P1: C-172
Kroon WL040/Bidgolio-Vol I WL040-Sample.cls June 20, 2003 13:9 Char Count= 0
APPLICATIONS 317APPLICATIONS
Two popular Internet applications of speech and audio
compression are telephony and streaming. In both cases
the IP data network is used to transport digitized (and
compressed) audio signals. The basic protocol is the IP
(Internet protocol), which is used to set up connections
between machines and sessions between applications. For
most data applications the next protocol layer would be
TCP (transmission control protocol), which guarantees a
reliable connection. Because this is accomplished
by acknowledgment of receipt and possible retransmis-
sion it cannot guarantee continuous throughput and is
therefore not suitable for telephony applications. Instead
the UDP (user datagram protocol) is used (see Figure 14).
UDP is a simple protocol that does not support retransmis-
sion. As a result the data transmission is based on a best
effort. For voice communications the use of UDP alone
is usually not sufficient. For example, all packets could
arrive but the order could have changed. Another proto-
col layer that is usually used on top of UDP is real-time
transport protocol (RTP). This protocol is used to provide
delivery services for real-time data such as time stamp-
ing, source identification, and sequence numbering. The
sequence count is used to determine the playback order
of packets and to identify if packets are missing. The time
stamp is used to determine the amount of delay variation
and could be used, for example, to adjust a jitter buffer.
The source identifier is used to allow different calls to be
combined into bigger packets. The RTP protocol is often
used together with the real-time transport control proto-
col (RTCP). This defines a mechanism for hosts conduct-
ing a RTP session to exchange information for monitor-
ing and control of packet counts, and number of packets
lost. The RTCP packets use the same header as RTP pack-
ets, but its payload contains the control information. The
RTCP packets are transmitted only several times per sec-
ond to reduce overhead, but at least once every 5 s. RTCP
can be used to monitor network performance, for example
by reducing the load introduced by non-real-time traffic
that is part of the same session (e.g., graphics) if too many
packets are lost. It could also control the compression al-
gorithm by increasing or decreasing its bit rate.
The use of the various protocols significantly increases
the overhead (i.e., the extra information beyond the use-
able (payload) information). The currently common stan-LAYER432PHYSICAL 1DATA LINKIPTCPPHYSICALDATA LINKIPUDPTYPICALLY NON
REAL-TIME DATATYPICALLY
REAL-TIME DATAFigure 14: Protocol stacks for non-real-time and real-
time sensitive data.dard IPv4 has a 20- to 24-byte header, whereas the new
proposed IPv6 header has a 40-byte header. UDP and RTP
add another 20 bytes, resulting in a significant packet over-
head. One way to minimize this overhead is by increasing
the packet size. For audio signals this is not always desir-
able, because a large packet size corresponds to a large
delay, or in case of packet losses, the removal of a large
time segment of audio information.
Another challenge in transmitting real-time sensitive
data over a packet data network is the fact that the Inter-
net is a best effort network. This means that one cannot
guarantee the on-time arrival of the audio data, resulting
in interruptions in the audio, or introducing unacceptable
delays. In other words,quality of service(QoS) cannot be
guaranteed for real-time communications. Many recent
efforts have focused on improving this situation. For ex-
ample, new protocols support differentiation of packets
in terms of being real-time critical or not and improved
routing schemes reduce the overhead needed for traffic
flow management. To give real-time sensitive data a bet-
ter guarantee on arrival it is necessary to prioritize these
packets. It is important to keep in mind that prioritization
can only work if other packets can get a lower priority and
if routers can handle these priority requests. The resource
reservations protocol (RSVP) tries to address this problem
by negotiating priorities with routers. Another approach
is based on the use of the differentiated services protocol
(DiffServ), which sets priorities in the header bits and as-
sumes multiple priority queues in the router. DiffServ is
considered to be a more manageable protocol when used
on a large scale, because it does not require routers to keep
track of all requests and streams. Both protocols assume
that the desired QoS result is achieved, but they will not
guarantee it.Internet Telephony
The use of the Internet for voice communications is now
commonly referred to asInternet telephonyorvoice over
IP(VoIP). Vocal Tech introduced the first consumer im-
plementation in 1985 and it was widely hailed as a way
to make free phone calls both domestically and interna-
tionally. As a result the concept of Internet telephony has
become a reality, especially when used over managed net-
works such as found in enterprise networks. It not only
will replace traditional circuit-switched phone systems in
terms of functionality but also will open up many new
opportunities. Examples are multimedia conferencing, in-
tegration of productivity tools on a PC with phone func-
tionality, and Internet call centers. Economically, it makes
sense to integrate voice and data services, not only from
the end-user’s perspective but also from the enterprise or
service provider perspective.
To transport voice data over the IP network the sig-
nals are digitized and compressed with any of the coders
described in the previous sections. The most commonly
used ones are 64 kb/s ITU-T G.711 or lower rate coders
such as the 8 kb/s ITU-T G.729 and G.729A. The media
streams are transported using the IP/UDP/RTP protocol
layering described before. To provide a complete phone
service it is necessary to support numbering schemes,
billing and call setup protocols, and servers that can acts