基于深度残差收缩网络的时空多通道信号调制方法

打开文本图片集

中图分类号：TN911.3 文献标志码：A

Abstract： The core of automatic modulation recognition（AMR） technology lies in the utilization of advanced signal processing techniques and deep learning algorithms to extract features and conduct pattern recognition of wireless communication signals，so as to accurately determine the modulation mode adopted by the signals. However，in the research of AMR technology，one of the main challeng es currently faced is the limitation of single-feature recognition and the insuficiency of the performance of a single modulation recognition network.To address this，a spatio-temporal multi-channel signal modulation method based on the deep residual shrinkage network （DRSN） for processng complex multi-channel signals is designed.An inteligent modulation recognition method based on multi-modal feature fusion is proposed.This method first converts the original IQ signal into an amplitude-phase （AP） bimodal feature space through orthogonal transformation，aiming to explore and analyze the differential features of diferent components of the signal.In the feature processing stage，a multi-branch parallel convolutional structure is adopted to extract features from the amplitude and phase signals respectively. Combining the multi-scale feature learning ability of the DRSN and the modeling advantage of the gated recurrent unit（GRU） for the long-term dependence of time series，a deep spatio temporal multi-channel feature fusion network is constructed.Experimental results show that this method performs excellently in the modulation recognition experiments conducted on the current opensource dataset RML2O16.1OA. In the complex electromagnetic environment with a signal-to-noise ratio（SNR） range of -18～20dB ，the classification accuracy is 7 percentage points higher than that of the traditional convolutional neural network （CNN） method， and the accuracy is 3.7 percentage points higher in the environment with a low SNR，verifying the effectiveness of the spatio-temporal multichannel design in complex electromagnetic environments.

Key words：AMR; feature fusion; data driven; deep learning

0 引言

自动调制识别（AMR）作为无线通信系统中的关键技术，其核心目标是从接收信号中快速、准确地辨识调制方式。（剩余10422字）