Data Independent Identification for Privacy Preservation

A potential data independent physical layer identification mechanism is proposed to defend against impersonation attacks and protect the privacy of the users with authenticated devices. A nonlinear function is derived from the impairments of a unique device to be used as the fingerprint in this mechanism and two methods are presented to learn this function. One is Kernel Regression which is commonly used for function modeling in Machine Learning and the other one is Neural Network which can theoretically approximate any continuous function arbitrarily well. In this paper, the implementation and experiment results of Kernel Regression are discussed in detail due to its higher efficiency. The experiment results show that the training and testing accuracy of nonlinear function modeling using kernel regression can reach nearly 90%, indicating that it is a potential scheme for data independent identification.