Identification of wireless users via power amplifier imperfections

Variations in the RF chain of radio transmitters can be used as a signature to uniquely associate wireless devices with a given transmission. Previous approaches, which have varied from transient analysis to machine learning, do not provide verifiable accuracy. Here, we detail a first step toward a model-based approach. In particular we exploit differences in nonlinearities of input/output (I/O) characteristics of power amplifiers modeled with Volterra series and develop algorithms for deciding the origin of a given message of interest based on these differences. We consider a generalized likelihood ratio test (GLRT) and a classical likelihood ratio test. For both tests, decision rules are derived and their performance is analyzed. Finally, to establish the viability of the proposed approach, the practical variations among power amplifiers are investigated through simulations and measurements. Results show that the methods can be very effective, when exploiting imperfections of commercially used RF power amplifiers (PAs).