PhyG-MoE: A Physics-Guided Mixture-of-Experts Framework for Energy-Efficient GNSS Interference Recognition

2026-01-19

AI Summary

Static deep learning models struggle to dynamically allocate computational resources in response to the physical complexity of GNSS interference signals, often leading to an imbalance between energy efficiency and accuracy. To address this challenge, this work proposes PhyG-MoE, a novel framework that introduces physics-guided dynamic sparse computation for GNSS interference identification. By leveraging spectral feature entanglement to drive a gating mechanism, the model adaptively routes inputs to either lightweight or high-capacity TransNeXt expert modules, achieving precise alignment between computational expenditure and electromagnetic environment complexity. Evaluated on a dataset encompassing 21 interference types, the proposed method attains an accuracy of 97.58% while significantly reducing computational overhead without sacrificing performance, offering an efficient solution for resource-constrained cognitive receivers.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Intelligent Robots: State Estimation

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Complex electromagnetic interference increasingly compromises Global Navigation Satellite Systems (GNSS), threatening the reliability of Space-Air-Ground Integrated Networks (SAGIN). Although deep learning has advanced interference recognition, current static models suffer from a \textbf{fundamental limitation}: they impose a fixed computational topology regardless of the input's physical entropy. This rigidity leads to severe resource mismatch, where simple primitives consume the same processing cost as chaotic, saturated mixtures. To resolve this, this paper introduces PhyG-MoE (Physics-Guided Mixture-of-Experts), a framework designed to \textbf{dynamically align model capacity with signal complexity}. Unlike static architectures, the proposed system employs a spectrum-based gating mechanism that routes signals based on their spectral feature entanglement. A high-capacity TransNeXt expert is activated on-demand to disentangle complex features in saturated scenarios, while lightweight experts handle fundamental signals to minimize latency. Evaluations on 21 jamming categories demonstrate that PhyG-MoE achieves an overall accuracy of 97.58\%. By resolving the intrinsic conflict between static computing and dynamic electromagnetic environments, the proposed framework significantly reduces computational overhead without performance degradation, offering a viable solution for resource-constrained cognitive receivers.

Problem

• GNSS interference recognition
• static deep learning models
• computational resource mismatch
• signal complexity
• dynamic electromagnetic environments

Innovation

• Physics-Guided
• Mixture-of-Experts
• Dynamic Routing
• Spectral Entanglement
• Energy-Efficient GNSS