WiDetect: Robust Motion Detection with a Statistical Electromagnetic Model论文阅读笔记

WiDetect: Robust Motion Detection with a Statistical Electromagnetic Model

FENG ZHANG, University of Maryland, College Park and Origin Wireless, Inc.
CHENSHU WU, University of Maryland, College Park and Origin Wireless, Inc.
BEIBEI WANG, University of Maryland, College Park and Origin Wireless, Inc.
HUNG-QUOC LAI, Origin Wireless, Inc.
YI HAN, Origin Wireless, Inc.
K. J. RAY LIU, University of Maryland, College Park and Origin Wireless, Inc.

Contribution

core contributions are three folds:
(1) We build the first statistical model of radio signal propagation indoors that takes all scattering multipath components into consideration for motion detection, which is fundamentally different from previous
geometrical models and underpins robust whole-home detection with a single link;
(2) We design an accurate and robust motion detector based on the proposed statistical model, which is independent of environments, locations, orientations, and subjects, and thus is generic to different scenarios without any training efforts;
(3) We implement a real-time system and deploy it at three different environments for a comprehensive evaluation and long-term study. Extensive experiments demonstrate WiDetect’s superior performance for practical applications.

STATISTICAL MODELING OF CSI

——comprehensively accounting for all possible multipaths indoors
——WiDetect only exploits the measured CSI amplitude for motion sensing

1. 基于观察，文中给出了（1）静态环境中，G可以看成不变的值u加上噪声；（2）动态环境中，G可以看成时变的值u加上噪声；（3）距离LoS近，动作对信号的干扰更大；（4）相同动作对不同子载波的干扰情况不同；

2. Modeling the Measurement Noise （in Sec 3.2）
   （1）静态环境下通过1小时的数据收集验证噪声的特征，发现：i符合均值为0，标准差为σ的高斯分布；ii 不同时刻的噪声相互独立；
   证明过程分三步：用quantile-quantile(Q-Q) plot验证正态分布，用ACF验证均值方差，使用相关系数验证不同子载波的独立性；

3. Modeling the Impact of Motion on CSI
   buildings and rooms can be viewed as reverberation cavities in which they exhibit internal multipath propagations.
   Remark：The sample ACF of the power response G(t, f ) of a CSI measurement H(t, f ) approximates 0 for a static environment without motion and inversely remains above zero in presence of motion, within a sufficiently short time lag. The result holds as long as the time delay is short enough, regardless of the specific environment, motion location and orientation, and subject
   Decomposition of the received electric field: 收到的信号能量可以分解为静态部分和动态部分；
   The intuition behind the decomposition is that each scatterer can be treated as a “virtual antenna” diffusing the received EM waves in all directions as shown in Fig. 3, and then, these EM waves add up together at the receiving antenna after bouncing off human bodies, walls, ceilings, furniture, windows, etc.
   As is known from the channel reciprocity, EM waves traveling in both directions will undergo the same physical perturbations (i.e. reflection, refraction, diffraction, etc.). Therefore, if the receiver were transmitting EM waves, all the scatterers would receive the same electric fields as they contribute to ERx(t, f ). Therefore, in order to understand the properties of ERx(t, f ), we only need to analyze its individual components Ei(t, f ), which is equal to the received electric field by the i-th scatterer as if the Rx were transmitting.
   ACF of the received electric field:
   
   The intuition behind is that it is assumed that the EM waves impinging each scatterer are uniformly distributed in the space and have equal energy. Integrating over all of the waves leads to the oscillating decay behavior of the distribution of the energy in the spatial domain around the scatterer.
   电磁波在传播过程中被均匀分布的障碍物所阻挡而衰减，将所有的波叠加就获得了这种震荡衰减的结果。
   
   通过detrended power responses 在ACF上的表现验证了前面的结论。

WIDETECT DESIGN

前面说了这么多，真正的模型就是通过上面的统计规律，以下面提到的式子作为标准，使用方便测量的物理量进行计算衡量。