COMPANY

L O G O

Error Resilient Concurrent Video Streaming Over Wireless Mesh Network

Author ： Danjue Li, Qian Zhang,Chen-Nee Chuah and S.J.Ben Yoo

學生：王志嘉指導教授：許子衡 老師

2

Introduction (i)Wireless mesh networks (WMNs) are ad-hoc networks with full or partial mesh topologies, where nodes can automatically establish and maintain mesh connectivity among themselves

In this paper, we will mainly focus on WMNs deployed inside residential areas and target providing high-quality concurrent Video-on-Demand (VoD) service to various clients over WMNs.

3

Introduction (ii)

In achieving this goal, there are a number of challenges.1. One of the main problems here is the reduction in total capacit

y due to the interference between multiple simultaneous transmissions.

2. It is difficult to maintain an end-to-end route with stable bandwidth and bounded delay in WMNs due to MAC layer contention.

Forward error correction (FEC) has been widely used to improve video streaming quality over the error-prone channels. Based on those observations, we propose to combine multi-source, multi-path diversity and FEC to design an error resilient concurrent video streaming system

4

Introduction (iii)

This paper integrate an interference model into our design and rely on the rate/interference-distortion optimization framework to design a route selection scheme for choosing the optimal source/path for each of our concurrent streaming sessions.

Toward this goal, we make the following contributions

1.We leverage both multi-path multi-source diversity and FEC to design an error-resilient video streaming system for supporting high quality concurrent VoD service over WMNs

5

Introduction (iv)2. We jointly take the media source characteristics, underlying

WMN conditions ,WMN connectivities, and applications requirements into consideration, and develop a more realistic analytical model for multi-source streaming over multiple paths.

3. Based on the analytical model, we formulate an interference aware route selection problem for the proposed streaming system using a joint rate/interference-distortion optimization framework and select the optimal sets of paths by solving this problem heuristically with a genetic algorithm (GA).

6

Error Resilient Multi-source Multi-path Video Streaming System Over WMN (i)

Fig.1 illustrates a residential WMN scenario that we consider in our design.

Fig.1

7

Error Resilient Multi-source Multi-path Video Streaming System Over WMN (ii)

If multiple clients are interested in the same video file at nearly the same time, multicasting can be leveraged to save bandwidth while maintaining satisfying video quality, but if clients are interested in various video files or their requests do not come closely enough in the time, building up multiple multicasting trees will not be cost-efficient

In this case, we would like to leverage the existing multi-path characteristics of WMNs and a hybrid server/P2P video streaming service model to find the best video source location/path(s)

8

Error Resilient Multi-source Multi-path Video Streaming System Over WMN (iii)

We propose an error resilient multi-source multi-path video streaming system as shown in Figure 2.

Fig.2

9

Error Resilient Multi-source Multi-path Video Streaming System Over WMN (iv)

To start video streaming, WMN nodes first send requests to the media server .When receiving new requests ,the media server sets up a sliding window to group incoming requests.

To serve a set of concurrent streaming requests, denoted by VV , the media server will then search to locate kk different descriptions for each of the requested video files among WMN nodes.

For a request, if there do not exist all kk descriptions, the media server generates missing descriptions for it using multiple description coding (MDC).

10

Error Resilient Multi-source Multi-path Video Streaming System Over WMN (v)

Based on the content locating results and collected path QoS parameters, the route selector will decide:

1. From where each description of the requested video file can be fetched.

2. What is the optimal path to stream it to the corresponding receiver.

An MDC decoder will attempt to reconstruct a video sequence from the k RS (Reed-Solomon)RS (Reed-Solomon) decoded descriptions.

11

Multiple description coding

Multiple description coding is a source coding technique that generates multiple, equally important bitstreams, called descriptions, for a single video file.

The advantage of doing this is that descriptions can be streamed to a receiver using disjoint streaming paths, which can potentially increase the resilience to packet loss.

12

Monitoring WMN status

The quality of WMN links vary over time due to channel fading, interference, and congestion.

We leverage some of the components proposed in link link quality source routing (LQSR)quality source routing (LQSR) for discovering neighbors of a node, measuring the link quality between a node and its neighbors, and propagating this information to other nodes in the WMN.

13

Modeling _ Source Model (i)

A set of concurrent video streaming sessions, denoted as VV..

The video clip vv (v V ∈ ) in each session.

dd00 be the distortion when bothboth descriptions successfully arrive in the receiver

σσ22 be the distortion when both descriptions do not successfully arrive in the receiver.

Let b bmmvv(m {∈ 1, 2}) defined in terms of the number of

bits per pixel (bpp) on the mmthth channel channel .

RRmmvv is the data rate of the mth description.

FFvv (frames/second) is encoded at a frame rate.

14

Modeling _ Source Model (ii)

W ·H (pixels/frame)W ·H (pixels/frame) is a resolution. CC is a known constant decided by the chroma subsampling format.

Dfs

σσ22 is equal to the variance of the source and αα is decided by the video codec

qqmm vv is the probability of successfully receiving the mth description of video clip v.

Average distortion perceived by the receiver can be computed as ：aa

15

Modeling_Network Model (i)

We model such a WMN using a time-varying directed time-varying directed graphgraph G = G (N, L), where NN is the set of vertices, representing nodes in the WMN, and LL is the set of directional edges.

