High-Reliability Built-in Self-Detecting/Correcting Architecture for Motion Estimation Computing Array

• Main Authors: Yu Liu, 劉羽
• Other Authors: Chun-Lung Hsu
• Format: Others
• Language: zh-TW
• Published: 2008
• Online Access: http://ndltd.ncl.edu.tw/handle/dbceck

碩士 === 國立東華大學 === 電機工程學系 === 96 === In the past year, following the capacity of multi-media storage and technology were improved, the applications of multi-media moved towards to a time of high quality and high resolution. Also in high quality of video compression, the MPEG-1 to MPEG-4 in 2003, all emphasized the high quality of video and low bit-rate. Not only the higher compression quality was appeared that was H.264 video coding but also the H.264 provided more outstanding compression quality than other standards of video coding. In the above mentioned, the motion estimation (ME) was a very important core block in video coding. Also, the core block was called motion estimation computing array (MECA) that was used to remove the redundant of frame to frame to achieve the high compression rate. But the computing of MECA had a great quantity and needed a lot of time. In order to increase the speed of computing, a lot of element array (PE array) was unable to avoid, especially in high video resolution such as Video CD, DVD, HDTV…etc. Hence, we face this so important and enormous process unit, the problem of testable design became more and more important. This thesis develops a built-in self-detecting / correcting (BISDC) architecture design for motion estimation computing array (MECA). Based on the error-detecting / correcting concepts of bi-residue arithmetic codes and low cost remainder and quotient (RQ) arithmetic codes, any error of each processing element (PE) in MECA can be effectively detected and corrected on-line by using the proposed built-in self-detecting (BISD) and built-in self-correcting (BISC) circuits, respectively. Performance analysis and evaluation show the proposed BISDC architecture has little area overhead, timing penalty and very low throughput loss.