| With the continuous development and evolution of wireless communication systems,computer technology and related hardware supporting industries,the future beyond 5th-generation communication system or 6th-generation communication system will based on a massive multi-input and multi-output(MIMO)systems as the infrastructure.And in the trend of joint sensing and communication,higher sensing requirements are proposed to support applications such as real-time target localization and tracking,human activity detection,and RF area imaging to achieve a human-centric intelligent living environment.Therefore,in this paper,facing the future joint sensing and communication scenario,based on the antenna array,and improves the real-time,reliability and expandability of the system by studying deep learning algorithms and multi-fusion algorithms for target localization.The following two aspects of multifusion indoor positioning methods are developed:1.This paper proposes an efficient direction of arrival estimation method based on deep learning for the future B5G/6G joint sensing and communication system,which requires high-precision real-time target position sensing.By dividing the algorithm into offline and online phases and organically combining them,we can achieve the improvement of adaptability for different deployment scenarios.Specifically,this paper uses a neural network-based model,using covariance and snapshot calculation for information extraction,and improves the process of model training and updating.The designed algorithm structure can effectively use the pre-trained model to effectively improve the speed and accuracy of estimating the direction of arrival.Simulation results show that the proposed method has lower computational complexity compared to conventional direction of arrival estimation schemes,higher estimation accuracy at low signal-to-noise ratios compared to existing machine learning methods,and can provide the capability of multi-target direction of arrival estimation for future communication systems.2.Considering the application of wireless headsets in VR scenes,which require high accuracy and real-time positioning,it is difficult to meet the demand by using only existing communication systems for positioning.Meanwhile,RGB-D cameras are widely used as motion-driven devices in VR scenes,which have the characteristics of high accuracy,miniaturization and low cost.Based on the existing equipment,this paper proposes a multi-fusion target localization method,which makes use of both visual and wireless domain information,and effectively improves localization accuracy as well as real-time performance by improving the visual domain information extraction process based on deep learning.Specifically,it first performs rough localization through the wireless domain,then extracts valid RGB information based on the target location,and finally reduces the amount of input data to the target detection network.This algorithm can effectively improve the positioning accuracy and speed.Simulations show that the proposed multi-fusion indoor localization scheme is faster than conventional machine vision methods,and can achieve a wider coverage area and high localization accuracy.In addition,the system architecture and implementation of the semi-physical simulation platform are presented to verify the usability of the complete system and the effectiveness of the algorithms in a semi-physical way.Finally,the overall research content is summarized and the postsequential research issues of target location sensing techniques and systems in indoor scenes are prospected. |
