posted on 2025-05-01, 00:00authored byNaveed Naimipour
The rapidly expanding volume and complexity of data in space systems pose significant challenges to traditional signal processing methods, necessitating innovative approaches that can efficiently adapt to these demands. This thesis explores the transformative role of machine learning in enhancing signal recovery for space signal processing, addressing critical challenges with computationally efficient and scalable solutions.
We begin by examining clustering algorithms tailored for large datasets, a cornerstone of modern data analysis. Two novel variants—2-Hard and 2-Soft clustering—are introduced to improve computational efficiency and adaptability, including extensions designed for scenarios with incomplete knowledge. These clustering methods are further demonstrated in practical space systems, where they enhance forward error correction processes, showcasing their utility in real-world applications.
Building on this foundation, we turn to the problem of phase retrieval, a key task in space signal recovery, e.g., in space telescope image formation. We propose a deep unfolded phase retrieval (UPR) framework, blending machine learning with model-based techniques to optimize both encoding and decoding processes. Our framework not only improves computational dynamics but also achieves superior performance compared to traditional algorithms, as demonstrated by extensive empirical evaluations.
Lastly, we explore the emerging frontier of quantum compressive sensing (QCS), integrating classical machine learning principles with quantum computing methodologies to advance signal recovery techniques. Using practical datasets such as LIDAR, we evaluate the robustness of the proposed QCS framework under both ideal and quantum noise conditions. The results highlight the potential of QCS for data recovery and its promising outlook in reshaping the landscape of space signal processing.