Side_1_360

(Dana P.) #1

The following issues have an impact on the QoS
obtained by the receiver; delay/latency, jitter,
digital sampling, voice compression, digital-to-
analogue conversion, tandem encoding, voice
activity detection, echo, packet loss and the pro-
tocols chosen. Even though the received quality
is subjective ITU have tried to make some stan-
dardized measures and from these it could be
concluded that the three main factors of the
impact of the received quality are latency, jitter
and packet loss. For latency the two main prob-
lems are echo and talker overlap. Echo becomes
a problem when the delay increases above about
50 ms, and the talker overlap becomes signifi-
cant and quite annoying when the delay is 500
ms and more. Both figures apply to the round-
trip-delay. The ITU specification G.114 [1] rec-
ommends that no more than 300 ms should
occur for the connection to be categorized as a
high quality connection. (It should be mentioned
that the one-way end-to-end delay for transmis-
sion to a satellite is approximately 270 ms.
Satellite links have been very common and are
still widely used for telephone conversations.)
For most cases the limit for echo to become
annoying will be reached and therefore VoIP
have to implement some kind of echo cancella-
tion procedures.


The delay is characterized as the total amount
of time it takes to transfer the voice from the
sender to the receiver. This time has mainly
three components; it is the propagation delay,
the serialization delay and the processing delay.
A description of the various components of the
delay is summarized in Figure 2, where all num-
bers indicate time in milliseconds. The figures
come from a study performed by Bell [2]. The
consumer objective (100 ms) and the business
objective (300 ms) are assumed values, which
are allocated to the six delay components. The
PSTN and the telephone client are neither
assumed to contribute to the total delay and are
therefore combined in the figure and marked
“negligible”. The column marked “today” is data
from actual measurements performed. The PC
Client and Gateway/POP values in the theoreti-
cal column may need some explanation. It is


assumed that two frames, with an encoding
delay of 30 ms each, are captured in one IP
packet, and that a look-ahead delay of 7.5 ms
is needed.

Packetized Voice


One of the problems of using real-time applica-
tions over packet switched networks is that
the timing of the sent packets never could be
obtained at the receiver end. This is due to the
fact that the IP networks are not isochronous, i.e.
the exact distance between two packets sent out
by the sender will most probably be changed dur-
ing the time they traverse the networks, since
they will experience different conditions on their
way to their destination. The conditions within
the network and along a route could be modelled
according to a probabilistic distribution to be able
to determine the expected delay variations. The
jitter is usually solved with a buffer; see Figure 3,
where the buffer has to be filled with Lbuffer
places before any packet is released to the
receiver. In this way, there will hopefully always
be a packet to release even though the arrival rate
of new packets has decreased momentarily. The
arrival rate to the buffer has a probabilistic distri-
bution (λ) while the release from the buffer is
done according to a deterministic distribution (d).

The packets traversing the network are not guar-
anteed in any way to reach their destination, for
datagram services not even to reach the destina-
tion in order. Individual packets could be lost
due to congestion on the links traversed. In nor-
mal cases the packet switched networks using IP
have retransmission algorithms implemented on
higher layers, i.e. the TCP on the transport layer.
The retransmission procedures are unfortunately

Figure 2 VoIP round-trip
delay allocation and current
performance in milliseconds
(ms), for the different delay
components

Figure 3 A buffer to com-
pensate for the jitter intro-
duced by the IP networks.
L is the number of packets
required in the buffer to start
releasing packets

Delay
component
PC client
Access
IP network
Gateway/POP
PSTN/phone
Total

Consumer
(objective)
100
70
50
80
Negligible
300

Business
(objective)
30
10
30
30
Negligible
100

Today
(actual)
150
150
96
160
Negligible
556

Theoretical
(minimum)
67.5
44
40
67.5
Negligible
159

Above
minimum
82,5
106
56
92,5
0
337

L

λ d
Free download pdf