6.2.1 The Local Exchange

Modern telephone systems in North America tend to be constructed in a relatively “flat” architecture, as illustrated in Figure 6.1.¹ Previously, a different architecture with a hierarchy of five different types of switches was used, and some of the terminology is left over from those architectures. In Figure 6.1, each telephone is connected to an end office, also called a central office, local exchange, or class-5 local exchange. Local exchanges are identified by the first three digits of the seven-digit local telephone number. If two telephones connected to the same end office talk to each other, then the signal goes no farther than that end office. End offices connect to other end offices and to tandem offices, or intraLATA toll offices. Tandem offices are the location at which operator services are normally provided (“provisioned” in telephone terminology). These tandem offices connect calls to other end offices and also to long-distance carriers, called interLATA carriers. Note the distinction between intraLATA and interLATA.

Figure 6.1 Modern telephone system architecture.

End offices employ so-called class 5 switches, which are designed to support subscriber features described later in the chapter. Class 4 switches are optimized for long-distance trunking and are used in tandem offices.

The term LATA stands for local access and transport area. It is the area in which a local telephone company transports long-distance calls itself without passing the call to a long-distance company. If a call is between telephones in the same LATA, then the local telephone operating company carries the call on its own facilities (intraLATA) rather than handing it to a long-distance carrier (interLATA).

In North America, switches communicate with each other using a protocol known as signaling system number 7 (SS7). This is an out-of-band signaling system used to provide basic routing information, call setup, and other call termination functions. It operates on a data network independent of all voice channels.

6.2.2 Switch Architecture

Figure 6.2 illustrates the architecture of a modern digital switch. The set of peripheral interfaces shown at the bottom of the figure convert external signals to digital form if they don’t already exist that way. They also perform some low-level call processing. Signals are passed to the switching matrix, which performs the function of connecting each call as appropriate. Within the peripheral interface set, line controllers interface with individual telephone instruments where they are routed into the exchange. These are ordinarily analog inputs, which must be converted to digital. DLC interfaces connect with digital loop carrier systems, a concentration system explained in Section 6.2.3. Trunk controllers interface with other end offices and tandem switches.

Figure 6.2 Architecture of modern switch.

Voice channels are handled digitally in modern switches. The basic digitization method converts each voice channel in each direction to a 64-kb/s datastream. This is explained in Section 6.3.1.

The input/output controller system provides access to the switch for maintenance, billing, routine operations and administration, and loading of software. The central processor is usually duplicated to ensure reliability. It controls call-processing activities, such as assigning time slots, and administering features, such as call forwarding.

6.2.3 Digital Loop Carrier

Traditionally, telephone instruments have been connected directly to the switch in the local exchange. However, this requires a lot of copper wire to run between homes and the exchange. A more efficient way of interconnecting telephones and switches has been developed in the form of digital loop carrier (DLC) systems. Bulk transport is used between the exchange and a point in the field closer to the users. From that point, individual copper pairs serve each subscriber.

Figure 6.3 may be compared with Figure 6.1, which showed a telephone connected directly to the switch at the exchange. A DLC system adds a concentrator in the field, which multiplexes a number of subscriber lines onto a single output line going to the switch. The concentrator interfaces with the DLC interface at the lower center of Figure 6.2. The DLC converts the analog telephone signals to digital, as do the line controllers in the switch. Rather than being located in the switch, the DLC is located in the field, close to the subscriber.

Figure 6.3 Integration of digital loop carrier into system.

Two interfaces are commonly used for DLC systems in North America; other interface specifications are used elsewhere. The more established interface is TR-08,* which is a DS1 interface (see Section 6.3.4) consisting of 24 time slots, each of which can handle one telephone call. The newer interface specification is TR-303, a wire line and fiber-optic (SONET) interface. It can support up to 2,048 lines and many telephone features.

Modification of the DLC Concept to Accommodate Cable Switched Circuit Telephony

Figure 6.4 illustrates the way in which a DLC architecture is modified for inclusion of switched circuit telephony over cable television systems. Later sections will cover more information relating to cable telephony. Section 5.6 will cover IP telephony.

Figure 6.4 Modification of DLC architecture for cable switched circuit telephony.

Compare Figure 6.4 with Figure 6.3. The concentrator of Figure 6.3 is located in the field and collects calls from a number of lines, concentrating them onto one cable for interface with the switch. The host digital terminal (HDT) of Figure 6.4, in the headend, serves the same function and “looks” to the switch as if it were a concentrator. In turn, the HDT interfaces with the telephones connected to the hybrid fiber-coax (HFC) network.

