Towards Evaluating Adversarial Attacks Robustness in Wireless Communication

Abstract

The emerging new technologies, such as autonomous vehicles, augmented reality, IoT, and other aspects that are revolutionising our world today, have highlighted new requirements that wireless communications must fulfil. Wireless communications are expected to have a high optimisation capability, efficient detection ability, and prediction flexibility to meet today’s cutting-edge telecommunications technologies’ challenges and constraints. In this regard, the integration of deep learning models in wireless communications appears to be extremely promising. However, the study of deep learning models has exhibited inherent vulnerabilities that attackers could harness to compromise wireless communication systems. The examination of these vulnerabilities and the evaluation of the attacks leveraging them remains essential. Therefore, this paper’s main objective is to address the alignment of security studies of deep learning models with wireless communications’ specific requirements, thereby proposing a pattern for assessing adversarial attacks targeting deep learning models embedded in wireless communications.