IPTV QoE:
Challenges for the IP network
End-to-End QoE/QoS Workshop. Geneva, 14-16 June
g
g
g
g
IPTV QoE framework and requirements
IP Transport requirements
IP backbone and access loss patterns
Error recovery
PG. 2
A Glance to Fastweb IPTV Services
g
g
g
Broadcast TV and VoD service offered commercially since 2001 (since
2003 over ADSL)
About 100 TV Channels (the infrastructure supports many more)
More than 5000 contents archived and available through VoD portals
g
g
g
g
g
g
g
Not only movies: fictions, news, music, documentaries
Network PVR
Video content (VoD or TV) up to 4 Mbps MPEG2 streams
Digital Dolby 5.1
TV: centralized multicast transport, VoD: unicast regionalised transport
180.000 Video customers
One of the largest IPTV operators in the world
PG. 3
QoE
QoE and QoS
Quality of Experience (QoE) describes the performance of a device, system,
service, or application (or any combination thereof) from the perspective of the
user.
QoE measures how well a network service satisfies the user’s expectations and
needs
•
QoS
•
•
Quality of Service (QoS) refers to a set of technologies (QoS mechanisms) that
enable the network administrator to manage the effects of congestion on
application performance as well as providing differentiated service to selected
network traffic flows or to selected users.
QoS mechanisms do not create bandwidth but instead manage available
bandwidth more efficiently, especially during peak congestion periods
QoS is a measure of performance at the packet level from the
network perspective. QoE is a measure of end-to-end
performance at the services level from the user perspective. PG. 4
Courtesy of: Nortel, Tim Rahrer
Video Service QoE Engineering
This is where
quality counts!
In-Home
Access
Infrastructure
Transport,
Voice, and Core
Content & Video
Headend
Voice Services
Local
Content
VOD /
Middleware
Content
Management
Digital
Content
Internet
g
g
Define the variables contributing to satisfactory service QoE – transport and video application
layers
Model end-to-end network to determine impacts on service quality
g
g
g
Model tradeoffs in service features with network capabilities
Define network performance requirements to achieve target QoE
Define measurement methods to verify QoE
Need a complete end-to-end view and user needs to ensure network
architecture and service success
PG. 5
Courtesy of: Nortel, Tim Rahrer
QoE and IP network requirements
g
A subset of user experience requirements for SD IPTV:
g
g
g
g
g
Same video quality as the user is normally used to through other means of distribution
(Satellite, terrestrial, cable)
High availability and continuity of the service
Very low visible impairment (error) rate – how low?
Channel Change time comparable to what the user is used to
These requirements will translate (also) into requirements for the network and IP
transport layers:
g
g
g
Minimum IP end-to-end bandwidth available for IPTV (Ex: 4 Mbps for SD)
IP QoS support to handle congestion
End-to-end IP Transport performance objectives for:
g Packet Loss
g Delay
g Delay variation
Based on field experience:
TV users may not complain or open a ticket for each error they notice… but they always compare what
they get through IP/DSL with traditional broadcasting service, and based on this may decide whether
to subscribe or confirm subscription:
g
g
Input in the network good enough quality video
Preserve it throughout all the IP delivery chain
PG. 6
IP Transport performance objectives
Delay
By far the most important parameter that affects the IPTV service quality is the IP packet
loss, indeed:
g
The delay introduced by a well performing IP network (say up to hundreds of milliseconds
for a huge geographical network) is usually negligible for the IPTV service considering
that:
g
g
g
g
IPTV is still mainly a unidirectional service with some loose interactivity requirements
(channel change, video portal and content access, pause, play, etc. for VoD)
Most of the delay from the live content will be added from the Head-end and video
acquisition/coding/transcoding chain
Current satellite broadcast TV service already may insert a delay up to a few seconds
Delay in Channel Change time due to network transfer delay signalling processing (IGMP
Leave/Join signalling and first packet reception of the requested channel for multicast
service) is negligible (up to few tenths of milliseconds) when compared to the time needed
to the STB to buffer enough video content for a smooth start for the play out .This time is
usually comparable to a GOP interval which for 4 Mbps MPEG2 video streams is about half
a second
PG. 7
IP Transport performance objectives
Channel change - Delay variation
Channel Change:
g
User requests
new channel
First packet of the
requested channel
received by the STB
-
STB processes request - IGMP Leave - IGMP Report -
-
IGMP Processing in L2 nodes (DSLAM, BRAS, BNG) -
I frame received,
STB decodes
TV flow queuing – DSL Interleaving – Transmission ETC
g
Delay Variation:
g
Because of the presence of a quite large play-out/de-jitter buffer at the receiving side (STB), all potential
jitter introduced by a well performing IP network is removed for a smooth play-out and virtually no
packets is lost due to late arrival.
