Daily Image

18-05-2021
PreviousNext
Click here or on the picture for a full size image.

Cobalt 2.0 phase 2 project finalised

Submitter: Chris Broekema on behalf of the Cobalt 2.0 team
Description: On May 6th we finished phase 2 of the Cobalt 2.0 project with a final demonstration. For an mixed audience of astronomers and personnel from telescope operations, we showed the features that were recently added to the Cobalt 2.0 software.

This demonstration marks the end of the Cobalt 2.0 software development project. We can proudly say that we implemented all features defined in the project plan, and many more to boot.

In the attached slideshow we have collected a number of slides from the various demonstrations given during this project. In order, we show how quantisation of beamformed data impacts data quality. This proofs that data is only slightly changed, while taking up significantly less data volume allowing more beams to be formed and transported to the processing cluster. The next slide shows how we implemented the ability to run multiple beamformer pipelines to run concurrently with an imaging mode observation. We now also allow each beamformer to be re-pointed during an observation.

We fixed two major data quality complaints in this project, shown in slides 3 through 6. In slides 3 and 4 we demonstrate the ability of our new code to track the Doppler shift across multiple channels that previously caused delay errors in the polyphase filterbank. Slides 5 and 6 show how we tackled spectral leakage in the beamformer pipeline. While this is still likely not an ideal solution (this would theoretically require either an infinite length FIR filter, or two additional filters to achieve), initial quality assessments are very promising.

In the course of this project we also added significant quality of life improvements for the developer to the code. the most visible of these is the addition of a GPU kernel performance regression test framework that allows us to easily and quickly identify if a newly implemented piece of code changes the performance of the correlator and beamformer. This is in addition the the functional tests that were already in place. Slide 7 shows an overview of that process.

The final slide in the slideshow gives an overview of the work packages as defined in the project and their current state. Looking back at the end of the project, we can only be very happy with the result, especially considering that the entire project implementation phase ran during COVID lock-down measures. We would like to thank Jan David Mol for his continuous availability answering questions about the system and for his timely and thorough review of the generated code.
Copyright: The Cobalt 2.0 phase 2 team
 
  Follow us on Twitter
Please feel free to submit an image using the Submit page.