U-Net for Indoor Pathloss Prediction from Sparse Measurements with Physics-Based Features
U-Net for Indoor Pathloss Prediction from Sparse Measurements with Physics-Based Features
Khoren Petrosyan,Rafayel Mkrtchyan,Hrant Khachatrian,Theofanis P. Raptis
2025 · DOI: 10.1109/MLSP62443.2025.11204276
International Workshop on Machine Learning for Signal Processing · 引用数 1
TLDR
This work proposes a physics-based feature engineering approach combined with a U-Net architecture featuring ResNet-34 encoder and Atrous Spatial Pyramid Pooling module to reconstruct indoor pathloss maps from extremely sparse ground-truth samples to achieve competitive performance across both uniform and strategic sampling scenarios.