RadioKMoE: Knowledge-Guided Radiomap Estimation with Kolmogorov-Arnold Networks and Mixture-of-Experts

TL;DR

This work tackles radiomap estimation in complex urban environments by proposing RadioKMoE, a two-stage, knowledge-guided framework. It first uses Kolmogorov–ArnolD Networks (KAN) to produce a fast, global coarse prior of the radio map, then refines this prior with a Mixture-of-Experts (MoE) module that partitions the scene into regions and specializes experts accordingly, aided by environmental cues and a physics-inspired depth map. The approach demonstrates superior accuracy over baselines in both multiband and single-band settings across sparse sampling regimes, validating the effectiveness of combining global propagation priors with region-wise specialization. The method offers robust, scalable radiomap reconstruction for dynamic spectrum management in next-generation networks.

Abstract

Radiomap serves as a vital tool for wireless network management and deployment by providing powerful spatial knowledge of signal propagation and coverage. However, increasingly complex radio propagation behavior and surrounding environments pose strong challenges for radiomap estimation (RME). In this work, we propose a knowledge-guided RME framework that integrates Kolmogorov-Arnold Networks (KAN) with Mixture-of-Experts (MoE), namely RadioKMoE. Specifically, we design a KAN module to predict an initial coarse coverage map, leveraging KAN's strength in approximating physics models and global radio propagation patterns. The initial coarse map, together with environmental information, drives our MoE network for precise radiomap estimation. Unlike conventional deep learning models, the MoE module comprises expert networks specializing in distinct radiomap patterns to improve local details while preserving global consistency. Experimental results in both multi- and single-band RME demonstrate the enhanced accuracy and robustness of the proposed RadioKMoE in radiomap estimation.

Figures (5)

• Figure 1: Structure of the Kolmogorov–Arnold Network (KAN): (a) KAN frameworks; (b) Shapes of spline kernels in KAN; (c) BEM representation of multiband spectrum strengths.
• Figure 2: KAN prediction results in free-space (top row) and in an urban scene with buildings (bottom row), where complex environments degrade the performance of KAN prediction.
• Figure 3: Overview of the proposed RadioKMoE architecture. Stage 1: Coarse resolution radiomap is generated by KAN as priors. Stage 2: The knowledge-guided priors (KAN-based coverage priors, radio depth map) and environment priors (building map, transmitter locations) are input into a MoE-based transformer for refinement and final radiomap estimation.
• Figure 4: Comparison for predictions of multiband radiomaps ($2.4$ GHz, $3.65$ GHz, and $4.90$ GHz) on the BRAT-LabW dataset at a 1% sampling ratio. Each row corresponds to a different frequency, and values after the slash indicate MSE ($\times10^{-4}$).
• Figure 5: Comparison in the RadioMapSeer dataset at a 1% sampling ratio. Values after the slash indicate MSE ($\times10^{-4}$).