Online Ski Rental for ON/OFF Scheduling of Energy Harvesting, Millimeter Wave Base Stations

Gilsoo Lee, Walid Saad, Mehdi Bennis, Abolfazl Mehbodniya, Fumiyuki Adachi

Introduction

As cellular networks evolve to meet exponential growth in data traffic, the deployment of small cell base stations (SBSs) has become a standard strategy for densification. However, this expansion brings a significant challenge: a massive increase in energy consumption. To mitigate environmental impact and operational costs, researchers are looking toward SBSs powered by renewable energy harvesting (EH), such as solar or wind power.

The Energy Harvesting Challenge

While EH-powered SBSs offer a sustainable path forward, they introduce a high degree of uncertainty. Unlike grid-powered stations, EH-SBSs rely on intermittent energy sources. An SBS must decide in real-time whether to remain ON to serve local users or turn OFF to conserve energy for future use.

System model of a macrocell network with energy-harvesting small cells.

If an SBS stays active until its battery is depleted, it may cause abrupt service interruptions, forcing users to hand over to a distant Macrocell Base Station (MBS), which increases network congestion and latency. Conversely, turning OFF too early results in underutilized capacity and wasted harvested energy. The core problem is making these ON/OFF decisions optimally without knowing when the next energy arrival will occur.

The Ski Rental Framework

To solve this, the researchers frame the scheduling problem as a classic "ski rental" problem. In the traditional analogy, a skier must decide each day whether to rent skis (a small daily cost) or buy them (a large one-time cost) without knowing how many days they will actually ski.

In the context of cellular networks:

  • Staying ON (Renting): The SBS incurs a continuous "rental" cost, which is the sum of its operational power consumption and the transmission delay experienced by its users.
  • Switching OFF (Buying): The SBS pays a one-time "purchase" price. This represents the cost of handing over all current traffic to the reliable, grid-powered MBS. Once "bought" (turned OFF), the SBS stays OFF for the remainder of the scheduling period.

Algorithmic Approach

The paper proposes two primary online algorithms to manage this trade-off:

  1. Deterministic Online Algorithm (DOA): The SBS switches to the OFF state at a fixed, predetermined time $t_j$. This approach is simple to implement and guarantees a competitive ratio of 2, meaning the total cost is never more than twice that of an "offline" optimal solution with perfect foresight.
  2. Randomized Online Algorithm (ROA): By selecting the power-down time based on a specific probability distribution, the ROA achieves an improved optimal competitive ratio of $e/(e-1) \approx 1.58$.
Flowchart and pseudocode for the Deterministic Online Algorithm (DOA).

To ensure the system is scalable, the authors developed a decentralized implementation. By approximating network interference at the start of each period, each SBS can act as an independent agent, making autonomous decisions without the need for a centralized controller or excessive signaling overhead.

Performance Evaluation

The algorithms were tested using extensive Monte Carlo simulations in a $0.5 \times 0.5 \text{ km}^2$ area. Energy arrivals were modeled as a Poisson process to simulate the stochastic nature of renewable sources. The results demonstrate significant improvements over baseline scheduling methods:

  • Energy Efficiency: The ROA reduced SBS energy consumption by up to 15.6%.
  • Latency Reduction: Per-SBS delay was improved by up to 20.6% due to better-timed handovers.
  • Cost Savings: The total operational cost (balancing energy and delay) was reduced by nearly 70% compared to simple scheduling.
  • Empirical Robustness: The observed competitive ratio was 1.86, confirming the algorithm's effectiveness in near-real-world conditions.

Conclusion and Future Directions

By applying the ski rental framework, this research provides a robust, decentralized method for managing energy-harvesting small cells under uncertainty. While the current model uses static cost approximations, future work could incorporate dynamic pricing based on real-time interference and expand the framework to multi-tier networks and specific millimeter-wave (mmWave) propagation characteristics. This approach paves the way for self-organizing, green cellular networks that balance performance with sustainability.

Summary updated

BibTeX
@article{lee2017online,
  author = {Gilsoo Lee and Walid Saad and Mehdi Bennis and Abolfazl Mehbodniya and Fumiyuki Adachi},
  title = {{Online Ski Rental for ON/OFF Scheduling of Energy Harvesting, Millimeter Wave Base Stations}},
  journal = {IEEE Trans. Wireless Communications},
  year = {2017},
  url = {https://openreview.net/forum?id=4EabsaWLhwK},
}
Figure