System Level Power and Energy Modeling for Signal Processing Applications

This article presents a new methodology of consumption and performance characterization of software's intellectual property (IPs) computing on DSPs. These IPs are generally submitted to various constraints especially the real time and energy. The proposed approach exploits parametric models representing the consumption's behavior of both DSP's architecture and algorithm. High-level parametric laws of consumption and performance are established. This methodology permits to deduce the energy and the adequate configuration of many applications written in an advanced language (Strict ANSI-C) for a given target. The approach is proved using signal processing applications: Fast Fourier Transform and Finite Impulse Response filter running on C6710 and C5510.