Neuro-Inspired Autonomous Data Acquisition for Energy-Constrained IoT Sensors

Abstract

The unprecedented pervasiveness of the Internet of Things (IoT) has unleashed an urgent need for autonomous IoT sensors that do not only autonomously operate, but more importantly autonomously also make intelligent decisions, including when and what data to acquire. Inspired by the autonomous nervous system (ANS) and its rapid de-centralized response to sensory stimuli, this article proposes an autonomous data acquisition approach for energy-constrained IoT sensors. The proposed approach achieves autonomy through rapid real-time event detection in the analog domain, which is then used to instantaneously trigger data acquisition from the sensor, without needing to consult the processor in making such a decision. Accordingly, the analog event-detection circuit would be the only circuit that is continuously ON, while all other system blocks remain in the sleep mode until an event is detected, significantly reducing the operation time of the overall system and the amount of redundant data it produces. A proof-of-concept circuit is designed and implemented, and its performance is verified and analyzed through extensive simulations and experiments, demonstrating event-detection speeds at the order of microseconds; orders of magnitude faster than the required limit for lossless data acquisition in many IoT applications. A case study on an Industrial IoT (IIoT) application is investigated through circuit-level implementation and simulations on real seismic data. The presented results demonstrate the feasibility of lossless autonomous active seismic data acquisition with a 95% reduction in the overall operation time of the sensor node as well as in the amount of data it produces compared to conventional data-acquisition approaches.