基于LM优化BP神经网络的5G室内定位

5G Indoor Positioning Based on LM Optimized BP Neural Network

随着物联网时代的到来,室内定位在人类生活中占有愈发重要的地位。传统的室内定位系统布设成本较高,室内复杂环境往往带来非视距(Non Line of Sight,NLOS)误差和多径干扰,导致系统定位精度较低。随着5G网络的日益普及,使用5G为物联网设备定位成为了可能。针对传统室内定位存在的问题,提出了将Levenberg-Marquardt(LM)算法改善的Back Propagation(BP)神经网络算法用于5G定位结果解算,通过搭建由5G网络基站、实验终端和时间同步系统所构成的实验平台,在室内真实场景中进行了实验。在实验场景下LM-BP神经网络算法静态定位精度达到了0.3457 m,动态定位精度达到了亚米级为0.4511 m,相较传统的Chan算法,LM-BP神经网络模型在提高系统抗NLOS误差能力的同时也能提高室内定位精度。

With the advent of the Internet of Things era,indoor positioning plays an increasingly important role in our lives.Traditional indoor positioning systems have high deployment costs,and complex indoor environments often bring Non-Line-of-Sight(NLOS)errors and multipath interference,resulting in low system positioning accuracy.With the increasing popularity of 5G networks,it is possible to use 5G to locate IoT devices.In view of the problems existing in traditional indoor positioning,a Back Propagation(BP)neural network model improved by the Levenberg-Marquardt(LM)algorithm is proposed to solve the 5G positioning results.An experiment is carried out in an indoor real scene by building an experimental platform consisting of a 5G network base station,an experimental terminal,and a time synchronization system.Experiments show that in the experimental scenario,the static positioning accuracy of the LM-BP neural network algorithm reaches 0.3457 m,and the dynamic positioning accuracy reaches sub-meter level of 0.4511 m.Compared with the traditional Chan algorithm,the LM-BP neural network model improves the system’s resistance to NLOS error as well as the indoor positioning accuracy.