A Secure Deep Autoencoder-based 6G Channel Estimation to Detect/Mitigate Adversarial Attacks

Abstract

Channel estimation (CE) is critical in wireless communications. However, it is vulnerable to adversarial attacks (AA) that are associated with the incorporated artificial intelligence (AI) functionality in 6G wireless communication systems/networks. The hazardous threat can compromise communications’ confidentiality and integrity due to the expected infrastructure, features, and AI models of the 6G paradigm. This paper proposed a deep autoencoder (DAE)-based 6G CE model to detect and prevent AA. It was trained using a dataset generated from the MATLAB toolbox for AA and incorporated a secure transmission protocol. Simulations were conducted to evaluate the model’s performance under different parameters (i.e., CE and DAE) with maximal epsilon values range (0.5-3.0). The results proved the model’s sufficiency of accuracy and security to detect AA compared to existing CE techniques. The proposal provided a promising solution for a secure 6G DAE-based CE and showed robustness against AA. Additionally, it offered a feasible solution for the deep learning training data required and avoids overfitting. Overall, the proposed model provides a valuable contribution towards enhancing the security of 6G networks, and its performance should be further validated in real-world scenarios.