Deep Neural Networks Meet CSI-Based Authentication

This approach helps network application developer, as it assigns different tasks to different layers. For instance network services are offered by higher layers while having lower layers providing the communication links. In this paradigm encryption or authentication are usually assigned to higher layers [2]. Most of the conventional security protocols are based on complexity of solving of an inverse problem and relying on attacker's limited computational capability. Thepower of different protocols, thus, depends on how time consuming is to solve that problem. The mechanism and the overhead of key management and distribution are also among the main features determining the power of a security protocol. Forwireless communication links, instead of using higher layer protocols, several techniques have been proposed that try to ensure security right at the Physical Layer of the OSI layering, [2], [3].These methods are designed to exploit the features of the wireless medium itself and usually use channel information to propose powerful and practical security protocols for encryption and authentication. Authentications based on non-cryptographic methods have been proposed by [4], [5], [6] and [7].