利用神经网络逐级缩小定位区域的低复杂度多级干扰源直接定位方法  

Low Complexity Multi-Stage Direct Position Determination of Interference Source Based on Neural Networks to Narrow the Location Area

作　　者：赵晓丹 张国梅[1] 尹佳文 李国兵[1] ZHAO Xiaodan;ZHANG Guomei;YIN Jiawen;LI Guobing(Faculty of Electronic and Information Engineering, Xi’an Jiaotong University, Xi’an 710049, China)

机构地区：[1]西安交通大学电子与信息学部,西安710049

出　　处：《西安交通大学学报》2022年第7期136-144,共9页Journal of Xi'an Jiaotong University

基　　金：地理信息工程国家重点实验室基金资助项目(SKLGIE2020-Z-2-1)。

摘　　要：针对利用直接定位(DPD)方法对全球导航卫星系统(GNSS)终端面临的干扰源进行定位时,在线计算复杂度高的问题,提出了一种利用神经网络逐级缩小定位区域的低复杂度多级干扰源直接定位方法。该方法首先使用多级全连接神经网络(FNN)逐级缩小干扰源所在的区域范围,每一级处理将目标区域均分为两个子区域,并利用预训练的以接收信号功率作为输入特征的神经网络选出干扰源所在的子区域,从而大幅缩小干扰源所在的目标区域;然后在锁定的最终子区域内划分网格并使用DPD方法对干扰源进行精细定位。由于逐级二分处理已将干扰源可能存在的区域大幅缩小,因此有效降低了DPD方法的网格搜索集合大小。仿真结果表明:在较高干噪比条件下(对于压制式干扰通常大于20 dB),所提方法能获得与DPD方法接近的定位性能,而在线计算复杂度相比于DPD方法可降低大约2^(M)倍(M为神经网络级数),定位误差相比于传统基于到达时间差(TDOA)的Chan氏定位方法可降低90%以上。To reduce the high online computation complexity of conventional direct positioning determination(DPD)used to locate interference sources of the global navigation satellite system(GNSS),a low complexity multi-stage direct position determination approach is proposed,in which the neural networks are applied to narrow the location area stage by stage.Firstly,multi-stage fully connected neural network(FNN)is applied to narrow the area where the interference source is located step by step.In each step,the target area is divided into two sub-areas,and the pre-trained neural network with received signal power as input feature is used to identify the sub-area where the interference source is located,thus greatly reducing the target area where the interference source is located.Then,the grids are divided in the locked final sub-area and the proposed DPD is used to fine locate the interference source.Since the possible area where the interference source is located has been narrowed significantly after the step-by-step binary processing,the set size for grid searching of the DPD can decrease efficiently.The simulation results show that compared with the conventional DPD approach,the proposed approach is close to it in performance at high jamming-noise ratio(usually greater than 20 dB for blanket jamming),and the online calculation complexity can be reduced by about 2^(M) times(M is the number of neural networks).Furthermore,the positioning error is more than 90%lower than CHAN algorithm based on the estimated time difference of arrival(TDOA).

关 键 词：全球导航卫星系统 干扰源定位 直接定位 神经网络 

分 类 号：TN972.4[电子电信—信号与信息处理]