[en] This work addresses the challenge of implementing an artificial intelligence-driven flexible payload onboard for next-generation satellites. Within the SPAICE project, we present the design and hardware deployment of hardware-optimized machine learning models for flexible payload and adaptive beamforming.
The models are restructured to reduce memory and parameter overhead, then quantized and compiled for the Versal ACAP AI platform. Optimization strategies, including Cross-Layer Equalization and Fast Fine-Tuning, mitigate quantization losses while maintaining near-floating-point accuracy. Experimental results demonstrate significantly faster inference than workstation implementations, confirming the feasibility of deploying
advanced machine learning models onboard satellites for real-time, reconfigurable payload operation with high computational efficiency.
Research center :
Interdisciplinary Centre for Security, Reliability and Trust (SnT) > SIGCOM - Signal Processing & Communications
Disciplines :
Aerospace & aeronautics engineering
Author, co-author :
GARCES SOCARRAS, Luis Manuel ; University of Luxembourg > Interdisciplinary Centre for Security, Reliability and Trust (SNT) > SigCom
CUIMAN MARQUEZ, Raudel ; University of Luxembourg > Interdisciplinary Centre for Security, Reliability and Trust (SNT) > SigCom
This work has been supported by the European Space Agency (ESA), which funded it under Contract No. 4000134522/21/NL/FGL, ”Satellite Signal Processing Techniques using a Commercial Off-The-Shelf AI Chipset (SPAICE).” Please note that the views of the authors of this paper do not necessarily reflect the views of ESA
