Research On Deep Learning Based Blind Detection Of Signal Type

With the continuous development of human society,the future society trends in diverse applications,requirements,business,and "connected intelligence".All of these require a more powerful,smarter,and deeper mobile network.In recent years,due to the vigorous development of artificial intelligence(AI)technology,the blind detection of signal types based on deep learning has very broad application prospects and it can provide strong support for future mobile communications.In this thesis,the smart blind detection algorithm of signal modulation type and index modulation are respectively studied for typical communication scenarios and intelligent reflecting surface(IRS)communication scenarios.In view of the traditional end-to-end communication scenarios,this thesis studies deep-learning-based automatic modulation classification(AMC).Considering the amplitude-frequency and modulation characteristics of wireless modulation signals,this thesis proposes a multi-scale convolutional neural networkbased(MSN)classifier,which employs one-dimensional convolution to extract multi-scale feature maps due to its merits of low computational complexity.Then these multi-scale feature maps are merged by repeated multi-scale fusions,to improve the classification accuracy performance and the robustness to varying SNR environments.Finally,the simulation analysis shows that MSN based classifier is a very cost-effective and robust algorithm.Furthermore,this thesis studies the optimization of the IRS matrix,the detection of the selected receive antenna index and signal.Considering the difficulty of channel estimation and high computational complexity in the IRSassisted communication scenario,this thesis proposes a fully connected neural network-based IRS matrix design algorithm.This algorithm can optimize the quantized passive phase shift of each element at IRS to maximize the downlink received signal-to-noise ratio at selected receive antenna without any channel state information.Then,a convolutional neural network(CNN)based detector is proposed to detect the selected receive antenna and the received signal,which can detect the signal and spatial modulation independently.Simulation results verify the efficiency and advantages of the proposed algorithm.