Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
1
Real-Time Architectures2003/2004
A real-time architecture formultimedia consumer terminals
(MCTs)
Reinder J. Bril03-05-2004
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
2
A real-time architecture for MCTs• Multimedia Consumer Terminals (MCTs):
– Trends;– Characteristics.
• Media processing in software:– A Digital Video Platform;– Scalable Video Applications;– Challenges & approaches.
• A co-operative QoS approach:– Multi-disciplinary team;– System.
• A retrospect on real-time issues
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
3
Multimedia Consumer Terminals
DVD CDxfront end
YC interface
IEEE 1394interface
DVB Tuner
Cable modem
CVBSinterface
VGA
RF Tuner Focus:Receivers in broad-cast environments
High-quality video applications
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
4
Trends• Digitization and compression:
– DVD, DTV
• Featurization:– Software services:
• TXT, OSD, menus, habit learning.– Processing:
• Sharpness enhancement, 100 Hz, PiP, dual screen.
• Convergence of services:– Delayed viewing (TV + VR/DVD-recorder);– Games (TV + game-console);– TV-commerce and interaction (TV + internet).
• Upgrading:– New standards for D-STB via satellite.
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
5
Characteristics
• High quality:– Including timeliness of audio and video.– Example: video enhancement.
Original (24 Hz)
Up-scaled (60 Hz)
Input stream: movie
Rendered stream: TV screen
QoS trade-off:Lesser picture quality often better than temporal incorrectness.
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
6
Characteristics
• High dependability:– Must work, immediately, and at all times.
• High Volume Electronics:– Global market with multiple standards;– Short product development cycles;– Product families.
• Low power:– No fan in a TV set …
• High pressure on price:– DVD recorder from $2000 to $200 in 2 years.
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
7
Traditional High-End TV Architecture
TXT CPUacq control
progROMRAM
PICNICenhancement
PALplusIC16:9 helper
MEM MEM
HFelectr.
standarddecoder
PICNICenhancement
picturecontrol
audiodemod.
Audioproces.
cableantenna
PAL/NTSC
display
YUV1fh
YUV2fh
RGB2fh
picturecontrol
standarddecoder
PiP+
Mem RGB2fh
CVBS
MEM
100HzFALCON
IC
MEM
MEM
NICAMdecod.demod
MPEGaudio
MPEGvideo
MEM
chan.decod.
transp.demux
channelbits
Traditional TVs, Set-Top Boxes:
• Fixed algorithms for fixed HW architectures
• Upgrade for new services/applications is problematic
• Systems are not flexible
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
8
Media processing in software
• Based on programmable platforms• Advantages:
– Upgrading;– Fast time-to-market of new features;– Scalable video applications → product families.
• Boundary conditions:– Cost-effectiveness;– Preserve existing qualities.
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
9
Media processing in software
• A Digital Video Platform• Scalable video applications• Challenges & approaches
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
10
A Digital Video Platform
Control + Services
Media processing
Bus
VLIWCPU
MIPSCPU
Memory
Coprocessors
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
11
HW Architectures vs SW Applications
Low-
end
Mid-
range
High-
end
Algorithm1 m
in max
Algorithm3
Algorithm2
Algorithm4
Resources
ProductFamilies
SW-Modules
Algorithm1
Algorithm2
Algorithm4
Algorithm3
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
12
Traditional vs Scalable Approach
Cost
Functionalitytarget limit
traditionalsystems
scalableapproach
Functionality
Quality
target limit
traditionalsystems
scalableapproach
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
13
Example: One High-Quality Application
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
14
HQ Video-Conference, Reduced Background Quality
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
15
Quality Levels in Algorithms
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
16
Scalable Algorithm
FUNCTION 1 FUNCTION 2
FUNCTION 4
FUNCTION 3
QUALITY CONTROL
control signal for
quality level
signal in
signal out
ALGORITHM FOR MEDIA PROCESSING
Quality
Resource needs
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
17
SVAs: Output Quality Mapping
Bandwidth CPU cyclesMemoryCo-processor cycles
Resources
Quality Levels1l 2l 3l 4l
13R
23R
12R
11R
14R
1l 2l 3l 4lQuality Levels
Output quality
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
18
Challenges & approaches• Maximize system utility, given available resources:
– Elements (semantically neutral):• Utility of individual applications;• Resource usage;• Relative importance.
