A reconfigurable wideband streaming channeliser for RF sensing applications: A multiple GPU-based implementation

Abstract

This paper describes the design and implementation of a reconfigurable, software-defined spectral channeliser for radio frequency (RF) spectrum sensing applications. The software-based design targets a parallel, multi-core processor architecture in the form of a graphics processing unit (GPU) and incorporates a polyphase filter bank design. Our implementation provides continuous channelisation of RF intercept data across a scalable number of GPU processing engines. On modern generation GPUs, such as the Tesla K40c, we found that our GPU-hosted channeliser satisfies real-time processing requirements for a current microwave intercept receiver with electronic warfare applications. Specifically, a system with an instantaneous collection bandwidth of 500 MHz and a digital sample rate of 1.333 GSa/s. The configurable nature of our channeliser was not done at the expense of performance, in that a similar computational efficiency was achieved across multiple channel sizes. Our analysis also includes a profiling of the computational load on the GPU and comparing it to that of a single core, high performance, CPU implementation. We found that we were able to achieve up to 40x improvement in performance with our implementation on a single GPU and this result scaled linearly when additional GPU resources were utilised.