多天线定位算法研究

发布时间：2018-08-17 10:59

【摘要】：基于Wi-Fi的室内定位常采用RSSI作为定位参量,而RSSI受环境和硬件设备的影响大。因此,本论文采用能有效的表征定位点频率和空间特征的CSI作为定位参量和机器学习算法中的kNN算法作为定位算法,来实现更为准确的室内定位。本文主要研究内容如下:(1)本文研究了Wi-Fi室内定位采用的定位方法,包括现有的室内定位估计方法和定位算法。在分析定位估计方法时,比较了在Wi-Fi室内定位中使用RSSI与CSI作为定位参量的优劣,给出本论文采用CSI作为定位参量的原因。在确定使用CSI作为定位参量后,研究和改进了基于CSI的定位估计算法。(2)在定位算法的选择上,不同于传统的三角质心法、双曲线法和最小二乘法这三种算法,介绍了机器学习算法中的kNN算法和Bayes算法,并对其在基于CSI时的定位性能进行了实验分析。实验结果表明,kNN算法的定位性能优于Bayes算法。其中,定位性能的评估包括平均定位误差和误差累计分布函数CDF。(3)建立离线阶段训练指纹库,这对系统的定位效果有着重要的影响。在CSI信号采集后,对CSI数据不同的处理方式和定位参量的提取是建立指纹库的重点。因此,本文提出了不同的算法对获取的CSI进行处理并采用PCA对CSI数据进行降维得到新的定位参量。对以上两种定位估计算法进行了比较分析,结果表明,这两种算法都较传统的处理方式具有更优的定位效果。同时,采用PCA处理后得到的CSI特征值作为定位参量时的定位性能达到最优。通过仿真工具和实验平台,讨论不同实验环境以及不同训练数据对最终定位性能的影响。本论文在理论研究的基础上,利用Matlab分析在不同定位算法和定位参量时,各个因素对定位性能的影响。最终实验结果表明,平均定位精度在论文给定实验条件下可以达到0.863m的定位精度,较传统基于CSI的算法提高了20%。
[Abstract]:RSSI is often used as the positioning parameter in indoor positioning based on Wi-Fi, and RSSI is greatly affected by environment and hardware equipment. Therefore, in this paper, CSI, which can effectively represent the frequency and spatial characteristics of the location points, is used as the location parameter and the kNN algorithm in the machine learning algorithm is used as the location algorithm to achieve more accurate indoor positioning. The main contents of this paper are as follows: (1) this paper studies the localization methods used in Wi-Fi indoor positioning, including the existing indoor location estimation methods and localization algorithms. When analyzing the location estimation method, the advantages and disadvantages of using RSSI and CSI as positioning parameters in Wi-Fi indoor positioning are compared, and the reason why CSI is used as location parameter in this paper is given. After using CSI as the location parameter, the location estimation algorithm based on CSI is studied and improved. (2) in the selection of location algorithm, it is different from the traditional tripod center method, hyperbolic method and least square method. This paper introduces the kNN algorithm and Bayes algorithm in machine learning algorithm, and analyzes the localization performance of the machine learning algorithm based on CSI. Experimental results show that the location performance of KNN algorithm is better than that of Bayes algorithm. The evaluation of location performance includes mean location error and cumulative error distribution function (CDF). (3) Establishment of off-line training fingerprint database, which has an important impact on the positioning effect of the system. After the acquisition of CSI signal, the key point of establishing fingerprint database is to extract different processing methods and location parameters of CSI data. Therefore, different algorithms are proposed to process the acquired CSI and to reduce the dimension of the CSI data by PCA to obtain the new location parameters. The comparison and analysis of the above two algorithms show that the two algorithms have better localization effect than the traditional methods. At the same time, the location performance is optimized when the CSI eigenvalue obtained by PCA processing is used as the location parameter. Through simulation tools and experimental platforms, the effects of different experimental environments and different training data on the final positioning performance are discussed. On the basis of theoretical research, Matlab is used to analyze the influence of various factors on location performance in different localization algorithms and parameters. The final experimental results show that the average positioning accuracy can reach 0.863 m under given experimental conditions, which is 20% higher than the traditional algorithm based on CSI.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN92

【参考文献】