ICT-ADAMANTIUM
WP4PQoS models and adaptation
mechanisms
ICT-ADAMANTIUM
OutlineOverviewObjectives, Progress and
AchievementsTasks
T4.1T4.2T4.3T4.4
Conclusions
ICT-ADAMANTIUM
WP4: PQoS Models and Adaptation Mechanisms
Leader: UOPDuration: M1-M24 Status: CompletedMMs Allocation Planned vs. Spent
ICT-ADAMANTIUM
Objectives and ProgressObjective Progress Report
Research and development of Voice Quality context-aware PQoS models Objective Fulfilled D4.1F
Research and development of Video Quality context-aware PQoS models Objective Fulfilled D4.1F
Research the subjective temporal tolerance of the end-user when she/he experiences degraded PQoS and subjective tests of voice/video models.
Objective Fulfilled D4.1F
Mapping NQoS parameters (including access and core networks) to voice and video PQoS for VoIP and IPTV applications
Objective Fulfilled D4.2F
Research and development of dynamic service adaptation mechanisms for optimizing the delivered PQoS level. Objective Fulfilled D4.3F
ICT-ADAMANTIUM
WP4: Innovation and Achievements
Developed and Subjectively evaluated novel voice/video PQoS models for VoIP and IPTV applications.
Developed simulation platforms of the overall ADAMANTIUM concept in OPNET and NS2.
Developed a Network QoS statistics to PQoS level mapping model for video.
Developed dynamic PQoS-driven adaptation algorithms for PQoS optimisation for VoIP and IPTV applications
ICT-ADAMANTIUM
Task 4.1:
Voice and Video Quality PQoS Models
ICT-ADAMANTIUM
...),,,,,( FRSBRcodecjitterdelaylossfMOS
Task 4.1: Voice and Video Quality PQoS Models (1/12)
Voice/Video Modelling and Adaptation Conceptual Diagram
ICT-ADAMANTIUM
Video Quality Modelling
Task 4.1: Voice and Video Quality PQoS Models (2/12)
Temporal Feature
Extraction
Spatial Feature
Extraction
Content Type
Estimation
PQoS
Model
CT, SBR, FR, …
Loss, Delay,
Jitter
Network
MOS
RNCSGSNGGSN
Video Modelling
ICT-ADAMANTIUM
Task 4.1: Voice and Video Quality PQoS Models (3/12)
Voi
ce M
odel
– U
oP
ed IIR 2.93
0 if 1)(
0 if 0)(
)3.177()3.177(11.0024.0
xxH
xxHwhere
dHddId
cbpaIe )1ln(
100 5.4
1000 107)100)(60(035.01
0 16
RforMOS
RforRRRRMOS
RforMOS
ICT-ADAMANTIUM
Task 4.1: Voice and Video Quality PQoS Models (4/12)
Vid
eo M
od
el -
Uo
PSimulation set up in Opnet
H.264 over UMTS
Analysis of Results st, sd and rd filesPSNR, MOS
Video Encoding
Original YUV videoRTP trace
.264 st file
SBR, FR
BLER, MBL
Simulation Methodology
ICT-ADAMANTIUM
Task 4.1: Voice and Video Quality PQoS Models (5/12)
10 15
0
0.5
1
Frame rate (fps)
Degre
e o
f m
em
bers
hip
in1mf1 in1mf2
0.2 0.4 0.6 0.8
0
0.5
1
Content type
Degre
e o
f m
em
bers
hip
in2mf1 in2mf2
50 100150200250
0
0.5
1
Sender bitrate (kbps)
Degre
e o
f m
em
bers
hip
in3mf1 in3mf2
1 1.5 2 2.5
0
0.5
1
Mean burst length
Degre
e o
f m
em
bers
hip
in4mf1 in4mf2
0.05 0.1 0.15
0
0.5
1
Block error rate
Degre
e o
f m
em
bers
hip
in5mf1 in5mf2
1 1.5 2 2.5 3 3.5 4 4.50.5
1
1.5
2
2.5
3
3.5
4
4.5
5
MOS-measured
MO
S-p
redic
ted
Vid
eo M
od
el -
Uo
Po ANFIS based Model over UMTS networks
R2 = 87.17%RMSE = 0.2812
ICT-ADAMANTIUM
Task 4.1: Voice and Video Quality PQoS Models (6/12)
α β γ δ ε ξ η μ
4.2694 -1.4826 x 10-9
0.0656 -0.9559 -0.0261 -2.4767 -5.3168 0.3327
R2 83.52% RMSE
0.2778
Vid
eo M
od
el -
Uo
P o Regression based Model
ICT-ADAMANTIUM
Task 4.1: Voice and Video Quality PQoS Models (7/12)
Vid
eo M
odel
- D
EM
Test Signal Logarithmic Function R2 factorMobile 0.1295ln(x)+0.1274 0.9759Imax 0.0563ln(x)+0.6411 0.9514M.I. 3 0.0668ln(x)+0.5747 0.9191
Da Vinci Code 0.0474ln(x)+0.6974 0.8833Warren 0.0738ln(x)+0.5210 0.9528Nasa 0.0950ln(x)+0.3892 0.9595
BBC – Africa 0.1098ln(x)+0.2702 0.9875Superman 0.0282ln(x)+0.8167 0.8859
1 2ln( )SSIMPQoS C BitRate C
where C1 and C2 are constants strongly related to the spatial and temporal activity level of the content.
