RSSI-based Passive Localization in the Wild, at Streetscape Scales

A modern Wi-Fi-enabled device (e.g., a smartphone) can spontaneously emit unencrypted and anonymized signals to the environment in search of an access point. This signal is called a probe request. Since it is freely available in the open air, one can build a sensor from a Wi-Fi adapter to capture the signal. Once captured, its signal strength can be measured in the form of a Received Signal Strength Indicator (RSSI).

RSSI is usually represented as a negative value in the unit of `dBm`. In theory, RSSI measurements are correlated with distance (the closer the Wi-Fi-enabled device is to a sensor, the larger the RSSI measurements). Thus, if multiple sensors are present near a Wi-Fi-enabled device when it emits probe requests, it is theoretically possible to predict the device's location based on statistical modeling of the RSSI measurements. And since smartphones are ubiquitous these days among pedestrians, RSSI-based localization via probe requests becomes a strong candidate for studying pedestrian mobility patterns on a city street in a low-cost, non-intrusive, and privacy-centric manner.

However, due to the fluctuating nature of RSSI measurements (RSSI can be affected by myriads of environmental factors such as humidity, obstruction, multi-path, interference, time, power variation, etc.), it is not trivial to model their behavior concerning device locations. In the literature, we have seen that past researchers had to build their own sensors and testbeds to evaluate new modeling ideas. This process is time-consuming and usually restricted to a controlled or small-scale outdoor environment. Neither is ideal to reveal the true performance of their models. It will benefit the research community if a dataset dedicated to real-world RSSI-based passive outdoor localization is publicly available. Unfortunately, to the best of our knowledge, such a dataset did not exist.

Not anymore! Here we present the West Palm Beach (WPB) dataset, whose sole purpose is for researchers to play with RSSI and location data collected in a real-world setting. We hope this dataset will allow faster iteration of model evaluation and benchmarking and invite more brainpower to tackle the RSSI-based passive outdoor localization problem.

现代支持Wi-Fi的设备（例如智能手机）会自发向环境发射未加密且已匿名化的信号，以搜寻可用接入点。此类信号被称为探测请求（probe request）。由于这类信号在开放空域中可被自由获取，研究人员可通过Wi-Fi适配器搭建传感器来捕获该信号。捕获后，可通过接收信号强度指示器（Received Signal Strength Indicator，RSSI）来测量其信号强度。

RSSI通常以负值形式表示，单位为dBm。理论上，RSSI测量值与设备距离存在相关性：支持Wi-Fi的设备距离传感器越近，其RSSI测量值便越大。因此，若某支持Wi-Fi的设备在发送探测请求时，其周边部署了多个传感器，便可通过对RSSI测量值开展统计建模，理论上实现该设备的位置预测。鉴于当前智能手机在行人群体中普及率极高，基于探测请求与RSSI的定位技术，便成为了一种低成本、非侵入式且注重隐私保护的城市街道行人移动模式研究的优选方案。

然而，由于RSSI测量值具有波动性（易受湿度、遮挡、多径效应、干扰、时间、功率变化等诸多环境因素影响），要建立其与设备位置的关联模型并非易事。在现有研究中，过往研究者往往需要自行搭建传感器与测试平台，以评估新的建模思路。该过程耗时耗力，且通常仅能在受控或小规模的室外环境中开展，均无法充分展现模型在真实场景下的实际性能。若能有面向真实室外RSSI被动定位任务的公开数据集，将对整个研究社区大有裨益。但据我们所知，此前并无此类数据集问世。

如今这一状况已得到改变！本文推出西棕榈滩（West Palm Beach，WPB）数据集，其唯一用途便是供研究者使用真实场景下采集的RSSI与位置数据。我们期望该数据集能够加速模型评估与基准测试的迭代进程，吸引更多研究者投身于基于RSSI的室外被动定位问题的研究当中。