Beruflich Dokumente
Kultur Dokumente
Video Streaming
2 R&D results
OLYMPIC project architecture
Video QoS techniques developed in the project.
Evaluation of developed solutions.
2
Heterogeneity
in Multicast Video Streaming
Network Heterogeneity
Networks have different resources (bitrate,
buffer capacity, management policies).
Receiver Heterogeneity
Receivers have different visual quality
requirements (screen dimensions /spatial
resolution, colours) and different processing
capabilities.
Consequently, different users could experience different
video quality reception.
4
255
PSNR = 20 log10
MSE
f ( i, j ) F ( i, j )
MSE =
N2
IntServ
DiffServ
Rate control
Window-based control (TCP, DCCPCCID2)
Rate-based control
- Source-based rate control (RTCP,
DCCP-CCID3)
- Receiver-based rate control
- Hybrid rate control
Rate shaping
Video frame Selective discard
Dynamic rate shaping
Selective Discard of
Video frames
Compression perspective
RTCP, UDP,
TCP,
Application
(A)
Transport
(T)
DCCP
IntServ,
DiffServ
Network
(N)
Rate shaping
Rate Adaptive encoding
2 Research results
OLYMPIC project architecture
Transmission of multimedia sport events (Audio/
Video) over IP in large scale, with large
heterogeneity of access network and terminals
Core
Network
Media Encoder/
Media Server
Core
SAS
Unicast/
Multicast
Access
Network
Peripheral
SAS
Unicast/
Multicast
Clients
7
Network
(IP)
Application
Application
Transport
UDP
UDP
TCP
DCCP
Discard
point
Router DS
Server
Server
Server
Discard
based on
DSCP/AF
RTCP
Sender
buffer
Sender
buffer
Discard layer
Transport
Protocol
Transport
layer
UDP
Network
layer
(MPEG4 DSCP)
Architecture of solution A
Selective Discard based on DiffServ
no drop
Access
Network
Core
Network
Encoder/
Media Server
QoS
Marker
Unicast/
Multicast
SAS
(QoS Marker)
drop B
Unicast/
Multicast
drop B+P
Tipo de Trama/
VOP MPEG4
Clientes
Classe AF
Diffserv AF
30,0
PSNR [dB]
25,0
20,0
15,0
10,0
5,0
0,0
0,0
5,0
10,0
15,0
20,0
PLR [%]
Conclusions of Solution A
13
(MPEG4 QOS-type)
Selective Dropping
RTCP RR
Buffer
Transport
layer
Network
layer
UDP
IP
(Best Effort)
Sent frames
RTCP feedback
from clients
Transport Layer
Discarded
frames
Outgoing Stream
GOV
Index i
I2
Pn
Pn-1
Pi
Pi-1
Discarded VOPs
P3
P2
P1
I1
first
Sent VOPs
15
Selective discard
30,0
PSNR [dB]
25,0
20,0
15,0
10,0
5,0
0,0
0,0
5,0
10,0
15,0
20,0
25,0
PLR [%]
16
Random discard
Selective discard
17
Conclusions of solution B
(Selective Discard based on RTCP)
Selective Discard based on RTCP presents high PSNR
gain in comparison with random discard
Allow dynamic adjust of discard level, optimizing the
available bandwidth;
The delay of RTCP loss reports (typical 5s) can be a
problem in case the network presents frequent variations of
congestion level.
18
Application
(MPEG4 QOS-type)
layer
Buffer
Selective Dropping
Transport
TCP
layer
Network
Buffer
IP
(Best Effort)
layer
Incoming
Stream
Transport Layer
Selective Discard
Algorithm
+
Application Buffer
Outgoing
Stream
TCP
buffer
Frames sent
Frames discarded
I9
Pn
P3
P2
P1
I8
Pn
P2
P1
Pn
P3
P2
P1
Discard
10
30,0
PSNR [dB]
25,0
20,0
15,0
10,0
5,0
0,0
0,0
5,0
10,0
15,0
20,0
25,0
PLR [%]
Conclusions of Solution C
(Selective Discard based on TCP)
Selective Discard based on TCP presents
significant PSNR gains in comparison with
random discard;
This solution can be useful when reliability is a
requirement (p. ex. Transcoder)
As this solution is TCP based, it can have high
delay in case of congestion.
The bitrate could increase due to TCP packet
retransmission.
22
11
layer
(MPEG4 QoS-type)
Transport
Selective Dropping
DCCP
layer
Model Implemented
MPEG4 QoS Classifier
(MPEG4 QoS-type)
Selective Dropping
DCCP
Buffer
Buffer
UDP
Network
IP
(Best Effort)
layer
IP
(Best Effort)
I
9
P
n
P
n
P
2
P
n
P
i
+
1
P
i
DCCP
buffer
Discard
24
12
PSNR [dB]
PSNR [dB]
10
15
20
25
30
35
26
25
24
23
22
21
20
19
18
17
16
15
14
13
40
10
15
20
25
30
35
40
Stefan
Akyio
25
1.8
1.6
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0
0
10
15
20
25
30
35
40
10
15
20
25
30
35
Akyio
40
Stefan
26
13
Conclusions of Solution D
(Selective Discard based on DCCP)
Selective Discard based on DCCP presents high
PSNR gain in comparison with random discard;
This solution presents the lowest delay, due to the
single buffer at sender (at transport layer)
Presents higher discard flexibility since it allow
frame discard at transport layer.
DCCP not deployed yet at large scale
27
Good results, works better with big buffers, but presents high end-toend delay.
14