一种基于神经网络的OFDM信号识别与符号估计改进方法  

An Improved Method for OFDM Signal Recognition and Symbol Estimation Based on Neural Networks

作　　者：熊刚[1] 张辉[1] 任祥维[1] 胡宗恺[1] XIONG Gang;ZHANG Hui;REN Xiangwei;HU Zongkai(No.30 Research Institute of CETC,Chengdu 610041,China)

机构地区：[1]中国电子科技集团公司第三十研究所,四川成都610041

出　　处：《舰船电子对抗》2023年第5期65-69,共5页Shipboard Electronic Countermeasure

摘　　要：针对正交频分复用(OFDM)信号调制识别与符号估计问题,提出了一种基于深度学习神经网络(DNN)的新方法。该方法通过深入分析OFDM信号的实际传输模型,同时采用优化的Dropout策略防止过度拟合,可适用于多种调制类型数据集的训练,增强了网络学习的泛化能力;另一方面,基于改进的OFDM导频数据训练结构,提高计算效率。该新思路增强了抗噪性能,无需大量的数据先验需求,具有良好的稳健性和工程实用性。仿真结果表明新方法的识别及估计性能比起过去传统思路更优,且可在低信噪比情况下成功实现识别及估计。Aiming at the problems of orthogonal frequency-division multiplexing(OFDM)signal modulation recognition and symbol estimation,a new method based on deep learning neural network(DNN)is proposed.This method deeply analyzes the actual transmission model of OFDM signals and adopts an optimized Dropout strategy to prevent over-fitting,which can be applied to train multiple modulation types of datasets and enhance the generalization ability of network learning;on the other hand,based on the improved OFDM pilot frequency data training structure,the computational efficiency is improved.This new idea enhances the anti-noise performance,does not need a large number of data prior requirements,and has good robustness and engineering practicality.The simulation results show that the recognition and estimation performance of the new method are better than that of the traditional ideas,and the recognition and estimation can be realized under the condition of low signal-to-noise ratio successfully.

关 键 词：正交频分复用信号 深度学习神经网络 调制识别 定时估计 Dropout策略 

分 类 号：TN918.91[电子电信—通信与信息系统]