ICT-ADAMANTIUM
Task 4.1: Voice and Video Quality PQoS Models (8/12)
Vid
eo M
odel
- q
PS
NR PQoS tool developed and used in
R&S protocol analyser to provide an estimate of the PSNR of H.264-encoded video.
The algorithm is based on the statistical analysis of encoded transformation coefficients.
Provide a non-reference analysis of the image quality of compressed video.
ICT-ADAMANTIUM
Task 4.1: Voice and Video Quality PQoS Models (9/12)
Comparison of Voice and Video Models
Model Content Service Codec Usage Resolution
Voice -- E-Model Voice VoIP PCM, G.723.1 and GSM
Terminal
Voice – UoP Voice VoIP AMR Terminal
Video – UoP Video VoIP/IPTV H.264 Terminal, Access QCIF
Video – DEM Video IPTV H.264 Core Network/Server
CIF
Video – qPSNR Video IPTV H.264 Terminal CIF
ICT-ADAMANTIUM
Aud
iovi
sual
Mod
ellin
g
(Sim
ulat
ion
Set
up)
Task 4.1: Voice and Video Quality PQoS Models (10/12)
ICT-ADAMANTIUM
Task 4.1: Voice and Video Quality PQoS Models (11/12)
The proposed low-complexity metric is based on PER and FPS
Aud
iovi
sual
Mod
el
22
10
)PER1(
FPS β β MOSAV
Coefficient Value
β0 2.284
β1 0.089
β2 1.537
R2 84.9%
RMSE 0.325
ICT-ADAMANTIUM
Summary
Developed Voice PQoS Prediction Models Developed Video PQoS Prediction Models Developed Audiovisual PQoS Prediction Model Voice/video PQoS models have been used for quality
monitoring and PQoS-driven adaptation mechanisms in the ADAMANTIUM Demonstrator for VoIP and IPTV applications.
Contributed to D4.1
Task 4.1: Voice and Video Quality PQoS Models (12/12)
ICT-ADAMANTIUM
Task 4.2: Subjective aspects, content-awareness and
subjective tests
ICT-ADAMANTIUM
PC-based (UOP) and Handset-based (EHU)video subjective tests
Task 4.2: Subjective aspects, content-awareness and subjective tests (1/8)
Subjective tests according to ITU-T P.910 for video
Used Absolute Category Rating (ACR) Created 6 websites for video
subjective tests 20 participants – over 3 days 90 test samples – 60 used for training
and 30 for validation
Video sequences FR (fps) SBR (kbps) BLER(%) MBL
Akiyo, Foreman, Stefan
10
48, 88,128
1, 20, 30, 40 1, 1.75, 2.5Suzie, Carphone,
Football90, 130
PC-based tests
Sample conditions
ICT-ADAMANTIUM
Task 4.2: Subjective aspects, content-awareness and subjective tests (2/8)
Mobile device based tests
6 sets of video. Different sequences, 15 per test. Different transmission conditions. Videos presented randomly. 20 participants. MOS scale used for evaluation
1 bad
2 poor
3 fair
4 good
5 excellent
ICT-ADAMANTIUM
Encoding Quality: Impact of Display
Task 4.2: Subjective aspects, content-awareness and subjective tests (3/8)
0 50 100 1501
1.5
2
2.5
3
3.5
4
4.5
5
SBR (kbps)
MO
S
0 50 100 1501
1.5
2
2.5
3
3.5
4
4.5
5
SBR (kbps)
MO
S
0 50 100 1501
1.5
2
2.5
3
3.5
4
4.5
5
SBR (kbps)
MO
S
akiyo Handsetakiyo PCsuzie Handsetsuzie PC
foreman Handsetforeman PCcarphone Handsetcarphone PC
stefan Handsetstefan PCfootball Handsetfootball PC
bSBRaMOS )log(*
Subjectively higher quality in PC. Lack of rescale of QCIF in mobile
annoying. Difference independent of content or
encoding level.