– Approach:• static: [Lee et al 99].• dynamic:
– Applications: local quality control;– System: global system utility control.
• Resource sharing:– Interference between applications
• Predictability;• System robustness
– Approach: resource reservation.
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
19
Challenges & approaches
• Mode changes:– Examples:
• System:– Start of new application: main → main + pip;– Change of user focus: main → pip;– Exchange of main and pip.
• Application:– movie (24 Hz film) to commercial (50 Hz camera);– Channel switching.
– Smooth transitions required– Approach: mode-change protocol(s).
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
20
Example: Main → Main + PiP
dmux
audiodec.
dec. sharpenh.main: scalable
DVD
mixer
mixer :non-scalable
audiorend
read
scaler
digitizer: non-scalable
digit
pip: scalable
(1) Initial: Main
(2) Intermediate: reduce quality
(3) Intermediate: start PiP(4) Final: Main + PiP
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
21
Mode change protocol
User focus change: Main →PiP
main
PiPQuality
Qualitydecrease
main
PiPBudget
timeBudget
configuration change
Qualityincrease
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
22
Example: Video enhancement modesVideo Enhancement - Main Window
Video Enhancement - PiP
AnArtRAnArtR
1D scaling
1D scaling
Error Conceal.
Error Conceal. 2D Scaling
2D Scaling Color / Gamma
Color / Gamma
DigArtRDigArtR
Scan-RateConvers.
Scan-RateConvers.
DigArtRDigArtR
2D Scaling
2D ScalingError
Conceal.
ErrorConceal.
DigArtRDigArtR
480p 16:9
Scan-RateConvers.
Scan-RateConvers.
Scan-RateConvers.
Scan-RateConvers. Color /
Gamma
Color / Gamma
Color / Gamma
Color / Gamma
ATSC 720p/16:9
ATSC 1080i/16:9
NTSC 480i/4:3
DVD 480p/16:9
2D scaling
2D scalingError
Conceal.
ErrorConceal.
2Dscaling
2Dscaling
Error Conceal.
Error Conceal. 2D Scaling
2D Scaling 2D
Scaling
2D Scaling
2Dscaling
2Dscaling
PiPColor / Gamma
Color / Gamma
Color / Gamma
Color / Gamma
Color / Gamma
Color / Gamma
2D Scaling
2D Scaling
NTSC 480i/4:3ATSC 1080i/16:9ATSC 720p/16:9DVD 480p/16:9
Note: Number and order of algorithms change for Main Window upon input change !
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
23
Example: Video enhancement modesInitial situation
AnArtRAnArtR
NTSC 480i/4:3 Main
PiPError Conceal.
Error Conceal. 2D Scaling
2D Scaling 2D
Scaling
2D Scaling
Color / Gamma
Color / Gamma
ATSC 1080i/16:9
Scan-RateConvers.
Scan-RateConvers. 1D
scaling
1D scaling
Exchange of Main and PiP
Final situation
Error Conceal.
Error Conceal. 2D Scaling
2D Scaling Color / Gamma
Color / Gamma
DigArtRDigArtR
Scan-RateConvers.
Scan-RateConvers.
ATSC 1080i/16:9
2Dscaling
2DscalingNTSC
480i/4:3PiP
Main
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
24
Challenges: dynamic load
time
load
structural load
running averagetemporal load
“worst-case” load
Close-to-averageresource allocation
MPEG decoding of DVD stream
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
25
A co-operative QoS approach
• Organization (multi-disciplinary):– Video domain experts;– System software specialists;– Mathematicians.
