User Capacity of Wireless Physical-Layer Identification

A wireless physical layer identification (WPLI) system aims at identifying or classifying authorized devices of different users based on the unique radio frequency fingerprints (RFFs) extracted from their radio frequency signals at physical layer. Most existing works mainly focus on demonstrating feasibility of system by presenting the classification performance of a fixed-user-number network. However, the real-world WPLI systems are expected to work in various scenarios with different scales of user numbers, dynamic changes in user numbers, and various choices of user combinations. Hence, an important question needs to be answered: what's the user number that a WPLI system can support, i.e. the user capacity, under the condition that the minimum system performance is guaranteed. In this paper, we theoretically characterize the user capacity of WPLI. A theoretical approach is proposed based on ensemble mutual information between RFF and user identity. With this approach and one-time RFF training, the user capacity of WPLI can be characterized under various real-world constraints. Extensive experiments are conducted to validate the accuracy and tightness of the theoretically derived WPLI user capacity.