Scalable Video Coding and Differentiated Error Protection The technology for sustainable future mobile TV
Authors: Alberto Morello (Rai) & Thomas Wiegand (Fraunhofer HHI) (Coverage estimations by Paolo Forni - Rai)
SINCE 60 YEARS... • Broadcast systems/networks have always been service-specific • • • • •
PAL/SECAM/NTSC for analogue video FM/AM for analogue radio DAB/DMB/DRM/IBOC/HDRadio/.... for Digital Radio DVB-S-T-C / ATSC /ISDB-S-T-C/,..... for Digital TV DVB-H-SH/MediaFlo/..... for Mobile-TV
• MANY TECHNOLOGIES= MANY NETWORKS= HIGHER COST
THE DVB SECOND GENERATION REVOLUTION With DVB SECOND GENERATION technologies: • Near-Shannon Physical layers • Scalable Video Coding • Very high flexibility • ... can we dream of a single technology and a single terrestrial broadcast network for TV, HDTV, MobileTV, Radio, ...????? • Can this offer a new paradigm for successful mobile media services in the future?
Where does current divergence come from? • Answer 1: different business scenarios / ecosystems: • • •
TV / HDTV: a Broadcasters' business (PSB, commercial or pay) Mobile TV: so-far a shared Mobile-Telco/Broadcaster business (or flop?) Radio: large national stations (sometimes owned by TV broadcasters), medium Regional stations and a galaxy of local stations: different needs & business models
• .. but for national broadcasters, integrated multiservice broadcast networks may be very cost effective
Where does current divergence come from? • Answer 2: different technical/functional requirements: • •
TV/HDTV fixed reception = high capacity & efficiency mobile reception = maximum ruggedness & synchronization speed
• Answer 3: The terrestrial transmission physics: huge SNR difference between outdoor reception with directive antennas and mobile & indoor reception with omni-antennas
The terrestrial physics E=85 dBÂľV/m at 10 m Band V
SNR@ 95% =48 dB
SNR@99% =11 dB rural -6 dB urban
In car
SNR@ 95% =-5 dB at ground floor (5 dB @ 70%)
SNR@ 99% =24 dB rural 7 dB urban
SNR@ 95% =11 dB urban (17 dB @ 70%)
The figures are confirmed by real tests EREMO TRANSMITTER (300W) 100%
99%
17
E=95 dBÂľV/m,
16
98%
100 % 99%
12
15 13
100% 7
11
74%
98%
100 %
90% 14
5
10
82%
100%
98%
3
9
98%
8
100%
Rai DVB-H test campaing: Torino-Eremo transmitter, UHF channel 29, ERP=4 kW 3500 indoor test points at ground floor, 500 locations
1
DVB-H Measured average Coverage Probability = 96%, in line with predictions (indoor SNR@ 95%=5 dB)
99%
6 4
2
97%
To achieve deep-indoor in every zone (3,10,11) at least 5 dB more are required – cellular network
A practical coverage example: Turin area
DVB-H urban portable outdoor & car antenna Broadcast DVB-H light-indoor (85%)
Torino - Eremo high power Transmitter (h=428 m above city level) Antenna mast: 50 m Tx=1 KW; ERP(max)=6 kW. UHF ch. 66
30Km E=85 dBµV/m 20Km E=90 dBµV/m Downtown
Cellular DVB-H urban deep-indoor (30-40 cells)
DVB-T Fixed
40Km E=80 dBµV/m
A single system for any application An impossible dream?
No unique broadcast system will ever cope with a 43-53 dB SNR difference: it's impossible to achieve from the same transmitter the same coverage area, for fixed HDTV reception with directive (outdoor) antennas and for handheld mobile-TV deep-indoor reception ...... but we have good weapons now to shoot against this problem, provided that .....:
ďƒ˜ we abandon the expensive "deep-indoor" (true >95% location) coverage criteria for a more reasonable "lightindoor" target (90% locations) ďƒ˜ we use second generation DVB technologies
The "second generation" DVB weapons Second generation DVB systems (T2, NGH,...) (will) allow NearShannon FEC and differentiated error protection, based on the Physical Layer Pipe (PLP) concept MIMO techniques provides important means to improve mobile/portable reception SVC (Scalable Video Coding) provides layered video bitstreams: a base layer fully MPEG-4 AVC compliant one or more enhancement layers (better spacial res, temporal res. or quality) 10
1
Second Generation (T2) Physical Layer Capacity Performance (Portable channel P1) 5.0000
DVB-T2 (and NGH) (QPSK 1/2) offers Vz DVB-H: • 3.5 dB SNR gain • the use of a double tuner/antenna (MaxRatioComb.) offers additional: 3-3.5 dB SNR gain
Effective bits per Cell
3.7500
2.5000
DVB-T2 DVB-T2 1.2500
0
• TOTAL=7 dB gain • With code rate 1/4: additional 3 dB
-5.0
-3.1
-1.2 3.5 dB0.7
2.6
4.5
6.5 DVB- 8.4
3.5 dBSNR [dB]
10.3
12.2
14.1
16.0
Perspectives of MIMO 2 techniques
Low X-polar discrimination for hand-helds (indoor: around 3 dB; out-door: around 7 dB): • for low-to-medium efficiencies (0.5 to 4 bit/s/Hz) (suitable for mobile applications) the use of Alamouti is considered (cross-polar MIMO 2x2 or SIMO 1x2): 3-4 dB SNR Gains over DVB-T2
DVB-T2 SIMO or Alamouti MIMO
• Spacial multiplexing (Goldencode,...): large gains for efficiencies larger than 4 bit/ s/Hz (high capacity, dense networks) but may give
3
Potential advantage of differentiated protection + Multi-layer video coding
It may be useful for very critical channels (e.g. large SNR variations in time or space), where it is difficult to ensure large service continuity (e.g. 99% of time and locations) and high picture quality: mobile terrestrial (or satellite) channels (heavy shadowing attenuations; deep- indoor building penetration loss...)
