Application of medium-grain multiprocessor mapping methodology to epileptic seizure predictor

In this paper we present a methodology that enables mapping and scheduling of a dynamic real-time medical signal processing application onto an MPSoC platform. We apply the Task Concurrency Management (TCM) methodology on Lyapunov Exponent calculator, which is a part of an epileptic seizure predictor. TCM requires a division of an application into thread frames and thread nodes. In particular, we demonstrate a new technique for thread node splitting so as to reduce execution time variance. This is necessary to meet stringent energy and performance requirements during mapping and scheduling. Through experiments we verify that the resulting model of the Lyapunov Exponent calculator fulfills the requirements of the TCM methodology.