Architecture Analysis and Trade Study of a Modular PPU for AI-Enabled Earth Observation Payloads

Earth observing satellites generate a lot of raw data from their sensors, whether it being from an optical or radar based sensor. This raw data rate exceeds the possible data rate for downlink, leading to the satellites needing large mass memories to store the data until it can be downlinked to a ground station. This introduces significant downlink latency and limits sensor operational duty cycles, wasting valuable orbital time. A Payload Processing Unit (PPU) can be used to avoid this. The Goal of the PPU is to move the ground based pre- processing and data filtering onboard the satellite using AI algorithms.
This thesis provides a trade study analysis of different architectural designs for the PPU.