Max-Min Throughput in RSMA-Aided SWIPT for Sustainable Internet of Underground Things: A Deep Reinforcement Learning Approach

Kaiqiang Lin, Mohamed-Slim AlouiniarXiv 2026 · arXiv

This paper, titled "Max-Min Throughput in RSMA-Aided SWIPT for Sustainable Internet of Underground Things: A Deep Reinforcement Learning Approach," addresses the unique challenges of powering and communicating with sensors buried beneath the earth. By combining advanced wireless access techniques with distributional reinforcement learning, the authors provide a blueprint for sustainable, long-term underground monitoring.

TL;DR

The authors develop a Deep Reinforcement Learning (DRL) framework to optimize resource allocation in UAV-powered underground sensor networks. By integrating Rate-Splitting Multiple Access (RSMA) with Simultaneous Wireless Information and Power Transfer (SWIPT), they achieve significantly higher throughput fairness and reliability in dynamic soil environments compared to traditional methods.

What Problem Does This Paper Address?

The Internet of Underground Things (IoUT) is critical for smart agriculture (soil moisture sensing), pipeline leak detection, and border patrol. However, it faces two massive hurdles:

  1. Extreme Attenuation: Soil is a harsh medium that absorbs radio frequency (RF) signals far more aggressively than air.
  2. Energy Scarcity: Replacing batteries in buried sensors is impractical.

While Unmanned Aerial Vehicles (UAVs) can act as mobile power stations—using Wireless Energy Transfer (WET) to charge sensors—managing the "conversation" is difficult. Most existing systems use simple Time-Division Multiple Access (TDMA), which is inefficient. Furthermore, soil moisture changes constantly due to rain or irrigation, causing the wireless channel to fluctuate wildly. The authors aim to solve a max-min fairness problem: maximizing the throughput of the "weakest" sensor while ensuring the UAV doesn't run out of battery and every sensor receives its necessary control commands (downlink).

Methods

The paper introduces a sophisticated architecture combining two key communication technologies and a novel DRL algorithm:

  • RSMA-Aided SWIPT: Instead of just sending data to one user at a time, the UAV uses Rate-Splitting Multiple Access (RSMA). It splits messages into "common" and "private" parts, allowing for better interference management. Simultaneously, it uses SWIPT, where sensors use a power-splitting (PS) ratio to harvest energy from the RF signal while decoding data.
  • PD-TQC (Primal-Dual Truncated Quantile Critics): Standard DRL often struggles with the high variance of wireless fading. The authors use Truncated Quantile Critics (TQC), a distributional DRL approach that learns the distribution of possible rewards rather than just the average.
  • Constraint Handling: To satisfy hard constraints (like the UAV's energy budget), they employ a Primal-Dual framework. This adaptively updates Lagrange multipliers (dual variables) during training to penalize the agent whenever it violates a constraint.

Datasets

The study is grounded in real-world data rather than just theoretical models:

  • In-situ Soil Measurements: The authors used volumetric water content (VWC) data from a center-pivot irrigation farm in Monahans, Texas.
  • Temporal Split: They used the full-year 2024 VWC sequence for training the DRL agent and the 2025 record for out-of-sample evaluation to ensure the model generalizes to future weather patterns.

Key Insights and Ideas

  • RSMA is a Game-Changer for IoUT: While RSMA is often discussed for high-speed 6G terrestrial links, this paper shows it is uniquely suited for underground networks. Its flexibility in managing interference allows the system to support more sensors in "near-far" scenarios where some sensors are buried deeper or further away than others.
  • Distributional DRL for Fading: The "aha moment" is that in wireless environments, the uncertainty of the channel is as important as the channel state itself. By using TQC to model the reward distribution, the agent becomes robust against "deep fades" (sudden signal drops) that would confuse a standard DRL agent.
  • Real-Time Feasibility: Despite the complexity of the optimization (non-convex and highly coupled), the trained DRL actor can make decisions in less than 1 millisecond, making it practical for deployment on actual UAV hardware.

Key Results

The proposed PD-TQC algorithm was compared against several benchmarks, including Greedy strategies, Proximal Policy Optimization (PPO), and standard Soft Actor-Critic (SAC):

  • Throughput Gains: PD-TQC achieved an annual average minimum uplink throughput of 45.05 kbps, outperforming the Greedy baseline by 108.8% and the PPO baseline by 17.3%.
  • Fairness: By focusing on the "max-min" objective, the system ensured that even the most disadvantaged sensors maintained reliable communication.
  • Reliability: The system satisfied downlink throughput and UAV energy constraints in 96.99% and 99.72% of evaluated frames, respectively.
  • RSMA vs. SDMA: The RSMA-based approach consistently outperformed Space-Division Multiple Access (SDMA), particularly as the number of sensors increased or channel correlation became stronger.

Limitations and Future Work

  • Fixed Hovering: The current model assumes the UAV hovers at a fixed point. Future iterations could integrate trajectory optimization, allowing the UAV to fly closer to specific clusters of sensors to improve energy transfer.
  • CSI Requirements: The model assumes perfect Channel State Information (CSI). In practice, obtaining accurate CSI from low-power buried sensors is difficult; future work could explore "CSI-free" or "limited-feedback" designs.
  • Burial Depth: The study focused on shallow burial (0.5m). For deeper applications (e.g., several meters), the signal attenuation might require different relaying architectures or lower-frequency bands.

Summary updated

Abstract

Unmanned aerial vehicle (UAV)-enabled wireless-powered underground sensor networks (WPUSNs) allow underground devices (UDs) to harvest radio-frequency energy from a UAV via wireless energy transfer (WET) techniques, enabling sustainable Internet of underground things (IoUT) applications. This paper integrates rate-splitting multiple access (RSMA)-aided simultaneous wireless information and power transfer (SWIPT) into UAV-enabled WPUSNs to deliver information and energy to UDs. Within this framework, we investigate a max-min fairness problem that jointly optimizes the UAV transmit power, power allocation and message splitting, power-splitting ratios, and time allocation, aiming to maximize the minimum uplink throughput across UDs subject to per-UD downlink throughput requirements and a UAV energy-consumption constraint. To address this non-convex problem under dynamic underground environments, we develop an off-policy deep reinforcement learning approach based on the primal-dual truncated quantile critics (PD-TQC), where a TQC-based reward critic evaluates the fairness objective, cost critics estimate constraint costs, and an actor learns the resource-control policy with adaptive dual-variable updates. Simulations using in-situ soil volumetric water content measurements from a center-pivot irrigation farm demonstrate that PD-TQC achieves the highest annual average minimum uplink throughput compared with four benchmarks while satisfying the per-frame constraints in over 96% of the evaluated frames, and that this throughput advantage is preserved under different Rician factors, numbers of UDs, and downlink throughput thresholds.
BibTeX
@misc{lin2026max,
  author = {Kaiqiang Lin and Mohamed-Slim Alouini},
  title = {{Max-Min Throughput in RSMA-Aided SWIPT for Sustainable Internet of Underground Things: A Deep Reinforcement Learning Approach}},
  year = {2026},
  eprint = {2611.19865},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.19865},
}
Figure