Explainability analysis based on attribution features for optimizing automatic modulation classification

doi:10.1016/j.comnet.2025.111696

Explainability analysis based on attribution features for optimizing automatic modulation classification

2025-09-16

0

PRE

AI

DOI：10.1016/j.comnet.2025.111696

Original

Original request for help

Save

Abstract

En 中文

• Evaluated four XAI methods (IG, DeepLIFT, SHAP, LIME) on both CNN and LSTM modulation classifier, identifying IG as the optimal interpretability method for wireless signals. • The reliability of the generated interpretability results in terms of sensitivity and invariance was verified through feature deletion and feature sliding tests. • XAI-guided attention mechanisms optimized the training process, enhancing the performance and accuracy of the AI-based AMC model. • Proposed an Explainability Metric for Modulation (EMM) to quantify XAI effectiveness.

Explainability analysis based on attribution features for optimizing automatic modulation classification

Abstract

Journal

Organization

Cited Papers