Spotting radio transients with the help of GPUs

Exploration of the time-domain radio sky has huge potential for advancing our knowledge of the dynamic universe. Past surveys have discovered large numbers of pulsars, rotating radio transients and other transient radio phenomena; however, they have typically relied upon off-line processing to cope with the high data and processing rate. This paradigm rules out the possibility of obtaining high-resolution base-band dumps of significant events or of performing immediate follow-up observations, limiting analysis power to what can be gleaned from detection data alone. To overcome this limitation, real-time processing and detection of transient radio events is required. By exploiting the significant computing power of modern graphics processing units (GPUs), we are developing a transient-detection pipeline that runs in real-time on data from the Parkes radio telescope. In this paper we discuss the algorithms used in our pipeline, the details of their implementation on the GPU and the challenges posed by the presence of radio frequency interference.