Side_1_360

best high medium low (very) poor

R-value 90 – 100 80 – 90 70 – 80 60 – 70 0 – 60

range

tion delay. Since the total queuing delay Tque,1of

the first packet is stochastic, the one-way mouth-

to-ear delay of eq. (4) also is. For static dejitter-

ing mechanisms the dejittering delay Tjitis usu-

ally chosen on the safe side, i.e. such that in the

worst case (when the first arriving packet hap-

pens to be a fast one) at most a fraction Ploss,jit

of the packets get lost. Hence,

Tjit= F-1(Ploss,jit) (5)

Second, we consider a dynamically dejittered

packetized phone call. When the dynamic dejit-

tering mechanism is set to tolerate a packet loss

of Ploss,jit, the dejittering delay is gradually ad-

justed to compensate for the total queuing delay

Tque,1of the first packet, so that after a transition

period the one-way mouth-to-ear delay tends to

TM 2 E=

Tpack+ TDSP+ Tnet,min+ F-1(Ploss,jit). (6)

Comparing eq. (6) with eq. (4) combined with

(5), we see that adaptive dejittering can (eventu-

ally) economize on the one-way mouth-to-ear

delay by an amount equal to Tque,1.

Note that for packetized phone calls the mouth-

to-ear delay in one direction is not necessarily

the same as that in the reverse direction as each

of the terms in eq. (4) (or eq. (6)) may differ

from one direction to the other.

Distortion stems from the encoding of the voice

signal and from packet loss Ploss,netin the trans-

port over the network or from the packet loss

Ploss,jitin the dejittering buffer, i.e.

Ploss= 1 – (1 – Ploss,net)(1 – Ploss,jit) (7)

Note that also the packet loss (and even the

codec format) may differ from one direction

to the other.

In the next section we determine how this one-

way mouth-to-ear delay and this distortion im-

pact the quality of the call.

3 Parameters Determining the

Quality of a Phone Call

3.1 The E-model

The E-model is a tool to predict how an “aver-

age user” would rate a phone call of which the

characterizing transmission parameters are

known. Similar proprietary models exist (see

the references in [16]), but the E-model has the

advantage that it is standardized in ITU-T Rec-

ommendation G.107 [4]. Based on an extensive

set of subjective experiments, a scale, referred

to as the R-scale, was defined in [8] upon which

impairments are approximately additive in the

range of interest. Four types of impairments and

an advantage factor were identified, that is

R= R 0 – Is– Id– Ie+ A (8)

The first term R 0 groups the effects of noise and

is amongst other things a function of the level of

the circuit noise and the (effective) level of the

room noise (present at both sides). The second

term Isincludes impairments that occur simulta-

neously with the voice signal, such as those

caused by quantization, by too loud or too soft

a connection and by a non-optimum side tone.

The third term Idcomprises delayed impair-

ments, including impairments caused by talker

and listener echo or by a loss of interactivity. It

is mainly a function of the level and the delay of

the echo with respect to the original signal and

the mouth-to-ear delays in both directions. The

fourth term Iecovers impairments caused by

what is referred to as “the use of special equip-

ment” in ITU-T Recommendation G.107 and

groups effects due to distortion. It is a function

of the type of low-bit-rate codec used and the

fraction of lost packets. The fifth term A,

referred to as the expectation factor, expresses

the decrease in rating a user is willing to tolerate

because of the “access advantage” that certain

systems have over traditional wire-bound tele-

phony. As an example, the expectation factor A

for mobile telephony (e.g. GSM) is 10.

Based on the rating Rsubjective user reactions

can be predicted, such as the Mean Opinion

Score (MOS) a judging panel would give to the

Table 1 Quality classes
according to ITU-T
Recommendation G.109 PSTN quality

Speech transmission

quality category