Poster
in
Workshop: Differentiable Almost Everything: Differentiable Relaxations, Algorithms, Operators, and Simulators
Differentiable Wireless Simulation with Geometric Transformers
Thomas Hehn · Markus Peschl · Tribhuvanesh Orekondy · Arash Behboodi · Johann Brehmer
Keywords: [ learning to simulate ] [ Equivariance ] [ Differentiable Simulation ] [ Geometric Deep Learning ] [ inverse problems ] [ electromagnetic signals ] [ wireless communication ]
Modelling the propagation of electromagnetic signals is critical for designing modern communication systems. While there are precise simulators based on ray tracing, they do not lend themselves to solving inverse problems or the integration in an automated design loop. We propose to address these challenges through differentiable neural surrogates that exploit the geometric aspects of the problem. We introduce the Wireless Geometric Algebra Transformer (Wi-GATr), a generic, equivariant backbone architecture for simulating wirelesspropagation in a 3D environment. Further, we introduce two datasets of wireless signal propagation in indoor scenes. On these datasets, weshow the data-efficiency of our model on signal prediction and applicability to inverse problems based on differentiable predictive modelling.