Sensing-Aided 6G Drone Communications: Real-World Datasets and Demonstration

TL;DR

This work tackles beam training overhead in mmWave/THz drone communications by introducing sensing-aided beam prediction and tracking that fuse vision and positioning data. By recasting beam selection as a multi-modal classification problem, the authors develop position- and vision-based ML architectures (including MLP and ResNet-50) and a GRU-based tracking framework, validated on Real-world DeepSense 6G Scenario 23 data that pairs RGB imagery, GPS, altitude, distance, and mmWave beams. The key contributions include formal problem formulations, a dual ML pipeline for current and future beam decisions, a new drone-focused dataset, and thorough evaluations showing vision-based methods achieve high accuracy (top-1 up to ~86%) and near-optimal power under realistic mobility, significantly reducing beam training overhead. The results demonstrate practical viability for robust, high-throughput 6G drone communications and motivate integrating additional sensing modalities and digital-twin concepts for further enhancements.

Abstract

In the advent of next-generation wireless communication, millimeter-wave (mmWave) and terahertz (THz) technologies are pivotal for their high data rate capabilities. However, their reliance on large antenna arrays and narrow directive beams for ensuring adequate receive signal power introduces significant beam training overheads. This becomes particularly challenging in supporting highly-mobile applications such as drone communication, where the dynamic nature of drones demands frequent beam alignment to maintain connectivity. Addressing this critical bottleneck, our paper introduces a novel machine learning-based framework that leverages multi-modal sensory data, including visual and positional information, to expedite and refine mmWave/THz beam prediction. Unlike conventional approaches that solely depend on exhaustive beam training methods, our solution incorporates additional layers of contextual data to accurately predict beam directions, significantly mitigating the training overhead. Additionally, our framework is capable of predicting future beam alignments ahead of time. This feature enhances the system's responsiveness and reliability by addressing the challenges posed by the drones' mobility and the computational delays encountered in real-time processing. This capability for advanced beam tracking asserts a critical advancement in maintaining seamless connectivity for highly-mobile drones. We validate our approach through comprehensive evaluations on a unique, real-world mmWave drone communication dataset, which integrates concurrent camera visuals, practical GPS coordinates, and mmWave beam training data...

Figures (16)

• Figure 1: An illustration of the mmWave base station serving a drone in a real wireless environment. The base station utilizes additional sensing data, such as RGB images and the GPS location of the drone, to predict the optimal beam indices.
• Figure 2: An illustration to highlight the future beam prediction/beam tracking task. The mmWave base station utilizes a sequence of past sensing data such as RGB images ans GPS locations of the transmitter to predict the future optimal beam indices.
• Figure 3: A block diagram showing the proposed solution for both the vision and position-aided beam prediction task. As shown in the figure, the camera installed at the base station captures real-time images of the drone in the wireless environment. A CNN is then utilized to predict the optimal beam index. For the other three sensing data, the base station receives the information, which is then passed through a fully connected neural network to predict the beam.
• Figure 4: A block diagram showing the proposed solution for the beam tracking task. It highlights the three different approaches for image, position, and beam sequences. There are two main stages in the proposed solution: (i) Feature extraction and embedding: For image sequences, this stage is responsible for extracting the feature vectors consisting of the bounding box coordinates of the transmitter and for the beam-based solution this stage embeds the input beam sequences into vectors of dimension $N \times 1$, and (ii) Recurrent prediction with RNN: this stage takes the embedded beam vectors, the feature extracted from the images and the GPS positional data to predict the future beams.
• Figure 5: The figure illustrates the experimental setup and location of the DeepSense 6G testbed deployment. Fig. (a) presents an aerial perspective from Google Maps of Thude Park, the selected site for data collection. Fig. (b) and (c) detail the hardware configuration of both communication endpoints: the mobile drone-based transmitter and the stationary base station. Fig. (b) specifically depicts the drone-mounted mmWave phased array operating in the 60 GHz band that establishes the communication link with the base station.