Nortel VoIP Gateway, Configuration Manual

Page: 23 / 220

Key IP telephony concepts 23
Norstar VoIP Gateway Configuration Guide
Jitter Buffer
Voice frames are transmitted at a fixed rate, because the time interval between frames is constant.
If the frames arrive at the other end at the same rate, voice quality is perceived as good. In many
cases, however, some frames can arrive slightly faster or slower than the other frames. This is
called jitter, and degrades the perceived voice quality. To minimize this problem, the VoIP
Gateway uses a jitter buffer for arriving frames.
The Norstar VoIP Gateway uses a dynamic jitter buffer that can be configured using two
parameters:
• Minimum delay (0 msec to 150 msec).
This defines the starting jitter capacity of the buffer. For instance, at 0 msec, there is no
buffering at the start. At the default level of 70 msec, the VoIP Gateway will always buffer
incoming packets by at least 70 msec worth of voice frames.
• Optimization Factor (0 to 12).
This defines how the jitter buffer tracks to changing network conditions. When set at its
maximum value of 12, the dynamic buffer will aggressively track changes in delay (based on
packet loss statistics) to increase the size of the buffer and then not decay back down. This
results in the best packet error performance, but at the cost of extra delay. At the minimum
value of 0, the buffer tracks delays only to compensate for clock drift and quickly will decay
back to the minimum level. This optimizes the delay performance but at the expense of a
higher error rate.
The default settings of 70 msec Minimum delay and 7 Optimization Factor should provide a good
compromise between delay and error rate. The jitter buffer "holds" incoming packets for 70 msec
before making them available to the codec for decoding into voice. The codec actually "takes"
frames from the buffer at regular intervals in order to produce continuous speech. As long as
delays in the network do not change (jitter) by more than 70 msec from one packet to the next,
there will always be a sample in the buffer for the codec to use. If there is more than 70 msec of
delay at any time during the call, the packet is too late. The codec will try to access a frame and
will not be able to find one. The codec must produce a voice sample even if a frame is not
available. It will actually create a voice sample to use that minimizes the effect of the loss. This
loss is then flagged as the buffer being too small. The dynamic algorithm then causes the size of
the buffer to increase for the next voice session. The size of the buffer may decrease again if the
gateway notices that the buffer is not filling up as much as expected. At no time will the buffer
shrink to less than the minimum size configured in the Minimum delay parameter.
This delaying of packets can provide somewhat of a communications challenge, as speech is
delayed by the number of frames in the buffer. For one-sided conversations, there are no issues.
However, for two-sided conversations, where one party tries to interrupt the other speaking party,
it can be annoying. In this second situation, by the time the voice of the interrupter reaches the
interruptee, the interruptee has spoken (2*jitter size) frames past the intended point of interruption.