Memory energy consumption reduction in video coding systems

Abstract

A very important component in video coding systems is the Motion Estimation (ME), that excels not only in computational cost, but off-chip memory bandwidth consumption as well. These two issues are considered critical constraints for video coding systems, especially for High Definition (HD) videos, since a large volume of data must be processed. This work presents an energy and memory bandwidth reduction approach by using two techniques for external bandwidth reduction from the literature: data reuse and reference frames compression. The proposed solution achieves high rates of memory bandwidth reduction for both writing and reading operations, reaching 31.69% and 86.32% respectively. Therewith, this scheme is able to reduce 80.67% of the ME energy consumption when compared to the traditional method, which presents no bandwidth reduction scheme. The ME search algorithm Full Search (FS) was considered for all evaluations. The solution presented is considered lossless, with no quality degradation for the coded video, and fully complied with current video coding standards.