DeepRx is one of the most successful machine learning solutions for PHY-layer processing, implementing a significant subset of receiver functionality within a single learned neural network. This has led to multiple adaptations, such as MIMO DeepRx and Hybrid DeepRx. However, DeepRx has not yet been explored in the context of Non-Terrestrial Networks (NTN).

This project investigates the applicability of DeepRx for NTN, more specifically on received signal processing on satellites. First, on a theoretical level, the functionality of DeepRx will be analyzed, to the align the findings with the unique challenges of NTN. Afterwards, using the OpenNTN and Sionna frameworks, end-to-end simulations of NTN systems will be conducted to develop a DeepRx-based architecture optimized for NTN scenarios. Additionally, to address the challenge of limited computational resources on satellites, the new architecture will be designed for minimal complexity while maintaining high performance.