Side_1_360

of the telephony system are used. Hence, no

echo cancellation is performed by the telephony

system even though long distance calls are made.

That forces VoIP to include echo cancellation

among the services supplied.

Voice Activity Detection

In the ordinary telephone network a two-way

simultaneous link is set up between sender and

receiver. This link carries voice at a rate of

64 kb/s in both directions. Usually only one of

the parties is active at any one time; even the

active part has breaks and pauses in a normal

speech pattern. Hence, the utilization of this

two-way link is most of the time less than 40 %.

This fact could be used in VoIP to enhance the

performance of the transmission and less band-

width is required to obtain better speech quality

using voice activity detection (VAD). A generic

outlook of the VAD algorithm is depicted in Fig-

ure 5, where it is shown that the algorithm works

by detecting the magnitude (dB) and then decid-

ing when the voice is inactive and thereby stop-

ping the transmission of packets in that direction

for the moment. To be on the safe side, when

cutting the transmission the algorithm waits a

fixed amount of time, hang-over time, after it

detects a drop in the voice magnitude before it

totally stops the voice sample packet transmis-

sion. The hang-over time duration is in the mag-

nitude of hundreds of ms (typically 150–250

ms). Another problem is to differ between voice

and background noise, and to calibrate itself the

VAD is disabled at the beginning of new calls.

However, even after that it could be cumber-

some to detect when a new voice spurt occurs.

The algorithm cut-offs the beginning of each

new voice spurt and waits until it is sure that it is

a new voice spurt and not, for example, a noise

peak. This phenomenon is called front-end

speech clipping, and is usually not noticeable

for the listener.

Standards

Interoperability among VoIP products has been a

major stumbling block to widespread acceptance

of the technology. The ITU’s H.323 umbrella

standard, shown in Figure 6, which was the first

posed for VoIP interoperability, proved complex

and difficult to implement. As a result, other

less-unwieldy standards were posed in its place

and until recently, we have seen little consensus

on which VoIP standards that would be the most

widely implemented. Even though the H.323

standard is the dominating standard at present,

most vendors foresee a coexistence of several

standards in the arena for quite some time. The

most supported standard is H.323 version 2, but

version 3 and 4 are rapidly catching up. (It

should be pointed out that H.323 version 1 is not

forward compatible with the latter standards of

H.323.) Other supported standards are SIP (Ses-

sion Initiated Protocol) by IETF, the Media

Gateway Control Protocol (MGCP) and H.248.

SIP is an application layer signalling protocol

that specifies call control for multiparty sessions,

IP phones or multimedia distribution. Unlike

H.323, which is based on binary encoding, SIP

is a text-based protocol that is usually easier to

implement. Further information regarding SIP

could be found in [6,7].

MGCP is designed as a simple mechanism to

mainly control the gateways. Its function is to

control the gateways while relying on external

call control intelligence for more complex func-

tions. With the MGCP model, the gateway

focuses on the audio signal translation function

while a call agent, external to the gateway, han-

dles the signalling and call processing functions.

By separating out the internal gateway functions

from the external signalling function, the imple-

mentation, upgrade and maintenance of the gate-

way are reduced to a minimum. This increases

the likelihood of widespread use of this technol-

Figure 5 The voice activity
detection (VAD) algorithm,
used to decrease the required
bandwidth for VoIP calls

dB Magnitude Hang-over Magnitude

Front-end

speech clipping

Time

Noise floor

Front-end

speech clipping