Jongsoo Kim (Korea Astronomy and Space Science Institute)
Topic: Lessons learned from recent digital correlator projects (LL), Emerging digital correlator concepts, technologies, and architectures (EC)
Lessons Learned from the Development of the ACA Spectrometer Based on GPU Technology
The development project of the ACA (Atacama Compact Array) Spectrometer based on GPU technology is led by the KASI (Korea Astronomy and Space Science Institute) in collaboration with the NAOJ (National Astronomical Observatory of Japan). The project successfully finished the CDMR (Concept Design and Manufacturing Review) on December 2019. The ACA Spectrometer is designed to provide auto-correlation power spectra for science targets and cross-correlation outputs for calibrations of the Total Power Array. The hardware of the ACA Spectrometer is composed of 4 GPU servers. Each server has 4 NVIDIA TITAN V cards and two data acquisition cards, all of which are plugged into the PCIe (Peripheral Component Interconnect Express) slots of the server. One GPU card has enough performance to process the 2 GHz bandwidth of one baseband pair. The prototype of the ACA Spectrometer successfully obtained SiO spectral lines using Nobeyama 45-m Telescope.
There are several lessons learned from the ACA Spectrometer project for a future ALMA correlator, which are itemized in the followings.
* Number of correlators: One ALMA correlator with flexible sub-array configurations
* Installation site: OSF
* Data Transport Protocol: a general-purpose protocol, for example, TCP/IP
* Sampling bit: At least 4-bit, and 5 or 6-bit may be needed for Band 1 and RFI mitigation
* Spectral leakage: need to implement the polyphase filter bank capability
In my talk, I will elaborate on each of the above items.