Derived mapping of SBR to MOS
Transmission Quality: Impact of Display
Video Seq
FR-SBR-BLER MBL
Tx Packets
Rx Packets MOSPSNR
MOSsubj
Handset
MOSsubj
PC
Stefan 10-88-20 1.75 175 172 1.40 1.0625 1.82.5 175 169 1.11 1 1.8
10-128-20 1.75 214 211 1.61 1.5625 2.1
2.5 214 208 1.24 1 2.1
Suzie 10-90-30 2.5 92 89 3.17 1.0625 2.210-130-
40 1.75 106 105 4.00 2.1875 2.5
Carphone 10-90-30 1.75 232 231 4.28 3.5 3.8
10-90-30 2.5 232 231 4.28 2.3125 2.410-130-
30 1.75 261 259 4.27 1.875 2.32.5 261 260 4.23 2.25 3.1
Football10-130-
30 2.5 95 89 1.78 1.0625 2.3
Subjectively higher quality in PC. Non-Bursty errors have no significant impact on
video quality, unless a whole I-frame is affected. PSNR provides good MOS estimation for high
motion videos and over-estimates MOS for low motion videos.
ICT-ADAMANTIUM
Long Videos
Task 4.2: Subjective aspects, content-awareness and subjective tests (4/8)
Video sequences FR (fps)-SBR(kbps)-BLER MBLNews 12-80-0
12-130-012-200-012-130-1512-130-2012-130-2512-200-0512-200-10
1.75
TV series 12-80-012-130-012-200-012-130-2012-130-2512-200-0512-200-10
Basketball 15-80-015-200-015-256-015-256-05
Users cannot tolerate degradation for a certain time. Beyond this threshold users will close their video as service is “unbearable”.
Encoding Quality: Impact of Spatial Resolution on Mobile Handsets
0 50 100 150 200 2501
1.5
2
2.5
3
3.5
4
4.5
5
MO
S
SBR (kbps)
Low Motion Video Sequences
QCIF-1QCIF-2SIF
0 50 100 150 200 2501.5
2
2.5
3
3.5
4
4.5
5Medium Motion Video Sequences
SBR (kbps)
MO
S
QCIF-1QCIF-2SIF
0 50 100 150 200 2500
1
2
3
4
5High Motion Video Sequences
SBR (kbps)
MO
S
QCIF-1QCIF-2SIF
PQoS for QCIF saturates – resolution limits PQoS – especially for low motion videos.
Better to use higher spatial resolution for higher send bit rate. QCIF and SIF MOS cross point
Below this point QCIF get better quality due to low bitrate. Above which SIF get better quality and spatial resolution
becomes the more relevant factor that influence PQoS for users
ICT-ADAMANTIUM
Task 4.2: Subjective aspects, content-awareness and subjective tests (5/8)
Transmission Quality
Video Seq FR-SBR-BLER MBL MOSsubj_Handset MOSsubj_PC IPLR
HM1 - Stefan 10-88-20 1.75 1,0625 1,8 2,86%
2.5 1 1,8 3,43%
10-128-20 1.75 1,5625 2,1 2,34%
2.5 1 2,1 3,27%
LM2 - Suzie 10-90-30 2.5 1,0625 2,2 3,26%
10-130-40 1.75 2,1875 2,5 1,89%
MM2 - Carphone 10-90-30 2.5 2,3125 2,4 0,86%
10-130-30 1.75 1,875 2,3 0,77%
2.5 2,25 3,1 0,38%
HM2 - Football 10-130-30 2.5 1,0625 2,3 6,32%
Video Seq FR-SBR-BLER MBL MOSsubj_Handset IPLR
LM 12-130-15 1.75 3,67 0,32%
12-130-20 1.75 1,88 0,77%
12-130-25 1.75 0,58 3,35%
12-200-05 1.75 2,98 0,32%
12-200-10 1.75 2,73 0,77%
MM 12-130-20 1.75 3,01 0,54%
12-130-25 1.75 1,74 2,47%
12-200-05 1.75 3,87 0,25%
12-200-10 1.75 2,47 0,59%
HM 15-256-05 1.75 3,79 0,33%
Long VideosShort Videos
Video SeqFR-SBR-
BLER IPLR_CN MOSsubj_Handset
LM 12-130-00 1% 3,25 12-130-00 3% 2,52 HM 15-200-00 2% 2,47 15-200-00 5% 1,97
Different Loss Patterns (UMTS + CN)
User tolerance to CN-like losses is much higher than the tolerance to UMTS-like losses.
