Paper detail

Solving the Corner-Turning Problem for Large Interferometers

The so-called corner turning problem is a major bottleneck for radio telescopes with large numbers of antennas. The problem is essentially that of rapidly transposing a matrix that is too large to store on one single device; in radio interferometry, it occurs because data from each antenna needs to be routed to an array of processors that will each handle a limited portion of the data (a frequency range, say) but requires input from each antenna. We present a low-cost solution allowing the correlator to transpose its data in real time, without contending for bandwidth, via a butterfly network requiring neither additional RAM memory nor expensive general-purpose switching hardware. We discuss possible implementations of this using FPGA, CMOS, analog logic and optical technology, and conclude that the corner turner cost can be small even for upcoming massive radio arrays.