博士学位論文

(1)

博

士

学

位

論

文

Doctoral Dissertation

内容の要旨及び審査結果の要旨

Dissertation Abstract and

Summary of the Dissertation Review Result

第

30

号

The Thirtieth Issue

平成

29

年

9

月

September, 2017

The University of Aizu

(2)

はしがき

博士の学位を授与したので、学位規則（昭和２８年４月１日文部省令第９号）第８条の規定に基づき、その論文の内容の要旨及び論文審査の結果の要旨をここに公表する。

学位記番号に付した｢甲｣は学位規則第４条第１項（いわゆる課程博士）によるものであることを示す。

Preface

On granting the Doctoral Degree to the individuals mentioned below, abstracts of their theses and the theses review results are herewith publicly announced, in according to the provisions provided for in Article 8 of the Ruling of Degrees (Ministry Of Education Ordinance No.9, enacted on April 1, 1953)

The Chinese character, “甲”, at the beginning of the diploma number represents that an

individual has been granted the degree in accordance with the provisions provided for in

Paragraph 4-1 of the Ruling Of Degrees (what is called “Katei Hakase,” or the Doctoral

Degree granted by the University at which the grantee was enrolled.).

(3)

- 1 -

Abstract

Ever since the early years of video streaming, there has been a growing demand for highquality video content. Thanks to today’s wireless broadband connections and the pervasiveness of high-performance mobile devices, video streaming has become a major Internet service. However, due to the best-effort nature of the Internet, providing users with high video quality is challenging. As a result, all new streaming standards developed since 2008 are based on Adaptive Streaming technology which allows the client/server to adapt the video bitrate to network fluctuations in order to enable smoother viewing experience with no playback interruption.

Over the past few years, HTTP adaptive streaming (HAS) has emerged as a de facto standard for streaming videos over the Internet. To adapt to network fluctuations, a streaming provider should generate multiple versions (with different video bitrates, for example) of an original video, each of which is chopped into short segments. During a streaming session, an adaptation method located at the client is responsible for deciding which video bitrate should be requested for each segment to maintain a good quality of service.

Despite the ongoing efforts to guarantee high video quality, recent HAS-related studies claim that the adaptation challenge has not yet been successfully resolved, especially for HAS over mobile networks where the connection throughput is naturally time-varying. Motivated by the untapped potential of HAS technology and the fast growth of mobile video streaming in recent years, I aim to develop adaptation methods that effectively cope with throughput variations of a mobile connection to improve user experience. The contributions to this area of research are outlined in the following.

First, we observed that existing adaptation methods in HAS mostly focus on CBR (constant bitrate) video streaming. However, compared to the CBR encoding, the VBR (variable bitrate) one has advantages in terms of video quality. Thus, in my first contribution, I present an adaptation method for HAS of VBR videos. To deal with variations of the video bitrate, a local average video bitrate is used as the representative bitrate of a video version. A buffer-based algorithm is then proposed to conservatively adapt video quality. Through experiments in the mobile streaming context, we show that our method can provide high video quality even under strong variations of connection throughput and video bitrate.

The recently ratified HTTP/2 protocol provides Server Push feature that helps reduce requestrelated overheads (e.g., in terms of energy, processing, bandwidth) for clients, servers, as well as network nodes. In my second contribution, I develop an adaptation method for HTTP/2-based adaptive streaming that not only leverages the Server Push feature of HTTP/2 but also provides buffer stability and gradual quality transitions. Since the method avoids playback interruptions and enables smoother viewing experience, it enhances the overall video quality.

In my third contribution, I focus on a particularly challenging use case of live streaming. Although the majority of the video content streamed over the Internet is video-on-demand (VoD), the amount of live streaming is growing rapidly. In live streaming, the client’s buffer is just 10 seconds or less (to enable a small capture-to-display delay); thus, this small buffer could be depleted easily due to strong throughput variations. To maintain seamless streaming, the client may switch to a very low video

(6)

- 4 -

version, leading to a drastic quality change. We develop a qualitydriven adaptation method to meet the tradeoff between the requirements of buffer stability and smooth quality transitions. Our experimental results show that the proposed method can provide high video quality under a small buffer size of 10 seconds.