PG. 8
IP Transport performance objectives
Packet Loss
g
IPTV service is highly sensitive to packet loss
g
Although the impact of a single packet loss depends of several factors such as:
g
g
g
g
g
Compression algorithm (MPEG2, H.264, etc.)
g
GOP Structure
g
Type of information lost (I,P,B frame, other MPEG information, etc)
g
Codec performance (encoding and decoding)
g
Complexity of the video content
g
Error concealment at the STB
It is highly likely that a single IP packet loss produce a visible impairment to the
user – Note that a single IP packet usually transports 7 188 Byte MPEG packets
Packet loss has to be minimized and strictly controlled:
Simple metrics as average Packet Loss Ratio are not enough to describe the packet loss
requirement
Æ what matters is the loss event, its rate (MTBE) and shape (loss period and distance), not
only the ratio lost_packets/received_packets.
PG. 9
IP Transport performance objectives
Packet Loss - 2
g
As an example:
4 Mbps SD stream
(MPEG2)
Æ 380 pps
12 Mbps HD stream
(H.264)
Æ 1140 pps
Quality
Single packet loss in 30
minutes
PLR= 1.46*10-6
PLR= 4.87*10-7
Likely Good
Single loss event of 10
lost packets in 30
minutes
PLR= 1.46*10-5
PLR= 4.87*10-6
10 single packet loss
events in 30 minutes
PLR= 1.46*10-5
Ideal target: 0 loss.
Typical target:
Likely Good
Loss distance:
Loss period:
Severe Loss distance:
Severe Loss period:
PLR= 4.87*10-6
Likely noticed by the user
1 per 30 minutes
few IP packets
1 per days/week
> TBD
Note that the rate of error events is more impacting the user experience than the weight of a single
event (at least for up to short errored events)
PG. 10
Input from Subject Video Quality
Testing
g
g
g
Subjective Tests on IPTV SD Service
One minute video sequences with sport (soccer) content
Different packet loss profiles introduced (taking into account ADSL
access loss profiles)
g
g
g
MPEG-2 4 Mbps streams
Very likely that a single packet loss is perceived by the user:
g
g
Single losses, burst losses, at different rates
On a separate ‘expert viewing’ analysis, only about 10% of single
packet losses event didn’t produce any visible artifacts (out of few
hundreds cases)
STB used for the testing does perform some error concealment
PG. 11
Video subjective testing - results
10
9
SINGLE Packet losses
8
15 packets lost, MOS: 2 to 5.5
Note: X errori singoli,
read X single packet loss events
7
10 packets lost, MOS: 3 to 6
6
5
4
3
2
1
0
1 errore singolo
2 errori singoli
4 errori singoli
6 errori singoli
10 errori singoli
15 errori singoli
25 errori singoli
10
9
Burst Packet losses
8
1 burst of 2,3,5,8 lost packets
2 bursts of 2,5 lost packets
7
3 bursts of 3,5 lost pckts
6
5
4
3
2
1
0
1 burs t da 2
1 burs t da 3
1 burs t da 5
1 burs t da 8
2 burs t da 2
2 burs t da 5
3 burs t da 3
3 burs t da 5
PG. 12
Backbone and Access contribution to Packet Loss
Aggregated View
Backbone and access contribution to Packet Loss
25
% of measurement samples with
packet loss > 0
Any backbone failure
contribution (fiber cut,
device fault, etc)
20
15
ADSL access
contribution
10
one day
5
0
Time
PG. 13
Access contribution to Packet Loss
Loop length, QoE and Service penetration
Adsl Contribution to packet loss
MTBEM vs loop attenuation - ADSL access
MTBEM
7000
Bitrate
100
Enough bandw idth to deliver SD IPTV and Triple play services
Minutes
90
System design goal: increase the MTBE for
long loops and increase the Service
penetation for a given QoE requirement.
