Frequency Domain Analysis and Convolutional Neural Network Based Modulation Signal Classification Method in OFDM System
Yue Hao, Xianpeng Wang, Xiang Lan · 2021 13th International Conference on Wireless Communications and Signal Processing (WCSP) · 2021
Automatic Modulation Classification (AMC) is widely used in many aspects and occupies a critical position in non-cooperative communication. Recently, deep learning (DL) based AMC algorithms attract more and more attention due to the outstanding performance in modulation recognition. In this paper, we propose a novel convolutional neural network (CNN)-based AMC method that employs frequency domain analysis (FDA) pre-processing and$l_{2}$regularization for the orthogonal frequency division multiplexing (OFDM) systems. Different from traditional algorithms, the proposed algorithm is superior in high accuracy of the classification even at low signal-to-noise ratios (SNRs) owing to pre-processing by FFT. Moreover, the adoption of$l_{2}$regularization effectively suppresses overfitting. Simulation results are given to illustrate that our proposed method has an evidently advantage over traditional methods in classifying BPSK, 4PSK, 8PSK and 16QAM modulation techniques.