The low-delay live streaming is a special case of live streaming, where the buffer size is limited to just few seconds. For specific services such as video surveillance, low delay is considered a mandatory requirement, but it is very challenging to guarantee seamless streaming in condition of low delay. In my final contribution, I present a probabilistic approach to adaptively decide the video bitrate by taking into account the instant buffer level. The experimental results with a buffer size of 4 seconds show that compared to traditional methods, the proposed one can significantly reduce the number of buffer underflows while providing similar video bitrates.

Summary of the Dissertation Review Result

This research investigates adaptation solutions to improve the quality of adaptive streaming over mobile networks. This research has four main contributions. The first contribution is an adaptation solution to adaptive streaming of VBR videos. To deal with bitrate variations in a VBR video, a local average bitrate is used as the representative bitrate of a version. A buffer-based algorithm is then proposed to conservatively adapt video quality. The second contribution deals with the problem of adaptive streaming over HTTP/2. This new version of HTTP provides the so-called Server Push feature that helps reduce request-related overheads. An adaptation method is proposed that not only reduces request-related overhead but also provides buffer stability and gradual quality transitions. The third contribution is about HTTP live streaming, where the buffer size is about 10 seconds or less.

Based on the knowledge of JND (Just Noticeable Difference), the authors develop a quality-driven adaptation method to balance the requirements of buffer stability and gradual quality transitions. The fourth contribution is about HTTP low-delay live streaming, where the buffer size is limited to just few seconds. In this context, it is challenging to guarantee a seamless streaming session. For this purpose, a probabilistic approach is proposed to adaptively decide the video bitrate by taking into account the instant buffer level.

The first, second, and fourth contributions have been published as three major journal papers and the third contribution as a major conference paper. The candidate is the first author of all these four papers.

In the final review, the candidate successfully improved his presentation and dissertation accordingly to the review committee's comments. In conclusion, the candidate has fulfilled all of the formal requirements for the doctoral degree.

(7)

- 5 - Name

氏名

Dang, Nam Khanh

（ダンナムカイン）

甲CI博第60号

論文題目

Development of On-Chip Communication Fault-Resilient Adaptive Architectures and Algorithms for 3D-IC

Technologies

3次元IC技術のための適応型耐障害チップ内通信アーキテクチャとアルゴリズムの開発

The University of Aizu, Prof. BEN, A. (Chief Referee)

The University of Aizu, Prof. MIYAZAKI, T.

The University of Aizu, Prof. TSUKAHARA, T.

The University of Aizu, Prof. KITAMICHI, J.

National Institute of Informatics, Prof. YONEDA, T 会津大学教授ベンアブデラゼク（主査）

会津大学教授宮崎敏明会津大学教授束原恒夫会津大学教授北道淳司

国立情報学研究所教授米田友洋

(8)

- 6 -

Abstract

Multicore processing is predicted to be the backbone of future complex embedded architectures. By distributing the tasks into multiple processing elements, the system's frequency and operation voltage can be reduced; thus, a decrease in total power consumption can be obtained. However, due to the high complexity in terms of organization, communication, and operation, multicore processing demands in high scalability, efficient bandwidth, and better power efficiency solution have become primordial.

Notably, wires have overcome gates to become the most dominant source of delay in the deep sub-micron era. Consequently, the power consumption caused by additional buffers and wires is considered as a critical obstacle. Moreover, conventional communication paradigms (e.g., bus, point-to-point) also encounter several scalability and latency issues. In the past few years, the benefits of 3D Integrated Circuits (3D-ICs) and mesh-based Network-on-Chips (NoCs) have been fused into a promising architecture, called 3D-Network-on-Chip (3D-NoC). In fact, the scalability and parallelism of NoCs can be enhanced in the third dimension thanks to the short wire length and the low power consumption of 3D-ICs interconnects. As a result, the 3D-NoC paradigm is considered to be one of the most advanced and auspicious architectures for the future of IC designs, as it is capable of providing extremely high bandwidth, efficient scalability and low power interconnects.

