Inter-Mobile-Device Distance Estimation using Network Localization Algorithms for Digital Contact Logging Applications

Figure by Lilly Clark


Mobile applications are being developed for automated logging of contacts via Bluetooth to help scale up digital contact tracing efforts in the context of the ongoing COVID-19 pandemic. A useful component of such applications is inter-device distance estimation, which can be formulated as a network localization problem. We survey several approaches and evaluate the performance of each on real and simulated Bluetooth Low Energy (BLE) measurement datasets with respect to both distance estimate accuracy and the proximity detection problem. We investigate the effects of obstructions like pockets, differences between device models, and the environment (i.e. indoors or outdoors) on performance. We conclude that while direct estimation can provide the best proximity detection when Received Signal Strength Indicator (RSSI) measurements are available, network localization algorithms like Isomap, Local Linear Embedding, and the spring model outperform direct estimation in the presence of missing or very noisy measurements. The spring model consistently achieves the best distance estimation accuracy.

In Connected Health: Applications, Systems and Engineering Technologies

Most modern day phones are equipped with hardware that allows for Bluetooth Low-Energy (BLE) communication with other phones. This can also be used to get a rough distance between two phones, something which many teams have tried to use to get contact tracing information. However, this approach has proven challenging, with the measurements being highly noisy and there being a good chance of two phones not having a measurement despite being within sensing range of each other.

To try to combat these challenges, the contact tracing problem can be considered a sensor network localization problem. This is a long-studied class of algorithms which attempt to localize members of a network using only pairwise distance measurements between network members. Modeling contact tracing as a network localization problem serves two distinct benefits. First, this allows for multiple measurements to be used to try to estimate the distance between phones, combatting the effects of noisy data. Secondly, this allows for the distance between two phones to be estimated even if there is no corresponding distance measurement between these two devices.

In this collaboration I worked with Lilly Clark to test out a number of sensor network localization approaches for contact tracing. See our results and read more about it here!

Alan Papalia
Alan Papalia
PhD Student | Autonomy | Robotics | Design

My interests include autonomous robotic systems and mechanical design