Implementation of a high-throughput low-latency polyphase channelizer on GPUs

Scott C Kim,Shuvra S Bhattacharyya

doi:10.1186/1687-6180-2014-141

Scott C Kim, Shuvra S Bhattacharyya

Open Access

https://doi.org/10.1186/1687-6180-2014-141

Copy DOI

Abstract

A channelizer is used to separate users or channels in communication systems. A polyphase channelizer is a type of channelizer that uses polyphase filtering to filter, downsample, and downconvert simultaneously. With graphics processing unit (GPU) technology, we propose a novel GPU-based polyphase channelizer architecture that delivers high throughput. This architecture has advantages of providing reduced complexity and optimized parallel processing of many channels, while being configurable via software. This makes our approach and implementation particularly attractive for using GPUs as DSP accelerators for communication systems.

Highlights

A modern communication transceiver contains two major components: a radio frequency integrated circuit (RFIC) and a baseband processor
3 Related work We introduced the notion of using the graphics processing unit (GPU) as a radio, as a front-end transceiver, in [7], and in this paper, we continue to explore the concept of a GPU front-end (GFE) receiver
6 Results and analysis For our experiments, we employed NVIDIA’s GeForce GTX 680 as the target GPU to implement our design. This GPU is based on the Kepler architecture, which has 1,536 Compute Unified Device Architecture (CUDA) cores (8 SMXs with 192 CUDA cores per SMX), 2 GB of GDDR5 memory, 64 kB of constant memory (CM), and 48 kB of Shared memory (SM)

Summary

Introduction

A modern communication transceiver contains two major components: a radio frequency integrated circuit (RFIC) and a baseband processor. An RFIC is responsible for conversion between analog and digital domain signals, and mixing signals up and down from baseband (BB) to some RF. In SWR systems, signal processing (SP) tasks, such as signal conversion, mixing, resampling, and filtering, are all done in BB using software in the discrete-time sample domain. We assume baseband operation and we are given a wide system bandwidth (BW) that contains multiple channels. These channels do not overlap and are spaced apart. 2.1 Polyphase channelizer A PCZ combines multiple operations into an all-in-one design. We will not derive or discuss the transformation of band-pass filters and mixers into a PCZ (for details on this, see [3]), but rather focus on design and implementation of an efficient PCZ architecture

Objectives

Methods

Results

Conclusion