Path Virtualization Using Fountain Code for Video Streaming

over Heterogeneous Networks

Sang-Chun HanHwangjun Song

Jun Heo

International Conference on Intelligent Hiding and Multimedia Signal Processing (IIH-MSP), Feb, 2008.

05/05 Feb 2009

2

OutlineIntroductionProposed AlgorithmExperimental ResultsConclusion

3

Introduction To satisfy required QoS in wireless network (1/4):

Fast Vertical Handoff:

4

Introduction To satisfy required QoS in wireless network (2/4):

When multiple wireless networks are available:Fast Vertical Handoff:

Involves changing the data link layer (Layer 2) technology. e.g. Between WLAN (802.11x) and UMTS

(CDMA2000) Different from Horizontal Handoff between different

AP. Between same technology only involved in Layer 3.

Drawbacks: Sophisticated architecture, implementation cost,

and a little transition time is unavoidable.

5

Introduction To satisfy required QoS in wireless network (3/4):

Path Diversity (provide a Virtual Path):

Virtual Path

※ This Virtual Path is consisted of 2 physical paths.

6

Introduction To satisfy required QoS in wireless network (4/4):

When multiple wireless networks are available:Path Diversity:

The Mobile Node must provide more than one network interface.

Objective of proposed algorithm:To provide a Virtual Path for video streaming by

combining Path Diversity Technology, H.264 codec Rate Control, and Fountain Code.

7

Proposed Algorithm

m: The number of physical paths consisting of a Virtual Path.

※ Feedback Info are sent: (i) periodically: pre-determined time is expired. (ii) when the condition of virtual path changes significantly:

8

Proposed Algorithm

PS: packet's payload size.

Bgop: the amount of output bits for a GOP.

9

Proposed Algorithm Problem Formulation:Find the maximum encoding rate ( VRmax )

and code rate ( C ) for video streamingsubject to , and

, where d: Delay dmax: Tolerable Maximum Delay

pblr: Block Loss Rate pmax: Tolerable Maximum Block Loss Rate

10

Proposed Algorithm Problem Formulation:Step 1. Initialization:

Derive Source Block Length:

PS: packet's payload size.

Bgop: the amount of output bits for a GOP.

11

Proposed Algorithm Problem Formulation:Step 2. Find Maximum Transmission Rate

(TRmax):The delay to transmit a packet though the ith

path by:

The number of packets that are transmitted though the ith path satisfying the delay constraint:

Therefore, the Maximum Transmission Rate of a virtual path:

fr: frame rate of video.

ngop: the number of frames in a GOP.

※ PS: packet's payload size.

12

Proposed Algorithm Problem Formulation:Step 3. Determine the Code Rate (C):

Now we can calculate BLR function P(K, C) as follows:

, where and .

We have to find largest C which satisfies:Consequently, the Maximum Encoding Rate

(VRmax) is determined by:

※ If: K=5, C=1/2 t = 10 Assume K’ = 6, Summation from i=5~10 means: All decoding failure

cases!

13

Experimental Results ※ Java is used to implement the proposed system.

14

Experimental ResultsThe authors assume there are only two

heterogeneous paths between a video server and a client.802.11b and 802.11g

Tolerable Maximum:Delay: 500msBlock Loss Rate: 0.01

Reference Software of H.264/AVC:JM12.4Original JM Rate Control Mechanism are used.

15

Experimental ResultsGOP:

Number of frames: 15Structure is “IPPP…”

Video sequence:CIF format HARBOUR.It consisted only 300 frames at 30fps.

Simply repeat the same video sequence to support the 25sec experiment time.

16

Experimental Results

※ Code rate decreases when PLR increases: Decoding Prob. is kept in the tolerable range at the cost of increased redundancy.

17The number of GOP

Experimental Results ※ The number of packets in a

source block is changed according to Delay.

Code Rate and Video Encoding Rate are adaptive to the wireless channel status. Thus, the number of packets received through Virtual Path is always higher than K‘.

18

Experimental Results

The authors also experimented in the same condition with Reed-Solomon code.(255, 223) RS code, for R = 1.14The same condition:

“Second path” is more reliable than “First path”.Always successfully decoded using only

“Second path”. Sometimes not decodable using two paths: 8th and

40th GOP are not decodable. Video freezes during 0.5 second when frame

repetition method is used to conceal the loss.

19

Experimental Results

20

ConclusionBy the experiment in the real wireless

environment, the authors showed that Virtual Path using Fountain Code consolidates multiple paths for video streaming.

This approach support higher bandwidth, lower delay, and lower BLR than traditional approach which use only one network at a time.

In addition, seamless video streaming service is possible without a Vertical Handoff Time.

Thank you!