An interface is required at each home subscribing to cable telephony services. This home interface supplies analog signals to the telephones in the home, just as would a switch or a DLC system. It is known by many names, but here we use the generic term network interface device (NID) to mean the equipment at the home that interfaces between the HFC network and the subscriber’s telephone instruments. Other names by which the interface is known include network interface unit (NIU), customer access unit, and wallbox.

NID is also sometimes used to mean a plastic or metal box, usually mounted on the side of a house, that contains protection equipment and the subscriber interface port to the telephone system. The term is used in the cable industry at times to describe a similar facility used to house the ground block, maybe a splitter, and maybe an amplifier. It is possible, but not mandatory, to place the telephony interface equipment in a NID according to this definition. In this chapter, we use the term to mean the equipment used to interface between the subscriber telephone equipment and the cable plant.

In general, you would expect to split the cable in front of the NID in order to provide other cable services. The splitter may be provided as part of the NID, or it may be supplied externally. It is possible, and some say desirable, to place a high-pass filter in the “other cable services” side of the splitter output to prevent leakage of spurious signals into the cable plant. This topic is covered in Chapter 16.

6.2.4 Backup Powering

The telephone industry tends to power from a −48-volt dc battery that is float charged from commercial power so that the batteries take over when commercial power fails. Some systems are powered from −130-volt dc. In addition, remote access vehicles are available that have on-board batteries to ensure the survivability of power if commercial power is lost for an extended time. Central offices and major field installations have generator backup and battery backup sufficient to hold the office during generator start-up.

6.2.5 POTS Through CLASS

A number of types of telephone service are available today. For residential service, we can talk about the services that are universally offered to all subscribers, as well as suites of upgraded services. The classical services, or plain old telephone services (POTS), include rotary and tone dialing, voice connection, operator services, access to long distance and international long distance, and in most locations in North America, 911 emergency service.(In most of North America, it is possible to summon police, fire, ambulance, or other emergency services by dialing 911. An emergency operator is presented with a screen showing the location from which the call is placed and pertinent information relating to that location.) We would expect any cable-based telephony system to provide POTS.

Advanced features allow more services and new revenue streams for telephone system operators. A second set of services, custom calling services (CCS), also called custom messaging services (CMS), work with any touch-tone tele-phone. The services include call waiting, call forwarding, three-way calling, speed dialing, added directory number (often called teen service) with distinctive ringing, multiline custom calling, and extension-bridged services. Some or all of these features may or may not be supported by cable-based telephony systems.

Custom local area signaling services (CLASS) are a set of advanced services provided by modern digital switches and the expanded intercommunication between them. These services include automatic callback, automatic recall, calling number delivery (also known as caller ID), customer-originated trace, distinctive ringing, selective call forwarding, and selective call rejection. Table 6.1 summarizes CLASS features.

Table 6.1 CLASS features

Many CLASS services depend on delivery of the number of the calling party. The number is delivered to the called phone before it is answered. This necessitates transmission of the calling number between applications of the ringing voltage.

Besides POTS, custom calling, and CLASS, other services are often provided by local telephone companies. Network-based voice mail takes the place of answering machines, with the added advantage of being able to answer incoming calls when the phone is in use. Voice messaging, voice menus, fax messaging, and voice forms are offered, primarily to provide small businesses with big-business types of features.

Integrated services digital network (ISDN) services allow digital transmission on the normal copper pair serving businesses and residential users. So far, ISDN has had most of its impact on businesses. It allows simultaneous voice and data transmission on the same copper pair. Basic rate ISDN operates at 144 kb/s and allows two 64-kb/s voice channels (b, or bearer channels) and one 16-kb/s data channel (d channel). Primary rate ISDN operates at nearly 1.5 Mb/s and offers 23 b channels and 1 d channel.

6.2.6 The Subscriber Loop

Here we present more information on the operation of the subscriber loop, the portion of the telephone system that connects to the telephone instrument.² This is of interest to the HFC telephony technologist because it describes the interface between the NID and the telephone instruments in the home.

Residential telephone service is provided using a single twisted pair line, which carries signaling as well as voice in both directions. The two wires are referred to as tip and ring — from the Model 310 phone jacks used to make circuit connections in the days when the telephone system was operated manually. The 310 jack is also known as a phone jack and is used for many audio applications. If a third connection is added for stereo, that connection is called the sleeve.

