【作者】 王志鹏；

【导师】 王慧强；

【作者基本信息】 哈尔滨工程大学 ， 软件工程， 2022， 硕士

【摘要】 近年来,由于移动网络的蓬勃发展,在生活、贸易和公用事业等方面,无线室内定位的相关技术带来了大量基于位置的应用,移动和普适计算的蓬勃发展激起了人们对准确、可靠的室内定位方案的强烈需求。如今,使用Wi-Fi信号进行室内定位是一种经济的技术。在目前主流的无线信号监测中,大部分的室内定位系统使用接收信号强度(RSS),然而,在复杂的情况下,这些系统由于一些因素比如多径效应,它们的定位效果会受到很大的影响。与RSS不同,来自物理层的信道响应可以用来解决多径效应的问题,由此室内定位中大量使用信道状态信息(CSI)来替代RSS,但其复杂的结构导致计算复杂度增加。因此,既具有RSS的简单性,又保留了丰富的统计位置相关信息(如CSI)的位置指纹是如今室内定位技术的研究重点。本文的研究内容如下:(1)针对粒子滤波算法对CSI信息进行预处理时,其计算开销过大的问题,本文提出了一种基于自适应方差和状态梯度来进行重采样的粒子滤波算法(Adapted Variance and Gradient Particle Filter),简称AVG粒子滤波算法。改进的粒子滤波算法是一种依赖于整个CSI数据集的后向递归算法,通过调节方差的值,并根据梯度,使其能够在接近真实位置分布或在高概率区域内产生粒子,使得重采样阶段可以删除更多不合逻辑的CSI位置采样(权重较小的CSI位置采样),并增加权重较大的粒子数,从而降低计算开销。(2)针对离线指纹数据库的存储量大、在线指纹匹配复杂度高的问题,本文提出了一种基于CSI幅度熵的自回归建模作为位置指纹,这种基于熵的信道使我们能够以新的视角感知室内统计多样性。由于避免了由精确概率密度函数(PDF)估计驱动的海量测量存储,此方式既保持了RSS的结构简单性,又充分利用了最具位置特异性的统计信道信息。这种简单的指纹结构有助于降低模式匹配的复杂度,其信息量丰富的统计实例也有助于位置估计的准确性。最后,本文对提出的定位方法进行了实验验证,对提出的AVG粒子滤波算法和基于CSI幅度熵的自回归建模作为位置指纹的方法,通过实验来测试其效果。本文采用的实验场景都是比较贴近生活的位置,在这些位置进行参考点设置,与现存的一些定位方法进行实验对比,验证了本文提出的定位算法以及指纹数据处理方法具有一定的有效性。更多还原

【Abstract】 In recent years,due to the vigorous development of mobile network,wireless indoor positioning technology has brought a large number of location-based applications in life,trade and public utilities.The vigorous development of mobile and pervasive computing has aroused people’s strong demand for accurate and reliable indoor positioning schemes.Nowadays,indoor positioning using Wi Fi signal is an economical technology.In the current mainstream wireless signal monitoring,most indoor positioning systems use received signal strength(RSS).However,in complex cases,the positioning effect of these systems will be greatly affected due to some factors such as multipath effect.Different from RSS,the channel response from the physical layer can be used to solve the problem of multipath effect.Therefore,channel state information(CSI)is widely used to replace RSS in indoor positioning,but its complex structure leads to increased computational complexity.Therefore,location fingerprint,which not only has the simplicity of RSS,but also retains rich statistical location related information(such as CSI),is the research focus of indoor location technology.The research contents of this paper are as follows:(1)Aiming at the problem that the computational overhead of particle filter algorithm in preprocessing CSI information is too large,this paper proposes an adaptive variance and gradient particle filter(AVG particle filter)for resampling based on adaptive variance and state gradient.The improved particle filter algorithm is a backward recursive algorithm that depends on the whole CSI data set.By adjusting the value of variance and according to the gradient,it can generate particles close to the real location distribution or in the high probability region,so that more illogical CSI location samples(CSI location samples with smaller weight)can be deleted in the resampling stage,and the number of particles with larger weight can be increased,this reduces the computational overhead.(2)Aiming at the problems of large storage capacity of offline fingerprint database and high complexity of online fingerprint matching,this paper proposes an autoregressive modeling based on CSI amplitude entropy as location fingerprint.This entropy based channel enables us to perceive indoor statistical diversity from a new perspective.Because the massive measurement storage driven by accurate probability density function(PDF)estimation is avoided,this method not only maintains the structural simplicity of RSS,but also makes full use of the most location-specific statistical channel information.This simple fingerprint structure helps to reduce the complexity of pattern matching,and its informative statistical examples also help to improve the accuracy of location estimation.Finally,the proposed location method is verified by experiments.The proposed AVG particle filter algorithm and autoregressive modeling based on CSI amplitude entropy are used as location fingerprints to test its effect.The experimental scenes used in this paper are close to life.The reference points are set in these positions.Compared with some existing positioning methods,it is verified that the positioning algorithm and fingerprint data processing method proposed in this paper are effective.更多还原