Hooman Sarvghadi, Juan Felipe Gutiérrez Gómez, Negar Halakou, Nasrin Razmi, Andrés Gómez
原文摘要
Batteryless Internet of Things (IoT) systems powered by energy harvesting require communication mechanisms that dynamically adapt to fluctuating energy availability while maintaining ultra-low-power operation. This paper presents a Wake-on-Radio (WoR) communication framework for low-power intermittent IoT systems in which the node's communication state adapts to its stored energy. The framework uses a batteryless IoT node that integrates a photovoltaic harvester, a 570 $μ$F capacitor, two voltage supervisors, and a microcontroller. The voltage supervisors drive an energy-aware finite state machine that switches the node between Shutdown, Low-Power Listening (LPL), Transmit, and Receive modes. The proposed system employs the integrated WoR capability of the transceiver, eliminating the need for additional wake-up hardware. For the evaluated traffic conditions and an LPL wake-up rate of 16 Hz, the measured power consumption is approximately 1.2 $μ$W in the Shutdown mode, 300 $μ$W during LPL, 531 $μ$W for reception, and 627 $μ$W for transmission. System-level experiments under solar-harvesting operation validate the proposed framework, demonstrating reliable bidirectional communication with no packet losses observed across repeated runs at a harvested input power of 1.2 mW.
Split Inference (SI) addresses the hardware resource limitations of Internet of Things (IoT) devices by distributing tiny Machine Learning (TinyML) model segments across multiple IoT devices. However, a device failure can interrupt the inference pipeline. This paper proposes a replica-aware recovery approach that deploys replicas of TinyML model segments originally hosted on failed devices onto surviving devices. We formulate the joint selection of replica hosts and communication protocols as a mixed-integer linear programming (MILP) problem that minimizes post-reconfiguration inference latency, accounting for computation and transmission latencies, under flash memory and runtime memory constraints. Simulations using MobileNetV2 on heterogeneous IoT networks compare the proposed approach with four baselines. We implement the model in a network of 15 IoT devices with a maximum link-rate reduction of 75%, and show that MILP approach achieves an average modeled end-to-end inference latency that is 6.8% lower than that of Computation Greedy and 18.3% lower than that of Random. Index Terms TinyML, split inference, Internet of Things, fault tolerance, replica placement.