. In this work, we present a deep learning (DL) based signal classifier able to differentiate among several Cyclic Prefix (CP) OFDM signals with different time parameters (useful and guard band times). The developed classifier employs a convolutional neural network (CNN) that operates with the FFT of the I -Q samples of the signals. With data augmentation techniques, varying the noise levels and phase offsets of the OFDM signals, the trained network can obtain classification accuracies better than 90 % at SNRs as low as 0 dB, for technologies with different OFDM symbol length.