The outbreak of the novel coronavirus (COVID-19) is unfolding as a major global health crisis whose influence extends to major aspects of our daily lives. There have been over 1,000,000 infections worldwide in early April 2020. To this end, tracking and predicting how the virus could spread among people and regions have become critical parts of public health response strategies and early warning and prediction systems.

We design a privacy-preserving crowdsensing system called A-Turf, an acoustic encounter detection solution for effective coronavirus tracking, by leveraging ubiquitous mobile devices. The system can accurately detect “encounters” between users within physical proximity, say the range of 6 feet (social distance). This work will overcome key challenges of infrastructure-based techniques (such as video monitoring systems) that are difficult to scale with broad coverage.

To the best of our knowledge, A-Turf is the first work on encounter detection purely using acoustic signals in the context of virus tracking. The benefits of using acoustic signals rather than other signals (such as Bluetooth radio signals) are mainly due to the former’s privacy-preserving nature—specifically, no unique MAC address disclosure and limited communication range—which is exactly what we need to avoid privacy concerns and “false encounters” between faraway users apart to avoid false alarms. The Bluetooth based solutions may use some mechanisms such as randomization to avoid unique MAC address disclosure. However, it is very difficult for them to limit the Bluetooth communication range within physical proximity, say the range of 6 feet. Further, our acoustic solution can be enhanced to capture more fine grained encountering detection such as face orientation, handshakes, conversations, etc.

A Simplified Working Scenario: A simplified working scenario of A-Turf is as follows. Bob has a mobile phone with A-Turf installed, as Figure 1 shows. His mobile phone broadcasts its randomly generated ID, using its speaker to send ultrasonic signals at a certain frequency f (or at multiple frequencies), either continuously or occasionally (at its discretion), to nearby phones. Note that the ID is randomly generated; it uses neither Wi-Fi nor Bluetooth MAC addresses. Meanwhile, Bob’s phone also listens on frequency f, trying to receive ultrasonic signals from surrounding phones. Alice does the same. Once these two people are in the same physical proximity, Bob’s mobile phone can receive Alice’s ultrasonic signals (i.e., Alice’s randomly generated ID) and vice versa. Note that both Bob and Alice may receive multiple such ultrasonic signals from each other, depending on how long they are nearby. Once Bob’s phone receives these signals, it will record them with the current timestamp in its local database. Bob’s phone can either automatically report these recordings to a central server (say at the CDC) or wait for Bob’s manual processing. Alice’s phone does the same.

Privacy Preservation: Information privacy means the information owner has a certain degree of control over information sharing with others. This means that the degree of privacy preservation is not measured by the amount of information shared or disclosed to others; instead, it is measured by the owner’s degree of control over such sharing or disclosure. In A-Turf, users have full control over their individual information. Our acoustic solution has no unique device MAC address disclosure. Besides, users have full control over information sharing or disclosure to their peers or public health authorities (e.g., the CDC). The above are two potential working scenarios with different degrees of information sharing or disclosure.

In working scenario 1, unaffected users will not disclose any information to the CDC. The tradeoff is that the CDC cannot proactively calculate and push risk-level information to users. In working scenario 2, users disclose their encounters to the CDC, which knows their real IDs and trajectories. The benefits are that the CDC can calculate their risk levels (with other information beyond that published in case 1), and push risk information to individual users. Also, the CDC has a better global view thanks to such data. Other working scenarios are possible at the discretion of individual users. For instance, some users may choose to upload their encounter information without timestamps and locations. Our system grants users fine-grained control over information sharing and disclosure. Also, a user may choose to just receive encountering peers’ acoustic random IDs without sending hers.

We are currently developing the A-Turf system. Our final deliverables will be (1) an interactive website showing a searchable map of virus spread and (2) a mobile app that can be downloaded to sense encounters and interact with the website for the encounter data as well as the infected information for risk level analysis. We are seeking collaborations in development and funding.

Main Faculty Contributors: Dong Xuan (Professor of CSE), Ness Shroff (Ohio Eminent Scholar in Networking and Communications, Chaired Professor of ECE and CSE), Zhiqiang Lin (Associate Professor of CSE), and Adam C. Champion (Senior Lecturer of CSE)
Main Student Contributors: Yuxiang Luo (B.S. student of CSE), Cheng Zhang (Ph.D. student of CSE), Yunqi Zhang (B.S. student of CSE), and Chaoshun Zuo (Ph.D. student of CSE)
Contact: aturfosu@gmail.com

Updated July 6, 2020 05:42pm by Yunqi Zhang
© Copyright 2020 The Ohio State University