Blind Modulation Classification of Wi-Fi 6 and 5G signals for Spectrum Sensing
Byung-Jun Kim, Christoph F. Mecklenbräuker, Peter Gerstoft · 2023
Classification of modulation of Wi-Fi~6 and 5G downlink (DL) user data signals for spectrum sensing is studied. First, the orthogonal frequency division multiplexing (OFDM) symbol duration and cyclic prefix (CP) length are estimated based on the cyclic autocorrelation function (CAF). We propose a feature extraction algorithm characterizing the modulation of OFDM signals based on the estimated parameters. The algorithm includes removing the effects of a synchronization error and converting the obtained feature into a 2D histogram of phase and amplitude. This histogram is input to a convolutional neural network (CNN)-based classifier. Our system works without knowledge of a carrier frequency, Wi-Fi preamble, or resource allocation of 5G physical channels. We evaluate the classifier's performance with data with various protocol-compliant configurations. Our classifier achieves at least 98% accuracy when SNR is above the value required for data transmission.