A Layered Hybrid ARQ Scheme for Scalable Video Multicast

over Wireless Networks

Zhengye Liu,

Joint work with

Zhenyu Wu

Outline

Motivations & challenges Review of error protection approaches Layered hybrid ARQ Operating point selection in multiple user

scenario A general game theoretic framework in operating

point selection Layered hybrid ARQ in video multicast

Conclusion

Motivations & Challenges

Motivation of video multicast over WLANs Utilize bandwidth efficiently

Challenges Error protection mechanisms are needed

Fading, channel interference, … Heterogeneity of channel conditions

Different channel conditions Overall system performance Individual user fairness

S

C C C

RR

RS

C CC

R

unicast multicast

Packet Loss Pattern

Burst packet losses Difficult to predict

Review of error protection approaches

Retransmission Inappropriate in multicast scenario

FEC Constant throughput and bounded delay Throughput is reduced in the good state

Adaptive FEC Prediction of channel conditions in the future

Hybrid ARQ [Majumdar 02]

Hybrid ARQ Scheme Generate parity packets Send source packets Send parity packets until all lost source packets can be

recovered

S 1 2 3 4 5 1 2 3

C

ACK

Hybrid ARQ Scheme

Use bandwidth efficiently Should have sufficient bandwidth

Layered Hybrid ARQ Scheme

Encode a video into multiple layers Temporal scalability

Transmit packets from more important layers to less important layers

For each layer, transmit source packets first and then parity packets, based on hybrid ARQ

Given a total transmission bandwidth, provide unequal protection Protect more important layers Selectively drop source packets from less important layers No overall rate expansion

An Illustration

S

C

Performance Evaluation

Single user scenario Only one user in the multicast group

Comparison Hybrid ARQ with single layer video (single hybrid

ARQ) Layered hybrid ARQ

Simulation Setup (1)

H.264 codec JM11 Football (720x480, 30 frame/sec) Average bitrate: 1400 kbps

Fix QPs

Temporal scalability in H.264

Layer 1 Layer 2 Layer 3

Simulation Setup (2)

RS coding (255, k) for each layer Frame copy in decoder Total transmission bandwidth: 2200 kbps Packet loss pattern: two-state Markov model

Packet Receive Ratio Percentage of received/recovered source packets over the total

encoded source packets

Layered hybrid ARQ can provide unequal protection for different layers

All packets from I and P frames can be received Most packets from Bs frames can be received

Average Channel Induced MSE

Layered hybrid ARQ can outperform hybrid ARQ significantly in received video quality

MSE Frame by Frame (p=30%)

Demo

Packet loss rate: 30% Single hybrid ARQ vs. Layered hybrid ARQ

Layered Hybrid ARQ in Multiple User Scenario

Heterogeneity of channel conditions Different preferred configurations (operating

points) of video multicast

S

C2C1

?

How to Select Operating Point?

Worst case Based on the user with the worst channel condition

Play a game Play “lottery” among users

C1 C2

Parity packet from layer 1

Source packet from layer 2

50% 50%

C1 C2

Parity packet from layer 1

Source packet from layer 2

Is This Game Fair? Two players, each owning a car, play lottery with each

other If a player wins the game, he/she can win the car from

the other player

Player 1 Player 2

50% vs. 50%? 99% vs. 1%

What are the probabilities for a fair game?

C1 C2

Parity packet from layer 1

Source packet from layer 2

λ1 vs. λ2 ?

Nash Bargaining Game

Proposed by John Nash in 1950 A cooperative game

Players have perfect knowledge of each other

Proved the existence of Nash bargaining solution (NBS) for this game Unique solution Pareto optimal

No other solution produces better utility for one player without hurting another player

Fair in the sense of cooperative game Satisfy the axioms of fairness

Formulation of Nash Bargaining Game Player:

N users in a multicast group Strategy:

M operating points, sm

Mixed game with mixed strategy S = λ1s1 + λ2s2 +,…,+ λMsM

Preference: The utility of each strategy for user i, ui(sm). Mixed utility

Ui = λ1ui(s1) + λ2ui(s2) +,…,+ λMui(sM) Initial utility:

di, user would like to at least achieve if they enter the game Ui>di, otherwise user i will not enter the game

Nash bargaining solution (NBS): λ*=(λ1, λ2,…, λM) Users consider it as a fair setting of the lottery

An Example

Player: Two users

Strategy: Three operating points, sm=“transmit a packet from layer m” Mixed game with mixed strategy, pm is the probability that a packet from layer m

will be chosen S = λ1s1 + λ2s2 + λ3s3

Preference: ui(sm): how much payoff user i can get when a packet from layer m will be sent The anticipation of payoff from the lottery (mixed game)

Ui = λ1ui(s1) + λ2ui(s2) + λ3ui(s3) Initial utility:

di

C1 C2

Parity packet from layer 1

Source packet from layer 2

Utility

If user i is requesting layer m, then only a packet from layer m is useful.

wm should represent the importance of a packet from layer m on video quality Use a channel distortion model to obtain wm

C1Parity packet from layer 1 u1(s1) = w1, u1(s2) = 0, u1(s3) =0

C2Source packet from layer 2 u2(s1) = 0, u2(s2) = w2, u2(s3) =0

Channel Distortion Model of Temporal Scalable Video Channel distortion model of single layer video

Channel distortion model of temporal scalable video

w1=1400, w2=650, w3=150

Initial Utility

Guarantee that the expected Ui>di

A flexible control parameter Select a higher di, if the “system” gives more

protection to user i User i subscribes more premium service It is more urgent for user i to win the game

If user i is requesting a packet from layer m

C1Parity packet from layer 1 C2

Source packet from layer 2

d1>d2

α=2 d1=w1/2,d2=w2/4

Obtain NBS

A Nash bargaining game Player, strategy, preference (utility), and initial utility

Solve an optimization problem

Exhaustive search for small M Convex programming for a large M

Procedure of operating point selection

Trace the state for each user From which layer the user is requesting a packet Based on the ACKs sent from the receivers

Play Nash bargaining game Obtain ui(sm) and di

Obtain the NBS λ*=(λ1, λ2,…, λm) Given λ*, play lottery to select a packet for

sending

Performance Evaluation (1)

Lead to NBS optimality and fairness in microscopic view (packet level)

The macroscopic affect of a strategy on received video qualities Overall performance: The majority of users are more

likely to obtain their preferred operating points than the minority of users

Individual fairness: No individual user is denied access to the multicasting system or overly penalized

Flexibility: Can be tuned to satisfy different requirements.

Performance Evaluation (2)

Comparison Worst case Nash bargaining game

Investigate the impact of initial utility di on system performance

Higher di leads to more protection to user i α=2, 4, and 100 By using a smaller α, guarantee a better basic video quality

for bad channel users

Simulation Setup

N users totally N-1 users are in a good channel condition

(p=1%) One user is in a bad channel condition (p=30%) N=2, 4, 8, 16, 32

Average MSE

(a) Good channel user (b) Bad channel user

Summery

Worst case Nash bargaining

Overall performance

Ignore the majority of users

Adapt the operating point to the majority of users

Individual

fairness

The good channel users are overly penalized

No user is overly penalized

Flexibility Change α to satisfy different requirements

Conclusion

Propose a layered hybrid ARQ scheme for video delivery over WLANs

Propose a game theoretic framework in operating point selection for video mulitcast

Examine the game theoretic framework with the proposed layered hybrid ARQ

Thanks!