Trusted Single-Source Sensors using SNARKs

The trustworthiness of sensor data readings is crucial for IoT applications. The trend of decentralized and distributed architectures give rise to multi-party scenarios where mutual trust between different parties might not be present. Current approaches to increase trust in sensor readings include crypto-graphic authentication, redundancy of sensors, and plausibility verification of received signals. However, these approaches can often only defend against certain types of attacks. In this paper, we propose a multi-layer approach to increase the trust in single data sources, such as wireless sensors, by using a trusted execution environment (TEE) and succinct non-interactive arguments of knowledge over authenticated data (AD-SNARKs). First, we bring several trust metrics as close to the sensor as possible to reduce the surface of attacks. Second, we develop an optimized constrained system for AD-SNARKs that allows offloading statistical operations on the sensor data, such as moving average, to a non-Trusted constrained device. By lowering the number of constraints to 6, our implementation is able to generate proofs in 60ms on a Raspberry Pi 3(B) offering 128 bit of security with all validation data fitting into 1023 bytes of payload. Compared to other security approaches, this is a small overhead for achieving provable sensing and processing of data from source to consumer, which is a major step towards distributed trust for IoT applications.