Explainable Artificial Intelligence for Intelligent Signal Detection and Interference Mitigation in Cognitive Radio Networks

Abstract

Cognitive Radio Networks (CRNs) enable dynamic and opportunistic utilization of the radio spectrum by allowing secondary users to sense and access underutilized frequency bands without causing harmful interference to licensed primary users. Deep learning models have substantially improved the accuracy of spectrum sensing and signal classification tasks in CRNs; however, their black-box nature limits transparency, trust, and regulatory acceptance in safety-critical wireless deployments. This paper proposes an Explainable Artificial Intelligence (XAI)-enabled framework that integrates a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) architecture with post-hoc interpretability techniques, namely SHapley Additive exPlanations (SHAP), Local Interpretable Model-Agnostic Explanations (LIME), and Gradient-weighted Class Activation Mapping (Grad-CAM), for intelligent signal detection and interference mitigation. The proposed system further incorporates a reinforcement-learning-based dynamic spectrum access controller that reallocates channels based on interpreted decisions, thereby closing the loop between detection, explanation, and mitigation. Simulation results on a synthetic RadioML-inspired dataset demonstrate that the proposed XAI-CNN-LSTM model achieves 96.7% detection accuracy, outperforming energy detection, support vector machine, random forest, and conventional black-box CNN baselines, while SHAP-based feature attribution reveals signal-to-noise ratio and spectral energy as the most influential features governing classification decisions. The results confirm that embedding explainability into cognitive radio signal processing pipelines improves both diagnostic transparency and interference mitigation efficiency without compromising detection performance.