Rtp Packet-Header Compression; Serialization Delay; Voice Activity Detection - Cisco VGD-1T3 Software Configuration Manual

Configuring Voice over IP

RTP Packet-Header Compression

Because of the repetitive nature of subsequent IP/UDP/RTP (network/transport/session-layer) headers,
you can compress them significantly. A recommended methodology is cRTP (Compressed Real-Time
Transfer Protocol), which, by tracking first-order and second-order differences between headers on
subsequent packets, compresses the 40-byte header to just 2 or 4 (without or with UDP checksum) bytes.
Other methodologies may be preferable if the cRTP high CPU usage causes delay. Use a compression
methodology on both ends of low-bandwidth (< 1.5 Mbps) WAN circuits, but not at all on high-speed
(> 1.5 Mbps) WANs.
For more information and configuration options, see the VoIP over PPP Links with Quality of Service
Tip
(LLQ / IP RTP Priority, LFI, cRTP) document, available online at
http://www.cisco.com/warp/public/788/voice-qos/voip-mlppp.html.

Serialization Delay

You can control packet (payload) size—which, in turn, controls how long one packet takes to be placed
on the system interface. Set this in bytes, ideally equaling no more than 20 ms (typically equivalent to
two 10-ms voice samples per packet). Increasing serialization delay increases end-to-end delay. You
want to incur no more than 150–200 ms of one-way, end-to-end delay.
Take care when you assign a payload size for your chosen codec. To assign a codec and payload size,
Note
you use the codec codec bytes payload-size command under the dial-peer voip command. Although the
codec command permits a wide range of payload sizes, the universal port and voice feature cards permit
a much smaller range of sizes, to help ensure that end-to-end delay for voice signals does not exceed
200 ms. If your network uses a variety of gateway and router types, you may need to ensure that payload
sizes are set both optimally (so as not to incur excessive end-to-end delay) and consistently.
Tip For more information and configuration options, see Voice over IP—Per Call Bandwidth Consumption,
available online at
http://www.cisco.com/warp/public/788/pkt-voice-general/bwidth_consume.html.

Voice Activity Detection

Because telephone users generally speak in turn, a typical voice conversation contains up to 50 percent
silence. A feature called VAD (Voice Activity Detection) causes the gateway to transmit when speech
starts and cease transmitting when speech stops. During silences, it generates white noise so that callers
do not mistake silence for a disconnected call. By suppressing packets of silence, VAD enables you to
handle more calls. For VoIP bandwidth planning, assume that VAD reduces bandwidth by 35 percent.
Enable VAD if you wish to allocate more bandwidth to other types of traffic.
A possible problem with VAD is that it tends to clip the start and end of speech. To avoid activation
during very short pauses and to compensate for clipping, VAD waits approximately 200 ms after speech
stops before stopping transmission. Upon restarting transmission, it includes the previous 5 ms of speech
along with the current speech.
VAD disables itself on a call automatically if ambient noise prevents it from distinguishing between
speech and background noise.
Cisco VGD 1T3 Voice Gateway Software Configuration Guide
Enabling QoS Features for VoIP
107