PhD thesis defense to be held on April 22, 2024, at 14:00 (virtually)


Picture Credit: Kapsis Theodore

Thesis title: MIMO Optical Satellite Systems: Power Allocation with Classical Optimization Methods and with Deep Learning Techniques

Abstract: Fifth-generation (5G) cellular networks are engineered to satisfy the unique requirements of a variety of emerging services, each with escalating demands: Broadband services characterized by low latency, security, high-speed, diversity, and high reliability. Because of their universal coverage, connectivity, scalability, and robustness, satellite communications and non-terrestrial networks are regarded as enabling technologies towards beyond 5G (B5G). High-throughput satellites (HTS) in the Ka-band (26–40 GHz) can be employed for Internet of Things (IoT) applications and backhauling because satellite technologies are now sufficiently mature for these usages. In addition, a number of countries and international space agencies take part in the exploration of Mars, the Moon, deep-space, and near-Earth. In the new space sector, optimizing satellite communications is another crucial field. Conventional radio frequency (RF) lines are unable to offer such high regulation because of the restricted spectrum available. In the satellite community, there has been a growing interest in free space optical (FSO) communication as an alternative to RF communication. Nonetheless, FSO is susceptible to the detrimental effects of scintillation, pointing errors, and clear air turbulence.

This Thesis focuses on the study of optical satellite communication systems and in particular on the modeling of the optical downlink (DL) channel and the optimization of power allocation in multiple-input multiple-output (MIMO) optical satellite networks. Emphasis is placed on the adverse effects faced by the laser beam from propagating in the Earth's atmosphere and on their mitigation techniques. First, for the accurate prediction of DL optical channel irradiance scintillations, a methodology was developed that generates time series using stochastic differential equations and validates them in terms of first and second-order statistics (probability density function, cumulative distribution function, power spectral density). Then, all-optical MIMO satellite system scenarios are considered, and optimal and efficient power allocation algorithms are proposed. The primary problem is formulated as a convex optimization problem with maximum allowed transmit power constraints to maximize capacity and then decomposed into independent convex sub-problems. The entire proposed methodology is based on the water-filling algorithm and satisfies the Karush-Kuhn-Tucker (KKT) conditions. Moreover, optimal power control in hybrid satellite relays is investigated. A dual-hop decode-and-forward optical DL link for a geostationary satellite source is considered. Spatial correlation, correlation matrices and, coefficients are included. Additionally, the power optimization problem is formulated and solved non-linearly for hybrid automatic repeat request (HARQ) schemes in low-earth orbit (LEO) satellites. Then, the power optimization problem incorporates the pointing errors, and a robust maximin methodology is proposed to deal with optical jitter uncertainty. Finally, the Dissertation concludes with the study of neural networks, deep learning, and deep reinforcement learning algorithms. A model-free learning algorithm is proposed that accurately predicts the optimal power distribution in each optical channel.

Supervisor: Professor Panagopoulos Athanasios D

PhD Student: Kapsis Theodore