Indoor radio maps improve as synthetic data and models scale up
Scaling Cheap Synthetic Pretraining Improves Indoor Radio-Map Prediction
Machine LearningArtificial Intelligence
Summary
Good indoor wireless signal maps need lots of real data, but that data is expensive and hard to get. The authors created a way to make many fake indoor building layouts and quickly simulate signal maps, producing a huge dataset. They trained deep learning models on this synthetic data and then improved them using real measurements, leading to better predictions of indoor signals. Their approach works well even though the fake and real data are not perfectly matched.
What this means in practice
- •For wireless network engineers: Predict indoor signal strength maps in various building types more accurately by pretraining models on large synthetic datasets generated with fast ray tracing.
- •For building automation teams: Create reliable indoor wireless coverage predictions to optimize sensor and device placement using pretrained models refined with scarce real measurements.
Authors
Khoren Petrosyan, Artashes Mkrtchyan, Rafayel Mkrtchyan, Hrant Khachatrian, Theofanis P. Raptis
Abstract
Real measurements are too limited to train an accurate indoor pathloss predictor that works across a wide range of buildings. Large-scale synthetic pretraining offers a way forward, but real building layouts are also scarce, and accurate ray-tracing simulators are slow, expensive, and often proprietary. We address both problems by introducing a generator that automatically creates diverse indoor layouts and a fast, open-source approximate ray tracer that computes radio maps for them. Using this pipeline, we generate 128M synthetic samples, pretrain ResNet-101 and Swin-T encoder-decoders, and fine-tune them on high-quality benchmark data. Despite simulator mismatch, our ResNet-101 model reduces error by 15% on average relative to the best known methods across five indoor radio-map prediction tasks and improves transfer on a small real-measurement dataset. Both architectures show predictable data-scaling trends: power laws fitted on 2M-16M samples predict the mean normalized downstream RMSE at 32M-128M with 3.33% mean absolute percentage error for ResNet-101 and 7.10% for Swin-T. These results show that pretraining on large amounts of cheap simulated data can improve indoor radio-map prediction, even when real measurements and high-quality simulations are limited.