<abstract xmlns="http://www.w3.org/1999/xhtml"> Signal processing of the global navigation satellite systems (GNSS) is a computationally demanding task due to the wide bandwidth of the signals and their complicated modulation schemes. The classical GNSS receivers therefore utilize tailored digital signal processors (DSP) not being flexible in nature. Fortunately, the up-to-date parallel processors or graphical processing units (GPUs) dispose sufficient computational power for processing of not only relatively narrow band GPS L1 C/A signal but also the modernized GPS, GLONASS, Galileo and COMPASS signals.The performance improvement of the modern processors is based on the constantly increasing number of cores. This trend is evident not only from the development of the central processing units (CPUs), but also from the development of GPUs that are nowadays equipped with up to several hundreds of cores optimized for video signals. GPUs include special vector instructions that support implementation of massive parallelism. The new GPUs, named as general-purpose computation on graphics processing units (GPGPU), are able to process both graphic and general data, thus making the GNSS signal processing possible.Application programming interfaces (APIs) supporting GPU parallel processing have been developed and standardized. The most general one, Open Computing Language (Open CL), is now supported by most of the GPU vendors. Next, Compute Unified Device Architecture (CUDA) language was developed for NVidia graphic cards. The CUDA language features optimized signal processing libraries including efficient implementation of the fast Fourier transform (FFT).In this paper, we study the applicability of the GPU approach in GNSS signal acquisition. Two common parallel DSP methods, parallel code space search (PCSS) and double-block zero padding (DBZP), have been investigated.Implementations in the C language for CPU and the CUDA language for GPU are discussed and compared with respect to the acquisition time. It is shown that for signals with long ranging codes (with 10230 number of chips - Galileo E5, GPS L5 etc.).<italic>Paper presented at the "European Navigation Conference 2012", held in Gdansk, Poland</italic></abstract>

Signal processing of the global navigation satellite systems (GNSS) is a computationally demanding task due to the wide bandwidth of the signals and their complicated modulation schemes. The classical GNSS receivers therefore utilize tailored digital signal processors (DSP) not being flexible in nature. Fortunately, the up-to-date parallel processors or graphical processing units (GPUs) dispose sufficient computational power for processing of not only relatively narrow band GPS L1 C/A signal but also the modernized GPS, GLONASS, Galileo and COMPASS signals.The performance improvement of the modern processors is based on the constantly increasing number of cores. This trend is evident not only from the development of the central processing units (CPUs), but also from the development of GPUs that are nowadays equipped with up to several hundreds of cores optimized for video signals. GPUs include special vector instructions that support implementation of massive parallelism. The new GPUs, named as general-purpose computation on graphics processing units (GPGPU), are able to process both graphic and general data, thus making the GNSS signal processing possible.Application programming interfaces (APIs) supporting GPU parallel processing have been developed and standardized. The most general one, Open Computing Language (Open CL), is now supported by most of the GPU vendors. Next, Compute Unified Device Architecture (CUDA) language was developed for NVidia graphic cards. The CUDA language features optimized signal processing libraries including efficient implementation of the fast Fourier transform (FFT).In this paper, we study the applicability of the GPU approach in GNSS signal acquisition. Two common parallel DSP methods, parallel code space search (PCSS) and double-block zero padding (DBZP), have been investigated.Implementations in the C language for CPU and the CUDA language for GPU are discussed and compared with respect to the acquisition time. It is shown that for signals with long ranging codes (with 10230 number of chips - Galileo E5, GPS L5 etc.).Paper presented at the "European Navigation Conference 2012", held in Gdansk, Poland

Gnss Signal Processing in Gpu

Czech Technical University in Prague, Faculty of Electrical Engineering

Artificial Satellites

{"article-title":"Gnss Signal Processing in Gpu"}

Signal processing of the global navigation satellite systems (GNSS) is a computationally demanding task due to the wide bandwidth of the signals and their...

Gnss Signal Processing in Gpu

Published Online: Jun 21, 2013

Page range: 51 - 61

DOI: https://doi.org/10.2478/arsa-2013-0005

This content is open access.