394
VOICE-OVER PACKETS IN A PACKET NETWORK
z
Queueing delay. Time spent in queue because it is a shared network. One method to
reduce this delay is to prioritize voice packets (vis `a vis data). Objective
<50 msec.
z
Propagation delay. Variable. Major contributor to total delay. Geostationary satellite
relay of circuits is a special problem. The trip to the satellite and back is budgeted
at 250 msec.
One way to speed things up is to increase the bit rate per voice data stream. To do
this, the aggregate bit rate may have to be increased. Or the number of voice streams
may be reduced on the aggregate bit rate so that each stream can be transmitted at a
faster rate.
15.4.4
Lost Packet Rate
A second concern of the VoIP designer is lost packet rate. There are several ways a packet
can be "lost." For example, Section 15.3.3 described a dejitterizing buffer. It has a finite
size. Once the time is exceeded by a late packet, the packet in question is lost. In the
case of G.711, this would be the time equivalent of 16 or 26 msec (duration of a packet
including its header). Another reason for a packet to be lost may be excessive error rate
on a packet whereby it is deleted. When the lost (discarded) packet rate begins to exceed
10%, quality of voice starts to deteriorate. If high-compression algorithms are employed
such as G.723 or G.729, it is desirable to maintain the packet loss rate below 1%. Router
buffer overflow is another source of packet loss.
IP through TCP has excellent retransmission capabilities for errored frames or packets.
However, they are not practical for voice-over IP because of the additional delay involved.
When there is a packet or frame in error, the receive end of the link transmits a request
(RQ) to the transmit end for a packet retransmission and its incumbent propagation delay.
This must be added to the transmission delay (and some processing delay) to send the
offending packet back to the receiver again.
15.4.4.1
Concealment of Lost Packets
. A lost packet causes a gap in the reception
stream. For a single packet we are looking at a 20- to 40-msec gap. The simplest measure
to take for lost packets and the resulting gaps is to disregard. The absolute silence of a
gap may disturb a listener. In this case, often artificial noise is inserted.
There are packet loss concealment (PLC) procedures which can camouflage gaps
in the output voice signal. The simplest techniques require a little extra processing
power, and the most sophisticated techniques can restore speech to a level approx-
imating the quality of the original signal. Concealment techniques are most effective
for about 40 to 60 msec of missing speech. Gaps longer than 80 msec usually have to
be muted.
One of the most elementary PLCs simply smoothes the edges of gaps to eliminate
audible clicks. A more advanced algorithm replays the previous packet in place of the
lost one, but this can cause harmonic artifacts such as tones or beeps. Good concealment
methods use variation in the synthesized replacement speech to make the output more like
natural speech. There are better PLCs to preserve the spectral characteristics of the talker's
voice and to maintain a smooth transition between estimated signal and surrounding
original. The most sophisticated PLCs use CELP (codebook-excited linear predictive)
or similar technique to determine the content of the missing packet by examining the
previous one [11].
Lost packets can be detected by packet sequence numbering.