 |
The E-model |
[ E-model Home] [Calculation tool]
E-model Tutorial
Background
The E-model (ITU-T Rec. G.107 [1]) is a transmission planning tool that provides a prediction of the expected voice quality, as perceived by a typical telephone user, for a complete end-to-end (i.e. mouth-to-ear) telephone connection under conversational conditions. The E-model takes into account a wide range of telephony-band impairments, in particular the impairment due to low bit-rate coding devices and one-way delay, as well as the "classical" telephony impairments of loss, noise and echo. It can be applied to assess the voice quality of wireline and wireless scenarios, based on circuit-switched and packet-switched technology.
The E-model is based on modeling the results from a large number of subjective tests done in the past on a wide range of transmission parameters. The primary output of the E-model calculations is a scalar quality rating value known as the "Transmission Rating Factor, R". R can be transformed into other quality measures such as Mean Opinion Score (MOS-CQE [2]), Percentage Good or Better ((GoB) or Percentage Poor or Worse ((PoW). However, caution should be exercised when comparing these transformed measures with values of MOS, %GoB or %PoW from other sources, which may not have been obtained under comparable conditions.
The E-model is based on a mathematical algorithm, with which the individual transmission parameters are transformed into different individual "impairment factors" that are assumed to be additive on a psychological scale. The algorithm of the E-model also takes into account the combination effects for those impairments in the connection which occur simultaneously, as well as some masking effects. To the extent that impairments are present for which psychological additivity is not maintained, E-model predictions may be inaccurate.
The relation between the different impairment factors and R is given by the equation:
R = Ro - Is - Id - Ie,eff + A
where:
|
The term Ro expresses the basic signal-to-noise
ratio (received speech level relative to circuit and acoustic noise). |
|
The term Is represents all impairments that
occur more or less simultaneously with the voice signal, such as: too loud
speech level (non-optimum OLR), non-optimum sidetone (STMR), quantization
noise (qdu), etc. |
|
The term Id sums all impairments due to delay
and echo effects. |
|
The term Ie,eff is an "effective equipment
impairment factor", which represents impairments caused by low
bit-rate codecs. It also includes impairment due to packet-losses of
random distribution. Values of Ie for specific codecs without packet loss
are given in ITU-T
Rec. G.113 App. I [3], and these values are
transformed to Ie,eff in case of random packet loss, using the E-model
algorithm. |
|
The term A is an "advantage factor",
which allows for an "advantage of access" for certain systems
relative to conventional systems, trading voice quality for convenience.
While all other impairment factors are subtracted from the basic
signal-to-noise ratio Ro, A is added and thus compensates other
impairments to a certain amount. It can be used to take into account the
fact that the user will tolerate some decrease in transmission quality in
exchange for the "advantage of access". Examples of such
advantages are cordless and mobile systems or connections into
hard-to-reach regions via multi satellite hops. |
Values of R fall in the range of 0 £
R < 100, with higher values indicating higher speech quality. Table 1, taken from
ITU-T
Rec. G.109 [4], relates the E-model Ratings R to categories of speech transmission quality and to user satisfaction.
Table 1 - Definition of categories of speech transmission quality
|
Range
of E-model Rating R
|
Speech
transmission quality category
|
User satisfaction
|
|
90
£
R <
100
|
Best
|
Very satisfied
|
|
80 £
R <
90
|
High
|
Satisfied
|
|
70 £
R <
80
|
Medium
|
Some users dissatisfied
|
|
60 £
R <
70
|
Low
|
Many users dissatisfied
|
|
50 £
R <
60
|
Poor
|
Nearly all users dissatisfied
|
|
NOTE
1 – Connections with E-model Ratings R below 50 are not
recommended.
NOTE
2 – Although the trend in transmission planning is to use
E-model Ratings R, equations to convert E-model Ratings R into
other metrics, e.g. %MOS, %GoB,
PoW can be found in ITU-T
Rec. G.107 Annex B [1].
|
Application of the E-model
Figure 1 shows the basic reference configuration used by the E-model for estimating end to end speech quality, with the parameters involved defined below:
|
- SLR |
Send Loudness Rating |
|
- RLR |
Receive Loudness Rating |
|
- OLR |
Overall Loudness Rating1 |
|
- STMR |
Sidetone Masking Rating2 |
|
- LSTR |
Listener Sidetone Rating2 |
|
- Ds |
D-value of telephone at send-side |
|
- Dr |
D-value of telephone at receive-side2 |
|
- TELR |
Talker Echo Loudness Rating |
|
- WEPL |
Weighted Echo Path Loss |
|
- T |
Mean one way delay of the echo path |
|
- Tr |
Roundtrip delay in a closed 4-wire loop |
|
- Ta |
Absolute one-way delay in echo free connections |
|
- qdu |
Number of quantization distortion units |
|
- Ie |
Equipment impairment factor |
|
- Ppl |
Random packet-loss probability |
|
- Bpl |
Packet-loss robustness factor |
|
- Nc |
Circuit noise referred to the 0 dBr-point |
|
- Nfor |
Noise floor at the receive-side |
|
- Ps |
Room noise at the send-side |
|
- Pr |
Room noise at the receive-side |
|
- A |
Advantage factor |
The connection is basically divided into a "send side" (subscript S) and a "receive side" (subscript
R) with a virtual centre referred to as a "0 dBr-point". One of the most important assumptions in the model is that the perceived quality is referred to the user on the "receive side", with this user experiencing conversational conditions (talking and listening). This is important when inputting parameters relating to listening (loudness, equipment impairment factor etc) and talking (e.g. sidetone, delay, echo etc).