Call Progress Signaling Tones

As the call progresses, a number of tones are used to effect signaling and to indicate the status of the call. Figure 6.5 illustrates the call progress tones commonly used in North America, and also includes the DTMF tones. Table 6.2 describes each call progress tone, as well as a few other tones of interest.

Figure 6.5 Call progress and DTMF tones.

Table 6.2 Call progress tones

6.3.1 Sampling of the Voice Signal

The first step in digitizing a voice signal is to sample it. The samples are taken in order to give an analog-to-digital converter a fixed level to sample. In order to promote good-quality voice transmission, it is known that you must transmit, as a minimum, frequency components from about 300 to 3,000 Hz. A somewhat wider range adds naturalness to the voice. It was shown by Harry Nyquist of Bell Laboratories that the minimum number of samples per second that could be taken and still allow the signal to be reproduced was equal to twice the highest-frequency component of the signal.⁴ It is essential to filter out any components exceeding one-half the sampling rate. If a higher-frequency component is sampled, then the spectrum of the sampled signal will fold back on itself, resulting in “aliasing” and a very distorted recovered signal.

The signal is sampled at some instant of time, and the sampled level is held in the capacitor, shown in Figure 6.6, for the time required to digitize the voltage level. Then a new sample is taken, and the process begins again. At the top, the signal is sampled, digitized, and sent through the “system” (in this case, the telephone system), then converted back to analog. The sampling process may be accurately modeled as an amplitude modulation process, in which the “carrier” is a square wave that goes between the value 0 (switch open) and 1 (switch closed). The rate at which the signal is sampled in the telephone system is 8,000 samples per second (8 ks/s), resulting in a Nyquist bandwidth, the bandwidth that can be reproduced, of 4 kHz. The practical upper frequency of the voice signal is lower than 4 kHz so as to allow for filter rolloff.

Figure 6.6 The sampling process.

The spectrum generated by the sampling process is shown at the bottom of Figure 6.6. The baseband spectrum is present even after sampling. In addition, the first harmonic of the “carrier,” at 8 kHz, is shown, surrounded by the normal modulation sidebands, which are due to the audio. The second harmonic, at 16 kHz, is also present with sidebands, though not shown here, and so on. Note that the spectra of the baseband signal and the harmonics do not overlap.

6.3.2 Analog-to-Digital Conversion

After sampling, the voice signal is converted to a digital representation. It has been found that a suitable minimum signal-to-noise ratio for voice transmission is approximately 30 dB. (The definition of signal-to-noise ratio is somewhat different from that used in video systems.) We can analyze the signal-to-noise ratio of a digitization process by considering the effect of an error in the least significant bit (LSB) of the digitized signal. It may be shown that a system for digitizing analog signals exhibits an encoding error (“quantizing error”) of one least significant bit. This is taken as the quantizing noise of the system. Quantizing noise acts somewhat the same as the random thermal noise familiar to cable engineers in that it masks the true value of the signal by creating an uncertainty in the value.

Telephone signals are digitized to eight bits. Since 2⁸ = 256, this process represents the analog signal as 1 of 256 levels. One minimum increment is 1 of these 256 levels, so the quantizing signal-to-noise ratio is

This provides an acceptable safety margin from the minimum deemed necessary. In order to further improve voice quality, the analog-to-digital conversion process utilizes so-called mu-law encoding, in which the digital states don’t represent constant voltage increments of the analog waveform. Low-level signals are encoded with less distance between states than are high-amplitude signals. By doing this, we can gain apparent signal-to-noise ratio with fewer bits. Various encoding laws (values of mu) are used in different countries. Outside North America, the A-law is used for encoding.

6.3.3 Time Division Multiplexing in Telephony Systems

As we stated, a voice channel is sampled 8,000 times a second, and each sample is encoded to 8 bits. The data rate required to transmit the signal is thus 8 × 8,000 = 64 kb/s (kilobits per second). This data rate is a basic rate of the time division multiplexing hierarchy used in digital telephony systems. The rate is called the DSO (pronounced “D S zero”), where DS stands for digital signal. Thus, when a telephone engineer speaks of a DS0, he or she means a 64-kb/s data signal.