While the 3D-NoC paradigm has been increasing in popularity with several commercial chips, it is threatened by the decreasing reliability of aggressively scaled transistors as they are approaching the fundamental physical limits. In deep sub-micron processes, gates have become more vulnerable to soft errors which can affect the operation accuracy of control logics and buffers in NoCs' routers; thus, leading to chip failure. In addition, low supply voltages enforce a very narrow noise margin, which makes the architecture more vulnerable and more sensitive to faults. In particular, hard faults, including both permanent and intermittent, can occur during the manufacturing stage or under specific operating circumstances. Because these faults do not permanently damage a given component, it can pass through several testing stages, but can still cause operating failures. Furthermore, and by shifting to 3D-ICs, 3D-NoCs are introduced to a new major challenge. That is, the high probability for TSV defects to occur. With high defect-rates and the clustering effect, TSVs need a proper fault-tolerance methodology to ensure the overall reliability. By accumulating all the failure sources, 3D-NoCs' reliability is expected to be one of the most critical issues in future System-on-Chips (SoC) designs.

Due to the numerous types of faults, many studies have proposed solutions for various individual aspects of on-chip reliability; however, a comprehensive approach encompassing soft errors, hard faults, and TSV defects pertaining to NoCs' reliability has yet to evolve. In addition, the error detection and diagnosis in 3D-NoC architectures have been studied thoroughly in the scope of offline-testing. On the other hand, with soft errors and intermittent faults becoming a dominant failure mode in modern NoCs and general VLSI systems, a widespread deployment of online-testing approaches has become crucial.

In addition to the variety and complexity of failure modes, the rapid development of fault-tolerance for NoC has become exposed to a new challenge: NoCs' reliability needs to be evaluated and quantitatively assessed in the early design stages. As a matter of fact, most of the existing evaluation

(9)

- 7 -

methodologies use the simple fault insertion and correctness-verification method. Such a method only ensures the functionality of a given technique. Moreover, this type of evaluation requires the complete design to be performed which may lead to a significant redesign time risk. To solve this issue, early reliability assessment is needed. After satisfying the performance requirements in the early assessment stage, the reliability of the design is also needed to be fully simulated and analyzed.

Starting from the above facts, fault resilient adaptive architectures and algorithms for 3D-ICs, especially for 3D-NoCs based systems, are developed in this research. With the aid of efficient detection, diagnosis and recovery mechanisms, and algorithms, the proposed system is capable of detecting and recovering from soft errors occurring in the routing pipeline stages. It also leverages configurable components to handle permanent faults' occurrences in links, input buffers, and crossbars by adopting our previous works. For integrating these hard error fault tolerant techniques, a detection, diagnosis, and recovery mechanism is proposed. This mechanism analyzes the transmitting operation and its failure state to determine the fault. Based on the position of the fault, it issues signals to handle it. Moreover, I also propose a dedicated fault-tolerant technique for TSV-cluster defects, which are the most vulnerable components of 3D-IC technology.

From another important perspective, this work presents a platform of reliability assessment for NoC systems. An analytical is used to help designers quickly estimate the efficiency of potential fault tolerant schemes. The result of this assessment can indicate the reliability enhancement of the evaluated technique. Later, the complete architecture is put into a netlist-based simulation process to estimate other results. The development of the reliability assessment, fault-tolerance architecture, and algorithms are integrated into the flow of Design for Reliability. In this flow, the analytical model helps the early assessment of the proposed fault-tolerant techniques, and the netlist simulations are conducted to confirm the reliability of the design.

The final goal of this dissertation is to propose a comprehensive fault-resilient architectures, algorithms, and a design methodology for highly reliable 3D-NoC systems development. In addition to providing the fault-tolerance techniques to deal with soft errors, hard fault, and TSV defects, a working flow is also presented. The complete working design stages are also provided to help designers understand their proposals, know how to approach the fault-tolerance challenge and complete a robust and graceful design.

