Transform-Invariant Generative Ray Path Sampling for Efficient Radio Propagation Modeling

TL;DR

A comprehensive machine-learning-assisted framework that replaces exhaustive path searching with intelligent sampling via Generative Flow Networks and applies a physics-based action masking strategy that filters out physically impossible paths before the model even considers them is proposed.

Abstract

Ray tracing has become a standard for accurate radio propagation modeling, but suffers from exponential computational complexity, as the number of candidate paths scales with the number of objects raised to the power of the interaction order. This bottleneck limits its use in large-scale or real-time applications, forcing traditional tools to rely on heuristics to reduce the number of path candidates at the cost of potentially reduced accuracy. To overcome this limitation, we propose a comprehensive machine-learning-assisted framework that replaces exhaustive path searching with intelligent sampling via Generative Flow Networks. Applying such generative models to this domain presents significant challenges, particularly sparse rewards due to the rarity of valid paths, which can lead to convergence failures and trivial solutions when evaluating high-order interactions in complex environments. To ensure robust learning and efficient exploration, our framework incorporates three key architectural components. First, we implement an \emph{experience replay buffer} to capture and retain rare valid paths. Second, we adopt a uniform exploratory policy to improve generalization and prevent the model from overfitting to simple geometries. Third, we apply a physics-based action masking strategy that filters out physically impossible paths before the model even considers them. As demonstrated in our experimental validation, the proposed model achieves substantial speedups over exhaustive search -- up to $10\times$ faster on GPU and $1000\times$ faster on CPU -- while maintaining high coverage accuracy and successfully uncovering complex propagation paths. The complete source code, tests, and tutorial are available at https://github.com/jeertmans/sampling-paths.

Figures (14)

• Figure 1: Comparison of a traditional ray tracing pipeline versus our machine-learning-assisted approach. The core modification replaces the Exhaustive Enumeration bottleneck with an efficient Generative Path Sampler, drastically reducing the validation workload.
• Figure 2: Representation of all possible path candidates from to in a scene with $N$ objects. Each path corresponds to a sequence of object interactions, forming a directed graph where exhaustive search explores all branches. An example path candidate, highlighted in red, is $\mathrm{\glsxtrshort{tx}}\to o_1 \to o_N \to \mathrm{\glsxtrshort{rx}}$.
• Figure 3: Illustration of pathfinding as a sequential decision process. A machine learning model learns to navigate the decision tree (right), sampling interaction sequences (e.g., $\mathrm{\glsxtrshort{tx}}\to o_3 \to o_1 \to \mathrm{\glsxtrshort{rx}}$) that correspond to valid geometric paths in the scene (left) while avoiding the expensive exploration of invalid branches.
• Figure 4: High-level representation of the proposed surrogate model and the computation of the reward for each sample path candidate. Each trapezoid represents a neural network module with learnable parameters, $\boldsymbol{\theta}$, which are different for each module. A different random key is used to sample each path candidate and each object within the path candidate, ensuring diversity in the generated paths.
• Figure 5: Illustration of the path generation process for second-order ray paths. Starting from the empty path candidate (), the model sequentially selects scene objects from a forward policy $\pi(\boldsymbol{p}' | \boldsymbol{p})$. The process terminates after $K$ steps, and connects to , forming a complete candidate path.

Theorems & Definitions (1)

• proof