基于CNN的超分辨率信道冲激响应室内指纹定位算法  

CNN-based super-resolution channel impulse response indoor fingerprint location algorithm

作　　者：罗开文 俞晖[1] LUO Kaiwen;YU Hui(School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240

出　　处：《上海师范大学学报（自然科学版）》2021年第1期92-100,共9页Journal of Shanghai Normal University(Natural Sciences)

摘　　要：针对室内环境中多径效应影响定位精度的问题,提出了一种基于卷积神经网络(CNN)的室内定位(PI-CNN)算法.以多重信号分类(MUSIC)算法处理后的信道状态信息(CSI)作为特征图像,基于室内环境中不同位置点具有独特多径信息的特点,利用各收发天线间所形成的子信道信息,获得具有更高时间分辨率的多径到达时间,将获取的伪谱信息组成伪谱图像,生成指纹库,再利用CNN进行训练和分类处理.仿真实验证明,在室内环境存在轻微扰动的情况下,该算法具有较好的抗干扰能力.Aiming at the problem that the multipath effect in the indoor environment affected the positioning accuracy,based on a deep convolutional neural network(CNN),pseudo spectral image-CNN(PI-CNN)algorithm was proposed in this paper.Using channel state information processed by multiple signal classification(MUSIC)algorithm as a feature image,based on the unique multipath information of different locations in the indoor environment,the sub-channel information formed between the transceiver antennas was utilized to process the channel state information(CSI)to obtain the multipath arrival time with higher time resolution.The pseudo-spectral information of all antennas at the same sampling point was constructed into pseudo-spectral images to generate a fingerprint library which were used to train the CNN.The simulation experiments showed that the PI-CNN algorithm performed well when dealing with slight disturbance in the indoor environment.

关 键 词：深度卷积神经网络(CNN) 多重信号分类(MUSIC)算法 信道状态信息(CSI) 指纹定位 

分 类 号：TN929.5[电子电信—通信与信息系统]