ICT-ADAMANTIUM
Task 4.2: Subjective aspects, content-awareness and subjective tests (6/8)
Evolution of User PerceptionLength of severe degradationsDepth of degradation
Single degradation that does not decrease from the first threshold is not severely penalized by users.
Deep scattered degradations of short duration do not represent a severe degradation of users’ perception.
0 10 20 30 40 50 60 70 80 90 1000
0.5
1
Time(s)
SS
IM
0 10 20 30 40 50 60 70 80 90 1000
0.5
1
Time(s)
SS
IM
0 20 40 60 80 100 120 140 1600
0.5
1
Time(s)
SS
IM
0 20 40 60 80 100 120 140 1600
0.5
1
Time(s)
SS
IM 1.5s
1.5s
3s
MOS = 4.42min score = 4
MOS = 3.25min score = 3
MOS = 3.86min score = 2.7
1s
7s
MOS = 3.74min score = 2.5
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
Time (s)
SS
IM
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
Time (s)
SS
IM
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
Time (s)
SS
IM
10s
5s
MOS=3.34
MOS=2.31
MOS=2.46
7s
6s
4s
5s
For deep degradations, the system must react with a maximum response time of 5 seconds.
For less severe degradations, the video can be maintained by users up to 7 seconds.
ICT-ADAMANTIUM
Task 4.2: Subjective aspects, content-awareness and subjective tests (7/8)
Audio-only, Video-only and Audiovisual subjective tests
Subjective tests according to ITU-T P.800, P.910 and P.911. Used Absolute Category Rating (ACR) Used a discrete 9-level quality scale 48 participants over 3 days Carried out audio-only, video-only and audiovisual tests A total of 60 samples Subjective test results were used for audiovisual modelling.
Subjective test website
ICT-ADAMANTIUM
Task 4.2: Subjective aspects, content-awareness and subjective tests (8/8)
Task 4.2: Summary Carried out PC based and mobile handset based video
quality subjective tests Carried out audio-only, video-only and audiovisual quality
subjective tests Subjectively evaluated impact of display method, video
length, spatial resolution, loss pattern etc. on PQoS quality Subjectively validated audio/video PQoS models Identified temporal tolerance threshold which can be used in
adaptation mechanisms
Contributed to D4.1
ICT-ADAMANTIUM
Task 4.3:
Mapping PQoS To DiffServ/MPLS and UMTS Traffic Classes
ICT-ADAMANTIUM
Delay Quality0 – 150 ms Acceptable for most calls.
150 – 400 ms Acceptable if callers are aware of impairment
> 400 ms Not acceptable
CODEC Algorithmic DelayG.711 without PLC 0.125 msG.711 with PLC 3.875 msG7.729 (compression-based) 15 ms
Packetization Delay ~20 msec
Propagation Delay <1msec
Algorithmic Delay
Delay Requirements
Task 4.3: Mapping PQoS to DiffServ/MPLS and UMTS Traffic Classes (1/12)
VoIP Mapping Framework -- Theoretical Approach
ICT-ADAMANTIUM
Task 4.3: Mapping PQoS to DiffServ/MPLS and UMTS Traffic Classes (2/12)
VoIP Mapping Framework -- Experimental Approach
0 5 10 15 20 25 301
1.5
2
2.5
3
3.5
4
4.5MOS (PESQ-LQO) vs. Packet loss rate
Packet loss rate (, %)
MO
S (
PE
SQ
-LQ
O)
AMR(12.2Kb/s)AMR(4.75Kb/s)G.729(8Kb/s)G.723.1(6.3Kb/s)iLBC(15.2Kb/s)iLBC(13.3Kb/s)
Parameters AMR (12.2Kb/s) AMR (4.75Kb/s) G.729 (8Kb/s) G.723.1 (6.3Kb/s) iLBC (15.2Kb/s) iLBC (13.3Kb/s)
a 4.1522 3.1923 3.8309 3.633 3.8551 3.8083
b 1.2190 0.8776 1.0964 1.0427 1.4981 1.3723
c 0.2508 0.2022 0.2493 0.2093 0.0799 0.1010
R2 factor 0.9971 0.9981 0.9988 0.9985 0.9996 0.9991
ICT-ADAMANTIUM
Task 4.3: Mapping PQoS to DiffServ/MPLS and UMTS Traffic Classes (3/12)
IPTV Mapping Framework – Theoretical Approach
Mapping of IPTV PQoS to SBR=f(BLER, MBL)
