GPU Synthesis of RF Channeliser Outputs for a Variable Bandwidth Microwave Digital Receiver

Abstract

This paper describes the design and implementation of a reconfigurable, software-defined synthesiser, when used in conjunction with a spectral channeliser forms an adaptive intercept receiver for radio frequency (RF) spectrum sensing applications. The software-based design of the synthesiser targets a parallel, multi-core architecture in the form of a graphics processor unit (GPU). Our implementation provides maximum flexibility in terms of the bandwidth and signal duration which can be synthesised. Specifically, the design supports the synthesis of any number of frequency channels that are output from the channeliser front-end and this is achieved on a signal-to-signal basis. Multiple NVIDIA Tesla V100 GPUs were used to benchmark the performance of our GPU synthesis design. A number of representative signal types with variable bandwidth and duration were input into the GPU synthesis framework and the performance measured by computing the maximum throughput of synthesised signals per second. Our analysis also included a comparison of the GPU implementation with that of a multithreaded central processing unit (CPU) implementation. Our GPU implementation was able to achieve a 100× improvement in throughput performance on a single GPU for one of the test signals, and a minimum of 25× improvement in throughput performance for all of the test signals when compared to the CPU implementation.