Network Architecture special function

1
IPTel 2000, April 12-13, 2000, Berlin
Delay and Distortion Bounds for Packetized Voice
Calls of Traditional PSTN Quality

Jan Janssen, Danny De Vleeschauwer, Guido H.
PetitAlcatel, Network Strategy Group, Antwerp,
Belgium
2
List of contents

• The E-model
• Description
• R0 , Is and A
• Delay impairment Id
• Talker and listener echo
• Standardized delay bounds for undistorted voice
• Results for undistorted voice
• Distortion impairment Ie
• for standardized low bit rate codecs
• Transcoding (matrix)
• Influence of packet loss
• Determination of R-factor of a packetized voice
  call
• Quality bounds
• Conclusions
• Questions Acknowledgments

3
The E-modelDescription
Objective network parameters
Impairments which occur simultaneously with voice
signal
Impairments caused by delay
Basic signal-to-noise ratio
Distortion Impairment
ExpectationFactor
E-model
R R0 - Is - Id - Ie A Rating Factor in
0,100
Predictions of user reactions
Mean OpinionScore
Good Or Better
Poor Or Worse
Ter- Minate Early
Subjective quality measures
4
The E-model R0 , Is and A

• The basic signal-to-noise ratio R0 (room noise,
  circuit noise, ...) and the impairments which
  occur simultaneously with the voice signal Is
  (too loud a connection, too loud a side tone, )
  are
• not fundamentally different for circuit-switched
  or packetized voice calls
• not considered
• The expectation factor A is defined as the amount
  of impairment tolerated because of access of
  advantage with respect to wire-bound telephony
• Example A10 for mobile telephony
• The quality of circuit-switched and packetized
  voice calls is compared in a fair way A0

5
The E-modelDelay impairment Id

• In the context of packet-based networks, the
  quality of voice calls is mainly determined by
• the impairment associated with the mouth-to-ear
  delay Id
• loss of interactivity
• talker echo
• listener echo
• the impairment associated with distortion Ie

one-way delay incurred from the moment the talker
utters the words until the listener hears them
6
Delay impairment Id Talker and listener echo

• In the PSTN, EL is typically 21 dB (due to 4-to-2
  wire hybrid echo)
• If the packetized voice call is terminated over
  the PSTN to a traditional phone, EL21 dB
• If the packetized voice call is terminated over a
  packet-based network on
• a PC, the EL is likely to be smaller (lt21 dB) due
  to acoustic echo in the PC
• an IP-phone, EL40 dB
• Echo control increases the EL by 30 dB, perfect
  echo control increases EL to infinity

EL (Echo Loss)
7
Delay impairment IdStandardized delay bounds for
undistorted voice

• ITU-T Recommendations G.114 and G.131 specify the
  following tolerable mouth-to-ear delay bounds for
  undistorted (analog or G.711_at_64 kb/s) voice with
  (hybrid or acoustic) echo

without echo control EL21 dB
echo controlneeded
acceptable
unacceptable
conditionally acceptable
25 ms
150 ms
400 ms
mouth-to-ear delay
8
Delay impairment Id Results for undistorted voice
intrinsicquality (R94.3)
traditional quality (R70)
9
The E-modelDistortion impairment Ie

• In the context of packet-based networks, the
  quality of voice calls is mainly determined by
• the impairment associated with the mouth-to-ear
  delay Id
• the impairment associated with distortion Ie
• compression by low bit rate codecs
• VAD (Voice Activity Detection)
• transcoding
• packet loss

10
Distortion impairment Ie for standardized low bit
rate codecs

• ITU-T RecommendationG.109 (draft)
• R-values below 50 are not recommended

PSTN quality

• VAD increases Ie by 0 to 1

11
Distortion impairment IeTranscoding

• Transcoding translation of voice from codec
  format X into codec format Y (via 8 kHz sampled
  voice)
• Might occur at the boundary between two networks
  that do not have the same codec bank
• In the E-model, impairments are (approximately)
  additive the distortion impairments Ie
  associated with codecs X and Y should be added to
  obtain the overall distortion impairment factor
  Ie Transcoding can be very
  harmful to the quality of a call and should be
  avoided if possible (see transcoding matrix on
  next slide with intrinsic quality levels, i.e.,
  for mouth-to-ear delayslt150 ms)

12
Distortion impairment IeTranscoding matrix
13
Distortion impairment IeInfluence of packet loss

• The distortion impairment Ie associated with a
  codec increases as the packet loss ratio
  increases
• 4 codecs (with specific packetization interval)
• G.711 (10 ms)
• GSM-EFR_at_12.2 kb/s (20 ms)
• G.729_at_8 kb/s with VAD (20 ms)
• G.723.1_at_6.3 kb/s with VAD (30 ms)
• PLC (Packet Loss Concealment)
• Built-in for all considered low bit rate codecs
• Can be implemented on top of G.711 codec

14
Distortion impairment IeInfluence of packet loss
(contd)
15
Determination of R-factor of a packetized voice
call

• Given mouth-to-ear delay, echo loss and
  distortion impairment Ie R-rating
  of a packetized voice call (from the G.711
  figure)
• Identify the curve corresponding to the given EL
  value
• Read rating R corresponding to given mouth-to-ear
  delay (R84)
• Subtract the distortion impairment Ie from this
  R-value (R68)
• Example on next slide
• Mouth-to-ear delay200 ms
• EL51 dB
• Distortion impairment Ie16 (GSM-EFR codec with
  1 packet loss)

16
Determination of R-factor of a packetized voice
call (contd)
17
Quality bounds

• Without echo control very small
  delay budgets if traditional quality is aimed for
• Assume perfect echo control
• Intrinsic quality (for mouth-to-ear delayslt150
  ms) depends solely on distortion impairment Ie
  (codec and packet loss)
• 94.3 (intrinsic quality undistorted voice) - 70
  (R-factor traditional quality) 24.3
  if we aim for traditional voice quality, we
  have a maximum impairment budget of 24.3 on the
  R-scale

18
Quality bounds (contd)

• Impairment budget of 24.3
• Part is consumed by impairment of (low bit rate)
  codec (see slide 11)
• Remainder can be consumed by
• allowing some packet loss (for mouth-to-ear
  delays lt 150 ms)

19
Quality bounds (contd)

• allowing a mouth-to-ear delay larger than 150 ms
  (when packet loss ratio 0 )
• a combination of both (trade-off between packet
  loss and mouth-to-ear delay)

NA (traditional quality is Not Attainable)
20
Conclusions

• Echo control is recommended
• For perfect echo control, quality remains equal
  to intrinsic quality up to a mouth-to-ear delay
  of 150 ms
• Intrinsic quality of some low bit rate codecs is
  lower than traditional PSTN quality
• Transcoding should be avoided
• Margin between intrinsic and traditional quality
  can be consumed by allowing a mouth-to-ear delay
  above 150 ms and/or by allowing some packet loss

21
QuestionsAcknowledgments

• Jan Janssen (jan.janssen_at_alcatel.be)
• Danny De Vleeschauwer (danny.de_vleeschauw
  er_at_alcatel.be)
• Guido H. Petit (guido.h.petit_at_alcatel.be)
• Alcatel, Network Strategy Group
• Qos, Traffic and Routing
  Technologies
• Francis Wellesplein 1
• B-2018 Antwerp
• Belgium

This work was carried out within the framework of
the project LIMSON, sponsored by the Flemish
institute for the promotion of scientific and
technological research in the industry