Reservoir Based Edge Training on RF Data To Deliver Intelligent and Efficient IoT Spectrum Sensors

Current radio frequency (RF) sensors at the Edge lack the computational\nresources to support practical, in-situ training for intelligent spectrum\nmonitoring, and sensor data classification in general. We propose a solution\nvia Deep Delay Loop Reservoir Computing (DLR), a processing architecture that\nsupports general machine learning algorithms on compact mobile devices by\nleveraging delay-loop reservoir computing in combination with innovative\nelectrooptical hardware. With both digital and photonic realizations of our\ndesign of the loops, DLR delivers reductions in form factor, hardware\ncomplexity and latency, compared to the State-of-the-Art (SoA). The main impact\nof the reservoir is to project the input data into a higher dimensional space\nof reservoir state vectors in order to linearly separate the input classes.\nOnce the classes are well separated, traditionally complex, power-hungry\nclassification models are no longer needed for the learning process. Yet, even\nwith simple classifiers based on Ridge regression (RR), the complexity grows at\nleast quadratically with the input size. Hence, the hardware reduction required\nfor training on compact devices is in contradiction with the large dimension of\nstate vectors. DLR employs a RR-based classifier to exceed the SoA accuracy,\nwhile further reducing power consumption by leveraging the architecture of\nparallel (split) loops. We present DLR architectures composed of multiple\nsmaller loops whose state vectors are linearly combined to create a lower\ndimensional input into Ridge regression. We demonstrate the advantages of using\nDLR for two distinct applications: RF Specific Emitter Identification (SEI) for\nIoT authentication, and wireless protocol recognition for IoT situational\nawareness.\n