Trans-REM: A Two Agent CNN-Transformer Based Approach for Indoor Radio Environment Mapping

Overview

This paper addresses the problem of indoor radio environment mapping (REM), with applications across wireless networks:

• Hybrid Architecture: Introduces Trans-REM, which combines a CNN for local spatial feature extraction with a transformer for capturing global context, leveraging the strengths of both models.
• Auxiliary Inputs: Develops two new input modalities—a line-of-sight (LoS) image and an antenna radiation pattern image—to guide the network in understanding complex indoor propagation characteristics.
• Performance Gains: Demonstrates through extensive numerical evaluations that Trans-REM outperforms state-of-the-art REM methods by at least 15 % in mean squared error on a comprehensive indoor dataset.

Recommended citation: S. Javid, S. Ghose, A. Dwivedi, and S. Sarkar. (2025). "Trans-REM: A Two Agent CNN-Transformer Based Approach for Indoor Radio Environment Mapping." 2025 36th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 2025).
Download Paper | Download Slides | Download Bibtex