Research on OFDM Modulation Recognition Method Based on High-Order Cyclic Cumulants and Neural Networks

• Published in: IEEE International Conference on Power, Intelligent Computing and Systems (Online) pp. 712 - 716
• Main Authors: Xiao, Qingxin, Xie, Wenjun, Li, Longqing, Yuan, Shudong, Zhao, Yongjie
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26.07.2024
• Subjects: Accuracy; Binary phase shift keying; Feature extraction; high-order cycle accumulation; Modulation; modulation recognition; neural network; Neural networks; OFDM; OFDM modulation; Robustness; Signal to noise ratio; Wireless communication; Wireless fidelity
• ISSN: 2834-8567
• DOI: 10.1109/ICPICS62053.2024.10796030

With the continuous development of wireless communication technology, orthogonal frequency division multiplexing (OFDM), as an efficient modulation technology, has been widely used in various communication systems, such as Wi-Fi, LTE 5G, etc. In the OFDM system, the receiver needs to accurately identify the modulation mode of the received signal for the correct demodulation and data resolution. Traditional modulation mode recognition methods usually rely on manually designed features and classifiers, and the performance of this method is limited by the feature selection and engineering complexity. To overcome the limitations of traditional methods, this paper proposes a method to identify OFDM signal modulation mode based on the combination of higher-order cyclic accumulative amount and neural network. Our goal is to use neural networks to learn manually extracted feature representations to improve the accuracy and robustness of identification and reduce the complexity of feature engineering. High-order cyclic accumulation can capture the periodic characteristics of signals and has a strong discriminatory ability to the OFDM signals with periodic structure. The neural network is used to learn the characteristics of higher-order extraction, and then make the modulation mode classification. This classical OFDM modulation mode (such as BPSK, QPSK, and 16 QAM) was used as experimental subjects to generate a series of OFDM signals with different modulation modes and signal-to-noise ratios. The performance of the proposed method is compared with the traditional feature engineering method and other machine learning methods, and it is experimentally verified under different SNR conditions. Experimental results show that the proposed method exhibits high discrimination accuracy and robustness under lower SNR and frequency-biased conditions. Compared with traditional methods, neural network-based methods are able to utilize the underlying features of the data more effectively, achieve better modulation recognition performance, and have lower computational complexity and better generalization ability.