Optimization of Calibration Algorithms on a Manycore Embedded Platform

Abstract

This paper presents the porting and the optimization of full polarization, direction independent calibration algorithm for radio-interferometry, on an embedded many-core platform. In astronomy, calibration algorithms consist of solving for the unknown complex antenna gains using a known model of the sky. Calibration is a key computation to provide images of the sky at good quality and high resolutions. In the context of the Square Kilometer Array (SKA) project, real-time and low power execution of the calibration is challenging. In this paper, we show that the CohJohnes algorithm provides good properties for being executed efficiently on the new generation of many-core embedded platforms. Experimental results are provided using the Kalray MPPA Bostan platform running 288 64-bit VLIW cores and delivering up to 845 GFLOPS at 12W.