Mapping of IPTV PQoS to SBR, FR
Mapping of IPTV PQoS to SBR=f(BLER, MBL)
Mapping of IPTV PQoS to CT=f(BLER, MBL)
Mapping of IPTV PQoS to BLER, MBL
Mapping of IPTV PQoS to FR=f(BLER, MBL)
ICT-ADAMANTIUM
Task 4.3: Mapping PQoS to DiffServ/MPLS and UMTS Traffic Classes (4/12)
0
0,05
0,1
0,15
0,2
0,25
0,3
0,35
0,4
0,45
0,5
0,55
0,6
0,65
0,7
0 0,02 0,04 0,06 0,08 0,1 0,12
Packet Loss Rate
Q E
xpec
ted
Su
cces
full
y D
eco
dab
le F
ram
es Packet Size 188 Bytes
Packet Size 250 Bytes
Packet Size 500 Bytes
Packet Size 1000 Bytes
P
P PI P P
I I p I B
jCN NC C jC C N C C C
GOP GOP GOP
j 1 j 1
1- *N 1- * 1- *N 1- 1- * M - 1 * 1- *N
Q
Q
dec dec-I dec- p dec-B
total - I total - P total - B total - I total - P total - B
to
p p p + p p p
N + N + NN
( N N N ) ( N N N )
( N
tal - I total - P total - B N N )
Packet Loss Rate
Expected Decodable Frame Rate Q
MOS
Expected Decodable Frame Rate Q vs. Packet Loss Rate for various packet sizes
ICT-ADAMANTIUM
Packet Loss Rate
Expected Decodable Frame Rate
Q
MOS
Task 4.3: Mapping PQoS to DiffServ/MPLS and UMTS Traffic Classes (5/12)
Expected Decodable Frame Rate Q vs. MOS for various packet sizes
For 0.01 0.05p , MOS becomes:
Pr1.01
53.0385.8
5621 ( )
Q1 0.05
-3.9204 + 1.0315
edictedMOS
p
For 0.05 0.1p , MOS becomes:
Pr1.01
53.0385.8
5621 ( )
1 Q
edictedMOS
ICT-ADAMANTIUM
Packet Loss Rate
Expected Decodable
Frame Rate Q
MOS
Task 4.3: Mapping PQoS to DiffServ/MPLS and UMTS Traffic Classes (6/12)
ICT-ADAMANTIUM
The generic framework/architecture for joint N- and P-QoS assessment
Task 4.3: Mapping PQoS to DiffServ/MPLS and UMTS Traffic Classes (7/12)
IPTV Mapping Framework -- Experimental Approach
ICT-ADAMANTIUM
The experimental signal set
Sequence no.
Source Spatial complexity Temporal dynamics
1 007 Quantum of Solace (MGM/Fox, 2008) Medium Very high
2 007 Quantum of Solace (MGM/Fox, 2008) Very high Very high
3 Marley & Me (Fox, 2009) High Medium
4 Lawrence of Arabia (Horizon, 1962) High Low
5 Lawrence of Arabia (Horizon, 1962) Low Low
Task 4.3: Mapping PQoS to DiffServ/MPLS and UMTS Traffic Classes (8/12)
ICT-ADAMANTIUM
Statistical Distributions…
PQoS = 100 - 3,92p
PQoS = 100 - 3,92p
PQoS = 100 – 3.92p
Task 4.3: Mapping PQoS to DiffServ/MPLS and UMTS Traffic Classes (9/12)
ICT-ADAMANTIUM
QoS PROPERTIES and mapping OF DIFFSERV CLASSES
DiffServ Class Type of Service
EF VoIP
AF1x A/V Content
BE Other data services
Diffserv Classes Mapping to Services
Task 4.3: Mapping PQoS to DiffServ/MPLS and UMTS Traffic Classes (10/12)
For both VoIP and IPTV, upper bound packet loss for acceptable quality is around 5%.
Statistical analysis of EF class by the ADAMANTIUM core network shows that the respective one-way delay is less than 50 ms and packet loss close to zero.
For AF1x classes at the core network, the respective IPLR has been measured less than 1% for all traffic schemes.
ICT-ADAMANTIUM
QoS PROPERTIES and mapping OF UMTS classes
• The conversational traffic class is marked as Expedited Forwarding to preserve the low-latency queuing behavior.
• The rate at which EF traffic is served at a given output interface should be at least the configured rate R, independent of the offered load of the non-EF traffic to that interface.
• Streaming class is marked as Assured Forwarding. Video traffic, due to its limited burst behavior and large packet size, is more problematic to manage than conversational Expedited Forwarding voice.
• The rest Assured Forwarding classifications are exploited by the Interactive traffic class.