Summary of the Dissertation Review Result

This thesis presents three contributions: (1) a soft error hard fault tolerant 3D Network-on-Chip system named as 3D-FETO; (2) a scalable TSV-cluster defect tolerant 3D Network-on-Chip architecture and algorithms; (3) a fast and accurate reliability assessment method for Network-on-Chip using analytical model.

In order to tackle both soft errors and hard faults in 3D Network-on-Chip systems, we proposed a soft

(10)

- 8 -

error resilient method to detect and recover soft errors in the pipeline stage. After that, by combining with Error Correction Code, Automatic Retransmission Request, and the previously developed hard fault tolerant techniques, the system has the ability to handle both types of faults. Moreover, we support detecting and recovery hard faults, which are incomplete in the previously developed works, by proposing a Detection, Diagnosis and Recovery Mechanism. The evaluation of 3D-FETO shows its ability to work with extremely high fault rates and to have reasonable values in terms of extra area cost and performance degradation. In summary, the 3D-FETO architecture can detect, diagnosis and recovery both soft errors and hard faults.

In order to handle TSV-cluster defects, we propose a scalable architecture and algorithms for 3D NoCs. TSVs of a router are organized in four clusters around it, and the clusters are shared between its original and the nearest router. The defect cluster is recovered by using a nearby cluster and routers arbitrate using a weighted system. To help reduce the area cost of TSVs, no redundancy is required and the sharing algorithms help the system ensure a high reliability. The evaluation demonstrates the high reliability of the proposed architecture and algorithms while having a reasonable area overhead and performance degradation.

Besides providing fault-tolerant techniques for 3D-NoCs, assessing them to classify their reliability is also important. In order to provide a quantitative value of reliability enhancement named as Reliability Acceleration Factor, we propose a reliability assessment method using an analytical model. A network is divided into sub-modules and the reliability of each sub-module is analyzed using Markov-state models. Later, the reliability of a network is synthesized from its modules. This method helps designers assess the reliability of their NoCs in the early design stages. The proposal is evaluated to show accurate and extremely faster results.

In the preliminary review, the candidate explicitly answered the questions that the reviewers had asked. Additionally, he addressed any of the comments to which they had on his paper and presentation, to improve the quality and understanding of the paper and the proposed system. In the final review, the candidate successfully changed his presentation and paper accordingly to the comments, which improved the dissertation.

In conclusion, the candidate has fulfilled all of the formal requirements for the doctoral degree.

(11)

- 9 - Name

氏名

Trinh, Viet Phuc

（チンヴィエットフック）

甲CI博第61号

論文題目

Channel Modeling for Hybrid Free-Space Optics/Radio Wireless Systems for Fifth-Generation(5G) Mobile Backhaul Networks

第5世代（5G）モバイルバックホールネットワークのためのハイブリッド光無線通信/ラジオ無線通信システムのチャネルモデル

The University of Aizu, Prof. PHAM, A. (Chief Referee)

The University of Aizu, Prof. MIYAZAKI, T.

The University of Aizu, Prof. LUBASHEVSKIY, I.

The University of Aizu, Associate Prof. TRUONG, C.T.

会津大学教授ファントゥアンアン（主査）

会津大学教授宮崎敏明

会津大学教授イゴールルバシェフスキー会津大学准教授コンタンチョオン

(12)

- 10 -

Abstract

Current cellular backhaul networks are mostly built with microwave links and fiber/copper-based links, which cannot cope with the capacity, latency, reliability, energy efficiency, and cost effectiveness required for the fifth generation (5G) of mobile networks. Therefore, the 5G backhaul research has been triggered, aiming at bridging the gap between the requirements stipulated by the 5G radio access network (RAN) and the realistic backhaul capabilities, from two different perspectives.

