Energy Efficiency Maximization in RSMA-Based LEO Satellite Networks Assisted by UAV-Mounted BD-Active RIS: A Hybrid DRL Strategy

arXiv:2505.04148 · eess.SP, cs.SY, eess.SY · Submitted 2025-05-07 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Electrical Engineering. The day in electrical engineering: signals, systems and the hardware that carries them.

Rosa: I'm Rosa.

Dev: And I'm Dev, here with Taro.

Rosa: We're looking at the paper: "Energy Efficiency Maximization in RSMA-Based LEO Satellite Networks Assisted by UAV-Mounted BD-Active RIS".

Dev: Here's the gist of the paper. This paper proposes an advanced non-terrestrial communication architecture that integrates Rate-Splitting Multiple Access (RSMA) with a Beyond-Diagonal Active Reconfigurable Intelligent Surface (BD-ARIS) mounted on a UAV under the coverage of a Low Earth Orbit (LEO) satellite.

Rosa: The proposed RSMA-BD-ARIS framework significantly outperforms conventional RIS-assisted designs and provides a scalable, energyefficient solution for 6G and massive IoT applications in nonterrestrial networks.

Dev: System Architecture and Components The system model considers a LEO satellite equipped with a uniform rectangular array (URA) serving as the primary signal source.

Rosa: A UAV-mounted BD-ARIS is deployed to act as an intelligent cooperative relay, dynamically adjusting its horizontal position to maximize energy efficiency.

Title and authors: Rosa: The title itself highlights that they are focusing on maximizing the energy efficiency of a specific setup involving RSMA, BD-ARIS, and a drone under a LEO satellite. It suggests they’re looking at a way to get better performance for massive IoT applications in non-terrestrial networks. Dev And the authors, Rahman Saadat Yeganeh and Hamid Behroozi, they seem to have deep roots in this area of research, which is good because it shows some experience with these kinds of complex system designs.

Taro: I’m interested in what this means for real-world deployment; is it just theoretical math or something that actually works when you deploy a drone and an active surface? Rosa That's the question, Taro. The implication is that they are proposing a hybrid strategy—combining the physical layer components with deep reinforcement learning to make the system adaptive in complex environments.

Dev: Exactly. They’re not just looking at one fixed setup; they’re optimizing several things simultaneously: how much power to use, where the drone should fly, and how much to split the messages between common and private users. Taro So it’s a very comprehensive optimization problem being solved by this learning approach, right?

Rosa: Right. And what I find interesting is that they are considering realistic models for the power consumption of both the drone and the BD-ARIS elements, which isn't always done in these kinds of theoretical studies Dev That’s important because energy efficiency isn't just about the signal transmission; it’s about everything consuming power.

Taro: I hope they address how this system recovers if something goes wrong, like the drone losing its connection to the satellite or some channel information getting bad. Rosa That uncertainty is actually baked into their approach by including realistic models for power consumption, which implies they're thinking beyond just perfect conditions.

The paper's summary: Dev: The core of the paper is that they propose this RSMA-BD-ARIS framework, and the key idea is to jointly optimize satellite beamforming, UAV positioning, power allocation coefficients, and rate-splitting ratios to maximize the overall energy efficiency. Rosa Essentially, they’re trying to find the best combination of all those variables so the system uses less power for a given performance level.

Taro: So they are aiming for that overall energy efficiency metric—EE—which is defined as maximizing a sum of achievable common and private rates, divided by the total power consumption Dev That’s a pretty standard way to look at it, but doing it with this specific combination of technologies makes it more interesting.

Rosa: And to handle the really tricky part of this optimization problem—which they call non-convex and high-dimensional—they employ three deep reinforcement learning algorithms: Trust Region Policy Optimization, TD3, and A3C Dev That’s where the AI comes in to navigate that huge space of possibilities.

Taro: I always wonder which one they landed on as the best tool for the job. They mention results show that TRPO consistently achieved the best performance in terms of both energy efficiency and sum rate, especially when transmit powers were high and deployment scenarios were challenging Rosa That suggests TRPO is pretty robust under those tough conditions.

Dev: TD3 was mentioned because it converges faster in moderate settings, but A3C had some instability due to its high variance, which is a common issue with actor critic methods Taro So the paper isn't just proposing a setup; it’s showing how to use specific AI tools—TRPO, TD3, and A3C—to tune the system intelligently.

