Scientists have long recognized the importance of modes in describing and utilizing the intricate properties of light, as modes are characterized by coherence and orthogonality. Any of the spatial, temporal, frequency, or polarization modes is considered an individual quantum degree of freedom (DoF). Building upon previous innovations, we introduce new perspectives on utilizing modes in the characterization of random media, LiDARs, and photonic processing units. First, we address wavefront distortions of light propagating through random media. We propose to characterize the transfer matrix of coupled multimode transmission channels by representing the wavefronts as superpositions of spatial modes and deploying naturally occurring Rayleigh scattering properties. Our method is beneficial for many applications such as imaging (e.g., endoscopy) and focusing inside random media where the distal end of the optical channel is inaccessible or non-cooperative. Although coherent distributed channel characterization can provide a powerful platform for LiDARs, the applications of spatial and frequency modes in improving LiDAR precision and measurement range will not stop here. We show that using a few-mode local oscillator (LO) with spatial modes at different frequencies at the receiver can significantly enhance the LiDAR detection range. The required signal-to-noise ratio (SNR) for the frequency-modulated continuous wave (FMCW) LiDAR decreases with the number of LO modes. In the few-mode frequency modulated receiver, every spatial mode contributes to the signal detection as an individual element resulting in an improved LiDAR performance by parallelizing the process. In general, optics is scalable and offers many dimensions to parallelize every function. This scalability can also be applied in other applications than LiDARs such as tensor acceleration to escalate the speed and computation power of the photonic processing units. Optics and photonics have great potential to further enhance the performance of neural networks by contributing to three major building blocks of ANNs and deep neural networks (DNNs) including interconnects, matrix multiplication, and nonlinearity. Here, as another application of DoF of light, we demonstrate a photonic tensor accelerator (PTA) based on multidimensional encoding, for the first time. The proposed PTA can perform matrix-vector, matrix-matrix, and batch matrix multiplications in a single clock cycle. The PTA can offer both significantly higher computing power and energy efficiency than state-of-the-art electronic or photonic accelerators.
| Identifer | oai:union.ndltd.org:ucf.edu/oai:stars.library.ucf.edu:etd2020-2908 |
| Date | 01 January 2023 |
| Creators | Fardoost, Alireza |
| Publisher | STARS |
| Source Sets | University of Central Florida |
| Language | English |
| Detected Language | English |
| Type | text |
| Format | application/pdf |
| Source | Electronic Theses and Dissertations, 2020- |
Page generated in 0.0021 seconds