In this project, we have developed new design methods and software tools for application of model-based signal processing optimization and graphics processing unit (GPU) acceleration to software defined radio and cognitive radio (SDR-CR) applications — in particular, SDR-CR applications that are developed using GNU Radio.
GNU Radio is a software development framework that provides software defined radio (SDR) developers a rich library and customized runtime engine to design and test radio applications. The applications targeted in this work were SDR-CR systems developed in GNU Radio, which applies graphical modeling principles based on signal processing-oriented dataflow models of computation. Specifically, GNU Radio employs the synchronous dataflow (SDF) model of computation, which is a form of dataflow that is widely used in the design of wireless communication systems.
In this work, we addressed critical challenges in the optimized mapping of SDF graphs in GNU Radio into efficient GPU implementations. First, SDR-CR applications must adhere to complex, multidimensional constraints involving throughput (data rate), latency (response time), energy efficiency, and memory requirements. Second, to leverage the capabilities of GPUs, signal processing kernels developed in GPU-oriented environments such as CUDA must be integrated systematically and efficiently into GNU Radio, which was not originally developed with support for GPU technology in mind. Third, high-level, SDF-based models of wireless communication systems in GNU radio must be analyzed and scheduled (mapped for execution into the targeted GPU) in a manner that systematically integrates the capabilities and performance characteristics of the GPU kernels involved in an SDR system design.
This project was a collaboration among the Maryland DSPCAD Research Group — through its affiliations with the University of Maryland Institute for Advanced Computer Studies (UMIACS) and the University of Maryland CUDA Center of Excellence — and the Laboratory for Telecommunication Sciences in College Park, Maryland.
We refer to this project as the DGIGR Project, which stands for Dataflow and GPU Integration with GNU Radio.
A list of publications from the DGIGR project can be found on the DGIGR Project Publications Page.
Selected presentations from the DGIGR project can be found on the DGIGR Project Presentations Page.
This research was supported in part by the Laboratory for Telecommunication Sciences.
This webpage was last updated on 03/08/2016.