A number of signals may be multiplexed to allow carriage of many signals on a single transmission path. In the standard North American hierarchy, 24 DSOs are multiplexed together to form a DS1, which operates at 1.544 Mb/s. The 24 DS0s require 1.536 Mb/s (megabits per second), whereas the remaining 8 kbps are overhead. This DS1 datastream is often called a T1 datastream, though technically T1 applies only when the transport medium is a twisted pair of wires.

European telephone companies tend to use a different multiplexing strategy in which 30 calls are multiplexed in a 2.048-Mb/s datastream, known as an E1 stream. The data rate of 2.048 Mb/s is fast enough to allow two more voice channels, but those channels are reserved for signaling (network control). The E1 data rate is sometimes used in North America internal to a system.

6.3.4 Asynchronous Digital Hierarchy

Figure 6.7 illustrates the hierarchy used in North America for digital multiplexing of asynchronous digital signals.⁵ The basic DS0 datastream may represent one data source, such as a telephone call, in which case it is called DS0-A.⁶ It may also comprise several lower-data-rate channels, in which case it is called DS0-B. (The terms DS0-A and DS0-B are used only for digital data service protocols, not for voice work. For voice work, we usually refer to simply DS0.) Twenty-four such signals are multiplexed, along with 8 kb/s of synchronization information, to form a DS1 signal. This 1.544-kb/s signal is very commonly encountered, but is slowly yielding to higher-rate asynchronous signals, and even to higher-rate synchronous signals. Issues related to handling DS1 signals are covered in Section 6.8. Four DS1s may be multiplexed to form a DS2. Only rarely will you encounter a DS2: it is usually just an intermediate multiplexing level in a system that ultimately multiplexes to an even higher level. Seven DS2s may be multiplexed to form a DS3, which can carry 672 simultaneous telephone conversations, or other digital data at the same rate. For example, television networks sometimes use a DS3 to transport a single digital video signal that has been compressed using algorithms most often based on linear predictive coding.

Figure 6.7 North American asynchronous digital hierarchy.

Notice that we have defined only a signal format. We have said nothing about the medium on which the signal is carried. The medium may be twisted pair (at least up through DS1, which is known as T-1 when carried on two twisted pairs — one in each direction), coaxial cable, microwave radio, optical fiber, or any other suitable medium.

The TR-08 interface to a telephone switch is based on DS1, several of which may be interfaced from one DLC multiplexer. In Europe and other places, a similar interface is known as the channel associated signaling (CAS) interface.

6.3.5 North American Synchronous Digital Hierarchy

A newer format used for optical signaling is known as synchronous optical network (SONET) in North America. In much of the rest of the world, the format varies slightly and is called synchronous digital hierarchy (SDH). The word synchronous in this context refers to transmission of signals that have a constant bit rate and are derived from a common clock. It may be that the signals are initially developed from different clocks (though this is unlikely), but then they would have to be synchronized before being multiplexed. It builds on the asynchronous hierarchy shown earlier but is extended to much higher speeds. Extra bits are included to allow the data to “float” in the time slot. This means that if the data arrives after the allocated time slot begins, it can be accommodated if it is not too late, in which case it would wait for the following allocated time slot.

Figure 6.8 illustrates the synchronous digital hierarchy. Although there are variations in the multiplexing options, it is convenient to think of the starting point as a DS3 signal at 44.736 Mb/s. This signal is up-converted to a 51.84-Mb/s datastream by adding some low-rate data and/or “stuffing” extra bits around the payload (the DS3), to add up to the desired number of bits per second. This datastream, 51.84 Mb/s, is known as STS-1 data, where STS stands for “synchronous transport signal.” When converted to optical form, the signal is known as an ОС −1 signal, where OC stands for “optical carrier.” As opposed to cable television practice, the baseband STS hyphen; signal on/off modulates the laser without use of an RF carrier.

Figure 6.8 Synchronous digital hierarchy.

Signals can be multiplexed further. If the signals to be multiplexed are in optical form, the multiplexer usually converts them to electrical form to multiplex them, then convert back to optical form. Optical switching and multiplexing is occasionally used. The next data rate is three times the ОС −1 data rate, 155.52 Mb/s, and is known as an OC-3. One OC-3 is capable of carrying up to 2,016 telephone calls or an equivalent amount of other data. It is possible to use even higher data rates. Commercial equipment is available at this writing to support OC-192 (9.953 Gb/s) transmission.

The TR-303 switch interface mentioned earlier is capable of operating with either synchronous or asynchronous interfaces. In Europe and some other places, the equivalent interface is known as the V5.2 interface. Unfortunately, V5.2 and TR-303 are incompatible.

