Learning Radio Maps for Physical-Layer Security in the Radio Access

While there is an evident trend towards “strengthening” the notion of secrecy in information-theoretic analyses, there remains the question of the assumptions under which the promises of physical-layer security may be delivered. A critical aspect here is the level of knowledge of the inherently uncertain wireless environment, required in order to be able to guarantee a certain security performance. A general observation is that the strictness of the notion of physical-layer security increases the focus on the design assumptions of the physical layer and the adequacy of secure transmission strategies. Guided by these insights, we propose a probabilistic secrecy characterization that captures the uncertainty in the communication model coming from the unknown channel state information of the eavesdropper's channel (e.g. due to Eve's unknown position in the wireless environment). Integral to our concept is the use of any-to-any radio maps as underlying context information for the probabilistic secrecy characterization. The radio maps are learned by using recent machine-learning advances in network channel gain cartography. The resulting secrecy maps provide the key ingredient in the integration of physical-layer security in the radio access, and represent a valuable tool for secure radio access system design. The statistical characterization relates to the concept of spatial availability of services in wireless networks, by indicating the locations in a given area where the secrecy-related performance would surmount a given threshold with a guaranteed level of confidence. In this sense, it provides a quality-of-service (QoS) notion for physical-layer security in a spatial context.