towards user-centric video transmission in next generation mobile networks

Towards User-centric Video Transmission in Next Generation Mobile Networks

Ali El Essaili

Klagenfurt, February 12, 2014

Technische Universität MünchenInstitute for Media TechnologyProf. Dr.-Ing. Eckehard Steinbach

Outline

Motivation and overview

QoE-driven adaptive HTTP video delivery

QoE-driven resource allocation for LTE uplink

Conclusions

February 12, 2014 2A. El Essaili: Towards User-centric Video Transmission in Next Generation Mobile Networks


Motivation

The majority of mobile data is streaming video and audio. The lion share of the mobile traffic is TCP/IP based.

[1] http://www.sandvine.com/downloads/documents/Phenomena_1H_2012/Sandvine_Global_Internet_Phenomena_Report_1H_2012.pdf

3February 12, 2014 A. El Essaili: Towards User-centric Video Transmission in Next Generation Mobile Networks


Overview

DASH (Dynamic Adaptive Streaming over HTTP) is standardized for mobile multimedia streaming (3GPP Rel-11, MPEG ISO/IEC 23009-1).

OTT DASH provides an end-to-end server-client adaptation. The client reacts to the resources assigned by the scheduler in the operator network.

Objective: Enhance the adaptive HTTP media delivery in next generation mobile networks.

QoE-driven traffic management Base

station

LTEnetwork

StandardDASH Client

AdaptationBit-rate estimation Segment switching

Media Presentation

DASH Server

End-to-end

4February 12, 2014

MPDHTTP Request

A. El Essaili: Towards User-centric Video Transmission in Next Generation Mobile Networks


OverviewDifference to RTP-based QoE-driven optimization

[1] S. Khan, Y. Peng, E. Steinbach, M. Sgroi, W. Kellerer, “Application-driven cross-layer optimization for video streaming over wireless networks,” IEEE Communications Magazine, 2006.[2] W. Kellerer, D. Svetoslav, E. Steinbach, M. Sgroi, S. Khan, EP1798897, granted June, 2008.

ApplicationServer

Base station

LTE network

QoE optimizerOptimal rate for

each user

Channel information

Traffic management

(e.g., transcoding, packet dropping)

Rate

Application utility function

MobileUsers

1) In-network content adaptation is costly, may react late2) DASH provides inherent adaptivity by encoding the same content at multiple bit-rates


Find the resource allocation which maximizes overall utility based on application and channel conditions.

… ∑


Outline




Conclusions



QoE-driven adaptive HTTP video deliveryProposed QoE-Proxy approach

DASHServer

Base station

LTE network

MPDHTTP Request


each client

Rate

Utility Media segmentsStandard DASH

Client

Channel information

ProxyRewrite client

requests

Target DASH Representation

The QoE optimizer returns the optimal rate of each user. The proxy rewrites the HTTP requests and forwards them to the DASH server. The DASH Server and the DASH clients are unaware of the proxy operation.

Resourceshaper

TCP rate

Rate



QoE-driven adaptive HTTP video deliveryQoE-Reactive scheme

DASHServer

Base station

LTE network

MPDHTTP Request


each client

Rate

Utility Media segmentsStandard DASH

Client

Channel information

The TCP throughput of each client is shaped according to the optimal rate. The streaming rate is determined by the standard DASH client.

Resourceshaper

TCP rate

Target DASH Representation



QoE-driven adaptive HTTP video deliveryUtility curves


[1] L. Choi, M. Ivrlac, E. Steinbach, and J. Nossek, “Sequence-level methods for distortion-rate behavior of compressed video,” IEEE ICIP’05 [2] S. Khan, S. Duhovnikov, E. Steinbach, and W. Kellerer, “MOS-based multiuser multiapplication cross-layer optimization for mobile multimedia communication,” Advances in Multimedia, 2007


QoE-driven adaptive HTTP video deliveryOptimization schemes

Scheme DescriptionQoE-Server The server encodes the video stream at the optimal

rate (e.g., managed content).QoE-Proxy Proposed proxy-based approach (OTT content).

QoE-d-Proxy Similar to the QoE-Proxy scheme. However, the optimizer uses discrete utility representations.

QoE-Reactive We shape the TCP throughput. The actual streaming rate, however, is determined by the client.

Non-Opt Standard OTT DASH streaming (Reference scheme)



QoE-driven adaptive HTTP video deliverySimulation parameters

Simulation parameters

LTE bandwidth 5 MHzNumber of PRBs 25

CQI update 2 secChannel model Urban macrocell

User speed 30 km/hrSimulation time 60 secSimulation runs 50

Application parameters

Video codec H.264 AVC, CIF, 30 fps

Segment size 2 secNumber of clients 8

PSNR-MOS mapping

Linear: {(1, 30 dB), (4.5, 42 dB)}

Client Standard MicrosoftSmooth Streaming



QoE-driven adaptive HTTP video deliveryExperimental results

CDF of the mean MOS for 8 users over 50 runs

3 3.5 4 4.50

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Mean MOS

CD

F

Non-OptQoE-ReactiveQoE-ProxyQoE-d-ProxyQoE-Server

Mean MOS gain

QoE-Reactive 0.19

QoE-Proxy 0.34

QoE-d-Proxy 0.49

QoE-Server 0.57



QoE-driven adaptive HTTP video deliveryExperimental results

Individual performance of 8 users over 50 runsSubstantial gains for resource-demanding videos



