Department: Electrical & Computer Engineering
Faculty Advisor(s): Sujit Dey

Primary Student
Name: Seyed Ali Mirtar
Email: smirtar@ucsd.edu
Phone: 858-534-7883
Grad Year: 2013

Dynamic Thermal Management (DTM) acts as a necessary tool for safe operation of systems and increases their lifetime; however, application of DTM affects system performance, and can significantly impact the quality of results of complex real-time applications such as real-time video encoders. In this paper, we propose a dynamic adaptation algorithm that can be used with an H.264 encoder to change its complexity in real time, and hence adapt its computational needs to the dynamic changes in system performance due to DTM, while minimizing the impact on the quality and bit rate of the encoded video. We formulate our dynamic adaptation approach as a multi-dimensional optimization problem that maximizes video quality and minimizes bit rate while ensuring that the video encoder can run in real-time in spite of the DTM effects. We have implemented our adaptation algorithm with x264, a very efficient and commonly used H.264 encoder. We evaluated the adaptive encoder on a computing platform with Intel® 1.8GHz CoreTM2 Duo processor, which employs DTM based on DVFS. Our measurements with several video clips reveal that because of the dynamic effects of DTM, the quality of videos encoded by x264 are affected significantly - an average 10 dB reduction. However, with the aid of our dynamic adaptation algorithm, the x264 encoder can encode all the streams in real-time, with an average video quality degradation of only 2.4 dB, and with only a nominal increase in bit rate of the encoded streams.

« Back to Posters or Search Results