Delay Optimal Scheduling for Energy Harvesting Based Communications

Juan Liu, Huaiyu Dai, Wei Chen

Abstract

This paper addresses the fundamental tradeoff in green communications: balancing unpredictable harvested energy with costly reliable grid energy. By modeling the system as a two-dimensional Markov chain, the authors derive a threshold-based scheduling policy that minimizes average packet delay under a strict power budget from a reliable energy source.

Introduction

Energy harvesting (EH) is a cornerstone of sustainable wireless networks, enabling nodes to power themselves using solar, vibration, or thermal energy. However, EH is notoriously bursty, with energy arriving at random intervals. Relying solely on EH can lead to massive delays or packet loss while waiting for battery recharge. To address this, hybrid systems utilize a Reliable Energy Source (RES), such as the power grid, as a backup. The challenge lies in optimizing the use of RES: over-reliance increases costs and reduces sustainability, while under-reliance causes latency to skyrocket. This research seeks the delay-optimal scheduling policy to manage this balance.

System Architecture

The system is modeled with two primary components: a data queue and an energy storage unit (battery). The data queue ($q_1$) tracks backlogged packets, while the energy queue ($q_2$) represents the amount of harvested energy stored in a finite-capacity battery. A packet scheduling block monitors the status of both queues and controls the switch between harvested energy and the RES.

System model for energy harvesting based communications with a reliable energy source.

Methodology

The authors represent the system state using a two-dimensional discrete-time Markov chain. The dimensions correspond to the data queue length and the battery level. Transitions between states are governed by stochastic processes for data arrivals and energy harvesting, as well as the scheduling policy's transmission probabilities.

Two-dimensional discrete-time Markov chains for general and specific cases.

To find the optimal policy, the steady-state equations of the Markov chain are transformed into a Linear Programming (LP) problem. The objective is to minimize the average packet delay, derived via Little’s Law, subject to a constraint on the average power drawn from the RES. The analysis covers several scenarios, including a general case and specific cases (Cases I, II, and III) that vary based on the granularity of data and energy units.

Key Insights: The Threshold Policy

A primary finding of the research is that the optimal scheduling strategy follows a threshold-based structure. Rather than complex varying probabilities, the node follows a simple rule: if the data queue length is below a critical threshold ($i^*$), it remains in a "green" state, waiting for harvested energy. Once the backlog hits the threshold, the system begins drawing from the RES to prevent excessive delay.

Furthermore, the study highlights that increasing battery capacity ($Q_2$) yields diminishing returns. The delay-power tradeoff is more sensitive to the arrival rate of energy than to the storage size itself. Stability is maintained only if the combined average harvested energy and maximum RES power exceed the average data arrival rate.

Results and Tradeoffs

Theoretical derivations were validated through Monte Carlo simulations over $10^6$ time slots, showing a perfect match. The results demonstrate a strictly decreasing, monotonic relationship between RES power consumption and average delay. As more power is allocated from the RES, average delay drops exponentially until it reaches a minimum floor. For simpler configurations (Case I), the authors provide closed-form expressions that allow for efficient real-time implementation.

Limitations and Future Directions

While the current model provides a robust framework, it utilizes Bernoulli processes which may not capture the full temporal correlation of real-world solar energy or variable bit-rate traffic. Additionally, the model assumes single-packet transmissions and static wireless channels. Future research will focus on extending these findings to multi-packet transmission scenarios and incorporating channel state information (CSI) to account for fading environments.

Summary updated

BibTeX
@article{liu2015delay,
  author = {Juan Liu and Huaiyu Dai and Wei Chen},
  title = {{Delay Optimal Scheduling for Energy Harvesting Based Communications}},
  journal = {IEEE J. Sel. Areas Commun.},
  year = {2015},
  doi = {10.1109/jsac.2015.2391972},
}
Figure