DocumentCode
2541498
Title
High performance VLSI architecture of fractional motion estimation in H.264 for HDTV
Author
Yang, Changqi ; Goto, Satoshi ; Ikenaga, Takeshi
Author_Institution
Graduate Sch. of IPS, Waseda Univ., Tokyo
fYear
2006
fDate
21-24 May 2006
Abstract
Fractional motion estimation (FME) on sub-pixels will occupy almost over 45% of the computation complexity of H.264 encoding process. Therefore a high performance VLSI architecture of FME is described in this paper to achieve the capacity of encoding the high-resolution real-time video stream for HDTV. Our design is improved from an existing work by involving a pipeline strategy in sub-pixel interpolation unit which can avoid the long delay paths in 6-tap ID FIR so as to increase the clock frequency up to 200MHz. Moreover, a 16-pixel search engine is adopted to remove the redundant interpolation area and parallelize the various block size search which can save more than half of the clock cycles in processing a macro block. Our design is implemented with only 189K gates at operating frequency of 200MHz in worst case (285MHz in typical case). It can provide the processing capacity of more than 250K MB/sec which is enough for 1080HD (1920times1088) video streams at frame rate of 30fps. It is a useful intellectual property (IP) design for multimedia system
Keywords
FIR filters; VLSI; computational complexity; high definition television; interpolation; motion estimation; multimedia systems; video coding; 200 to 285 MHz; H.264 encoding; HDTV; VLSI; computation complexity; fractional motion estimation; intellectual property; multimedia system; Clocks; Computer architecture; Encoding; Frequency; HDTV; Interpolation; Motion estimation; Pipelines; Streaming media; Very large scale integration;
fLanguage
English
Publisher
ieee
Conference_Titel
Circuits and Systems, 2006. ISCAS 2006. Proceedings. 2006 IEEE International Symposium on
Conference_Location
Island of Kos
Print_ISBN
0-7803-9389-9
Type
conf
DOI
10.1109/ISCAS.2006.1693157
Filename
1693157
Link To Document