Abstract
We present Silmarillion (PanCast), a novel, inclusive system for digital contact tracing and epidemic risk notification, which simultaneously provides utility as well as security. Silmarillion relies on a low-cost infrastructure of strategically placed beacons, inexpensive and low-maintenance user devices like dongles (if smartphones are inaccessible), and a backend that assists in epidemiological analysis and risk dissemination. Unlike today's smartphone-based contact tracing systems, Silmarillion records encounters between users' devices and beacons installed in well-known and strategic locations, which enables capturing contextual information relevant for epidemiological analysis, individual risk prediction, and prevention of spread of false risk information. Silmarillion keeps a user's encounter history local to their device and allows the user to control what information they share with the backend. Furthermore, it provides differential privacy for patients in risk dissemination and information-theoretic privacy for users receiving the risk information. We have evaluated a prototype of Silmarillion using small IoT boards and show that the battery and speed of risk dissemination is adequate for a practical deployment. Furthermore, we ran a small-scale deployment within a university building, demonstrating Silmarillion's practicality.
Introduction
The COVID-19 pandemic highlighted a significant "utility-privacy" gap in digital contact tracing. Centralized systems often collected excessive data, raising surveillance concerns, while decentralized peer-to-peer (P2P) systems—such as the Google/Apple Exposure Notification (GAEN) framework—protected privacy but offered limited epidemiological context. Without knowing if an encounter occurred in a well-ventilated outdoor space or a cramped indoor room, public health authorities lack the data necessary to understand transmission dynamics. Furthermore, P2P systems require modern smartphones, effectively excluding vulnerable populations like children and the elderly.
PanCast (also referred to as Silmarillion) bridges this gap by shifting from a phone-to-phone model to a Person-to-Infrastructure (P2I) architecture. By utilizing fixed Bluetooth Low Energy (BLE) beacons, the system provides high-utility environmental data while maintaining robust individual privacy.
System Architecture
The PanCast architecture consists of three primary components: fixed BLE beacons, user devices (smartphones or low-cost dongles), and a backend server.
- Passive Listening: User devices act as passive listeners. They do not transmit signals, making them invisible to eavesdroppers and significantly extending battery life. Beacons are placed in strategic "hotspots" (e.g., transit hubs, classrooms) and broadcast ephemeral IDs (ephIDs) along with static metadata such as location type and ventilation levels.
- Encounter Data Collection: When a user is near a beacon, their device records the beacon's ephID and the associated environmental context. This history remains stored locally on the user's device.
- Risk Dissemination: If a user tests positive, they can choose to upload their encounter history to the backend. This data is routed through a Mixnet to decouple the user's identity from their trajectory. The backend aggregates these reports to create a risk database.
- Private Risk Querying: Healthy users check for exposure by querying the backend. To maintain privacy, they use Information-Theoretic Private Information Retrieval (IT-PIR), allowing them to download risk data for specific regions without the server learning which regions were queried.
Privacy and Security Mechanisms
PanCast employs a multi-layered approach to security:
- Mixnets: Based on the Vuvuzela protocol, the mixnet shuffles and shreds encounter data from many users, preventing the backend from linking IDs to reconstruct a single person's movements.
- Differential Privacy (DP): The system adds "noise" or dummy IDs to the risk database. This ensures that an adversary cannot determine if a specific individual is infected by observing changes in the number of reported cases in a specific building.
- Oblivious Authentication: To prevent false reports, the system uses a token-based exchange. A health authority provides a sick user with a token that can be traded for signing keys without revealing the user's identity to the backend.
Evaluation and Results
The system was evaluated through hardware benchmarking on IoT kits (Nordic nRF52832) and a 16-day pilot deployment in a university building with 15 volunteers.
- Efficiency: BLE beacons can operate for over three years on a single coin-cell battery. User dongles, despite frequent scanning, last approximately 8.8 days on a single charge.
- Scalability: For a city-scale deployment of 100,000 users, the privacy overhead results in roughly 1.8 MB of data per day per device, which is well within the capacity of modern mobile networks.
- Practicality: The pilot study confirmed that the system accurately matched encounters for simulated "sick" users. Participants reported that both the hardware dongles and the companion app were intuitive and easy to use.
Conclusion
PanCast demonstrates that it is possible to reconcile the competing needs of public health utility and individual privacy. By leveraging a P2I model with fixed beacons, the system provides the environmental context necessary for advanced epidemiological analysis while ensuring that user movements remain private and the system remains accessible to those without smartphones.
Summary updated