Traditionally, back-ends for radio telescopes have been built using a hardware-based approach with ASICs, FPGAs, etc. With advancements in processing power of CPUs, software-based systems have emerged as an alternative option, which has received additional impetus with the advent of GPU-based computing. We present here the design of a hybrid system combining the best of FPGAs, CPUs and GPUs, to implement a next generation back-end for the upgraded GMRT. This back-end can process 400 MHz bandwidth signals from 32 dual-polarized antennas, for both interferometry and beamformer applications, including narrowband spectral line modes for the interferometer, incoherent array and phased array mode of operations for the beamfomer, and also a voltage mode attached to a real-time coherent dedispersion system for the beamformer. We describe in detail the design and architecture of this system, including the novel features and capabilities. We also present sample results from the system that validate its performance in conjunction with the entire receiver chain of the upgraded GMRT.

No abstract

:The GMRT Backend Group is in the process of developing wide-band digital backends as part of a major upgrade of the GMRT. The main requirements are for processing 400 MHz bandwidth signals from 30 antennas for interferometry and array mode operations, along with additional features like narrow band modes and RFI cancellation schemes. Both FPGA and GPU based designs have been explored and this paper presents the design and implementation details along with results from testing of prototype units. A comparison of the resources needed and performance achieved on FPGA and GPU based designs is presented, along with an extrapolation of these designs to a 30 station backend for the final GMRT system.