Error Correction strategies allow to reach this
objective:
- Phisycal Layer (Interleaving + RS)
- Application Layer
- Other approaches (ex. retransmission for VoD
unicast services)
Loop Qualification and control
80
MTBEM = Mean
Time Between
Errored Minutes.
70
60
An Errored Minute is
a minute with at least
one packet loss
50
6000
5000
kbps
4000
3000
40
Example of IPTV QoE Target
30
2000
20
1000
10
0
23
-24
22
-23
21
-22
20
-21
19
-20
18
-19
17
-18
16
-17
15
-16
14
-15
13
-14
12
-13
11
-12
10
-11
9-1
0
8-9
7-8
6-7
5-6
4-5
3-4
2-3
1-2
0
Upstream attenuation - dB
PG. 14
Access contribution to Packet Loss
Single line with High attenuation
ADSL Error correction
39 db Att DW 23 Db Att UP
occurrencies of lost packets per minute detail
16
0 lost packets / min ~
98% with interleaving
12
10
MTBEM changes from 3 to
50 Minutes with ADSL
interleaved path (16 msecs)
g
8
6
4
2
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
0
0
% of total minutes
14
no intl
intl
lost ppm
Error recovery for IPTV
g
Goal
g
g
Extending the reach of the IPTV service for ADSL/ADSL2+ access
Meet QoE requirement
g
No current recognised Best Practice or solution wide spread in the industry
g
Possible options:
g
DSL Physical Layer (Interleaving, RS, ADSL2/2+ improvements)
g
g
Multiple bearer channels capabilities from ADSL2+ standard
Application Layer FEC
MPEG CoP3, Other codes, (DVB IPI is evaluating different alternatives)
g Need to be supported on the Service end points: Head end, STB
g Currently no or limited availability in the industry
g
g
Possible use of both of a PHY Layer and Application Layer FEC
g
g
Care must be used: two non linear processes
Retransmission
g
Likely feasible for Unicast services
Measure and study carefully packet loss patterns in the network to choose the
best approach, dimension the best error correction strategy and optimize
investments
PG. 16
Some clues for network engineering
to improve IPTV QoE
g
Backbone
g
g
g
g
Redundancy
High Availability
Minimize routing convergence time
IP QoS-enabled to handle possible congestions
g
g
Note that for an IP triple play broadband operator might be challenging to always guarantee that every
link will transport less than the 50% of its capacity (to avoid congestion in case of failure of a redundant
link).
Access
g
QoS in bottlenecks points:
At the Residential Gateway for upstream traffic
g At the DSLAM and/or BRAS/BNG for downstream traffic
g
g
g
g
g
g
Error recovery design (might span across the whole network in case of an Application Layer
error correction or retransmission)
DSL Line qualification and monitoring
Understand user requirements, sometimes it is also useful to save money!
Use in service, passive, end-to-end measurements to assess service
performance
DSL Forum WT-126 Provides QoE performance guidelines for IPTV services.
PG. 17
Thank you
riccardo.fiandra@fastweb.it
PG. 18