Therefore it may be reasonable to reduce network cost or increase efficiency (more programs) by offering: high quality signals (e.g. CIF resolution) + high spectrum efficiency for fair percentages (e.g. 70-90%) of time and locations lower quality signals (e.g. QCIF resolution) + high protection for large 13 service continuity (e.g. from 90% to 99% of time and locations)
Differentiated protection: DVB-T2 PLP approach Capacity Performance Portable
PLP2 DVBT2 TX
6.0000
Effective bits per Cell
PLP1 DVBT2PLP2 RX
8.0000
Spectrum efficiency
PLP1
channel
Each Physical Layer Pipe provides independent LDPC coding, time interleaving and mapping to constellation
4.0000
2 Tuner/Antenna
2.0000
DVB-T2
0 -5.0
PLPn
PLPn n < 256
0
5.0
10.0 SNR
15.0
1 Tuner/Antenna
20.0
SNR
For each PLP, a different "point" on the Capacity/SNR plane can be selected: 2 x tuners: from -5 dB SNR (0.5 bit/s/Hz) to 19.5 dB SNR (6 bit/s/Hz)
SVC: Scalability of Video Scalability: a functionality for removal of parts of the bitstream while achieving a performance comparable to single-layer H.264/AVC coding at that particular resolution.
Temporal: change of frame rate
30 fps 15 fps 7.5 fps
Fidelity: change of quality (e.g. SNR)
Spatial: change of frame size (i.e., resolution)
QCIF
CIF
TV
Temporal Scalability with H.264/AVC N=1 Temporal Scalability
I P P P P P P P P Base-layer
N=2 I S0 P S0 P S0 P S0 P
N=4 I S1 S0 S1 P S1 S0 S1 P
N=8 I S2 S1 S2 S0 S2 S1 S2 P
• • •
S stands for P or B picture Works for mobile TV General rule: Layers with higher temporal resolution shall not be used for prediction of layers with lower temporal resolution
Spatial (resolution) Scalability: Typical Encoding
Hierarchical MCP & Intra prediction
Spatial decimation
texture motion
Base layer coding
Multiplex
Inter-layer prediction: • Intra • Motion • Residual
H.264/AVC MCP & Intra prediction
texture motion
Base layer coding H.264/AVC compatible encoder
H.264/AVC-compatible base layer bit-stream
Scalable bit-stream
SVC Loss Typical SVC loss (vz single layer AVC) =10% at baselayer and at enhancement layer
It is possible to trade-off base-layer and enhanced-layer losses: same loss (Δ< 10%) no loss in the base-layer, larger loss in the upper layer
COMBINING DIFFERENTIATED PROTECTION AND VIDEO SERVICES Two independently-coded AVC video streams (Multi-layer simulcasting) at two different quality levels R
1
E1
Coder 1 (base layer)
MODULATOR
Coder 2 R
2
E2
(enhanced layer)
SIMULCAST
Problems • synchronization: "quasi-seamless" stream switching during fading • efficiency loss: since the "scene" is the same in the two layers, the base-layer information is "duplicated"
DIFFERENTIATED PROTECTION OF VIDEO SERVICES: SVC To solve the above problems of Multi-layer simulcasting, a single R1' scalable codec (SVC) may be adopted: Base layer
SVC
R1'
Coder (e.g. 10% loss compared to single coding)
E1 MODULATOR
R2'
E2
SVC R1'+R2' Enhanced layer
SVC BW GAIN VERSUS SYMULCASTING R2
R2/E2
R1
High efficiency
R1 E
Coder 1
Low Efficiency
1
T2/NGH
(base layer)
MODULATOR
Coder 2 (enhanced
R1/E1
E
R2
2
layer) SIMULCAST
BW= (R / E )+( R / E ) 1
1
2
2
Loss Δ=1.1 (=10%) R'1
SVC
E
Coder (e.g. 10% loss compared to single coding) SVC
1
T2/NGH
E
R'2
MODULATOR 2
GAIN
GAIN BW'= (R ' / E )+( R ' / E ) 1
1.1(R2-R1)
1.1R1
1
2
2
1.1(y-x)/E2
1.1R1/E1
Limits and opportunities of SVC What's the formula (R2/R1) + (E2/E1) = [1- (G Δ)-1 ]-1 Examples:
To achieve G=21% : To achieve G=13%: To achieve G=9% :
saying?
