The Role of Network Packet Loss Modeling in Reliable

The Role of Network Packet Loss Modeling in Reliable
The Role of Network Packet Loss
Modeling in Reliable Transport
of Broadcast Audio
April 14, 2015
NAB Show 2015
Featuring
GatesAir’s
Junius Kim
Hardware Engineer
Keyur Parikh
Architect / Software Lead
Copyright © 2015 GatesAir, Inc. All rights reserved.
The Role of Network Packet Loss Modeling
in Reliable Transport of Broadcast Audio
Junius Kim and Keyur Parikh
GatesAir
Mason, OH
Overview
Encoder
•
•
•
•
•
Network Impairments
Packet Loss Modeling
Packet Loss Analysis
Packet Loss Simulation
Packet Loss Mitigation
Audio
In
RTP Data +
FEC Packets
Network
Audio
Out
Decoder
AoIP Application
Network Impairments
• Jitter
• Out-of-Order Packets
• Duplicate Packets
Well defined solution for above impairments…
• Packet Loss
More difficult problem to solve…
Packet Loss
• Causes of IP packet loss: route flapping, transmission errors,
congestion
• Unmanaged vs. managed network services
• Packet loss concealment methods: energy interpolation, noise
substitution, replaying previous frame
• Concealment works well at very low packet losses
• Need to use correction techniques along with concealment for
higher level packet losses
• Correction techniques are based on standard RTP over UDP
protocol
Packet Loss Patterns
• Random vs. Burst Packet Loss
• Random Losses
• Uncorrelated
• Appear to be spread out
• Burst Losses
• Correlated
Random
Burst + Random
Four State Markov Model
• Multi-state model
• Transition between states with a transition probability
• 4-state model represents burst periods, during which packets are
received and lost according to a first 2-state model and gap
periods during which packets are received and lost according to a
second 2-state model
Packet Loss Model
Four-State Markov Model
Four-State Markov Model
State 1 - Packet is successfully received
State 2 - Packet is received within a burst
State 3 - Packet is lost within a burst
State 4 - Isolated packet lost within a gap
For example, using the loss pattern:
0000011001010101101100000000100
where 1 represents a lost packet and 0 is a good
received packet, correlates to the state pattern of…
1111133223232323323311111111411
Burst Packet Loss Model
•
•
•
•
•
•
•
Gap state: isolated or random losses
Burst state: correlated packet losses in burst
Density: probability of loss
Period/Length: time within a state
Gap Length and Gap Density
Burst Length and Burst Density
Objective of modeling is to characterize and fit “real world”
network behavior
RFC 3611
• RFC 3611 defines an Extended Report (XR) packet type for the
RTP Control Protocol (RTCP)
• RTP Control Protocol (RTCP) is a companion protocol of the
Real-time Transport Protocol (RTP)
• XR supplements the reports in RTCP
• Intended for VoIP applications for monitoring of performance
metrics
RTCP
• RTCP specifies report PDUs exchanged
between sources and destinations of
multimedia information
– Receiver reception report
– Sender report
• Reports contain statistics such as the number
of RTP-PDUs sent, packet jitter, packet loss,
round trip delay
• Used to provide feedback on QoS by
periodically sending statistics
RFC 3611 Metrics
The VoIP Metrics payload is shown below:
Network Packet
Loss
Packets discarded
due to jitter
Average duration of burst periods (mS)
RTP Round Trip Delay (mS)
Signal Level (dBm)
Noise Level (dBm)
R factor
External R factor
PLC and Jitter Buffer Reserved
Config
Current max jitter buffer delay (mS)
Density of lost/
Density of lost/
discarded packets in discarded packets in
burst periods
gap periods
Average duration of gap periods (mS)
End System Delay (mS)
Residual Echo
Gmin - typically 16,
Return Loss (dB)
which classifies >5%
loss as burst
