Designing a GPU-parallel algorithm for raw SAR data compression: A focus on parallel performance estimation

When a Synthetic Aperture Radar (SAR) acquires raw data using a satellite or airborne platform, it must be transferred to the ground for further processing. For example, SAR raw data need a so-called ’focusing’ signal processing to render it into a visible image. Such processing is time and computin...

Full description

Saved in:
Bibliographic Details
Published in:Future generation computer systems Vol. 112; pp. 695 - 708
Main Authors: Romano, Diego, Lapegna, Marco, Mele, Valeria, Laccetti, Giuliano
Format: Journal Article
Language:English
Published: Elsevier B.V 01.11.2020
Subjects:
CUDA
ENVISAT ASAR image mode level 0
GPU computing
Parallel performance evaluation
Raw SAR image compression
CUDA
Raw SAR image compression
Parallel performance evaluation
GPU computing
ENVISAT ASAR image mode level 0
ISSN:0167-739X, 1872-7115
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:When a Synthetic Aperture Radar (SAR) acquires raw data using a satellite or airborne platform, it must be transferred to the ground for further processing. For example, SAR raw data need a so-called ’focusing’ signal processing to render it into a visible image. Such processing is time and computing consuming, and it is commonly carried out in computing centres. Since the data transfer rate is a typical limitation when communicating with the ground station, compression is necessary to reduce transmission time. So far, this procedure has been implemented in application-specific hardware, but recent adoption of avionic computational GPUs opened to new high-performance onboard perspectives. Due to the limited availability of avionic GPUs, we focused on parallel performance estimation starting from measures relative to a similar off-the-shelf solution. In this paper, we present a GPU algorithm for raw SAR data compression, which uses 1-dimensional DCT transforms, followed by quantisation and entropy coding. We evaluate results using ENVISAT (Environmental Satellite) ASAR Image Mode level 0 data by measuring compression rates, statistical parameters, and distortion on decompressed and then focused images. Moreover, by evaluating the Algorithmic Overhead induced by the parallelisation strategy, we predict the best thread-block configuration for possible adoption of such a GPU algorithm on one of the most available avionic hardware. •Raw SAR images can be compressed through GPU on-board satellites and aircrafts.•No significant quality degradation is measured when focusing decompressed images.•Algorithm performance on GPU can be predicted by means of algorithmic overhead estimation.
ISSN:0167-739X
1872-7115
DOI:10.1016/j.future.2020.06.027