Hardware design of motion data decoding process for H.264/AVC

In H.264/AVC, motion data can be basically derived by the following two schemes: one is a typical spatial prediction scheme based on the DPCM and the other is a sophisticated spatiotemporal prediction scheme for the skipped motion data, formally referred to as a direct mode. We verified through instruction level profiling that when these schemes are combined with various H.264/AVC coding techniques, the computational burden to derive the motion data could be considerably aggravated. Specifically, its computational complexity amounts to maximally 55% of that of the overall syntax parsing process. In this paper, we aim at an efficient hardware design of the motion data decoding process for H.264/AVC, for which all the key design considerations are addressed in detail and respective rational answers are presented. As comparing the resulting hardware design with the processor-based solution, its effectiveness was clearly demonstrated. The proposed design was implemented with 43.2 K logic gates and three on-chip memories of 3584 bits using Samsung Semiconductor’s Standard Cell Library in 65 nm L6LP process technology (SS65LP), and was capable of operating the H.264/AVC high-profile video bitstream of 1080p@60fps at 100 MHz consuming 843 μW.