Age-Minimal Online Policies for Energy Harvesting Sensors with Incremental Battery Recharges

Ahmed Arafa, Jing Yang, Şennur Ulukuş, H. Vincent Poor

Abstract

This paper investigates the minimization of the long-term average Age of Information (AoI) for a sensor node powered by an energy harvesting source. The sensor is equipped with a finite battery of size $B$ that recharges incrementally in discrete units according to a Poisson process. We characterize the optimal online update policy, demonstrating that it follows a renewal structure. For the specific case of a two-unit battery ($B=2$), we explicitly derive an energy-dependent threshold policy where the sensor's decision to update depends on both the current age of information and the remaining battery energy.

Introduction: The Freshness Challenge in IoT

In modern Internet of Things (IoT) applications, the value of data is often tied to its "freshness" rather than just the volume of throughput. This freshness is quantified by the Age of Information (AoI), defined as the time elapsed since the most recent update was generated and successfully received at the destination.

The sawtooth evolution of Age of Information (AoI) over time. Updates at times $s_i$ reset the age, while $x_i$ denotes the inter-update intervals.

Managing AoI is particularly challenging in energy-harvesting environments where power is intermittent. Unlike previous models that assume infinite energy or instantaneous full recharges, this study focuses on a more realistic "incremental recharge" model. Here, energy arrives in discrete units, and the sensor must decide when to "spend" a unit to transmit an update, balancing immediate freshness against future energy availability.

System Model and Problem Formulation

We consider a sensor node with a battery capacity of $B$ units. Energy units arrive according to a Poisson process with rate $\eta$. Each transmission consumes exactly one unit of energy. The sensor operates in an online setting, meaning it only has causal knowledge of energy arrivals—it knows when energy has arrived in the past but cannot predict future arrivals.

The goal is to find an online policy $\pi$ that minimizes the long-term average AoI: $$\bar{\Delta} = \limsup_{T \to \infty} \frac{1}{T} \mathbb{E}\left[ \int_0^T \Delta(t) dt \right]$$ subject to energy causality constraints, ensuring the sensor never attempts to transmit without at least one unit of energy in the battery.

Methodology: Renewal Theory and Epoch Optimization

The authors prove that the optimal policy follows a renewal structure. This implies that the system's state resets at specific points—specifically, whenever an update is sent and the battery is left with $k$ units of energy. This allows the complex long-term average minimization to be simplified into an optimization problem over a single "renewal epoch."

Visualization of update timing variables $y_1$ and $\bar{y}_2$ within a renewal epoch, illustrating how the sensor schedules transmissions based on energy arrival times $\tau_i$.

By analyzing these epochs, the authors use Lagrangian optimization to determine the functional form of the optimal policy. For a battery size $B=2$, the problem involves determining the optimal waiting times before sending an update, which are functions of the energy arrival times within the epoch.

Key Insights: Energy-Dependent Thresholds

The analysis reveals that the optimal policy for $B=2$ is an energy-dependent threshold policy. The sensor should only transmit an update if the current AoI exceeds a certain threshold, and this threshold changes based on the battery level:

  1. Low Energy ($E=1$): The sensor employs a higher threshold $x_1$. It acts "stingy," waiting for the data to become older before risking its last unit of energy.
  2. High Energy ($E=2$): The sensor employs a lower threshold $\lambda$. With a full battery, it becomes more "aggressive" in sending updates to keep the AoI low.

Mathematical derivation shows that $x_1 > \lambda$, confirming the intuition that energy scarcity necessitates more conservative update behavior. Furthermore, the study finds that incremental recharges are superior to bulk recharges for AoI. Because energy arrives unit-by-unit, the sensor has more opportunities to spread updates uniformly, which is more effective at reducing the average age than sporadic, high-energy bursts.

Conclusion and Future Directions

This research provides a rigorous framework for managing AoI in energy-constrained sensors. While the $B=2$ case is explicitly solved, extending this to larger batteries ($B \geq 3$) introduces significant combinatorial complexity. Future work may explore near-optimal heuristics for larger batteries and consider the impact of noisy wireless channels, where transmission failures would introduce additional energy management challenges.

Summary updated

Abstract

A sensor node that is sending measurement updates regarding some physical phenomenon to a destination is considered. The sensor relies on energy harvested from nature to transmit its updates, and is equipped with a finite <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">B</i> -sized battery to save its harvested energy. Energy recharges the battery <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">incrementally</i> in units, according to a Poisson process, and one update consumes one energy unit to reach the destination. The setting is online, where the energy arrival times are revealed causally after the energy is harvested. The goal is to update the destination in a timely manner, namely, such that the long term average <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">age of information</i> is minimized, subject to energy causality constraints. The age of information at a given time is defined as the time spent since the latest update has reached the destination. It is shown that the optimal update policy follows a renewal structure, where the inter-update times are independent, and the time durations between any two consecutive events of submitting an update and having k units of energy remaining in the battery are independent and identically distributed for a given <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">k</i> ≤ <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">B</i> - 1. The optimal renewal policy for the case of <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">B</i> = 2 energy units is explicitly characterized, and it is shown that it has an <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">energy-dependent threshold</i> structure, where the sensor updates only if the age grows above a certain threshold that is a function of the amount of energy in its battery.
BibTeX
@inproceedings{arafa2018age,
  author = {Ahmed Arafa and Jing Yang and Şennur Ulukuş and H. Vincent Poor},
  title = {{Age-Minimal Online Policies for Energy Harvesting Sensors with Incremental Battery Recharges}},
  booktitle = {ITA},
  year = {2018},
  doi = {10.1109/ita.2018.8503180},
}
Figure