Português English
Contato

Divulgação da Defesa da Dissertação de Mestrado do aluno BRUNNO ALVES DE ABREU


Detalhes do Evento


O Programa de Pós-Graduação em Microeletrônica – PGMICRO, da Universidade Federal do Rio Grande do Sul, tem a satisfação de convidar a Comunidade Universitária para assistir à defesa pública da Dissertação de Mestrado do(a) aluno(a) BRUNNO ALVES DE ABREU a realizar-se:

Data: 16/05/2019 às 09h:30min

Local: Instituto de Informática 43412(65) – sala 218

Aluno: BRUNNO ALVES DE ABREU

Título: “Exploring Partial Distortion Elimination Techniques in the Sum of Absolute Differences Architecture for HEVC Integer Motion Estimation ”

Orientador: Prof. Dr. Sergio Bampi

Coorientador: Prof. Dr. Mateus Grellert da Silva

Banca Examinadora:

Prof. Dr. Bruno Zatt (UFPEL)

Prof. Dr. Altamiro A. Susin (PGMICRO/UFRGS)

Prof. Dr. Luciano Volcan Agostini (UFPEL)

Abstract:

Managing the energy requirements of digital video services has been a challenging task throughout the years, especially when dealing with battery-powered systems, e.g., smartphones, tablets, camcorders. Digital videos are among the applications that have been given the most importance in the recent years, due to the increase of video traffic on the Internet and the popularization of streaming and other real-time services — especially in embedded devices. High Efficiency Video Coding (HEVC) is a popular digital video coding standard, and it was designed as a successor to H.264/AVC. Integer Motion Estimation (IME) is one of the most costly steps in the HEVC coding process, as it is responsible for exploring the temporal redundancies in video encoders. Two parameters are considered when calculating the redundancies (distortion) between video blocks: Sum of Absolute Differences (SAD) and motion vector cost (MVCost). Partial Distortion Elimination (PDE) techniques may be used to optimize the calculation of the SAD unit itself, in order to avoid the computation of candidates that will certainly not be selected in the IME. This work explores PDE techniques in the distortion cost with and without the consideration of the MVCost. Cycles analyses are presented showing the benefits of the PDE without the MVCost, comparing this number to an analysis without any PDE technique. Then, another exploration is presented by including the MVCost in different ways, in order to be more faithful and precise to the implementation of software encoders, and save even more cycles when compared to standard PDE usage. Different architectures that calculate SAD and the total distortion cost were also proposed and implemented, in order to determine the actual energy values that are saved with the proposals. The architectures were synthesized for ASIC with a 65 nm standard cells library. Power analyses were performed using real-input vectors from video sequences to obtain more accurate results. The use of PDE without the MVCost achieves an average cycles (and energy, consequently) reduction of 16.41% and 11.64%, for 1080p and 2160p sequences, respectively, considering the encoder software x265, when compared to an implementation that does not use PDE. When taking MVCost into account, an improved analysis was performed considering both the HEVC Test Model (HM) reference software and x265. The use of PDE including the MVCost after the SAD calculation achieves an average reduction of 28.21% and 33.74% in the energy consumption when compared to an implementation without PDE, for HM and x265, respectively. When taking advantage of the MVCost in PDE before the SAD calculation, approaches accumulating MVCost with a multiplexer and with a Carry-Save Adder (CSA) were analyzed. Average additional reductions of 17.51% and 5.05% were obtained when using a multiplexer for HM and x265, respectively, when compared to the case that accumulates MVCost after SAD. Even though the solution accumulating MVCost with a CSA does not reduce the number of cycles when compared to the one with a multiplexer, the architecture using a CSA obtained slightly reduced power results, compensating its higher number of cycles and becoming the best solution in terms of total energy. Thus, additional reductions of 1.94% and 2.27% were obtained from using a CSA when compared to the multiplexer implementation — which already reduces in energy when compared to the implementation accumulating MVCost after SAD –, for HM and x265, respectively. Besides, the versions with MVCost increase the quality of the video by up to 1.53%, when compared to implementations that do not employ MVCost.

Keywords: High Efficiency Video CodingInteger ,Motion Estimation,Sum of Absolute Differences