Some wireless SIP based features such as Push to Talk may have slightly dif-
ferent specifications for delay.
Burst vs. Average Packet Loss
The average packet loss figures quoted by most legacy telecom sources are not
very meaningful for either data or for voice, and do not accurately reflect the
behavior of IP networks and IP applications, since:
■■ Network congestion or route flaps produce long bursts of packet loss.
■■ Distributed packet loss is, therefore, not meaningful and is also easier to
compensate for when transmitting either data or voice.
Internet voice codecs have a high tolerance for distributed packet loss over
time, but long bursts of packet loss cannot be compensated for and translate
into loss of speech syllables or even entire words and sentences, similar to that
experienced with mobile phones. For this reason, the Extended Report (XR)
has been defined for the RTP Control Protocol (RTCP) in RFC 3611 [3].
Acoustics and the Network
Voice is probably the most demanding real-time application for the Internet,
though networked games are quite close, or even more demanding, when it
comes to delay and packet loss. Figure 18.1 shows the main phenomena [4] at
issue for real-time voice communications.
Figure 18.1 Acoustics and the network
Network (delay and packet loss)
Adaptive Filter
Decoder
4 wire to
2 wire
Far end echo
Near end
echo
(sidetone)
Encoder Σ
304 Chapter 18