Adaptive Playout Scheduling in Packet Voice Communications

1
Adaptive Playout Scheduling in Packet Voice
Communications
2
Outlines

• QoS concerns and tradeoffs
• The jitter adaptation scheme as a playout
  scheduling algorithm
• Setting the playout schedule
• Packet scaling and loss concealment
• Results of listening tests and performance
  comparison

3
QoS Concerns at the Receiver
Delay Jitter
Obstructs proper reconstruction of voice
packets at the receiver
Impairs interactivity of conversations
Delay
Packet Loss
Impairs speech quality
4
Different Scheduling Algorithms (1)
5
QoS Tradeoffs (1)
Playout Jitter
Delay
Packet Loss
6
Jitter Absorption vs. Jitter Adaptation
Jitter Absorption
7
Different Scheduling Algorithms (2)

1. Method which uses fixed playout time throughout
   the whole session
2. Method which estimates delay dynamically but only
   adjusts playout time during silence periods
3. Method which dynamically estimates and adjusts
   playout time, and scales packets within
   talkspurts using time-scale modification.

8
QoS Tradeoffs (2)
Playout Jitter
Delay
Packet Loss
9
Problems

• How to set the playout schedule?
• How to scale the packet?
• How does playout jitter affect audio quality?

10
Setting the Playout Schedule
11
Packet Scaling
12
The Jitter Adaptation Procedure
13
Jitter Adaptation
14
Loss Concealment
15
QoS Tradeoffs (3)
Playout Jitter
Delay
Packet Loss
16
Three Algorithms under Comparison

1. Method which uses fixed playout time throughout
   the whole session
2. Method which estimates delay dynamically but only
   adjusts playout time during silence periods
3. Method which dynamically estimates and adjusts
   playout time, and scales packets within
   talkspurts using time-scale modification.

17
Performance Comparison

• Traces measured
• between a host at
• Stanford and hosts
• in
• Chicago
• Germany
• MIT
• China

18
Test on the Quality of Scaled Audio (1)

• Three short network traces with different jitter
  statistics
• Six samples are simulated over each trace
  Samples are given and evaluated in pairs, in the
  format of reference sample processed sample
  to provide higher sensitivity
• Eighteen people participated the tests
• Score for each condition is obtained by
  averaging 6x18 ratings

19
Test on the Quality of Scaled Audio (2)
5 Degradation is inaudible                        
  4 Degradation is audible but not annoying 3
Degradation is slightly annoying               2
Degradation is annoying                         
1 Degradation is very annoying                 
20
Quality of Speech Processed by Different
Algorithms (1)

• Four network traces with different jitter/loss
  statistics
• Four samples are simulated over each trace using
  two different playout scheduling algorithms (4x2
  conditions)
• Score for each condition is obtained by
  averaging all the ratings

21
Quality of Speech Processed by Different
Algorithms (2)
MNRU (dB) 10 18 24 30 Source
MOS 1.4 2.7 3.6 4.1 4.4
Quality      Score Excellent      5
Good           4 Fair             3
Poor            2 Bad             1
22
Conclusions
Being the results of the new tradeoffs

• Small jitter can be traded for lower delay and
  lower loss rate
• Infrequent packet scaling does hardly impair
  sound quality
• Using adaptive scheduling, both delay and loss
  rate can be significantly reduced improved
  overall performance