UMTS Traffic Class
Conversational
(Real Time)
Streaming
(Real Time)
Background
(Best Effort)
Example Applications VoIP IPTV -
Diffserv Class / Map to DSCP
Expedited Forwarding Assured Forwarding Best Effort
Task 4.3: Mapping PQoS to DiffServ/MPLS and UMTS Traffic Classes (11/12)
ICT-ADAMANTIUM
Task 4.3: Mapping PQoS to DiffServ/MPLS and UMTS Traffic Classes (12/12)
Task 4.3 Summary
Carried out theoretical and experimental research on mapping of NQoS to PQoS for VoIP and IPTV
Developed mapping of Diffserv classes to PQoS degradation Developed mapping of UMTS traffic classes to PQoS
Contributed to D4.2
ICT-ADAMANTIUM
Task 4.4:
Real Time Dynamic Content Adaptation Mechanisms
ICT-ADAMANTIUM
General adaptation procedure
Dec
isio
nm
akin
g
Network awareness(TNMM, ANMM)
Service awareness(MSMM, TAM)
PQoSmodels
Networkmodels
Servicemodels
Variable ParametersC
urr
ent
sta
tus
New
sta
tus
Mo
dif
iab
le p
aram
ete
r v
alu
es
Network adaptations(TNAM, ANAM)
Service adaptations(MSAM, TAM)
Task 4.4: Real Time Dynamic Content Adaptation Mechanisms (1/14)
ICT-ADAMANTIUM
Decision Making
ADAMANTIUM enhanced decision making PQoS-driven Network-and service-aware Centralized approach Combined adaptations
QoE-driven Bitrate-based Service Adaptations Validation of QoE-driven approach Source of degradations: network bottleneck (+encoding) Adaptation action: application bitrate
QoE-driven Cross-layer Multi-parameter Adaptations Sources of degradations: encoding, AN, CN Adaptation actions
Different parameters at application-layer BLER control at AN (bearer switching) IPLR control at CN (promotion of CoS)
Task 4.4: Real Time Dynamic Content Adaptation Mechanisms (2/14)
ICT-ADAMANTIUM
QoE-driven Bitrate-based Service Adaptations VoIP
44
dMOS MOS
POOR ACCEPTABLE GOOD
Z BI SI Z
S SI Z Z
B Z SD SD
IF MOS IS Good AND IF dMOS IS Z then Ctrl IS Z.
),,( ModeinstDelayPLRfinstMOS
avgMOSMOSdMOS max
IT-2 Fuzzy Controller
CtrlModeMode ni 1
Network-aware QoE-driven
AMR mode switching
Task 4.4: Real Time Dynamic Content Adaptation Mechanisms (3/14)
ICT-ADAMANTIUM
45
Control surface
50 VoIP sessions Competing for available
network bandwidth
Number of users served at 1Mbps
MOS Threshold
Non-Adaptive aVoIP FLC
2.0 70 110 1182.5 65 85 1083.0 62 70 1003.5 55 62 854.0 55 58 60
Users above the threshold
Task 4.4: Real Time Dynamic Content Adaptation Mechanisms (4/14)
QoE-driven Bitrate-based Service Adaptations VoIP
ICT-ADAMANTIUM
Define delta (delta=50)
if (0 < D < 0.4) and (0<C<0.2) then SBR = SBRsame (Maintain SBR)
else if (D ≥ 0.4) and (C≥0.2) then SBR = SBR – delta (Decrease SBR)
end if
end if
46
Encoder Packetizer DecoderDe-
Packetizer
Feedback Mechanism
Reference-free QoE prediction
model
Video Sender Bitrate
Adaptor
Content Classifier
Video quality measurement (PSNR/MOS)
Feedback
Raw Video Degraded Video
Network
s
l
B
BC
PQoS-driven Video Adaptation Algorithm
MOSmMOSD maxPQoS-driven
Network-aware
Task 4.4: Real Time Dynamic Content Adaptation Mechanisms (5/14)
QoE-driven Bitrate-based Service Adaptations Mobile Video
ICT-ADAMANTIUM
47
0 0.5 1 1.5 2 2.5 3 3.5 40
0.5
1
1.5
2
2.5
3
3.5
4
Time (seconds)
MO
S
Football-Not adaptedFootball-adapted
150 200 250 300 3501.5