The paper's improvements: Rosa: One of the main improvements they point out is using TRPO as a policy optimization method because it ensures monotonic policy improvement. Dev That means during training, the AI won't just jump around wildly; it will make smoother and more stable progress toward that energy-efficient solution in these high-dimensional RSMA scenarios.

Taro: Stability is crucial when you have tightly coupled decision variables like drone position and beamforming settings interacting with each other Rosa I think that stability directly translates to reliability in the physical system, meaning it finds a solution that’s actually achievable without crashing the hardware.

Dev: They also incorporate realistic channel state information uncertainty into their deep reinforcement learning framework Taro This is important because we don't always have perfect knowledge of how signals are traveling; modeling that error lets the AI learn policies that are robust even when the observations are imperfect.

Rosa: And they talk about scaling—how this system handles a lot of users and a large number of RIS elements Dev The scalability analysis shows that TRPO demonstrates the highest robustness and generalization ability in complex multi-user environments, even maintaining good performance up to eleven users.

Taro: I noticed they also look at how the AI can dynamically adapt resource allocation based on environmental factors like drone altitude and user distance Rosa That means if the drone gets lower or a user moves further away, the system adjusts its strategy in real time to keep the sum rate high.

Conclusion: Dev: So to wrap up, this paper on "Energy Efficiency Maximization in RSMA-Based LEO Satellite Networks Assisted by UAV-Mounted BD-Active RIS: A Hybrid DRL Strategy" shows how integrating RSMA and BD-ARIS with a deep reinforcement learning approach can lead to a system that optimizes power consumption across the board. Rosa It suggests that using TRPO, TD3, or A3C—with TRPO showing the best results in terms of energy efficiency—can find near-optimal solutions for these complex non-convex problems.

Taro: From my side, what stands out is how they model the power consumption of both the drone and the BD-ARIS elements realistically, which grounds their theoretical optimization in something more practical than just assuming perfect hardware performance Dev That makes it relevant for real deployment scenarios where you have real power constraints.

Rosa: And they show that this approach scales reasonably well, with TRPO being particularly robust to environmental dynamics and CSI errors across different user densities Taro It’s a solid architecture for thinking about how we can make these non-terrestrial networks more energy efficient in the coming 6G era <ref:2505.04148#pg3>.

Dev: Overall, the implication is that this kind of intelligent control system could lead to significantly better performance in massive IoT applications where power is a major constraint Rosa We’ll keep an eye on how this hybrid DRL strategy performs when we move from simulation to actual hardware trials.

Sharif University of Technology

eess.SP, cs.SY, eess.SY

Submitted: 2025-05-07

Updated: 2026-11-04

DOI: 10.1109/TCOMM.2026.3747993

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 83/100

The gist: This paper proposes an advanced non-terrestrial communication architecture that integrates Rate-Splitting Multiple Access (RSMA) with a Beyond-Diagonal Active Reconfigurable Intelligent Surface…

Key concepts

Rate-Splitting Multiple Access (RSMA)
RSMA allows multiple users to share the same frequency resource by splitting the data stream into different rate components. This technique enables simultaneous communication with various users, improving spectral efficiency and overall system throughput in a network.
Beyond-Diagonal Active Reconfigurable Intelligent Surface (BD-ARIS)
This is an intelligent surface mounted on a UAV that actively adjusts its properties to reflect signals optimally. Unlike simpler surfaces, the BD-ARIS uses a group-connected structure, allowing it to dynamically change how it reflects signals based on the environment.
Deep Reinforcement Learning (DRL) Strategy
The paper employs DRL algorithms like TRPO to automatically find the best operational settings. The AI learns how to adjust parameters—such as UAV location and RIS configurations—to maximize energy efficiency without needing explicit, complex mathematical formulas for every scenario.
Energy Efficiency (EE)
EE measures how much data can be transmitted per unit of power consumed by the system. The optimization goal is to maximize this ratio by intelligently managing the power used by both the LEO satellite and the UAV-mounted RIS.

Terminology

Summary

This paper proposes an advanced non-terrestrial communication architecture that integrates Rate-Splitting Multiple Access (RSMA) with a Beyond-Diagonal Active Reconfigurable Intelligent Surface (BD-ARIS) mounted on a UAV under the coverage of a Low Earth Orbit (LEO) satellite. p. 0 The proposed RSMA-BD-ARIS framework significantly outperforms conventional RIS-assisted designs and provides a scalable, energyefficient solution for 6G and massive IoT applications in nonterrestrial networks. p. 0

System Architecture and Components

The system model considers a LEO satellite equipped with a uniform rectangular array (URA) serving as the primary signal source. A UAV-mounted BD-ARIS is deployed to act as an intelligent cooperative relay, dynamically adjusting its horizontal position to maximize energy efficiency. p. 2 The BD-ARIS utilizes a group-connected structure with a group size of 2, where the overall reflection matrix Φ is modeled as a block-diagonal matrix. p. 3 The UAV's position PosUAV = [xUAV, yUAV, hUAV] is dynamically determined to maximize the system’s energy efficiency.

