Simplifying Multi-Domain Specific Emitter Identification

Wireless communications systems are a seamless, ubiquitous part of daily life, which necessitate their security; however, consistent, pertinent, and effective security remains an ongoing challenge. Physical layer security mechanisms, such as Specific Emitter Identification (SEI), are an alternative security mechanism that can complement bit-level security mechanisms and can assist in addressing this challenge by providing the "something you are" to Multi-Factor Authentication's "something you have", and "something you know". Currently, Deep Learning (DL) based SEI is a prominent physical layer security approach; however, it faces a fundamental challenge known as the multi-domain problem. The multi-domain problem is when a DL trained on one set of data suffers severe performance degradation when classifying data from another set it did not see during training. This work introduces a multi-domain SEI approach addressing this performance degradation by mitigating waveform collection artifacts and procedures while simultaneously reducing the DL architecture's complexity, resulting in a computationally efficient methodology that generalizes well across emitter types and wireless protocols while outperforming more complicated domain adaptation techniques. The presented multi-domain SEI approach uses a singlelayer Long Short-Term Memory (LSTM) to achieve accuracies up to 99.37% when identifying eight Wireless-Fidelity (WiFi) emitters and 98.1% when identifying eight ZigBee emitters with single waveform decisions made in approximately 1ms and 0.89ms, respectively.