Beamforming and deep learning techniques for communication in remote sensing satellites

In this dissertation, a Conformal Phased Array Antenna (CPAA)-based solution is proposed to overcome the temporal and coverage limitations encountered in the downlink of imagery acquired by Low Earth Orbit (LEO) Remote Sensing (RS) satellites to ground stations. Within this scope, beamforming mechanisms for different conformal antenna geometries were modeled and evaluated to enable data transmission from a LEO satellite to a Geosynchronous Orbit (GEO) satellite. For the computation of beamforming weights, deep Feed Forward Neural Networks (FFNNs) were employed. When compared with traditional Second-Order Cone Programming (SOCP)-based optimization techniques, the FFNN model achieved similar accuracy while reducing inference time by approximately 1500 times, demonstrating its potential for real-time applications. Additionally, the study presents a detailed evaluation of antenna steering and interference suppression performance by taking into account the gain, power, and interference constraints of communication systems used in GEO satellite links. The proposed method contributes to the timely and secure transmission of RS data from critical regions, while also offering a novel perspective on the applicability of deep learning in space-based communication systems.