Explainable Quantum Learning for Robust Interference Classification in Next-Gen O-RAN

Abstract

As open radio access networks (O-RAN) progress into an intelligence-driven era, artificial intelligence (AI)-powered core functions and applications have recently attracted much attention. The AI-driven techniques enable near-real-time tasks such as signal interference classification for large-scale networks. Accordingly, accurate identification of interference types enables more effective successive interference cancellation, reducing performance degradation and strengthening reliability in O-RAN-based mobile networks. Since quantum computing remains energy-intensive and limited by current hardware capabilities, hybrid quantum–classical machine learning offers an efficient alternative. However, adversarial robustness and explainability remain critical challenges for such hybrid models, as emerging perturbation-based attacks can easily distort their decision boundaries. This work presents Q-SHIELD, a quantum-assisted defense framework that enhances robustness and explainability via trusted-manifold-guided quantum state transformations and fidelity-based boundary reinforcement. Q-SHIELD is evaluated under various poisoning attacks and adversarial attacks. Using a public dataset from real O-RAN testbeds, Q-SHIELD attains 98.2% accuracy in common cases and maintains 84.7% accuracy in worst-case conditions, even under strong adversarial attacks, e.g., quantum projected gradient descent. For explainability, Q-SHIELD emphasizes causal compactness, achieving a drift correlation of 0.988 and sufficiency/comprehensiveness scores of 0.871/0.007 on the top 20% features. These results outperform adversarial-training baselines by up to 35% in defense performance. The results show that Q-SHIELD can deliver robust quantum-assisted services for the intelligent O-RAN.