Subjective testsSettings

SAMVIQ (Subjective Assessment of Multimedia Video Quality) method [1] 20 test subjects after screening procedure 2 scenarios:□ Scenario 1: users move at a speed of 30km/h□ Scenario 2: users move at a speed of 120 km/h

8 users in the cell 10 seconds of video (without starting phase) Differential MOS is used as subjective quality rating [2]□ DMOS(sequence) = R(sequence) – R(hidden ref) + 100

[1] ITU-T Rec. BT.1788 Methodology for subjective assessment of video quality in multimedia applications, 2007[2] ITU-T Rec. P.910 Subjective video quality assessment methods for multimedia applications, 2008



Subjective testsUser interface



Subjective testsOverall results

Higher mean MOS, improvements more notable in dynamic scenario. Fair user experience, up to 35% MOS increase for worst-case user. February 12, 2014 16

QoE-Proxy QoE-Reactive Non-Opt QoE-Proxy QoE-Reactive Non-Opt40

50

60

70

80

90

100

mea

n D

MO

S (8

use

rs)

mean DMOS rating

Scenario 1: 30 Km/h Scenario 2: 120 Km/h



QoE-driven adaptive HTTP video deliveryVideo demo

17February 12, 2014

QoE-Proxy vs QoE-Reactive



Outline




Conclusions



QoE-driven resource allocation for LTE uplinkChallenges

February 12, 2014 A. El Essaili: Towards User-centric Video Transmission in Next Generation Mobile Networks 19


Use-case 1:□ Service‐centric approach for uplink distribution of real‐time user-

generated video content. □ Based on QoE and popularity of the video content.

Use-case 2:□ Joint upstream of live and time‐shifted video content under scarce

uplink resources.□ Transmit a basic quality in real‐time and upload a refined quality

for on‐demand consumption.

QoE-driven resource allocation for LTE uplinkUse-cases



QoE-driven resource allocation for LTE uplinkProposed approach

Motivation: Mobile networks will have to deal with a vast increase in user-generated video content given limited resources in the wireless uplink.

Contribution: We propose a QoE-driven approach for jointly optimizing the uplink transmission of live and on-demand videos.



QoE-driven resource allocation for LTE uplinkApplication model

Utility function for video streaming is defined on the MOS scale.

A Group of Pictures (GoP) is encoded into a set of video layers with a common deadline (i.e., base layer (BL), enhancement layer (EL)).

Joint optimization of video layers with playout deadline d1 (live) and cached layers with playout deadline d2 (on-demand).

deadlines all ofset LS



QoE-driven resource allocation for LTE uplinkOptimization functions

Network-based QoE optimization that maximizes the sum of utilities of K users:

Distributed QoE optimization for each user k:

s.t.

timprovemen QoE additional :MOSsize cache:

capacityuplink ousinstantane :

lk

k

Hc

esdependencilayer describes }1,0{otherwise 0 scheduled, if 1:

transmit totime:rate ousinstantane:,size:

l

l

l

ll

bat

RL

dL dLdLdLll S Sl

llllkSlSRa

RMOSbaU, ,,,

)(maxarg,,,

dL dLS Sl

kll cRa, ,

dL dLS Sl

kll HaL, ,

)1(

dLSl

ll SdtadL

,,,

K

kkk

cccU

K 1),...,()(maxarg

1



QoE-driven resource allocation for LTE uplinkSimulation results

Simulation parameters

LTE bandwidth 5 MHz

Channel model Urban macrocell

Scheduling metric Proportional Fair (PF)

Number of upstreamusers

4: Bus, Football, Soccer, Foreman

Video codec H.264/SVC, CIF, 30 fps, 4 layers

Relative cache size 0.3

(Live, VoD) delay (266 ms, 20 sec)

Simulation time, runs 100 sec, 20

Simulator LTE OPNET 16.0

Scheme DescriptionMOS-CLO QoE distributed and QoE-

driven network optimization

MOS-PF QoE distributed and PF

1,51,75

22,25

2,52,75

33,25

3,53,75

4

MOS-CLOLive

MOS-CLO VoD

MOS-PF Live

MOS-PFVoD

Mea

n M

OS

Scheme



QoE-driven resource allocation for LTE uplinkDemo



Outline




Conclusions



Conclusions

QoE-driven adaptive HTTP video delivery Exploit the benefits of QoE-driven traffic management:□ Propose a proactive approach for rewriting the client requests.□ Requires no adaptation of the media content, suitable for OTT services.□ Main results: improved QoE, fairness, and network awareness.

Playout buffer-aware traffic and resource management.

QoE-driven resource allocation for LTE uplink Service-centric approach for uplink resource allocation:□ Higher QoE compared to standard scheduling mechanisms in LTE.□ Improved video quality and efficient usage of the network resources by

considering the consumers' consumption patterns.



Acknowledgments

Prof. Eckehard Steinbach (Institute for Media Technology, TUM)

Damien Schroeder (Institute for Media Technology, TUM)

Prof. Wolfgang Kellerer (Institute for Communication Networks, TUM)

Dr. Dirk Staehle (DoCoMo Euro-Labs)

Dr. Liang Zhou (Nanjing University of Posts and Telecommunications)
