Arjun Radhakrishnan

"If you think you can, or if you think you cannot, either way, you are right." - Henry Ford


About Me

I was born on the 4th of September in 1985 in a small (at the time) city in India called Bangalore. Having finished most of my primary and secondary education in India I decided to do my undergraduate studies at the University of Cape Town in South Africa, studying Electrical and Computer Engineering. On completing that I continued on at UCT pursuing an MSc in Electrical Engineering under the supervision of Prof. Michael Inggs. I am currently in the second year of that course and aim to finish this year.




Past Work

My undergraduate thesis was entitled "Signal Processing on a Graphics Card - An Analysis of Accuracy and Performance", and is available here. It investigated the possible increases in performance offered when using Graphics Cards to run common signal processing primitives such as the Fast Fourier Transform and Finite Impulse Response Filters. I have since continued working with GPUs, looking to accelerate various other signal processing tasks.

MSc Research

For my masters dissertation, I am going to be implementing two signal processing algorithms on a GPU for the Square Kilometre Array Project.

The first application is a part of the imaging process and is called gridding. Gridding is the process used to assign the visibility values obtained from the telescope onto a regular, rectangular matrix or ‘grid’. This is done by using an interpolation algorithm to assign values at the grid points corresponding to each observed data point. The reason we need to place the data on a regular grid is so that we can then use the fast Fourier transform (FFT) to evaluate the output image. I will be implementing this on a GPU and benchmarking it with various grid sizes against a CPU implementation.

The second application is for pulsar dedispersion. Pulsars are rapidly rotating neutron stars and they emit regular pulses of electromagnetic radiation. As this radiation moves through space, it gets affected by the Interstellar Matter (ISM) that causes the lower frequency components of the wave to be delayed more than the higher ones. Thus when the signal arrives on Earth the pulse appears to be spread out over a longer period of time than it actually is. I aim to correct for this error in real time using a method called coherent dedispersion and will benchmark its performance against that of the dedispersion program SIGPROC on the CPU.

Finally I will be making recommendations on the potential to use GPUs for these and other similar problems for the SKA project.

Contact Details

Tel             : (+27)730750738

Email         : arjun53[at]gmail[dot]com