Reservoir-Based Distributed Machine Learning for Edge Operation of Emitter Identification

This paper has several contributions, all motivated by the operational aspects of in-situ retrainable Specific Emitter Identification (SEI) for authentication of mobile emitters at the Edge, tactical or IoT. The paper first provides a review of the prior work (DLR) that uses our design of reservoir delay loops (DL) to implement low-power, high accuracy and high-reliability classifiers of signals represented as time series of samples, capable of in-situ training at the Edge. We analyze those DLR properties that enable seamless authentication of mobile emitters on a larger scale using radio frequency (RF) fingerprints. Delay loops project the SEI inputs into a space where different input classes are linearly separable, allowing the use of a linear classifier for emitter identification. Moreover, the architecture of split loops enables a more effective linear separation, constraining the number of weight coefficients, which is important for efficient integration of locally trained DLRs into a global SEI model (D-DLR). D-DLR enables mobile edge platforms to authenticate and then track emitters. To authenticate mobile devices across large regions, D-DLR is trained in a distributed fashion with very little additional processing and a small communication cost, all while maintaining accuracy. We illustrate how to merge locally trained DLR SEI classifiers, and how to reliably detect unseen emitters using a simple multi-layer perceptron to which the DLR weights have been transferred.