Projects:2018s1-177 Radio astronomy with software-defined radio

Project name

Radio astronomy with software-defined radio

Project introduction

Background

Radio astronomy is a combination of electronic field and astronomy. Because the light wave cannot penetrate interstellar dust, optical astronomy has its limitation on the galaxy level observation. However, from 1932, scientists started to use radio frequency system to catch the signal from celestial bodies. Some frequency bands are able to penetrate across the whole galaxy. Therefore, antennas can be used in astronomy observation as radio telescopes.

Software defined radio (SDR) devices has been developed in the last ten years. It provides convenience for students or electronics enthusiasts to build RF receiving system. Graphics processing unit (GPU) is a professional parallel structure processor which becomes more and more popular in recent years, for some algorithm with redundant computations. This project shows a promising application in radio astronomy which combines SDR and GPU.

Due to the atomic transition of electrons, the "spin-flip" line of hydrogen emits from the universe. The frequency of these hydrogen lines is 1420. 406 MHz and they were firstly discovered by Ewen and Purcell in 1951. In recent years, as the emergence of SDR --- a new and relatively inexpensive device, amateurs can re-explore the spectrum of the hydrogen line with a small budget.

System overview

This project aims to design and build up a system to observe the spectrum of hydrogen line from the universe. The system consists of two parts --- radio frequency (RF) and signal processing.

The RF part consists of a dish antenna, a low-noise amplifier, a band-pass filter, and a SDR device.

For signal processing part, a new signal processing method, PFB algorithm, is used to replace Welch’s method for a lower spectral leakage in power spectrum estimation. A mathematical analysis from plain DFT to PFB algorithm is shown. The spectral leakage of plain DFT, windowed DFT and PFB is compared, which shows a huge improvement of the PFB result. Then, we gives an answer that when the frequency of the sinusoid interference is close to the hydrogen line frequency band, the anti-interference performance of PFB algorithm is definitely better than Welch’s method. This algorithm is implemented on GPU, with a parallel computing structure.

Motivation

This is a follow-up project. The experiment results from last year have some drawbacks that need to be improved in this year. Four major drawbacks are listed as follows. 1) Spurs exist in the power spectrum density plots. 2) Weak signals could be swamped. 3) The resolution is not high enough. 4) The system cannot process data in real time.

Objectives

According to the limitations of last year's project, the objective of this year's project is to improve the performance of the system which has been built by the students in last year. The first goal is to reduce the spectral leakage in power spectrum estimation, so as to to improve signal resolution and reduce the effects of radio interference. The second goal is to accelerate the calculation, so as to estimate power spectrum in real time. As a solution, the PFB algorithm is proposed with parallel computing acceleration.

Spectrum estimation

Spectral leakage

Because we cannot calculate data with infinite length, spectral leakage is unavoidable. Therefore, we can only ‘estimate’ the spectrum for time domain data. However, we can take actions to reduce the leakage, which is one of the goals for this project.

Welch's method

When calculating power spectrum density (PSD), the general method consider the square of DFT results as the power for each frequency bin, and divide it by the number of DFT points to be the differential ‘density’. Based on this process, Welch’s method adopt Hamming window to replace the rectangular window in N- point DFT calculation, and average the spectrum results among many N-point segments.

PFB algorithm

PFB is a computational efficient filter structure to reduce the power leakage of the frequency bins of Fast Fourier Transform (FFT). The anti-interference performance of PFB is proved to be better than FFT when the sinusoid interference is close to the hydrogen line. We have compared the spectrum leakage between Welch's method and PFB method, and the PFB result is much better than Welch's result.

Parallel computing hardware

Floating point operations per second

Floating point operations per second (FLOPS) is the quantitative criterion for this performance. FLOPS shows the maximum computation capacity for a device when the algorithm structure is optimal. We compared the FLOPS performance of CPU, FPGA, Raspberry Pi and GPU. Then we choose GPU as the signal processing hardware for our project.

GPU acceleration

Although PFB can significantly improve the spectrum leakage, it brings a big challenge to process massive signal data. In this study, we propose a GPU-based parallel architecture to implement the PFB based signal processing algorithm. The PFB algorithm are separated into several parallel task, and then GPU handles those multiple tasks simultaneously to speed up to signal processing time.

Headline text

GPU acceleration

Team members

Hongwen Qu, Hanwen Chang

Supervisors

Dr Withawat Withayachumnankul, Dr Brian Ng