Use this URL to cite or link to this record in EThOS:
Title: GPU accelerated onboard data processing for downlink optimisation
Author: Davidson, Rebecca
ISNI:       0000 0004 7972 3374
Awarding Body: University of Surrey
Current Institution: University of Surrey
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
The dimensionality and volume of raw payload data generated onboard Earth Observation (EO) satellites has increased beyond the capabilities of satellite downlink technologies, as a result a bottleneck in the data delivery chain has developed. This data bottleneck must be alleviated in order for EO satellites to efficiently deliver the quality and quantity of payload data now expected by its reliant applications. In this thesis, hardware architectures, processing algorithms and software design are aspects explored towards a solution. As a result, a new onboard satellite data processing architecture is proposed. The key novelties of the proposed system are the use of a Graphical Processing Unit (GPU), to facilitate state-of-the-art image processing, and the highly flexible nature of the architecture, enabling an adaptive processing chain that can be deployed across numerous platforms and missions. In addition, the research documented in this thesis aims to demonstrate the viability and evaluate the advantages of using low-power GPUs in an onboard data processing system. Onboard suitable GPU optimised software development approaches are proposed and practically assessed by leveraging the state-of-the-art image compression algorithm, CCSDS-123, as a case study. Firstly, application development for maximised processing throughput is investigated using hyperspectral and multispectral EO data sets. The processing throughput, compression ratio and power consumption of the new CCSDS-123 image compression GPU application are assessed and characterised for a desktop GPU and the onboard representative low power NVIDIA Jetson TX1 GPU platform. Secondly, software based error injection experiments are leveraged to investigate the error resilience of the CCSDS-123 GPU application. This is a vital area of research which is required to facilitate the wider acceptance and use of GPU devices in space and safety critical applications, where errors are possible and cannot be tolerated. Using these results new error mitigation techniques are also proposed and evaluated.
Supervisor: Bridges, Christopher Sponsor: Surrey Space Centre
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral