Voice over ATM

Voice over ATM (VoATM) can be supported as standard pulse code modulated (PCM) voice via circuit emulation (AAL1, described later) or as variable bit rate voice in ATM cells as AAL2 (also described later). ATM offers many advantages for transport and switching of voice. First, quality of service (QoS) guarantees can be specified by service provisioning or on a per-call basis. In addition, call setup signaling for ATM switched virtual circuits (SVCs), Q.2931, is based on call setup signaling for voice ISDN, Q.931. Administration is similar to circuit-based voice networks.

However, VoATM suffers from the burden of additional complexity and incomplete support and interoperability among vendors. It also tends to be more expensive because it is oriented toward all optical networks. Most importantly, ATM is typically deployed
as a WAN Layer 2 protocol and therefore does not extend all the way to the desktop. Nevertheless, ATM is quite effective for providing trunking and tandem switching services between existing voice switches and PBXs.

Voice over Frame Relay (VoFR) has become widely deployed across many networks. Like VoATM, it is typically employed as a tie trunk or tandem-switching function between remote PBXs. It benefits from much simpler administration and relatively lower cost than VoATM, especially when deployed over a private WAN network. It also scales more economically than VoATM, supporting links from T1 down to 56 kbps. When deployed over a carefully engineered Frame Relay network, VoFR works very well and provides good quality. However, voice quality over Frame Relay can suffer depending on network latency and jitter. Although minimal bandwidth and burstiness are routinely contracted, latency and jitter are often not included in service level agreements (SLAs) with service providers. As a result, voice performance can vary. Even if quality is good at first, voice quality can degrade over time as a service provider's network becomes saturated with more traffic. For this reason, many large enterprise customers are beginning to specify latency and jitter, as well as overall packet throughput from carriers. In these situations, voice over Frame Relay can provide excellent service.

Voice over IP (VoIP) has begun to be deployed in recent years as well. Unlike voice over Frame Relay and Voice over ATM, Voice over IP is a Layer 3 solution, and it offers much more value and utility because IP goes all the way to the desktop. This means that in addition to providing basic tie trunk and tandem-switching functions to PBXs, VoIP can actually begin to replace those PBXs as an application. As a Layer 3 solution, VoIP is routable and can be carried transparently over any type of network infrastructure, including both Frame Relay and ATM. Of all the packet voice technologies, VoIP has perhaps the most difficult time supporting voice quality because QoS cannot be guaranteed. Normal applications such as TCP running on IP are insensitive to latency but must retransmit lost packets due to collisions or congestion. Voice is much more sensitive to packet delay than packet loss. In addition to normal traffic congestion, QoS for VoIP is often dependent on lower layers that are ignorant of the voice traffic mingled with the data traffic.

Voice Networking

Basic voice technology has been available for more than 100 years. During that time, the technology has matured to the point at which it has become ubiquitous and largely invisible to most users. This legacy of slow evolution continues to affect today's advanced voice networks in many ways, so it is important to understand the fundamentals of traditional voice technology before emulating it on data networks.

Traditional analog telephone instruments used for plain old telephone service (POTS) use a simple two-wire interface to the network. They rely on an internal two-wire/four-wire hybrid circuit to combine both transmit and receive signals. This economical approach has been effective but requires special engineering regarding echo.

Basic Telephony

Three types of signaling are required for traditional telephony: supervision, alerting, and addressing. Supervision monitors the state of the instrument—for example, allowing the central office or PBX to know when the receiver has been picked up to make a call, or when a call is terminated. Alerting concerns the notification of a user that a call is present (ringing) or simple call progress tones during a call (such as busy, ringback, and so on). Finally, addressing enables the user to dial a specific extension.

In addition to signaling, telephony services also provide secure media transport for the voice itself, analog-to-digital conversion, bonding and grounding for safety, power, and a variety of other functions when needed.

Analog voice interfaces have evolved over the years to provide for these basic functions while addressing specific applications. Because basic POTS two-wire analog interfaces operate in a master/slave model, two basic types of analog interfaces are necessary for data equipment to emulate: the user side and the network side. The user side (telephone) expects to receive power from the network as well as supervision.

