基于神经网络的室内单站精确被动定位技术研究

Research on indoor single-station precise passive positioning technology based on neural network

为了提高室内环境下对目标的定位精度,提出一种室内单站精确定位技术.该技术利用室内电波传播多径效应构成的复杂信道信息,基于机器学习,构建卷积神经网络架构,通过卷积提取不同位置目标到达接收传感器的多径时延特征信息;然后通过多层全连接层深度神经网络的模型训练,将基于复杂信道的定位问题转化为回归模型的问题,建立信道指纹与位置之间的非线性关系来完成被动定位.训练和仿真结果表明,在室内复杂电波传播环境下,基于神经网络的室内单站精确定位技术能够实现单接收站情况下对目标的精确定位.本文主要对3×3网格大小的金属散射体进行定位,接收站位于室内时,平均定位误差为0.621个网格(12.42 cm);接收站位于室外时,能够分别实现信噪比20 dB、30 dB、40 dB情况下44.09 cm、21.42 cm、20.96 cm的平均定位误差.本文方法为室内复杂环境下的目标定位提供了一种新的定位方法.

In order to improve the positioning accuracy of targets in the indoor environment,a single-station precise passive positioning technology is proposed in this paper.Based on the complex channel information formed by the multipath effect of indoor radio wave propagation,a convolutional neural network(CNN)architecture is constructed,which can extract the multipath delay characteristic information from different location targets to the receiving sensor.Then,a deep network with several dense layers is designed to train the convoluted data,in which way,we transform the location problem based on complex channels into the problem of regression model and establish the nonlinear relationship between signal fingerprint and position,and finally the passive location of the target can be realized.The training and simulation results show that in the indoor environment of complex radio wave propagation,the method we proposed can accurately locate the target in the case of a single receiving station.In this paper,we mainly locate the metal scatterer with the size of 3×3 grids.When the receiving sensor is located indoors,the average positioning error is 0.621 grid(12.42 cm).When outdoors,the average positioning error is 44.09 cm,21.42 cm,20.96cm under the SNR of 20,30,40 dB,respectively,which provides a new method for indoor target positioning of complex environments.