Figure 1 - Basic reference configuration of the E-model
Values of SLR, RLR and Nc must be referred to a 0 dBr-point (i.e. any gains or losses taken into account). For the basic end to end loudness, use
SLRS and RLRR.
To evaluate the impairment due to talker echo, the E-model expects two parameters, the mean one-way echo path delay (T), and the Talker Echo Loudness Rating (TELR) of the echo path. It is very important to note that talker echo is referred to the user on the receive side. The value for TELR is obtained in a pre-calculation according to the basic formula:
TELR = SLRR + EL + RLRR
where: EL (Echo Loss), represents the effective echo return loss seen by the receive side.
To evaluate the impairment due to one-way delay in the absence of echo, two approaches may be used:
| i.
|
Input the required value of Ta and set T = Tr = 0. This effectively turns "off" the echo algorithm and can be considered the absolutely ideal echo-free case. |
| ii.
|
It may be more realistic to input the required value of Ta and set T = Tr/2 = Ta, with TELR and WEPL set to their default values. This corresponds to the practical case of a near-ideal echo canceller. |
Default values for all E-model parameters and
nominal parameter ranges3 are listed in Table 2. With these default values, the resulting value of R is 93.2.
Table 2 - Default values and recommended ranges for the parameters
|
Parameter
|
Abbr.
|
Unit
|
Default value
|
Recommended range
|
Notes
|
|
Send Loudness Rating
|
SLRS
|
dB
|
+8
|
0 to +18
|
1
|
|
Receive Loudness Rating
|
RLRR
|
dB
|
+2
|
-5
to +14
|
1
|
|
Sidetone Masking Rating
|
STMR
|
dB
|
15
|
10 to 20
|
2
|
|
Listener Sidetone Rating
|
LSTR
|
dB
|
18
|
13 to 23
|
2
|
|
D-value of telephone, send side
|
Ds
|
–
|
3
|
-3
to +3
|
|
|
D-value of telephone receive side
|
Dr
|
–
|
3
|
-3
to +3
|
2
|
|
Talker Echo Loudness Rating
|
TELR
|
dB
|
65
|
5 to 65
|
|
|
Weighted Echo Path Loss
|
WEPL
|
dB
|
110
|
5 to 110
|
|
|
Mean one-way delay of the echo path
|
T
|
ms
|
0
|
0 to 500
|
|
|
Round trip delay in a 4-wire loop
|
Tr
|
ms
|
0
|
0 to 1000
|
|
|
Absolute delay in echo free connections
|
Ta
|
ms
|
0
|
0 to 500
|
|
|
Number of Quantization distortion units
|
qdu
|
–
|
1
|
1 to 14
|
|
|
Equipment impairment factor
|
Ie
|
–
|
0
|
0 to 40
|
|
|
Packet-loss Robustness Factor
|
Bpl
|
–
|
1
|
1 to 40
|
3
|
|
Random Packet-loss Probability
|
Ppl
|
%
|
0
|
0 to 20
|
3
|
|
Circuit noise referred to 0 dBr-point
|
Nc
|
dBm0p
|
-70
|
-80
to -40
|
|
|
Noise floor at the receive Side
|
Nfor
|
dBmp
|
-64
|
–
|
3
|
|
Room noise at the send side
|
Ps
|
dB(A)
|
35
|
35 to 85
|
|
|
Room noise at the receive side
|
Pr
|
dB(A)
|
35
|
35 to 85
|
|
|
Advantage factor
|
A
|
–
|
0
|
0 to 20
|
|
|
NOTE
1 – Total values between microphone or receiver and 0
dBr-point.
NOTE
2 – Fixed relation: LSTR =
STMR + D.
NOTE
3 – Currently under study.
|
The Calculation Tool that is provided here implements the basic E-model calculation, based on the most recent version [1].
References
[1] "The E-model, a computational model for use in transmission planning" ITU-T Rec. G.107 (March 2005)
[2] "Mean Opinion Score (MOS) terminology" ITU-T
Rec. P.800.1
[3] "Provisional planning values for the equipment impairment factor Ie and packet-loss robustness factor Bpl"
ITU-T Rec. G.113
App. I
[4] "Definition of categories of speech transmission quality" ITU-T Rec. G.109
Recommended bibliography
The following references provide more details on the E-model and its application.
"Application of the E-model - A Planning Guide" ITU-T Rec. G.108
"Guidance for assessing conversational speech transmission quality effects not covered by the E-model"
ITU-T Rec. G.108.1
"Transmission and Multiplexing (TM); Speech communication quality from mouth to ear for 3.1 kHz handset telephony across networks" ETSI ETR 250, July 1996
"The ETSI Computation Model: a Tool for Transmission Planning of Telephone Networks", N.O. Johannesson, IEEE Communications Magazine, January 1997
"Voice Quality Recommendations for IP Telephony" TIA/TSB-116-A (March 2006)
http://www.tiaonline.org/standards/technology/voip/documents/TSB116-Afinalforglobal.pdf
1 No direct input value; calculated as OLR = SLRS + RLRR.
2 These parameters have a fixed relation by: LSTR = STMR + Dr.
3 The E-model is not validated outside of these ranges
Last update: 2008-05-29
|