(R2/R1) + (E2/E1)=4 (R2/R1) + (E2/E1)=5 (R2/R1) + (E2/E1)=6
SVC gains if: the two streams are not be too different in bit-rate and protection : • Of course, same protection or same bit-rate are nonsense • But there are interesting use cases where SVC can offer reasonable bandwidth gains (10% to 20%) over simulcasting •
For extreme cases (very critical channels), where very large SNR differences 22 between layers are needed, SVC offers marginal gain over simulcasting
Examples of NGH potential How effective may NGH be to achieve mobile-TV / multimedia from purely broadcast networks?
Example 1 NGH with scalable video coding: DVB-H: 11 programs (CIF) at SNR=5.5 dB DVB-NGH: 18 programs at SNR=- 1.5 dB (QCIF) and at SNR=7 dB (CIF)
SVC gain=16%
Simulated coverage example 1: Turin area If we accept to target indoor at reduced quality (QCIF), the overall system capacity increases significantly: DVB-H: BW=500 KHz/channel --> 11 mob-TV programs DVB-NGH + AVC: BW= 378 KHz / channel --> 18 mob-TV prog. 30Km E=85 dBµV/m
Light-indoor QCIF 90% prob. Deep-indoor
20Km E=90 dBµV/m
QCIF quality, 97% probability Downtown CIF quality, 65% probability
40Km E=80 dBµV/m
Example 2 (maximum protection) NGH with FEC 1/4 and video simulcasting DVB-H: 11 programs (CIF) at SNR=5.5 dB DVB-NGH: 11 programs at SNR=- 4.5 dB (QCIF) and at SNR=7 dB (CIF)
SVC not used
Simulated coverage example 2: Turin area Further increase of NGH protection (LDPC rate 1/4) DVB-NGH: 11 mob-TV programs as for DVB-H
30Km E=85 dBµV/m
20Km E=90 dBµV/m
Deep Indoor QCIF quality, 95% probability Downtown CIF quality, 65% probability
40Km E=80 dBµV/m
Example 3 (HDTV & Mobile-TV Coexistence) HDTV: 3 programs (10.5 Mbit/s each) at SNR=23 dB DVB-NGH (in Future Extension Frames) : 4 programs at SNR=- 1.5 dB (QCIF) and at SNR=7 dB (CIF)
SVC applied to Mobile-TV only gain=16%
Simulated coverage example 3: Turin area 3 HDTV programs using DVB-T2 4 mobile-TV programs using DVB-NGH+SVC
HDTV Mob-TV: HH Urban outdoor Rural in car
30Km light indoorE=85 QCIF dBµV/m 90% prob.
20Km E=90 dBµV/m
Deep-indoor QCIF , 96% probability Downtown CIF, 65% probability
40Km E=80 dBµV/m
Conclusions (1) "Gain budget" of DVB-NGH Vs DVB-H Tool
Gain (rough estimation from examples)
comment
DVB-2 physical layer
3.5 dB SNR (QPSK 1/2) 6.5 dB SNR (QPSK 1/4)
(OFDM, LDPC FEC , time interleaving and rot.constell.)
Double receive tuners/ antennas
3.5 dB SNR
(SIMO or Alamouti MIMO)
Layered video coding & error protection
SVC: 30% BW (or same BW but 3 dB SNR *)
loss: QCIF instead of CIF in critical locations (indoor/incar)
TOTAL
7 dB SNR + 30% BW
"light-indoor" coverage * Note: QPSK + LDPC 1/4
(or 10 dB SNR and same BW *) 90% of locations: 3 dB SNR
instead of 95% deep-indoor
Conclusions (2): How to transform the SNR gain into € saving budget
First generation Mobile-TV was unsuccessful: requiring very expensive cellular networks to reach "deep-indoor" broadcast coverage in towns based on a mixed Mobile-Telco / Broadcaster business concept: only pay-content may justify network investments people is not ready to pay for broadcast-type content people may be ready to pay for specific content and services, but occasionally
Why not changing the business/cost paradigm of mobile media? Broadcast & Premium services
Broadcast services DVB-T2 / DVB-NGH Free broadcast networks (cheap, light-indoor&cars) or low-cost
DVB-H Cellular & Broadcast networks Pay First generation
Premium services
UMTS / HSDPA/LTE cellular networks (expensive, deep-indoor)
Second generation
Mainly Pay
Thank you for attention
The capacity gain of SVC PLP Spectrum effic. ratio (E1/E2)
2.0000 Same protection
G
1.5000 1.2500
1.5 2 2.5 3 4 5
1.0000 1.0
Ideal SVC (no coding loss) E1/ E2=1
Increasing Î&#x201D; protection
1
1.7500
1.6
2.3
2.9
3.5
=R2/R1 bit-rate ratio
K
33 The two streams must not be too different in bit-rate and protection