Trustworthy Deep Learning for Wireless Channel Estimation: The XAI-CHEST Framework

Abstract

The advent of 6G networks heralds a new era in connectivity, designed to support a diverse array of critical applications such as autonomous driving and remote surgery, where artificial intelligence (AI)-based decisions must be executed in real-time. Key decisions in these applications include resource allocation, localization, and channel estimation. However, the inherent opacity of AI-based models presents significant challenges in understanding and trusting these decisions, especially in safety-critical applications. To address this, explainable AI (XAI) techniques are crucial for elucidating the logic behind these models. This paper introduces a novel XAI-based channel estimation (XAI-CHEST) scheme that enhances the interpretability of deep learning (DL) models used in doubly-selective channel estimation. The XAI-CHEST scheme aims to identify relevant model inputs by applying high noise to irrelevant inputs, allowing for a deeper analysis and evaluation of DL-based channel estimators based on the resulting interpretations. Simulation results demonstrate that the XAI-CHEST scheme provides valid and insightful interpretations across various scenarios, thus paving the way for more trustworthy DL-based solutions in 6G networks.