Optimization Formulation

The main objective is to maximize the overall EE of a LEO satellite communication system assisted by a UAV-mounted group-connected BD-active RIS. The optimization problem is formulated to maximize the sum of the achievable common and private rates for all I users under the RSMA scheme, normalized by the total power consumption. p. 5 The objective function is defined as maximizing EE = Rc(wc,Φ,xUAV,yUAV)+PIi=1 Ri(w i,…,Φ,xUAV,yUAV) Ptotal(ac,…,ai). p. 5

Deep Reinforcement Learning Strategies

Three state-of-the-art deep reinforcement learning (DRL) algorithms are employed to solve the resulting non-convex and high-dimensional problem. These algorithms include Trust Region Policy Optimization (TRPO), Twin Delayed Deep Deterministic Policy Gradient (TD3), and Asynchronous Advantage Actor-Critic (A3C). p. 0 Simulation results reveal that TRPO consistently achieves the best performance in terms of EE and sum rate, especially under high transmit powers and challenging deployment scenarios. p. 0 TD3 converges faster and performs competitively in moderate settings, while A3C suffers from instability due to its high variance. p. 0

Performance Analysis and Robustness

The simulation results show that TRPO consistently achieves the highest final reward, converging around episode 810 to a stable value near 1100. The superior convergence speed of TD3 stems from its twin-critic architecture and delayed deterministic policy updates. p. 8 TRPO again proved to be the most robust to moderate CSI errors, highlighting its resilience in partially observable environments. p. 0 Furthermore, TRPO consistently achieves the highest sum rate across all distances. p. 9

Scalability and Comparison

The scalability analysis shows that TRPO demonstrates the highest robustness and generalization ability in complex multi-user environments. The performance continues to grow with RIS size, eventually approaching saturation around 56 elements when using the BD-ARIS architecture with TRPO. p. 8 Overall, the combination of BD-ARIS and RSMA, when enhanced by intelligent learning-based optimization, offers a promising architecture for future 6G and IoT communication systems. p. 0

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Model-free deep reinforcement learning algorithms such as TD3, A3C, and TRPO can be integrated into AI systems to autonomously find near-optimal solutions for complex non-convex optimization problems in dynamic wireless environments. This allows the AI system to dynamically determine optimal satellite beamforming vectors, UAV positioning, power allocation coefficients, and rate-splitting ratios simultaneously.

  2. The improved AI system can achieve superior energy efficiency by optimizing the total power consumption across all components of the RSMA-BD-ARIS framework. This optimization directly maximizes EE by jointly considering the realistic energy consumption of both the UAV and the BD-ARIS elements, as modeled in equation 16.

  3. By utilizing TRPO as a policy optimization method, AI systems can ensure monotonic policy improvement, leading to smoother and more stable convergence during training in high-dimensional RSMA scenarios. This stability is crucial for reliably finding energy-efficient long-term solutions when dealing with tightly coupled decision variables.

  4. The system can enhance communication reliability by incorporating realistic channel state information uncertainty into the DRL framework. This involves modeling the actual channel as a sum of an estimated component and a stochastic error term, allowing the AI to learn robust policies resilient to imperfect observations.

  5. The improved AI system can be designed for scalable and adaptive operation across varying network conditions, including changes in user density and number of RIS elements. Simulations show that TRPO demonstrates the highest robustness in complex multi-user environments, maintaining better performance as the number of users increases up to 11.

  6. The AI system can dynamically adapt its resource allocation based on environmental factors such as UAV altitude and ground user distance to maximize sum rate performance. This allows the system to adjust transmission strategies in real-time, leveraging the insights from Figure 4 regarding how UAV altitude influences total throughput in RSMA-enabled networks.

  7. The AI system can be configured with a group-connected BD-ARIS architecture that captures inter-element interactions, leading to enhanced signal amplification and flexible beamforming. This structure is modeled by block-diagonal matrices in equation 5, providing more practical and general modeling of active reflection behaviors than conventional diagonal RIS.

Sources

Related papers