Maximizing Audio Quality A Linear Programming Problem

1
Maximizing Audio Quality - A Linear Programming
Problem

• CSC591 Topics in Algorithms
• Winter 2003
• Final Project
• Lopa Roychoudhuri

2
Problems

• Current Internet still provides best-effort
  service
• no guarantee of performance for real-time
  multimedia applications
• Multimedia typically uses UDP
• not reliable
• no congestion control
• Multimedia traffic is normally subject to
• Restricted available bandwidth
• delay, delay jitter
• loss of packets

3
Control Mechanisms

• Mechanisms which
• Dynamically adapt the behavior of the audio
  application to maximize the audio quality under
  the constraints of
• restricted bandwidth
• Delay
• packet loss
• jitter
• present in the network at that point in time

4
Audio Compression Techniques (codecs)

• Current codecs have a diverse range in terms of
  degrees of compression (bitrate) and underlying
  technologies
• Thus the quality of an IP telephony call is
  highly dependent on the codecs and their reaction
  to available bandwidth, link delays and packet
  loss

5
Mean Opinion Score (MOS)

• Described in ITU recommendation P.800
• Formal subjective measure of voice quality
• real number between 1 and 5
• Toll quality - quality with MOS between 4 4.3
• Communication quality - between 3.5 and 4
• lower bound for acceptability of a speech 3.5
• MOS has been determined for every codec under the
  ideal conditions of no loss

6
Examples of codecs
7
Bandwidth constraint

• End-to-end available bandwidth - the maximum rate
  that the path can provide to a flow
• depends on the utilization of various links in
  the path in presence of cross-traffic
• Less than or equal to capacity of the path the
  maximum rate a path can provide to a flow, when
  there is no other traffic in the path

8
Bandwidth constraint cont.

• In underutilized network we can use high bitrate
  codecs
• which will consume more bandwidth
• but will guarantee higher quality
• But switch to low bitrate codecs when available
  bandwidth gets tighter
• It is possible to mix multiple codecs in a
  certain ratio for bandwidth optimization ensuring
  that the audio quality provided is optimum for
  the user 

9
Delay constraint

• Delay of the path
• Propagation delay of individual links
• Queueing delay at individual hops/routers
• Delay inherent to the codec
• Tcod Tenc Tdec TLA
• Total delay must be under the constraint of
  tolerable Mouth-to-ear (M2E) delay
• the time that elapses between the moment the
  talker utters the words and the moment the
  listener hears them
• Must be under 400ms (ITU recommendation G.114
  G.131)

10
The LP Problem

• Maximize the audio quality under the constraint
  of available bandwidth and link delay
• Maximize c1x1 c2x2 cnxn
• Subject to
• b1 x1 b2 x2 bn xn lt B / available
  bandwidth constraint /
• d1 x1 d2 x2 dn xn lt D / delay
  constraint /
• c1x1 c2x2 cnxn lt 4.3 / best
  possible score attainable by a codec/
• c1x1 c2x2 cnxn gt 3.5 / lower bound
  of acceptable MOS scores /
• x1 x2 xn 1
  / total of all percentages /

11
LP problem cont.

• Where
• x1, x2, .. xn amount/percentage of each
  codec in the transmission mixing
• c1, cn MOS values for each codec
• B available bandwidth
• D 400 - link One Way Delay
• b1, b2 .. bn bit rate
• d1, d2, .. dn (packet size in
  bytes)(encoding/decoding delay to create/decode
  1 byte)

12
Implementation
The delay values for each codec have been
determined based on existing literature and
experiments
13
Implementation cont.

• The formulation of this particular linear
  programming problem is thus
• Maximize 4.3x1 4.0x2 3.7x3 4.0x4 3.9x5
• Subject to
• 64 x1 32x2 13x3 8x4 5.6x5 lt B
• 200.5x1 200x2 1982.42x3 2002.5x4
  2106.0x5lt D
• 4.3x1 4.0x2 3.7x3 4.0x4 3.9x5 lt 4.3
• 4.3x1 4.0x2 3.7x3 4.0x4 3.9x5 gt 3.5
• x1 x2 x3 x4 x5
  1
•  
• xi gt 0, i1..5

14
Results

• Two software programs were used to test the
  linear programming problem
• Maple V, a mathematical software which has
  options to solve a simplex problem
• a Java applet found on the Internet at
• http//www-ftp.mcs.anl.gov/otc/Guide/CaseStudies/
  simplex/applet/SimplexTool.html

15
Results cont.
16
Future Work

• Need to consider the situation of packet loss,
  which directly affects the MOS scores for the
  various codecs
• Need to consider the effect of packet size while
  formulating the problem
• Need to consider the effect of adding redundancy
  in terms of FEC (Forward Error Correction)
• This has direct implication in terms of consuming
  more bandwidth, but ensuring higher quality at
  the receiver side

17
References

• 1 Olivier Hersent, David Gurle Jean-Pierre
  Petit, IP Telephony Packet-based multimedia
  communications systems, Addison Wesley, 2000.
• 2 ITU-T Recommendation G.114 (05/93) One way
  Transmission Time
• 3 ITU-T Recommendation G.711 (11/88) - Pulse
  code modulation (PCM) of voice frequencies
  http//www.itu.int/itudoc/itu-t/rec/g/g700-799/g71
  1.html
• 4 ITU-T Recommendation P.800 (08/96) - Methods
  for subjective determination of transmission
  quality.
• 5 D. De Vleeschauer, J. Janssen, G. H. Petit,
  Delay and Distortion Bounds for Packetized Voice
  Calls of Traditional PSTN Quality, IPTEL,
  Berlin, 12-13 April, 2000.
• 6 W.T.K. Wong, R.M. Mack, B.M.G. Cheetham, X.Q.
  Sun, Low Rate Speech Coding for
  Telecommunications, BT Technol J Vol 14, No.1
  January 1996.