Event
Proposal Exam: Benjamin Klimko
Tuesday, June 6, 2023
10:30 a.m.
ERF 1207
Maria Hoo
301 405 3681
mch@umd.edu
ANNOUNCEMENT: Ph.D. Research Proposal Exam
Name: Benjamin Klimko
Committee:
Professor Yanne Chembo (Chair)
Professor Thomas Murphy
Professor Thomas Antonsen
Date/time: Tuesday, 6 June 2023 at 10:30am
Location: ERF 1207
Title: Development of photonic reservoir computers for radio frequency signal classification
Abstract:
Machine learning is an alternative information processing paradigm to traditional deterministic tools which use known rules to manipulate input
data to come to an output. In contrast, machine learning models instead use input data and its known output states to determine a set of rules
that may then be applied to infer the state of input data without a known output. This makes machine learning a very powerful tool for working
with complex, real-world data, but it can severely underperform without significant quantities of ground truth data. Depending on the
phenomenon under study, such a quantity of data may not be available. The time and energy needed to train a machine learning system is
also nontrivial, with large systems often taking days to reach acceptable performance levels and the process occurring on power-hungry
graphics cards. Reservoir computing aims to overcome these drawbacks by having the vast majority of connections within the system fixed
and random, instead of fully training the system end to end. Significantly less training data and computational resources are required and the
time to train is decreased because only a readout layer is optimized. A common theme to further reduce computational needs is to replace the
fixed, random connections, or "reservoir," with a physical system that information can be coupled into and out of, with only the readout layer
requiring traditional computing resources.
The optoelectronic oscillator is a nonlinear system comprised of optical and electronic components capable of generating pure radio frequency
signals and is often used in the study of chaos in dynamic time-delayed systems. They have been used in reservoir computing experiments for
over a decade, but to date no experimental work has utilized the native radio frequency operation of the optoelectronic oscillator to perform
computational tasks on communications signals in the passband.
In this proposal we use nonlinear dynamics tools to express an envelope equation for a narrowband optoelectronic oscillator when an
IQ-modulated radio signal is injected into the feedback loop. We then introduce results from recent optoelectronic oscillator-based reservoir
computing experiments showing excellent agreement between theory and simulation. Finally, a series of proposed works are presented.
Due to the interplay of frequency and use of specific radio signals, there are multiple options for extending this work into classification tasks
that are derived from real world needs. Possibilities include radar waveform identification across a wide frequency range and modulation
recognition in the mmWave for 5G and beyond dynamic spectrum access.