6.4.1 The Home Interface

A transition is necessary between the signals on the HFC network and those required to work with standard analog telephone instruments. This equipment may be located in the home, but is often mounted on the side of the house, where it forms an interface between the cable plant and all cable services located in the home. The box on the side of the home is called a network interface device (NID). Other names have also been given to the box.

Figure 6.9 illustrates the basic functions of an HFC telephone NID. Also shown is a portion of the HFC network, from the optical node, through a bidirectional RF amplifier, the tap at the subscriber’s home, and the NID. Cable signals are routed through the NID, where a directional coupler extracts some energy for the telephony system. The rest of the power in the signal continues to the home to provide other cable services. In this path, the manufacturer may choose to provide a service disconnect switch, an amplifier, and/or a filter to remove upstream energy coming out of the home.

Figure 6.9 Telephone interface at home.

From the directional coupler, the signal is routed to a diplex filter, which separates the downstream frequencies above 54 MHz, from the upstream frequencies, usually below 40 MHz. The high-frequency output from the diplexer is supplied to a receiver, which receives the downstream telephony signal. The automatic gain control shown is conventional and corrects for any variations in signal level as a result of system operating point.

The low-frequency side of the diplexer is used as an input for the upstream signal that comes from the NID transmitter. The attenuator, AT4, controls the output level from the transmitter. As explained in Chapter 16, a long loop ALC system is used, in which the level received at the headend is compared with the expected level, and attenuator AT4 is commanded to adjust the transmitter output level until the level received at the headend equals that expected.

The modem logic is responsible for tuning the transmitter and receiver to the correct frequencies, using time slots (or equivalent) assigned from the HDT, adjusting transmit level according to instructions received from the headend, format conversion, and control of the NID. The interface circuitry provides standard analog signals to the telephone instruments in the home. It provides 24- or 48-volt dc battery voltage, ringing voltage, and dial tone, as well as handling conversion of voice signals between the analog and digital domains.

One or two lines are typically supported in a NID. It is possible in some systems to replace the analog telephone function for one line with a 64-kb/s data channel. A different subscriber interface is required, as is a different interface at the switch.

6.4.2 Powering the NID

A very important consideration involves powering the NID. At the time of this writing, four powering philosophies are being considered by the industry:

A principle embedded in the first three powering philosophies is that telephone service should be maintained in the event of commercial power failure, as is done by the incumbent phone company. This requires some sort of battery power and possibly generator power also. The fourth philosophy implies that backup power is too hard and too expensive to supply, so the cable company will not attempt to provide lifeline service and will concentrate on serving the second-line market.

Early generations of NIDs drew 5–10 watts of power. The industry would like to see power drain reduced to 1 or 2 watts. As the power drain decreases, network powering of some sort becomes more palatable. Typically, a NID will draw less power at idle, more while in use, and the greatest power when ringing.

6.4.3 Headend Telephony Equipment

Figure 6.10 illustrates a headend configuration to support a switched circuit HFC-based telephony system. Variations abound. The downstream HFC plant receives signals from a transmitter, typically part of a bank of transmitters and receivers (modems) dedicated to the telephony system. Typically, the downstream signal will be located toward the higher end of the downstream spectrum, perhaps between 350 and 750 or 870 MHz.

Figure 6.10 Headend elements of telephony system.

A cable operator may own and operate his or her own switch, or may lease access to the incumbent telephone operator’s switch. The transmitter and receiver interface to a telephone switch using one of the interface protocols (TR-08 or TR-303 in North America) described in this chapter. The switch may be co-located with the HFC telephony equipment, or the two may be interconnected with a suitable link, such as a DS1 (TR-08) or SONET (TR-303) interconnection.

6.4.4 Channel Sharing

For the sake of efficiency, it is often desirable to share one communications channel between a number of users. This may be done in many ways, which we describe in more detail in Chapter 4. The review here is oriented toward a timesharing protocol, which is common and related to that used in baseband data communications, as outlined in Section 6.3. In the downstream direction, time division multiplexing (TDM) is used. Each of the multiple (typically 24 or 30) lines associated with the data carrier is assigned one time slot. Data is transmitted in time slot 1 to line 1, then to line 2 in time slot 2, and so on.

In the return direction, a related concept is used, except that the multiplexing must be done by turning off one transmitter, then turning on another. This is called time division multiple access (TDMA).

