High-efficiency and cost-sharing architecture design of fast algorithm based multiple 4×4 and 8×8 forward transforms for multi-standard video encoder

摘要

In this work, fast algorithm-based multiple transforms with a hardware-sharing design are exploited for 4×4 and 8×8 forward transforms in H.264/AVC, VC-1, HEVC, and AVS standards, and for 8×8 forward transforms in MPEG-1/2/4 schemes. The 4×4 VP8 and AVS-M fast forward transforms are developed by cost-sharing hardware for multi-standard video encoding applications. By matrix factorizations, our proposed 1D hardware-sharing transform architecture is realized by only shifters and adders. Compared with the directly combined fast algorithms without hardware-sharing functionality, our proposed architecture reduces the number of shifters and adders by 24.5% and 73.4%, respectively. Compared with existing multi-standard transform designs, our proposed architecture reveals larger normalized hardware efficiency. Our proposed hardware sharing based 1-D forward transform supports the Full HD (1920×1080@60Hz) specification with the 110.8MHz working frequency.