MOS LQ
MOS CQ
Average jitter buffer delay (mS)
Max jitter buffer size (mS)
Packet Loss Simulation
• Linux module netem for network
emulation
• Simulation of packet delay, drops,
duplicates, corrupted packets, and lost
packets
• Linux router
• Usage of netem for WAN emulation
• Evaluation of performance of AoIP in a
lab environment
AoIP Analysis Tool
AoIP Codec
IP Network
IP Network
Packet
Receiver
AoIP Stream
Processing
Input
Audio
Decode
Output
Data
Collection
Network Analyzer
Random Packet Loss
Packet Loss Rate
Gap Density
Burst Density
Burst Packet Loss
Packet Loss Rate
Burst Density
Burst Size
Network Analytics Flow
Loss Rate, Burst and Gap Density,
Burst and Gap Size
Packet Loss
Modeling
Decision
Tree
Recommend
FEC if Losses
Random
Time Delayed
streams
Calculate
Time Delay
Recommend
multiple
Networks If
excessive losses
Time Delay
Value
Report Generation
Report Generation
Stream Selection
Mitigation of Packet Loss
• Mitigation methods
• Forward Error Correction (FEC)
• Interleaving
• Redundant streaming
• Network diversity
• Mitigation of random packet loss
• Mitigation of burst packet loss
RTP Forward Error Correction (FEC)
• FEC packets are generated from a matrix of RTP data
packets
• Both data and FEC packets are sent to the receiver
• FEC attempts recovery of lost data packets at the receiver
• Unrecovered packets are considered lost and concealment
is applied
• Effectiveness of recovery depends on the packet loss model
FEC Matrix
Col 1
Col 2
Col 3
Col 4
Row 1
1
2
3
4
XOR(1,2,3,4)
Row 2
5
6
7
8
XOR(5,6,7,8)
Row 3
9
10
11
12
XOR(9,
10,11,12)
Row 4
13
14
15
16
XOR(13,
14,15,16)
XOR(1,5,9,13 XOR(2,6,
)
10,14)
XOR(3,7,
11,15)
XOR(4,8,
FEC(x)
12,16)
FEC(x)
FEC Correction Capability for Random
Loss
FEC Correction vs. Burst Density
Average Packet Loss = 1%, Burst State Length= 16 packets
0.8
0.7
0.6
0.5
E
P
L 0.4
,
% 0.3
4x6
4x4
3x3
2x2
0.2
0.1
0.0
10
20
30
40
50
Burst Density, %
60
70
80
FEC Performance for 5% Random Loss
No FEC
Router Configuration:
5% Random Loss
25% FEC
50% FEC
Packet Interleaving
Redundant Streaming
• Two time diverse streams
• Time diversity value is set based on receiver’s calculation of
burst length
1% Avg PL, 80% Burst
Density Network Loss
Two Time Diverse streams:
400 msec. EPL 0.07
Network Diversity
• Streams of the group are split across multiple diverse
networks
• Provides “hitless” protection against failure of any single
network
• Provides higher level of packet loss protection due to
uncorrelated network paths
Loss Distribution
Two systems connected via Time Warner at home and Verizon in the lab
400
350
Isolated Losses
P
a 300
c
k 250
e
t
200
s
Burst Losses
wY2dX5hB2
150
L
o
s 100
t
50
0
3/12/2014
3/13/2014
3/13/2014
3/13/2014
3/13/2014
3/13/2014
3/14/2014
3/14/2014
3/14/2014
3/14/2014
3/14/2014
AoIP Case Study
• Two streams with time diversity
• Stream 1 time offset = 0, with 4x4 FEC
• Stream 2 time offset = 1.25 secs
1.6
Loss Rate(%)
1.4
Loss Rate After Correction(%)
L 1.2
o
s 1
s
wY2dX5hB2
0.8
R
a 0.6
t
e 0.4
0.2
0
3/12/2014
3/13/2014
3/13/2014
3/13/2014
3/13/2014
3/13/2014
3/14/2014
3/14/2014
3/14/2014
3/14/2014
3/14/2014
Summary
•
•
•
•
•
•
•
Network impairments: jitter, duplicate, out-of-order, and lost packets
Real world packet loss tends to occur in bursts
Packet loss modelling can be used to characterize network behaviour
Usage of an analytics tool to measure and characterize packet loss in
an AoIP application
Packet loss can be mitigated using FEC, interleaving, stream and
network diversity
Mitigation methods have bandwidth, delay, and network resources
trade-offs
An optimized mitigation strategy can be deployed based upon your
network characteristics
Thank You
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertising