周承復臺灣大學：資訊工程學研究所徐明煒Hsu, Ming-WeiMing-WeiHsu2007-11-262018-07-052007-11-262018-07-052004http://ntur.lib.ntu.edu.tw//handle/246246/53966近年來串流服務已成為網路的主要應用之一，加上無線網路快速普及，在無線網路上提供串流服務已成為一項重要的議題。然而由於無線傳輸媒介的特性，如較低的頻寬和較高的位元錯誤率，使得在無線網路上提供串流服務比在有線網路上，更具有挑戰性。 當我們經由一資源有限的網路傳輸串流影片時，有時候因為頻寬的不足，導致部分的資訊遺失是不可避免的。為了提高網路頻寬的使用率，我們必須先傳送對影片品質有較大影響的訊框(frame)。因此我們提出一以相似度為基礎的訊框丟棄演算法(Similarity-based Frame Discard)，依相似度決定各影片訊框的重要性。 目前串流連線普遍採用TCP Friendly Rate Control (TFRC)為其壅塞控制機制以達到和傳統TCP連線公平地競爭網路資源的目的。如同TCP，TFRC在遇到封包遺失時，會認為網路已經進入壅塞狀態並調降傳送速度，以免網路的負荷量持續增加。但在無線網路上，由於較高的位元錯誤率，封包遺失可能是由無線傳輸媒介的不穩定所造成。因此，盲目地遇到封包遺失就調降傳送速度將會造成頻寬使用率的低落。為了解決此一問題，我們提出了以趨勢和遺失密度為基礎的遺失原因區分演算法，利用封包傳送時間的改變趨勢和封包遺失的密集程度，判斷封包遺失的原因是否為網路壅塞，抑或為無線媒介所造成。Streaming is more challenging in a wireless network than in a wired network, because wireless networks have their own characteristics such as low bandwidth and high bit error rate. We propose two schemes to address these issues. When we deliver a video stream across a bandwidth-constrained network, loss of frames may be unavoidable. To maximize the perceived quality-of-service at the client side, it is desirable to find a server transmission schedule that can transmit important frames first and utilize the network resources efficiently. To address this issue, we develop Similarity-based Frame Discard (SimFD). Unlike previous works, we select frames into transmission schedule by taking both frame similarity and inter-frame dependency into account. The frames that are less similar with their previous frame will get higher priority. TCP Friendly Rate Control (TFRC) is a widely accepted rate control method for streaming in the Internet. It assumes that packet losses are primarily due to congestion, and reduces sending rate in the present of packet losses. But this is not applicable to wireless networks in which the bulk of packet losses are due to wireless channel error and not relative to congestion. To help TFRC differentiate the losses caused by wireless channel error and avoid reducing sending rate blindly, we develop an end-to-end approach, Trend and Loss Density based Loss Differentiation Algorithm (TD). Our approach is based on the observation that congestion losses often occur after the ROTT peak and always appear in burst.感謝 II Abstract III 中文摘要 IV Table of Contents V List of Figures IX Chapter 1 Introduction 1 1.1. Video Streaming 1 1.1.3 Real-time Service 1 1.1.2 Introduction to video streaming 2 1.1.3 Introduction to wireless multimedia service 2 1.2 Problem 2 1.2.1 Problems in wired networks 2 1.2.2 Problems in wireless networks 3 1.3 Our Approaches 4 1.3.1 Similarity Based Frame Discard 4 1.3.2 Trend and Loss Density Based Loss Differentiation Algorithm 4 1.4 System Architecture 5 1.5 Thesis Organization 5 Chapter 2 Background and Related Works 7 2.1 Background 7 2.1.1 Background of Frame Discard Algorithm 7 2.1.1.1 Providing Streaming Service