In typical telephony practice, the data in either direction is divided into packets, or data sets that are transmitted at one time. Each packet will contain several samples (frequently eight) of a single phone call. The transmission of an entire set of packets from each allocated user is called a frame (analogous to the television usage, where a frame is one complete picture). Commonly, frames are repeated every 1 millisecond, a frame rate of 1,000 per second. In a frame, there will be 24 packets (if the system accommodates 24 lines), each packet containing 8 samples of 8 bits each, plus overhead.

6.4.5 A Sample HFC Telephony Marshaling Protocol

This section deals, in a fairly general manner, with some of the issues that are addressed when a new NID is added to the system. The exact details of what must be done, and the order in which the steps are taken, are a function of the particular system. This section is intended only to illustrate what must be done. The frame of reference is a TDM/TDMA system, in which the downstream data consists of data supporting, typically, 24 to 30 lines simultaneously.

TDM/TDMA Summary

Although not all systems will use TDM/TDMA for transmission on the cable, many early entry systems are using it. This section summarizes and illustrates some of the concepts described previously.

Figure 6.11 illustrates several significant aspects of a typical TDM/TDMA signal. The downstream TDM signal is illustrated in Figure 6.11(a). In this example, it consists of 24 time slots, each of which can accommodate one telephone call in the downstream direction. The frame starts with a header, which marks the start of the frame. The frames are 1 ms long, for a total of 1,000 frames per second. Since each telephone call is sampled 8,000 times each second, each time slot must contain a packet of 8 samples, each 8 bits in length. In addition, the time slot must include overhead for operating the NID. Instructions to ring the phone, supply dial tone, adjust upstream level, reconfigure (reprovision) itself, accept a new software download, and other instructions must be sent to the NID. Because of this, a protocol that supports 24 lines will likely operate at a higher data rate than 1.544 Mb/s, the DS1 rate described earlier, which supports 24 lines. The next standard rate available is the E1 rate of 2.048 Mb/s.

Figure 6.11 TDM and TDMA transmission. (a) Downstream TDM signal stream. (b) Upstream TDMA signal stream with time offset measurement.

Note that in a TDM system, it is not necessary to allow time between individual packets (time slots). The header defines where each bit occurs between it and the following header. It is necessary only to count the number of bits that have been received since the header to know which bit is now being received.

Contrast this to the TDMA datastream used in the upstream direction (see Figure 6.11(b)). Packets from individual return transmitters “parade” back to the headend. The same packet structure as in the downstream data may be used, but it is necessary to allow some small time between the packets. This time allows one transmitter, which has sent its packet, to turn off, and the next transmitter to turn on and stabilize before it begins transmitting its packet. Because of the need for dead time, it is necessary either to transmit at a higher instantaneous bit rate during a packet or to not transmit some bits in the upstream direction.

Shown in the TDMA signal is a marshaling time slot, which may be reserved, at least at times, for marshaling a new NID. Some options for marshaling the NID were discussed earlier. In one possibility, a new NID transmits in the time slot reserved for marshaling. The problem is that the range (signal transport delay) from the NID to the headend is not known initially. If the assumed range is in error by as little as 1 mile or so of fiber, it is possible that the response from the marshaling NID will fall outside the allocated time slot, interfering with another telephone conversation.

6.4.6 Implication of Diverse Routing

In order to improve the availability of cable telephony systems, some operators are installing redundant node receivers and transmitters, and connecting them to the headend using fiber-optic cable having diverse routing. Figure 6.12 illustrates use of a redundant node, which has two optical receivers and two optical transmitters. (Chapter 18 includes more information about plant architectures.) Normally, the selection switches are in the positions shown, coupling the primary receiver and transmitter to the headend. If the received signal is lost, this fact is detected, and the switches thrown to the opposite position, to couple to the backup equipment. The loss of signal may be based on either loss of light or loss of RF signals. Normally, it is assumed that if downstream light or RF is lost, the reason is a cut in the fiber cable, so the reverse path is also affected. For this reason, most redundant systems cause the upstream switch to follow the downstream switch.

Figure 6.12 Redundant node with diverse fiber routing.

Typically, a telephone system will tolerate loss of the link for much more than 50 ms before the call is dropped, so it is an accepted practice to detect the failure and effect switching in less time than this. The backup path is almost certainly a different length from that of the primary path. This means that the marshaling of all NIDs will be wrong when the backup path is selected. The problem is only a translation in time if all the NIDs are similarly affected by the change in propagation delay. However, it is possible to split a headend telephone modem, with some of its NIDs in one node and some in another. If this has been done, then complete remarshaling will be necessary.

