The Internet Encyclopedia (Volume 3)

P1: C-172

Kroon WL040/Bidgolio-Vol I WL040-Sample.cls June 20, 2003 13:9 Char Count= 0

318 SPEECH ANDAUDIOCOMPRESSION

H.323

MCU

H.323

TERMINAL

H.323

TERMINAL

H.323

TERMINAL

H.323

GATEKEEPER

PACKET

BASED

NETWORK

H.323

GATEWAY

PSTN ISDN

Figure 15: An example network topology using the H.323

protocol.

as intermediaries between end points. Two protocols that

are widely used are H.323 and SIP. Both of these protocols

build on existing protocols. The H.323 protocol suite was

defined by ITU-T to support multimedia communications

over IP independent of network topology. It contains four

endpoints: (1) the terminal, which serves as the interface

to the user, (2) the gatekeeper, which supports services

such as billing and authentication, (3) the gateway, which

supports connection to other networks, and (4) a multi-

point control unit (MCU), which supports teleconferenc-

ing. Figure 15 shows an H.323 network topology. A simple

end-to end call can be accomplished with two H.323 ter-

minals, without the need for the functional modules.

Although we can build a telephony network indepen-

dently of the traditional PSTN (public switched telephony

network), in most cases we need to be able to interface

with the legacy networks, for example when making a

call from an IP phone to a traditional PSTN phone. The

interface function between the Internet-based telephony

and PSTN telephony is served by a so-called gateway. This

gateway will facilitate the protocol translation and will

also convert the audio data into a format suitable for the

PSTN network.

The session initiation protocol (SIP) was proposed by

IETF. It is modeled after the simple mail transfer proto-

col (SMPT). It is independent of the underlying packet

protocol, and it leverages the Internet and Web structure.

Although most existing IP phones use H.323, SIP is gain-

ing momentum due to its simplicity. It is expected that

both protocols will coexist for a long time and that most

equipment will support both protocols.

Despite the use of RTP, there are still no guarantees that

all packets will arrive on time, or will arrive at all. Buffer-

ing the incoming packets can compensate for late arrival,

at the expense of an increase in delay. Hence it is common

to make this buffer, the so-called jitter buffer, adaptive and

to have it increase in size if the throughput becomes less

reliable. Proper error mitigation is also important, and

some coders have this built in, whereas others such as

G.711 need external concealment. Due to buffering and

other transport delay mechanisms, quite often there is a

need for echo cancellation, especially if a speakerphone is

used (or the speakers on a PC).

Although it is hard to believe that Internet telephony

over the public Internet will replace PSTN services soon,

it is a very likely to become a dominant application in

corporate networks. Besides the economical advantage of

only needing one network structure, these networks are

also more carefully managed, thereby making it possible

to guarantee a minimum quality of service.

Audio Streaming

There are two popular ways of distributing music over

the Internet.Downloadingis a copying operation from

some server containing the original or compressed ver-

sion of the material. The downloading will typically use

the IP/TCP protocol, and one has to wait till all the ma-

terial has been downloaded before being able to play the

music. Once downloaded, the same material can be played

repeatedly. The other common form of music distribution

isstreaming, which is very much like radio broadcasting.

The material is transmitted to the end user, but the signal

is played out almost instantaneously without local stor-

age. Similarly to the VoIP scenario, it is necessary to buffer

the packets to compensate for late arrival. Because this

is a broadcast or one-way communication scenario, the

amount of buffering can be large (a few to 30 s). Although

in principle the buffers could be made even larger, this

would create large delays before a player produced an au-

dible signal. From a user interface perspective this is un-

desirable. Hence, sophisticated buffer control has been

developed to make sure that continuous throughput is

maintained, while maintaining a relatively small buffer.

Some streaming services accomplish this by reducing the

coder rate temporarily if the average available connection

rate cannot support the initial streaming rate.

The compression techniques used are typically based

on audio coding algorithms, because most material that

is being streamed will be music. For lower rates it is pos-

sible that speech-coding algorithms are used instead. A

commonly used format is the MP3 (MPEG-1 Layer III) for-

mat running at bit rates varying from 64 to 128 kb/s. For

streaming applications, it is possible to use proprietary

coders, as long as the decoders are available as down-

loads. However, most content providers will only support

one or two formats, and as a result a couple of proprietary

standards have become de facto standards. Examples are

Apple’s Quicktime, Microsoft’s Windows Media Player,

and RealNetworks’ RealPlayer. All of these proprietary

coders have a reasonable quality vs. bit rate performance,

while trading off other parameters such as delay, com-

plexity, and audio bandwidth. As should be clear from the

previous sections, the best quality is obtained at higher bit

rates (e.g., 96 to128 kb/s), whereas at lower rates (e.g., 24

to 64 kb/s) tradeoffs will be made by reducing the audio

bandwidth, or even switching to mono. It should also be

noted that even at the same bit rate and using the same

format, differences in quality could exist due to the qual-

ity of the source material and the encoder used. For most

streaming applications it is important that a variety of

rates can be accommodated to support the various con-

nection speeds. It is also important that the decoder have

a relative low complexity to allow it to run on the host pro-

cessor. With the advances in computing speed, this has be-

come less of an issue. However, if these formats are used

for downloading in portable players, complexity becomes

an issue because it is connected to battery life and cost.