On resource estimation of MPEG-4 video decoding for a multiprocessor architecture.

This paper addresses an efficient implementation of new emerging video algorithms like the coding of arbitrarily shaped video objects in the new MPEG-4 standard. This type of advanced multimedia applications pose challenging requirements on embedded systems design with respect to decomposition and scalability, in order to meet real-time constraints. We study the design of networks-on-chip (NoC), which intrinsically satisfies these requirements [5]. A job scheduler needs to know the worst-case execution time (WCET) of a starting job to ensure that the job can meet its timing constraints. For the purpose of timing analysis, such as computing the WCET, a timing model has been applied which has a linear dependence on a set of inputdependent data parameters. We derive a linear timing model for MPEG-4 video object decoding from a running executable specification. Our timing model is computed and verified with an instruction-set simulator of a RISC processor element containing a flat local memory model. The derived model is accurate within 6% for the average execution time