Hofs, Jocelyn
[UCL]
Takasumi Tanabe
[Keio University]
Optical wavemeters and spectrometers are devices that take light as input and are able to recognize the different frequency components in the input signal. These devices have been used in various applications including optical communication and food monitoring. The main issue is that they usually consist of grating structures, which make the devices bulky and expensive. By using nanophotonic technologies, it becomes possible to miniaturize these devices and make it possible to easily integrate them into optical systems. Therefore, there have been some trials using photonic crystals (PhC), but the operation usually relies on precisely controlled resonant structures which are difficult to achieve because of the presence of fabrication errors in such nanophotonic devices that often makes them difficult to put into practical use. Here, we describe a way to overcome this problem by combining light localization (caused by fabrication randomness) with deep learning. We use a simple chirped PhC waveguide (WG) in which light will exhibit different pattern depending on its frequency that will allow us to detect the frequency of the input light by recognizing the associated pattern. This structure also takes advantage of the random localization of light resulting from fabrication errors to increase the resolution of the device. We reconstruct the spectrum by feeding our algorithm with training images of the wavelength sensitive pattern formed by the scattering of light in the structure on which we perform learning. Using this approach, we can obtain a resolution even higher than the wavelength resolution of the fabricated device, which is limited by the resolution of the fabrication technique. By using deep learning, we show that we are not limited by the resolution of fabrication and that we can even use it at our own advantage to increase the resolution of our system. Chapter 1 briefly introduces this work by describing the context and the objectives of this research. Chapter 2 explains the theoretical background of photonic crystals and the photonic crystal structure that we are going to use for our design. Chapter 3 explains the theoretical background regarding deep learning. Chapter 4 describes the experimental procedure that has been made in order to acquire data. Chapter 5 focuses on the main part of this work: how can we predict the frequency components of the input signal with deep learning. Chatper 6 summarizes this work and shows the possible methods for improving the performances of the system.
Bibliographic reference |
Hofs, Jocelyn. Light frequency detection in a chirped photonic crystal waveguide spectrometer using deep learning. Ecole polytechnique de Louvain, Université catholique de Louvain, 2020. Prom. : Takasumi Tanabe. |
Permanent URL |
http://hdl.handle.net/2078.1/thesis:27653 |