2
2.5
3
3.5
4
Link Bandwidth (Kbps)
MO
S
SBR-Football @88kbpsSBR-Football @128kbpsSBR-Football @200kbpsSBR-Football adapted
Time evolution under UMTS degradations
Network-aware QoE-driven
H.264 SBR switching
Task 4.4: Real Time Dynamic Content Adaptation Mechanisms (6/14)
QoE-driven Bitrate-based Service Adaptations Mobile Video
ICT-ADAMANTIUM
Multi-parameter Cross-Layer
Decision Making
Dec
isio
nm
akin
g
Network awareness(TNMM, ANMM)
Service awareness(MSMM, TAM)
PQoSmodels
Networkmodels
Servicemodels
Variable Parameters
Cu
rren
t st
atu
s
New
sta
tus
Mo
dif
iab
le p
aram
ete
r v
alu
es
Network adaptations(TNAM, ANAM)
Service adaptations(MSAM, TAM)
Task 4.4: Real Time Dynamic Content Adaptation Mechanisms (7/14)
ICT-ADAMANTIUM
491
23
45
67
8
0
2
4
6
8
100
5
10
15
20
25
30
35
AN statesCN states
VoI
P c
onfig
urat
ions
Service-level decision map
Task 4.4: Multi-parameter service-level decision making (8/14)
ICT-ADAMANTIUM
50
Task 4.4: Multi-parameter cross-layer decision making (9/14)
1 1.5 2 2.5 3 3.5 4 4.50
0.5
1BLER = 0.1%
1 1.5 2 2.5 3 3.5 4 4.50
0.5
1BLER = 0.5%
1 1.5 2 2.5 3 3.5 4 4.50
0.5
1BLER = 1%
1 1.5 2 2.5 3 3.5 4 4.50
0.5
1
EC
DF
(M
OS
) BLER = 2.5%
1 1.5 2 2.5 3 3.5 4 4.50
0.5
1BLER = 5%
1 1.5 2 2.5 3 3.5 4 4.50
0.5
1BLER = 10%
1 1.5 2 2.5 3 3.5 4 4.50
0.5
1BLER = 20%
1 1.5 2 2.5 3 3.5 4 4.50
0.5
1
MOS value
BLER = 30%
MOS<3.5 MOS<3.1 MOS<2.5min max min max min max
BLER=0.1% 0.71% 1.52% 0% 0.70% 0% 0%BLER=0.5% 4.27% 12.61% 0.67% 2.71% 0% 0%BLER=1% 9.72% 25.83% 2.63% 6.10% 0.37% 0.89%BLER=2.5% 29.85% 53.69% 12.86% 20.92% 1.72% 4.74%BLER=5% 63.93% 80.96% 38.99% 55.22% 12.27% 22.38%BLER=10% 87.92% 97.18% 80.57% 88.65% 42.61% 62.71%BLER=20% 98.42% 99.84% 98.42% 99.34% 91.43% 94.07%BLER=30% 99.46% 100% 99.46% 100% 98.49% 99.50%
ECDF of best VoIP configurations for combined AN/CN states
POORACCEPTABLEGOOD
IT-2 Inference Rules
ICT-ADAMANTIUM
51
0 5 10 15 20 25 301
1.5
2
2.5
3
3.5
4
4.5
5
BLER (%)
Exp
ect
ed
MO
S
LM, SBR=200kbpsLM, SBR=130kbpsMM, SBR=200kbpsMM, SBR=130kbpsHM, SBR=256kbpsHM, SBR=200kbpsHM, SBR=130kbps
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 51
1.5
2
2.5
3
3.5
4
4.5
IPLR (%)
Exp
ecte
d M
OS
HM, SBR=200kbpsLM, SBR=130kbps
0 5 10 15 20 250
51
1.5
2
2.5
3
3.5
4
4.5
BLER (%)
IPLR (%)
Exp
ect
ed
MO
S
50 100 150 200 2501
1.5
2
2.5
3
3.5
4
4.5
5
SBR (kbps)
Exp
ect
ed
MO
S
LM, SIFMM, SIFHM, SIFLM, QCIFMM, QCIFHM, QCIF
Task 4.4: Real Time Dynamic Content Adaptation Mechanisms (10/14)
Optimization Function: Integral PQoSH.264 Baseline Profile @ Level 1.2
- CT = {LM, MM, HM}- SBR = {48-256} kbps- FR = {10-15} fps- SR = {320x240, 176x144}
D4.1
D4.1
“Interactive or Background / UL:64 DL:384 kbps / PS RAB”
RLC-level 2-state Markov loss model
D4.2
D4.1
CN model:
DiffServ/MPLS CoS
IPLR model
ICT-ADAMANTIUM
52
Multimedia session 1
Application-layer
Multimedia session 2
Parameter Allowed Values VariableCT LM=Low Motion, MM=Medium
Motion, HM=High MotionNo
SBR {80, 130, 200, 256} kbps YesSR 1 (320x240), 2 (176x128) YesDLBR {384, 128} kbps YesBLER No Restrictions IndirectCoS 1 (BE), 2 (AF22), 3(AF11), 4(EF) YesIPLR No Restrictions Indirect
Task 4.4: Real Time Dynamic Content Adaptation Mechanisms (11/14)
