Horn Berthold K P
Department of Electrical Engineering and Computer Science, MIT, Cambridge, MA 02139, USA.
Sensors (Basel). 2020 Jul 20;20(14):4027. doi: 10.3390/s20144027.
The IEEE 802.11mc WiFi standard provides a protocol for a cellphone to measure its distance from WiFi access points (APs). The position of the cellphone can then be estimated from the reported distances using known positions of the APs. There are several "multilateration" methods that work in relatively open environments. The problem is harder in a typical residence where signals pass through walls and floors. There, Bayesian cell update has shown particular promise. The Bayesian grid update method requires an "observation model" which gives the conditional probability of observing a reported distance given a known actual distance. The parameters of an observation model may be fitted using scattergrams of reported distances versus actual distance. We show here that the problem of fitting an observation model can be reduced from two dimensions to one. We further show that, perhaps surprisingly, a "double exponential" observation model fits real data well. Generating the test data involves knowing not only the positions of the APs but also that of the cellphone. Manual determination of positions can limit the scale of test data collection. We show here that "boot strapping," using results of a Bayesian grid update method as a proxy for the actual position, can provide an accurate observation model, and a good observation model can nearly double the accuracy of indoor positioning. Finally, indoors, reported distance measurements are biased to be mostly longer than the actual distances. An attempt is made here to detect this bias and compensate for it.
IEEE 802.11mc WiFi标准提供了一种协议，用于让手机测量其与WiFi接入点（AP）的距离。然后，可以根据报告的距离以及AP的已知位置来估计手机的位置。有几种“多边测量”方法在相对开阔的环境中有效。在典型的住宅环境中，信号会穿过墙壁和地板，这个问题就变得更加困难。在这种情况下，贝叶斯网格更新显示出了特别的前景。贝叶斯网格更新方法需要一个“观测模型”，该模型给出在已知实际距离的情况下观测到报告距离的条件概率。观测模型的参数可以使用报告距离与实际距离的散点图来拟合。我们在此表明，拟合观测模型的问题可以从二维简化为一维。我们进一步表明，也许令人惊讶的是，一个“双指数”观测模型能很好地拟合实际数据。生成测试数据不仅需要知道AP的位置，还需要知道手机的位置。手动确定位置会限制测试数据收集的规模。我们在此表明，使用贝叶斯网格更新方法的结果作为实际位置的替代进行“自展法”，可以提供一个准确的观测模型，而一个好的观测模型可以使室内定位的精度几乎提高一倍。最后，在室内，报告的距离测量往往存在偏差，大多比实际距离长。本文尝试检测这种偏差并进行补偿。
