An efficient GPU implementation of an arbitrary resampling polyphase channelizer

A channelizer is a part of a receiver front-end subsystem, commonly found in various communication systems, that separates different users or channels. A modern channelizer uses advantages of polyphase filter banks to process multiple channels at the same time, allowing down conversion, downsampling, and filtering all at the same time. However, due to limitations imposed by the structure and requirements of channelizers, their usage is limited and poses significant challenges due to inflexibility using conventional implementation techniques, which are intensively hardware-based. However, with advances in graphics processing unit (GPU) technology, we now have the potential to deliver high computational throughput along with the flexibility of software-based implementation. In this paper, we demonstrate how this potential can be exploited by presenting a novel GPU-based channelizer implementation. Our implementation incorporates methods for eliminating complex buffer managements and performing arbitrary resampling on all channels simultaneously. We also introduce the notion of simultaneously processing many channels as a high data rate parallel receiver system using blocks of threads in the GPU. The multi-channel, flexible, high-throughput, and arbitrary resampling characteristics of our GPU-based channelizer make it attractive for a variety of communication receiver applications.