Optimization system variables
ICT-ADAMANTIUM
Task 4.4: GA-based decision making (12/14)
53
Update system status
PQoS Alarm
InitialPopulation
Fitness Function
Scaling Selection
ReproductionCrossOver
Max. GenerationsStall Generations
no
Optimal Individual
yes
Select Best SBR = f {CT, DL_BR, BLER}
Promote CoS of degraded sessions
Change SBR of degraded sessions
Decrease SBR and promote CoS of
degraded sessions
Initial Individual
InitialPopulation
Update IPTV structures
Assignment Matrix
Estimate new CN conditions
(D4.2)
Compute expected MOSi
(D4.1)
Compute fitness score
Genetic Algorithm
Initial Population
Fitness Function
min, MOSMOSfreFitnessSco avg
ICT-ADAMANTIUM
Optimization Results
54
0 500 1000 15001
2
3
4
5
Background Traffic Load
Ave
rage M
OS
0 500 1000 15000
1
2
3
4
5
Background Traffic Load
Min
imum
MO
S
No AdaptationCN-driven AdaptationAN/CN-driven AdaptationADAMANTIUM Adaptation
No AdaptationCN-driven AdaptationAN/CN AdaptationADAMANTIUM Adaptation
20 mobile IPTV sessions competing for available
network bandwidth
CT SBR (kbps) BLER (%)LM 130, 200 0,10,25MM 130, 200 0,10,25HM 256 0,10HM 130,200 0,10,25
Additional Background Load (Mbps)0 0.5 1 1.5
Load (kbps) 3482 3982 4482 4982IPLR (%) 0.056
20.8132 11.7571 29.45
Task 4.4: Real Time Dynamic Content Adaptation Mechanisms (13/14)
0 2 4 6 8 10 12 14 16 18 201
1.5
2
2.5
3
3.5
4
4.5
5
IPTV Session
MO
S
Background Traffic = 0 Mbps
0 2 4 6 8 10 12 14 16 18 201
1.5
2
2.5
3
3.5
4
4.5
5
IPTV Session
MO
S
Background Traffic = 500 Mbps
0 2 4 6 8 10 12 14 16 18 201
1.5
2
2.5
3
3.5
4
4.5
5
IPTV Session
MO
S
Background Traffic = 1000 Mbps
0 2 4 6 8 10 12 14 16 18 201
1.5
2
2.5
3
3.5
4
4.5
5
IPTV Session
MO
S
Background Traffic = 1500 Mbps
No AdaptationCN-driven AdaptationAN/CN AdaptationADAMANTIUM Adaptation
No AdaptationCN-driven AdaptationAN/CN AdaptationADAMANTIUM Adaptation
No AdaptationCN-driven AdaptationAN/CN AdaptationADAMANTIUM Adaptation
No AdaptationCN-driven AdaptationAN/CN AdaptationADAMANTIUM Adaptation
ICT-ADAMANTIUM
Carried out research and developed Service-level adaptation mechanisms for VoIP and IPTV
Developed per-segment network-aware cross layer adaptation mechanism
Carried out research and developed decision support for AEM Adaptation rules for AEM Intelligent AEM based on Fuzzy Controller and Genetic
Algorithm
Contributed to D4.3
Task 4.4: Real Time Dynamic Content Adaptation Mechanisms (14/14)
Task 4.4 Summary
ICT-ADAMANTIUM
WP4 Conclusions
Developed voice/video PQoS models. Developed PQoS-driven voice/video adaptation
mechanisms both in simulation and in test-bed Developed full ADAMANTIUM simulation
environment in Opnet and NS2 Conducted extensive subjective tests and models
evaluations. Developed a framework for mapping of Network QoS
statistics to PQoS level for VoIP and IPTV
ICT-ADAMANTIUM
WP 4PQoS Models and Adaptation Mechanisms
Planned Tasks [M25 – M30]
ICT-ADAMANTIUM
WP4 Planned Tasks
Perceptual PQoS Modelling. Intelligent audiovisual model for audiovisual over UMTS
networks.
Adaptation Mechanisms Computationally Intelligent EARLY adaptation mechanisms. Initial results very promising, a mechanism beyond
ADAMANTIUM
ICT-ADAMANTIUM
THANK YOU