Addressing the challenges facing deep learning based Specific Emitter Identification via preamble based waveforms

The purpose of this study is to conduct in-depth experiments that analyze the effects on Deep Learning (DL) based Specific Emitter Identification (SEI) and address three issues facing the field. SEI is targeted as a physical-layer security measure that can identify radios within an Internet of Things (IoT) deployment without the need of digital credentials. In the current space, DL SEI is still in its infancy, and has not had the incubation time for a tailor-suited approach to solve issues facing SEI. This thesis introduces methods of improving DL SEI using transforms to allow the networks to learn features that reduce computational cost and improve security. Overall, this thesis highlights the introduction of (i) the natural logarithm as a computationally inexpensive transform of preamble-based waveforms, (ii) assessment of the impacts signal energy has on DL SEI, and (iii) an approach to improving the multi-day classification performance of IEEE 802.11a OFDM emitters