Packet networks have tremendous advantages because
they allow data, video, and voice to be carried over a single network with
optimal efficiency. They deliver true multi-service capabilities that give
us voice, video, and data all in one unit. Latency is a major problem with
packets carrying real-time data such as voice and video because packet
networks are, by definition, variable latency transmission devices.
Latency is the time it takes for a packet of data to travel from one point
on a network to another. Delayed, disordered, duplicate, and missing packets
cause latency. Most latency issues occur in the transmission across the
virtual circuit, where the connection appears to be a direct link, but can
actually involve routing information over a defined, but longer path in the
network. It is the major voice quality problem for packetized networks.
Unfortunately, almost every component of a VoIP system introduces latency.
The challenge is in minimizing latency to acceptable quality levels.
The quality of service (QoS) levels needed for real-time processing of
phone calls are dependent upon network conditions and can vary
substantially. Networks are defined by their bandwidth, indicating how much
information can be carried under the best network conditions. However,
bandwidth is shared by the users of a network and is subject to variations
in call performance. In addition, different voice compression/decompression
algorithms used in VoIP have different bandwidth requirements. Finally,
routers can cause quality problems in networks due to induced delays because
routing is not a continuous process.
In the gateway, the interfaces from the telephone network and IP network
can cause processing delays associated with the different speech algorithms.
Buffering and echo cancellers can also cause delays. Different speech
vocoding (voice coding/decoding) introduce latency as well. These latency
conditions are foreign to the circuit-switched world where information
physically arrives in order and on time. In the packet-based network, there
is neither fixed time nor fixed order. Additionally, the latency will be
variable and often unpredictable. There is always the possibility of packets
that are lost, out of order and delayed in IP networks.
What can be done? At the gateway where the transition between the user
and the network takes place, the transition from circuit-switch to data
networks to multi-service networks also takes place. Properly constructed
and configured gateways can minimize the effects of network latency.
Jitter is distortion caused by a lack of synchronization of signals. Jitter
can be caused by packets traveling via different paths if a router is
congested -- the higher the congestion, the greater the jitter. The speech
codec portion of a VoIP system requires data be delivered at a constant
rate. If the intervening network has a variable delivery rate, then the VoIP
system must buffer data received from the network before sending it to the
codec. This buffering increases the latency.
The effects of jitter are mitigated by the use of a jitter buffer that
stores packets and identifies their order. It interprets the network
behavior and determines the holding time of the packet. A good jitter buffer
design is adaptive to the network. The jitter buffer should be small and
flexible so that it can expand and adapt if network conditions change since
network size affects both jitter and latency. Jitter buffer tracks the rate
of change and adapts accordingly.
Gateway design can minimize the amount of buffering needed. The gateway
defines the packet size small enough to reduce latency problems, but large
enough so sufficient information is transmitted to minimize packet overhead.
Two of the more common vocoder (voice coder) standards are G.729AB and
G.723.1A, which have packet loss concealment procedures. These coders model
speech production mechanisms where a small set of parameters represent a
vocal tract model, which results in low bit-rate. The maximum latency
incurred is dependent upon the vocoder used in a particular IP system.
With lost packets, the speech vocoder will interpolate speech from past
speech frames. However, there are no clear instructions on how to use the
packet loss concealment feature, and it must be adequately called to conceal
a lost packet and hence improve speech quality. Vocoders only contribute to
a part of the latency in the gateway. There are also contributors to latency
that are not controllable in the gateway including the codec. The codec
causes latency, but there is nothing that can be done about it because it is
inherent to the applicable VoIP standards. However, gateways can be
configured to run low-latency Vocoders like G.711 as the default and to
negotiate during call configuration for the lowest latency Vocoder.
The Bottom Line
Different system implementations in the voice gateway architecture can
achieve higher quality, lower latency, and lower variance throughput. It is
important to understand how latency affects speech quality and how a gateway
can be configured to reduce latency. The main areas to look at include the
implementation of the data buffer, echo canceller, and the jitter buffer.
The jitter buffer needs to be adaptive (smart), the data buffering delays
should be minimized, and the echo canceller should not introduce any
Scott Robertson is product line manager for Remote Access Products and
Fabian Lis, senior Voice Software engineer at Analog Devices. Analog Devices
is a semiconductor company that develops, manufactures, and markets
high-performance integrated circuits (ICs) used in signal-processing
applications. Please visit their Web site at www.analogdevices.com.
To The March 2001 Table Of Contents ]