JPEG-2000 Workload Prediction for Adaptive System on Chip Entropy Coders Architecture

Multimedia applications are quickly becoming the most common workload for embedded systems and portable devices. Video, sound, image applications including digital TV, Web access through wireless transmission are among the various possible tasks to be handled by next generation portable devices. In contrast with traditional workloads mostly static in nature, multimedia workloads exhibit high variability depending on the data processed which when coupled with real time processing requirements make difficult to dimension hardware resources when designing systems on chip. In this paper, we propose the use of a neural network for workload forecasting in order to adapt hardware resources during JPEG-2000 based image compression. The major performance bottleneck in JPEG-2000 being the entropy coder our aim is to adapt in real time the number of concurrent entropy coders through workload forecasting.