An improved LFS engine for physical layer security augmentation in cognitive networks

Security and privacy within existing wireless architectures remain a major concern and may be further compounded when considering multi-node wireless cognitive networks. However, the same computational capabilities that enable cognitive transceiver operation can also be used to enhance physical-layer security at each node. The approach here uses RF Distinct Native Attribute (RF-DNA) features that embody unique statistical properties of received RF emissions. The baseline system uses a Multiple Discriminant Analysis, Maximum Likelihood (MDA/ML) process to classify devices by exploiting RF-DNA uniqueness that enables serial number discrimination. MDA/ML limitations, to include a lack of feature relevance indication, are addressed using a previously investigated Learning From Signals (LFS) process. Of significance here is the expansion of LFS capability which will be readily implementable in envisioned cognitive network architectures. By coupling Kernel Regression (KR) with a Differential Evolution (DE) genetic algorithm, LFS is able to “learn” an improved model of the signal environment. Results here for experimentally collected 802.11a WiFi signals demonstrate recent improvements to the LFS engine that enable it to operate more effectively within a higher-dimensional RF-DNA feature space. The addition of a fractional Euclidean Distance (ED) similarity metric and vector class labeling provide improvement of 9 % to 23 % in average percent correct classification over the earlier LFS implementation.