Reconfigurable Intelligent Surface-Assisted Aerial-Terrestrial Communications via Multi-Task Learning

Xuelin Cao, Bo Yang, Chongwen Huang, Chau Yuen, Marco Di Renzo, Dusit Niyato, Zhu Han

TL;DR

This paper proposes an adaptive transmission protocol for aerial-terrestrial (UAV-to-ground) communications that leverages Reconfigurable Intelligent Surfaces (RIS) to bypass urban blockages. By employing a Multi-Task Learning (MTL) framework to jointly optimize RIS element allocation and phase shifts, the authors achieve a staggering four-order-of-magnitude reduction in computation time compared to traditional optimization methods, enabling near-real-time link enhancement.

What Problem Does This Paper Address?

Aerial-terrestrial communication—using Unmanned Aerial Vehicles (UAVs) as flying base stations—is a promising way to provide ubiquitous coverage. However, in dense urban environments, the crucial Line-of-Sight (LoS) links between a high-altitude UAV and a ground user are frequently blocked by buildings.

While Reconfigurable Intelligent Surfaces (RIS) can "bend" wireless signals to create alternative paths, configuring them is a nightmare. Specifically, the system must decide which users get which RIS elements (a discrete allocation problem) and what the phase shift of each element should be (a continuous beamforming problem). This results in a Mixed-Integer Non-Linear Program (MINLP) that is too computationally heavy to solve on the fly as UAVs and users move.

Methods

The authors introduce two major innovations:

  1. An Adaptive RIS-Assisted Protocol: They propose a frame-based structure divided into a Negotiation Phase (NP) and a Communication Phase (CP). The NP handles synchronization, channel estimation, and optimization, while the CP handles actual data transfer. Crucially, they divide the RIS into "groups" so it can serve multiple UAV-user pairs simultaneously without inter-group interference.
  2. Multi-Task Learning (MTL) Framework: Instead of solving the MINLP iteratively, they use a Deep Neural Network (DNN). The network features shared hidden layers that branch into two specialized "heads":
  3. Classification Head: Uses a Softmax function to predict the optimal RIS group allocation (which users get the RIS).
  4. Regression Head: Uses a Sigmoid function to predict the continuous phase shift values for the RIS elements.

By training this model offline using data from exhaustive searches, the RIS controller can perform "inference" in milliseconds during the online phase.

Datasets

The researchers did not use a public dataset. Instead, they generated a custom dataset using Algorithm 1: Dataset Collection.

  • Generation: They simulated random 3D deployments of UAV and user clusters.
  • Labeling: They used exhaustive search to solve the MINLP for thousands of scenarios, providing the "ground truth" optimal strategies.
  • Scale: The dataset covers a wide range of channel conditions and geometries to ensure the model generalizes well to unseen environments.

Key Insights and Ideas

  • Exploiting Correlation: The core "aha moment" is that RIS allocation and phase-shift configuration are deeply intertwined. By using MTL with shared hidden layers, the model learns the underlying physics of the wireless environment that affects both tasks, leading to better accuracy than training two separate models.
  • Complexity Shift: The paper successfully shifts the heavy lifting from online computation (solving a hard math problem every millisecond) to offline training (teaching a neural network once).
  • Grouping for Scalability: By dividing $N$ RIS elements into $L$ groups, the system can support multiple users at once. This makes the RIS a shared infrastructure rather than a single-user luxury.

Key Results

  • Massive Speedup: The MTL approach reduced computation time by four orders of magnitude compared to the conventional Spatial Branch and Bound (sBB) method (e.g., ~0.017 ms vs ~3.8 ms for 8 users).
  • Throughput Gains: In scenarios where LoS was blocked, the proposed RIS-assisted system increased throughput from a measly 0.06 Mbps to 28 Mbps.
  • Efficiency: The system significantly reduced the required transmit power for UAVs while maintaining high Signal-to-Noise Ratios (SNR), which is critical for battery-constrained aerial platforms.
  • Inference Accuracy: The MTL model maintained high classification accuracy (above 90% for small user counts) and very low Mean Square Error (MSE) for phase shifts.

Limitations and Future Work

  • Perfect Channel Estimation: The paper assumes "perfect" channel state information (CSI). In reality, estimating the channel for a passive RIS is notoriously difficult and often noisy.
  • Static Snapshots: While the protocol handles mobility frame-by-frame, the current optimization doesn't explicitly account for the UAV's trajectory.
  • Future Directions: The authors suggest that future work should integrate UAV trajectory planning with RIS configuration to further optimize long-term system performance. Additionally, exploring how the system handles "imperfect" CSI would be a vital next step for real-world deployment.

Summary updated

Abstract

The aerial-terrestrial communication system constitutes an efficient paradigm for supporting and complementing terrestrial communications. However, the benefits of such a system cannot be fully exploited, especially when the line-of-sight (LoS) transmissions are prone to severe deterioration due to complex propagation environments in urban areas. The emerging technology of reconfigurable intelligent surfaces (RISs) has recently become a potential solution to mitigate propagation-induced impairments and improve wireless network coverage. Motivated by these considerations, in this paper, we address the coverage and link performance problems of the aerial-terrestrial communication system by proposing an RIS-assisted transmission strategy. In particular, we design an adaptive RIS-assisted transmission protocol, in which the channel estimation, transmission strategy, and data transmission are independently implemented in a frame. On this basis, we formulate an RIS-assisted transmission strategy optimization problem as a mixed-integer non-linear program (MINLP) to maximize the overall system throughput. We then employ multi-task learning to speed up the solution to the problem. Benefiting from multi-task learning, the computation time is reduced by about four orders of magnitude. Numerical results show that the proposed RIS-assisted transmission protocol significantly improves the system throughput and reduces the transmit power.
BibTeX
@article{cao2021reconfigurable,
  author = {Xuelin Cao and Bo Yang and Chongwen Huang and Chau Yuen and Marco Di Renzo and Dusit Niyato and Zhu Han},
  title = {{Reconfigurable Intelligent Surface-Assisted Aerial-Terrestrial Communications via Multi-Task Learning}},
  journal = {IEEE J. Sel. Areas Commun.},
  year = {2021},
  url = {https://openreview.net/forum?id=sOy59QRe71},
}

Loading related content...

Figure