Context-awareness for Dependable Low-Power IoT

TL;DR

The paper addresses dependable operation in large-scale, energy-constrained IoT by introducing context-aware protocols that integrate four key dimensions: energy status, information freshness, task relevance, and physical/medium conditions. It presents a two-step, edge-centric protocol design that first encodes hardware constraints and then app-specific parameters, enabling a cross-layer mapping from context to dependability targets. Through three use cases, it demonstrates substantial improvements in detection latency, accuracy, and availability while incurring minimal control-plane overhead, leveraging an edge-based coordinator to balance centralized reliability with local responsiveness. This framework advances practical, scalable dependability for low-power IoT by aligning sensing, scheduling, and transmission decisions with dynamic context and energy budgets, supporting sustained operation under harsh conditions with energy neutrality potential.

Abstract

Dependability is the ability to consistently deliver trusted and uninterrupted service in the face of operational uncertainties. Ensuring dependable operation in large-scale, energy-constrained Internet of Things (IoT) deployments is as crucial as challenging, and calls for context-aware protocols where context refers to situational or state information. In this paper, we identify four critical context dimensions for IoT networks, namely energy status, information freshness, task relevance, and physical/medium conditions, and show how each one underpins core dependability attributes. Building on these insights, we propose a two-step protocol design framework that incorporates operation-specific context fields. Through three representative use cases, we demonstrate how context awareness can significantly enhance system dependability while imposing only minimal control-plane overhead.

Figures (5)

• Figure 1: Illustrative example of context-awareness for dependable .
• Figure 2: Energy-limited/-powered network where the coordinate dynamic processes monitoring such as event-driven activations (left) and uplink updates (right) from $N$ . The red circles indicate the full detection area for a given event, while the red cross indicates failed polling/uplink update.
• Figure 3: Probability of mean event detection and event detection latency in ms (top), and in seconds (bottom) for a context-aware approach and a context-agnostic benchmark as a function of the IoTD density. The benchmark relies solely on spatial correlation for wake-up requests, while duty-cycle parameters are fixed via brute-force tuning according to the event statistics.
• Figure 4: Illustration of , total energy consumed at , and average query response time during the test run of the task-aware device scheduler and the Monte Carlo scheduler from raghuwanshi2024goal. Here, E$_0$ is the energy consumed, per transmission, at the device. Moreover, the latency of the device-to-edge, edge-to-device, and edge-to-server transmission is $1$ ms.
• Figure 5: Comparison of three AoI-based approaches as function of the probability of activation $p$ for a network with $N=32$ , $F=2$ channels, and $\varepsilon=0.5$. The optimal approach guarantees low at the cost of increased communication between and . The autonomous scheduling also decreases the , specially at higher values of $p$, but are required to listen to the channel and communicate among themselves. The threshold-based achieves a higher , but no signaling between any entities is required.