Co-Processor Aiding for Real-Time Software GNSS Receivers

Increasing interest exists in replacing hardware components of GNSS receivers with software. Moving the digitization closer to the antenna has the major benefit of added flexibility. Algorithms can be developed and tested quickly. Implementations can be modified with a simple software update rather than a hardware change. General purpose radios can be used for GNSS, requiring only specific software, not necessarily hardware. The major drawback, however, is the need for large amounts of processing power to perform Doppler removal and correlation (DRC). This is especially true with the introduction of new, higher bandwidth signals which require sampling rates upwards of 40 Msps. With each channel being tracked requiring six multiplications and four additions per sample per code phase, to achieve realtime operation it becomes necessary to offload some of the processing from the central processing unit (CPU) onto a coprocessor. To the best of the authors knowledge, the only functional Graphics Processing Unit (GPU) aided GNSS receiver in development is in the PLAN group (Petovello et al 2008). The receiver is based on GSNRxTM – the PLAN group’s software GNSS receiver – and is capable of performing DRC processing for eight satellites on 1 ms of 25 Msps data in less than 1 ms, suggesting real-time capability, although no real-time capability is implemented yet. This research starts with a complete re-design of the DRC module and extends the real-time capability of GSNRxTM to 40 Msps. The new stand-alone DRC module for the GPS L1 C/A signal has been developed using the NVIDIA CUDA software development kit. The module employs a multithreaded asynchronous design and as such utilizes CPU resources only when initiating correlation tasks. All DRC operations are performed on an NVIDIA GeForce 8800 GTX. The DRC module has been integrated into GSNRxTM. Timing as well as tracking and navigation solution results are given.