A Disentangled Representation Learning Framework for Low-altitude Network Coverage Prediction

TL;DR

This work tackles low-altitude network coverage (LANC) prediction when BS antenna beam patterns are inaccessible, by introducing a two-pronged framework: expert-knowledge-based feature compression to mitigate imbalanced sampling, and a propagation-guided, disentangled representation network with separate branches for distance fading, frequency fading, and antenna gain. Theoretical analysis shows the disentangled, model-guided design can achieve comparable learning capability to traditional deep models with fewer parameters, reducing overfitting under data scarcity. Empirically, the approach yields a 7% MAE reduction over strong baselines and demonstrates practical reliability with MAE around $5\,\mathrm{dB}$ in real-network tests, including multi-BS scenarios and transfer to unseen AAU types. These results support robust, cross-region LANC prediction using readily available BS operational parameters, with clear implications for incremental network planning and optimization in the growing low-altitude economy.

Abstract

The expansion of the low-altitude economy has underscored the significance of Low-Altitude Network Coverage (LANC) prediction for designing aerial corridors. While accurate LANC forecasting hinges on the antenna beam patterns of Base Stations (BSs), these patterns are typically proprietary and not readily accessible. Operational parameters of BSs, which inherently contain beam information, offer an opportunity for data-driven low-altitude coverage prediction. However, collecting extensive low-altitude road test data is cost-prohibitive, often yielding only sparse samples per BS. This scarcity results in two primary challenges: imbalanced feature sampling due to limited variability in high-dimensional operational parameters against the backdrop of substantial changes in low-dimensional sampling locations, and diminished generalizability stemming from insufficient data samples. To overcome these obstacles, we introduce a dual strategy comprising expert knowledge-based feature compression and disentangled representation learning. The former reduces feature space complexity by leveraging communications expertise, while the latter enhances model generalizability through the integration of propagation models and distinct subnetworks that capture and aggregate the semantic representations of latent features. Experimental evaluation confirms the efficacy of our framework, yielding a 7% reduction in error compared to the best baseline algorithm. Real-network validations further attest to its reliability, achieving practical prediction accuracy with MAE errors at the 5dB level.

Figures (20)

• Figure 1: Disentangled representation learning framework.
• Figure 2: The process of low-altitude coverage prediction based on operational parameters.
• Figure 3: The design of our disentangled representation network, where distance fading, frequency fading, and antenna gain networks can be a simple MLP for feature representation, respectively.
• Figure 4: Full model and prior constrained model.
• Figure 5: Average MAE for different algorithms across 20 random splits (random seeds 0-19) as a function of the BS sampling rate in ablation experiments.

Theorems & Definitions (13)

• Definition 1: Decomposable data
• Definition 2: Full model
• Definition 3: Constrained model
• Remark 1
• Definition 4: Learning capability
• Remark 2
• Lemma 1: Learning capability bounded by mutual information xu2017information
• Remark 3
• Definition 5: Equivalent indicator
• Proposition 1: Comparable learning capability of constrained model