Rapid Prototyping Framework for Intelligent Arrays with Heterogeneous Computing

The availability of highly capable RF platforms with integrated heterogeneous compute has necessitated the need for software development frameworks that allow application designers to take advantage of these newly available capabilities. By leveraging open-source libraries and frameworks including GNU Radio, PyTorch, and Vitis AI, we have developed a framework that enables rapid prototyping and development of intelligent array processing applications and other SDR applications. This framework enables researchers and developers to deploy applications that make use of CPU, GPU, and Xilinx DPU resources, without the need to be a GPU or FPGA expert. Our initial kernel performance metrics show that this new ability will allow system designers to efficiently leverage trade-offs in system performance by effectively allocating resources across several processors.