VVC Interpicture Prediction Using SAD with Imprecise Subtractors: A Quantitative Analysis.

Real-time video encoding demands intense computational effort resulting in large power dissipation and energy consumption. Therefore, energy efficiency becomes mandatory especially for energy-constrained systems such as battery-powered mobile devices. In this context, approximate computing arose as a promising approach to improve energy efficiency. This work presents an analysis of the SAD distortion metric using approximate subtractors in interpicture prediction considering the VVC reference encoder. The approximate 10-bit subtractors were evaluated considering four levels of imprecision (in number of approximate bits). The results achieved show the potential in the adoption of approximate subtractors both in coding efficiency and in power reduction. The SAD architecture can achieve a power reduction of 16.45% when compared to SAD using RCA subtractor, while not degrading the compression efficiency (less than 0.1% BD-Rate increase).