Sensing-Based Feature Engineering and Asynchronous OFDM Blind Modulation Classification Using SMOTE-DNN

Blind modulation classification (BMC) is a key technology for communication perception intelligence, cognitive radio, and electronic countermeasures. With the wide applications of orthogonal frequency division multiplexing (OFDM) technology in 5G communication and unmanned aerial vehicle (UAV) communication, the BMC of OFDM signals is of great significance. The existing modulation classification (MC) methods of OFDM signals mainly focus on incomplete blind classification, and the receiver still needs to obtain some prior information. The BMC of asynchronous OFDM signals with completely unknown signal parameters and channel information still has some technical challenges. This article proposes a blind classification method for asynchronous OFDM signals and a feature engineering mechanism based on normalized statistical dispersion of amplitude (NSDA) and high-order statistics by designing a perceptual processing method combining discrete Fourier transform (DFT) and self linear convolution (SLC). This helps solve such a problem that the recognizability of OFDM signals is not obvious at a low SNR. At last, a synthetic minority over-sampling technique—deep neural network (SMOTE-DNN) classifier is designed to significantly enhance the classification accuracy of OFDM blind classification. By building a software radio experimental platform, BMC experimental verification is conducted on the OFDM RF signals whose subcarriers are modulated by binary phase shift keying (BPSK), quadrature phase shift keying (QPSK), minimum frequency shift keying (MSK), 16-quadrature amplitude modulation (QAM), 64-QAM, 4-PAM, and 8-PAM. The experimental results indicate that the proposed algorithm can realize BMC of asynchronous OFDM signals without prior information in various scenarios, and the comprehensive classification accuracy reaches 87.5% at a SNR of 0 dB.