Earth observation satellites are continuously evolving due to the advances in digital
technologies and sensor performances, and both the commercial and defence
markets are requiring for instruments able to provide for high-resolution images,
such as synthetic aperture radars and hyperspectral imagers. However, the classic electronic
system of an Earth observation satellite presents some bottlenecks that do not
allow fully exploiting the potential of this high-performing payloads. A first limitation
is represented by the on-board data handling subsystem. Indeed, due to the continuous
growth of the data rate requirement, current solutions results inadequate for supporting
high-speed instrumentation. In particular,
this work will present a detailed analysis of SpaceFibre, the newest link technology
promoted by the European Space Agency.
Downlink capability is another limiting factor for the Earth Observation
market. Deep neural network are arising as a promising technology for optimizing the
downlink bandwidth of Earth Observation satellites. The European Space Agency has
already launched the first in-orbit demonstrator for testing the feasibility of the deep
neural network technology in space, employing an Intel Movidius Myriad-2 visual processing
unit. This work proposes to exploit FPGAs as target
technology for accelerating this class of algorithms, providing an extensive comparison
with the Myriad-2 and showing the advantages and disadvantages of both solutions in
different mission scenarios.