A foreign exchange service (FXS) interface is used to connect an analog telephone, fax machine, modem, or any other device that would be connected to a phone line. It outputs 48 vdc power, ringing, and so on, and it accepts dialed digits. The opposite of an FXS interface is a foreign exchange office (FXO) interface. It is used to connect to a switching system providing services and supervision, and it expects the switch to provide supervision and other elements. (Why "foreign"? The terms FXS and FXO were originally used within telephone company networks to describe provision of telephone service from a central office other than normally assigned.)

Within FXS and FXO interfaces, it is also necessary to emulate variants in supervision. Typical telephones operate in a loop start mode. The telephone normally presents a high impedance between the two wires. When the receiver goes off-hook, a low-impedance closed circuit is created between the two wires. The switch, sensing current flow, then knows that the receiver is off-hook and applies a dial tone. The switch also checks to be sure that the receiver is on-hook before sending a ringing signal. This system works well for simple telephones, but it can cause problems on trunks between PBXs and COs with high activity. In that situation, the remote end and the CO switch can both try to seize the line at the same time. This situation, called glare, can freeze the trunk until one side releases it. The solution is to short tip or ring to ground as a signal for line seizure rather than looping it. This is called ground start.

After the line is seized, it is necessary to dial the number. Normal human fingers cannot outrun the dial receivers in a modern switch, but digits dialed by a PBX can. In that case, many analog trunks use a delay start or wink start method to notify the calling device when the switch is ready to accept digits.

Another analog interface often used for trunking is E&M. This is a four- or six-wire interface that includes separate wires for supervision in addition to the voice pair. E&M stands for "ear and mouth" or "Earth and magneto" and is derived from the early telephony days. The E&M leads are used to signal on-hook and off-hook states.

Analog voice works well for basic trunk connections between switches or PBXs, but it is uneconomical when the number of connections exceeds six to eight circuits. At that point, it is usually more efficient to use digital trunks. In North America, the T1 (1.544 Mbps) trunk speed is used, consisting of 24 digitized analog voice conversations. In other parts of the world, E1 (2.048 Mbps) is used to carry 30 voice channels. (Engineers refer to the adoption of E1 and T1 internationally as "the baseball rule"—there is a strong correlation of countries that play baseball to the use of T1. Therefore, the United States, Canada, and Japan have the largest T1 networks, while other countries use E1.)

The first step in conversion to digital is sampling. The Nyquist theorem states that the sampling frequency should be twice the rate of the highest desired frequency. Early telephony engineers decided that a range of 4000 hertz would be sufficient to capture human voices (which matches the performance of long analog loops). Therefore, voice channels are sampled at a rate of 8000 times per second, or once every 125 ms. Each of these samples consists of an 8-bit measurement, for a total of 64000 bits per second to be transmitted. As a final step, companding is used to provide greater accuracy of low-amplitude components. In North America, this is u-law (mu-law), while elsewhere it is typically A-law. For international interworking purposes, it is agreed that the North American side will make the conversion.

To construct a T1, 24 channels are assembled for a total of 1.536 Mbps, and an additional 8 bits are added every 125 ms for framing, resulting in a rate of 1.544 Mbps. Often, T1 frames are combined into larger structures called SuperFrames (12 frames) and Extended-SuperFrames (24 frames). Additional signaling can then be transmitted by "robbing bits" from the interior frames.

Basic T1 and E1 interfaces emulate a collection of analog voice trunks and use robbed bit signaling to transfer supervisory information similar to the E&M analog model. As such, each channel carries its own signaling, and the interface is called channel associated signaling (CAS). A more efficient method uses a common signaling channel for all the voice channels. Primary Rate Interface for ISDN is the most common example of this common channel signaling (CCS).

If voice/data integration is to be successful, all of these voice interfaces must be supported to provide the widest possible range of applications. Over the years, users have grown to expect a certain level of performance, reliability, and behavior of a telecommunications system, which must be supported going forward. All these issues have been solved by various packet voice systems today so that users can enjoy the same level of support to which they have become accustomed.