The first consists of evolving the current backhaul (microwave, optical fiber, copper, etc.) to meet 5G expectations and encompassing new wireless technologies such as millimeter-wave (mmWave) and free-space optics (FSO). The other backhaul research perspective looks at adapting the 5G RAN to the available backhaul with realistic performance, such as investigating intermediate RAN architectures between the centralized RAN and the distributed RAN to fit the backhaul/fronthaul capabilities. In this thesis, the main focus is on the channel modeling of the disruptive wireless technologies, including mmWave and FSO, for 5G backhaul networks.

To enable the mixture of mmWave and FSO, how to accurately capture the channel characteristics of both mmWave and FSO under various transmission conditions for performance evaluation is very important. To do so, the following problems need to be addressed. Firstly, the analysis with different channel models and effects induced by physical layer impairments would result in highly complex analytical problems. Secondly, the effectiveness of different signal processing and coding techniques needs to be investigated to optimize the performance of the mixed systems. Finally, comprehensive experimental implementation should be conducted to validate the analytical models and performance results.

To derive the analytical model for performance analysis, we strive to formulate mathematical models describing the effects of transmission channels and study the statistical characterization of the end-to-end signal-to-noise ratio (SNR) of the mixed systems. On the other hand, improvement techniques suitable for the mixture of mmWave and FSO systems are proposed, including including two-way communications, relaying techniques, and diversity reception with generalized selection combining (GSC). Furthermore, comprehensive computer-based experiments using practical channel data and Monte-Carlo simulations are implemented to evaluate the accuracy of the derived analytical model and performance results.

Summary of the Dissertation Review Result

The main contribution of this dissertation is the development of analytical models for performance analysis of different hybrid and relay-assisted systems using millimeter-wave (mmWave) and free-space optics (FSO) that could be employed in the future fifth-generation (5G) cellular backhaul networks. To analyze the performance of engineering systems, it is usually difficult to identify the key parameters driving system behavior because they are buried deep within systems. Therefore, analytical models play an important role as they describe systems using mathematical equations, showing exactly how different parameters affect system behavior. Analytical models can offer important insights into

(13)

- 11 -

system that are difficult to gain with purely numeric modeling approaches. Consequently, developing analytical models lays out the foundation for engineering implementations. This dissertation focuses on the formulation of mathematical models describing the effects of transmission channels and study the statistical characterization of the end-to-end signal-to-noise ratio (SNR) of the hybrid systems.

Performance metrics that reflect system behaviors under the impact of physical channels are comprehensively studied. Furthermore, improvement techniques suitable for the hybrid systems are investigated. Analytical results are experimentally verified by computer-based simulations.

The candidate has an excellent scholastic aptitude which is reflected in his strong record of publications and achievements. The candidate has excellent English command. Both review sessions went smoothly. Having carefully evaluated the submitted dissertation by the applicant, the committee unanimously agrees that the contribution of the dissertation is significant to the field of communication networking. In overall, the candidate is fully qualified for the conferment of PhD degree considering the contribution of the dissertation, his publication record and his scholastic ability.

(14)

- 12 -

博士学位論文 Doctoral Dissertation

内容の要旨及び審査結果の要旨 Dissertation Abstract

and

Summary of the Dissertation Review Result

第30号 The Thirtieth Issue

平成29年9月 September, 2017

発行会津大学

〒965-8580 福島県会津若松市一箕町鶴賀 TEL: 0242-37-2600

FAX: 0242-37-2526 THE UNIVERSITY OF AIZU Tsuruga, Ikki-machi Aizu-Wakamatsu City

Fukushima, 965-8580 Japan

博 士 学 位 論 文

Doctoral Dissertation

Dissertation Abstract and

Summary of the Dissertation Review Result

30

The Thirtieth Issue

29

9

September, 2017

The University of Aizu

Preface

The Chinese character, “甲”, at the beginning of the diploma number represents that an

individual has been granted the degree in accordance with the provisions provided for in

Paragraph 4-1 of the Ruling Of Degrees (what is called “Katei Hakase,” or the Doctoral

Degree granted by the University at which the grantee was enrolled.).

Contents

Abstract

Summary of the Dissertation Review Result

Abstract

Summary of the Dissertation Review Result

Abstract

Summary of the Dissertation Review Result

博士学位論文