• System:– Adaptive applications:
• Scalable video applications;• Local quality control.
– QoS-based resource management (QRM):• Resource reservations;• Global system utility control.
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
26
Co-operative QoS approach: system
Adaptive applications
Provide quality levels + estimated resource req.
Resource manager Provides guaranteed resource budgets
Local quality control
SVAs…
Global system utility controlOptimizes system utility, sets quality levels + allocates resources
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
27
QoS approach: local quality control
• Synchronous (3D graphics):– Estimate `work’ for next frame and set operational
quality level according to available resources.• Asynchronous (high quality video):
– Through buffering and work-ahead;– Progress (i.e. amount of budget [= virtual time] till
deadline);– Quality selection based on penalties & reward:
• Deadline miss (−);• Quality level fluctuation (−);• Quality level (+).
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
28
Co-operative QoS approach• Shared responsibilities:
– QRM: enabling;– Applications: executing.
• Cost-effectiveness:– QRM:
• resource allocation according to relevance;• Low system overhead;• Close to average-case resource allocation.
– Applications: effective and efficient media processing.• Robustness:
– QRM: resource reservations;– Applications: “get by” with budget.
• Seamless switching between modes:– QRM: mode-change protocol;– Applications: smooth transitions.
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
29
A retrospect on real-time issues• Real-time is a QoS parameter !
– Deadline miss: a penalty for local quality control.• Modes of a system:
– Number of applications (main, pip, disk);– Modes of applications (input, output);– Quality settings and budgets of applications.
• Hierarchies:– Control:
• Global system utility control;• Local quality control (of an RCE).
– Scheduling:• Periodic budgets;• Asynchronous tasks (sharing a budget).
Reinder J. Bril, [email protected]/e Informatica, System Architecture and Networking
30
References• [Bril et al 01] R.J. Bril, C. Hentschel, E.F.M. Steffens, M.
Gabrani, G.C. van Loo, and J.H.A. Gelissen, Multimedia QoS in consumer terminals (invited lecture), In: Proc. IEEE Workshop on Signal Processing Systems (SIPS), pp. 332 – 343, September 2001.
• [Lee et al 99] C. Lee, J. Lehoczky, R. Rajkumar, and D. Siewiorek, On quality of service optimization with discrete QoS options, In: proc. 5th IEEE Real-Time Technology and Applications Symposium (RTAS), pp. 276 – 286, 1999.
• [Steffens et al 03] L. Steffens, S. van Loo, and C. Otero Perez, Trends in Operating Systems – Resource Management for Future CE Systems (invited lecture), ESTIMEDIA Workshop, http://www.idt.mdh.se/estimedia/steffens.pdf, October 2003.
• [Wüst et al 04] C.C. Wüst and W.F.J. Verhaegh, Quality control for Scalable Media Processing Applications, Journal of Scheduling, 7(2): 105-117, March 2004.
http://www.idt.mdh.se/estimedia/steffens.pdf
Real-Time Architectures2003/2004A real-time architecture for MCTsMultimedia Consumer TerminalsTrendsCharacteristicsCharacteristicsTraditional High-End TV ArchitectureMedia processing in softwareMedia processing in softwareA Digital Video PlatformHW Architectures vs SW ApplicationsTraditional vs Scalable ApproachExample: One High-Quality ApplicationHQ Video-Conference, Reduced Background QualityQuality Levels in AlgorithmsScalable AlgorithmSVAs: Output Quality MappingChallenges & approachesChallenges & approachesExample: Main Main + PiPMode change protocolExample: Video enhancement modesExample: Video enhancement modesChallenges: dynamic loadA co-operative QoS approachCo-operative QoS approach: systemQoS approach: local quality controlCo-operative QoS approachA retrospect on real-time issuesReferences