The path for delivering the mth description of video clip vv be Lm

v

Each link {i,j} we define an index variable:

Ddda

16

Modeling_Network Model (ii)

End-to-end packet loss rate and delay:pij = 1 − (1 − eij )S is the packet loss on link {i,j}The maximum allowable packet size on each link is identically set to be SThe end-to-end packet loss rate associated with path LLmm

vv can be defined as: Dss

λij is jointly decided by the link capacity and the traffic load on {i,j}.

PDF( probability density function )

ttijij =

17

Modeling_Network Model (iii)

The end-to-end delay for path LLmmvv can be defined as:

DD ….(7)

When the number of hops along Lmv is not large enoug

h, we will use the moment generating function (MGF) to find a solution.Base on (7) and PDF of tij , we find the MGF of Tm

v as ： dsf

If we assume that in the wireless mesh network there are no two identical wireless links,according to (9) can be expanded as ：

　　 dfs

18

Modeling_Network Model (iv)

Therefore, the PDF of Tmv can be computed as:

Dfsf

Using (11), we can get P( TTmmvv > > ΔΔmm

v v ) when the number of hops on the path is small, as:

DSFS

19

Modeling_Network Model (v)

Error Protection Using Forward Error CorrectionError Protection Using Forward Error Correction ：：Forward error correction based on Reed-Solomon (RS) codes , where the codewords are formed across kkmm

vv information packets and nnmm

vv− k− kmmvv redundancy packets.

The higher kkmmvv is, the less protection we add. Here, we

choose kkmmvv based on the end-to-end packet loss rate ppmm

vv on the delivery path Lm

v. If ppmmvv is high, we need heavy

protection, and as a result we will choose smaller kkmmvv

An (nn, kkmmvv) RS code is implemented, the probability of

successfully receiving this frame will be:

20

Modeling_Network Model (vi)

Interference model ：This protocol model is mainly concerned with the interference requirement on the receiver side. For a single wireless channel, node i can successfully finish a transmission to node j only if it can satisfy both of the following requirements:

1. dij ≤ ri

2. Any node nk, which satisfies dkj ≤ R`k , does not transmit when node i is transmitting.

21

Interference Aware Route Selection (i)

We must be solving route path selection problem.There are several issues we have to consider during the problem formulation.

1. All the concurrent streaming sessions will co-exist in the system and compete for wireless bandwidth resources, and as a result, they will affect each others’ performance.

2. Each link has limited bandwidth available for supporting streaming service. Overloading the link will cause network congestion and degrade the system performance.

3. We have mentioned in previous discussions, the wireless interference can adversely affect the performance of multi-source multi-path diversity due to the fact that conflicting links cannot be used simultaneously.

22

Interference Aware Route Selection (ii)

The total traffic on link {i,j}, ρij , will be:

edfs

Using (1), (14) and (15), this route selection problem can be mathematically formulated as:Minimize:

Subject to ：

的

23

Interference Aware Route Selection (iii)

The distortion minimized route selection problem defined by (16)-(22) can be treated as a joint optimization of several single-streaming session route selection problems

24

Performance Evaluation (i)

We simulate an IEEE802.11-based WMN by randomly placing 12 wireless nodes in a 200m × 200m square region as shown in Figure 3.

Fig.3Fig.3

25

Performance Evaluation (ii)

We use three standard QCIF video sequences Foreman(176×144), Mother(176×144) and Mobile(176×144) in our simulations, and they are requested by three randomly selected WMN nodes simultaneously with a delivery delay tolerance of 150ms.

We choose the quantization step-size to Foreman Foreman sequence sequence is 2424. Mother sequenceMother sequence is 2020, and 3232 for Mobile sequenceMobile sequence.

Figure 4 shows the rate-distortion curves produced by the H.264 encoder and the fitted values based on (1) (α = 2). From this figure, we can see that the source model we have defined is a valid estimation of the MDC distortion.

26

Fig.4

27

Performance Evaluation (iii)

When enhance the error resilience ,we choose different levels of FEC protection based on the estimated end-to-end packet loss ratio.

We consider two levels of FEC for each SD (single SD (single description)description) and MD (multiple description)MD (multiple description) encoded stream, one for heavy loss scenario and the other for light loss scenario. The parameters for SD and MD encoded streams with different FEC level are shown in Table I.

28

Table I

29

Simulation results and analysis (i)

We compare our proposed scheme with a single source single path scheme, and plot the instantaneous PSNR ( Peak Signal to Noise Ratio) value over time for different streaming schemes.

In Fig.5 and Fig.6 , the curve labeled as SD is the performance when only single source single path streaming is supported.

The curve MDC: desc1+ desc2MDC: desc1+ desc2 shows the performance when both descriptions from multi-source multi-path streaming are used to recover the video sequence

30

Fig.5 & Fig.6

31

Simulation results and analysis (ii)

The simulation results of all three concurrent test sequences are shown in Table II, Table III, and Table IV, respectively

32

Simulation results and analysis (iii)

Fig. 7 compares the PSNR values of a set of 300 frames for the test sequence foreman when different streaming methods are used

33

Conclusion

這篇 paper 主要提到 multi-source multi-path 有較多的多樣性，也因為如此，可以設計出 forward error resilient concurrent video streaming system 並提供高品質的 VoD 服務在 WMN 上這篇 paper 也提到 multi-source multi-path 比 single-source single path 能夠減少干擾對於效能的影響