One proposed solution is to build two marshaling tables into the modem logic and to notify it when the backup route is in use. This means that, at some time before the backup route is needed, it will have to be activated and a second marshaling table built. If changes are made in the backup routing, then the process will have to be repeated. If the same modem serves more than one node, then backup tables must be implemented for each node served, and the modem logic will have to be notified of which node has transferred to its backup path.

6.4.7 Alternative Transmission Methods

Most of the preceding discussion was based on TDM/TDMA transmission protocols. These are currently in widespread use, but other methods have been introduced to the marketplace as well. Alternatives include SCPC, CDMA, and PFDM. They and more are covered in Chapter 4.

6.5.1 Call Concentration and Blockage

In order to achieve optimum performance in terms of cost and service level, it is important to concentrate calls. That is, since not every telephone is used all the time, the time slots on the HFC plant and the time slots in the interface with the switch may be shared between users. Design of concentration is part of the job of the network engineer. The complete job of network engineering is beyond the scope of this chapter, but information should be available from the chosen telephony vendor. Only the basic principles are presented here.

In HFC telephony, concentration may be applied independently to two different parts of the network. The interface between the HFC telephony system and the switch (if it is a TR-303 interface) can be concentrated. This path is typically shared by a number of headend modems, so the universe over which concentration can be achieved is large.

The other point of concentration is in the RF plant. It is not necessary to assign a permanent (“nailed up”) time slot to each line, but rather, the time slot each line occupies can be assigned on a call basis. By allowing all lines to have access to the entire universe of time slots, significant RF concentration may be achieved. When a line needs a time slot (that is, is involved in a phone call), any time slot not previously in use can be assigned to it. Lines not currently active in making a phone call are not tying up a time slot.

Figure 6.13 illustrates call concentration. The concentration group illustrated is too small to be of practical significance, but it illustrates the concept. Each call is represented as a rectangle, with time on the horizontal axis. The first number in each call identifies the telephone: telephones 1 through 5 are involved. The number after the dash is the sequential call number from that telephone. Telephone 1 made two calls, telephone 2 made one, and so on.

Figure 6.13 Call concentration. (a) No concentration. (b) 5:3 concentration.

In Figure 6.13(a), each telephone is assigned its own time slot (or spreading code, or frequency, or copper pair, or whatever resource is to be shared). Concentration is not practiced in this case. Since five time slots are dedicated, the total capacity is 5 times 36, or 180 CCS, the resource dedicated to handling calls from these five phones. The total call time occupied is 68.4 CCS, so the efficiency is 68.4/180 = 38%.

Figure 6.13(b) illustrates what happens if only three time slots are dedicated to these five telephones. Note that call 3-1 was assigned to time slot 1, the only slot open when that call was initiated. When call 3-2 was initiated, it was assigned to time slot 3. Call 5-1 was assigned to time slot 3, but when call 5-2, involving the same phone, was placed, it was assigned to time slot 1. In this example, since only three lines are dedicated to serving these same five phones, the total resources tied up were only 108 CCS, a savings of 40% in resources. Because of this reduction in resources, the efficiency with which the resources were utilized increased from 38% to 58.3%.

The penalty, though, is that call 1-1 was blocked. When it was initiated, all three lines were busy, so instead of a ring, the caller got an all-trunks-busy signal (see Table 6.2) even though the phone was free. Of the eight calls considered, one was blocked, so the blocking rate was 1/8 = 12.5%.

Real concentration groups should be larger than that illustrated in Figure 6.13 in order to provide the best concentration. The larger the concentration group, the better the concentration because statistics are working for the network engineer: the more lines available, the greater the likelihood that one of them will be available when needed. Figure 6.14 illustrates this principle.⁷ This curve plots the percent blockage accepted on the horizontal axis and the percent of maximum capacity utilized on the vertical axis. Percent of maximum capacity is computed by dividing the number of CCS actually carried by the number of CCS of capacity represented by the number of channels available. In order to convert this to the number of subscriber lines, the network engineer must determine the average CCS load offered by each line.

Figure 6.14 Traffic load versus blocking for differing capacities.

