Vision Transformers for Efficient Indoor Pathloss Radio Map Prediction

2024-12-12

Electronics

AI Summary

Indoor path loss prediction remains challenging due to environmental complexity and scarcity of labeled data. This paper proposes the first end-to-end radio map prediction method leveraging DINO-v2 self-supervised pre-trained Vision Transformers (ViTs), taking floor plans with wall semantics as input and directly generating high-resolution path loss heatmaps. We introduce geometric-aware data augmentation—including wall perturbations—and multi-channel feature fusion, empirically demonstrating their critical role in few-shot settings. Experiments show that under a <50-scenario few-shot regime, our method reduces mean absolute error (MAE) by 37% compared to CNN-based and ray-tracing baselines, while accelerating inference by 8×. Moreover, it achieves superior generalization across buildings and layouts. To our knowledge, this is the first successful transfer of self-supervised ViTs to radio propagation modeling, establishing a novel paradigm for resource-constrained wireless network planning.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Intelligent Robots: Motion and Path Planning
• Computer Vision: Multi-modal Vision

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Indoor pathloss prediction is a fundamental task in wireless network planning, yet it remains challenging due to environmental complexity and data scarcity. In this work, we propose a deep learning-based approach utilizing a vision transformer (ViT) architecture with DINO-v2 pretrained weights to model indoor radio propagation. Our method processes a floor map with additional features of the walls to generate indoor pathloss maps. We systematically evaluate the effects of architectural choices, data augmentation strategies, and feature engineering techniques. Our findings indicate that extensive augmentation significantly improves generalization, while feature engineering is crucial in low-data regimes. Through comprehensive experiments, we demonstrate the robustness of our model across different generalization scenarios.

Problem

• Predict indoor pathloss using vision transformers
• Address data scarcity in radio map prediction
• Improve generalization with augmentation and features

Innovation

• Uses Vision Transformer (ViT) with DINO-v2 pretraining
• Processes floor maps with wall features
• Employs data augmentation and feature engineering