in a resource constrained network 7 2.1.1.2 Burst of video complicates the design of streaming mechanisms 7 2.1.1.3 Why Smooth is not suitable 8 2.1.1.4 Selective Frame Discard 8 2.1.1.5 Introduction to MPEG 9 2.1.1.6 Streaming MPEG 10 2.1.2 Background of Loss Differentiation Algorithm 11 2.1.2.1 TCP-Friendly Rate Control (TFRC) 11 2.1.2.2 The performance problem of TCP and TFRC in wireless networks 12 2.1.2.3 Schemes to improve the performance of TCP and TFRC in wireless networks 12 2.2 Related Works 13 2.2.1 Related Works of Frame Discard Algorithm 13 2.2.1.1 Selective Frame Discard 13 Cost Function 13 Gain Function 14 Heuristic Algorithms 14 2.2.1.2 QoS-Competitive Video Buffering 15 2.2.1.3 Frame-Induced Packet Discard ( FIPD) 15 2.2.1.4 MPEG based Frame Discard Filter 16 2.2.1.5 Comparison 16 2.2.2 Related Works of Loss Differentiation Algorithm 17 2.2.2.1 Local Retransmission 17 2.2.2.2 Split Connection 17 2.2.2.2.1 I-TCP 17 2.2.2.2.1 MTCP 17 2.2.2.3 End to End Mechanism 18 2.2.2.3.1 WMSTFP 18 2.2.2.3.2 RCM 18 2.2.2.3.3 MULTFRC 18 2.2.2.3.4 Biaz 18 2.2.2.3.5 Spike 19 2.2.2.3.6 Zigzag 19 Chapter 3 Similarity based Frame Discard Algorithm 21 3.1 Basic Idea 21 3.2 Similarity 21 3.2.1 Peak Signal Noise Ratio (PSNR) 21 3.3 Video Similarity 22 3.4 Preprocessing 23 3.5 Decision Window: the decision unit of SimFD 23 3.6 SimFD (Similarity-based Frame Discard) 23 3.6.1 Preprocessing 24 3.6.2 Online Policy 26 3.6.2.1 Discard B frames 26 3.6.2.2 Discard P frames 26 3.6.2.3 Discard I frames 26 3.6.2.4 PBP Heuristic 27 3.6.2.5 Adjust the size of the decision window dynamically 27 3.6.3 SimFD with Local Optimization (Lopt) 28 Local Optimization 29 Preprocessing 29 Online Policy 29 3.6.4 SimFD Algorithm 29 Chapter 4 Trend and Loss Density based Loss Differentiation Algorithm 31 4.1 Basic Idea 31 4.2 Loss Density 32 4.2.1 Autocorrelation 32 4.2.2 Relative Distance and Relative Packet Sequence 32 4.3 Trend of ROTT 33 4.4 TD Algorithm 34 Choosing the thresholds 35 Choosing the parameter: Relative Distance and Weight 35 Chapter 5 Performance Evaluation 37 5.1 Similarity Based Frame Discard Algorithm 37 5.1.1 Simulation Environment 37 5.1.2 Simulation Topology 37 5.1.3 Trace File 37 5.1.4 Simulation Metrics 38 5.1.5 Schemes in the simulation 38 5.1.6 Simulation Parameter 38 5.1.7 Simulation Result 38 5.1.7.1 SimFD 38 5.1.7.2 Decision Window Size 39 5.1.7.3 Lopt 40 5.1.7.4 PBP 43 5.2 Trend and Loss Density Based Loss Differentiation Algorithm 46 5.2.1 Simulation Environment 46 5.2.2 Simulation Topology 46 5.2.2 Simulation Metrics 47 5.2.3 Schemes 47 5.2.4 Simulation Result 47 5.3 Integration of SimFD and TD 51 Chapter 6 Conclusion and Future Work 55 6.1 Conclusion 55 6.2 Future Work 55 Reference 57844675 bytesapplication/pdfen-US無線網路串流wirelessstreamingframenetworkthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53966/1/ntu-93-R91922064-1.pdf