If the concentration group is served by only one DS1 (capable of 24 simultaneous conversations), then, depending on the blocking tolerated, the efficiency of utilization of that resource ranges from 50% to 68%. On the other hand, if the concentration group is large enough to support 12 DS1s (288 simultaneous calls), then the efficiency will range from 85% to 92% for the same range of tolerated blocking. The reason is that, as the number of lines increases, the probability of blocking goes down simply because there are more possibilities of finding an unused time slot somewhere in the group.

This efficiency speaks toward designing larger telephone nodes. Large nodes introduce other problems, however. First, large nodes allow more noise concentration, reducing the probability that the reverse path will work properly. Second, the larger the node, the more reverse spectrum will be required to operate the telephone system. A typical practical number for first-generation telephony systems is that a group of 24 to 30 channels will require about 2 MHz of both upstream and downstream spectrum.*In the preceding example, if only a single DS1 is required to service a telephony node, then 2 MHz of reverse spectrum will suffice (not including backup spectrum for use if the primary spectrum becomes corrupted). On the other hand, if 12 DS1s are used, then 24 MHz will be required. With some spectrum held in reserve (to allow for ingress), this could easily consume all the usable reverse spectrum. Network engineering on the HFC network involves trade-offs in the size of the nodes, the tolerated noise, the spectrum allowed for telephony, and the quality of service (percent blocking tolerated).

6.6.1 A Simplified Description of VoIP

Figure 6.15 offers a simplified comparison of the functionality of a switched circuit telephone system and a VoIP system. The VoIP system shown is what is known as a client-server system, the type of system that the cable industry is installing. There is another type of VoIP system, known as a peering system, which will be described briefly later.

Figure 6.15 Simplified comparison of switched circuit (a) and VoIP (b) telephone systems.

In the switched circuit system of Figure 6.15(a), the switch handles all call functions and also the signals to and from the end telephones. As described earlier in this chapter, the telephones may be connected either directly to the switch or through a digital loop concentrator (DLC). In either event, a dedicated pair of wires is assigned to each phone number from the switch or DLC to the phone.

As an example of a switched circuit call, consider a call placed by subscriber A to subscriber B. A picks up the phone, which notifies the switch to place a dial tone on the line. A then dials B’s number. The switch has to know which of thousands of wire pairs, or circuits, is B’s connection. It then places a ringing voltage on B’s line and a ring-back tone on A’s line. When Вpicks up the phone, the switch removes the ringing voltage on B’s line and removes the ring-back from A’s line. It then completes a voice circuit in both directions from A to Вand Вto A and maintains this connection for the duration of the call.

Now consider the client-server VoIP system shown in Figure 6.15b. Several devices are involved in completing the call, as will be explained in more detail later. The softswitch (server) sets up the call, but then it is not involved again until the call is completed. The router shown is the same router that’s handling other data packets to and from customers. There is no separate wire pair connecting each subscriber’s telephone to a central point. Rather, the signal is carried on coaxial and fiber cables, along with data and television signals.

The customer gateway may consist of a cable modem plus specialized circuits (the client) to connect to the telephone. Data from that subscriber’s client is multiplexed with other data from the subscriber plus data from other subscribers and sent to the headend. The telephone signal is multiplexed with other data, frequently but not always using TDM/TDMA (see Chapter 4). This is the first point of simplification for VoIP, as compared with switched circuit telephony. No separate circuits are used to connect the subscriber to the central office (in fact, in classical terms there is no central office). Rather, the telephony data traverses the same circuit as does all other data. The inherent multiplexing capability of cable modems drastically reduces the number of router ports needed.

When A wishes to call B, A picks up the phone. This sends a packetized signal via the router to the softswitch (the server), which takes on some of the functionality of a normal telephone switch. If the subscriber is authorized, the softswitch sends a packet telling the customer gateway (the client) to apply a dial tone to A’s telephone. A dials B’s phone number, which is sent to the softswitch. The softswitch obtains the IP address for Вand sends that address to A’s customer gateway. The softswitch then sends a packet to B’s gateway, telling it that A is calling. B’s gateway applies a ring voltage to B’s phone and provides caller ID if that service is included. At the same time, the softswitch instructs A’s gateway to play ring-back to A’s phone. When Вanswers, B’s gateway removes the ringing voltage and sends a packet informing the soft-switch that Вhas answered. A and Вthen communicate directly by sending voice packets back and forth over the IP network represented by the router. Note that the softswitch is not involved in the voice call itself and that A and Вnever communicate directly with each other during call setup. The softswitch retains information that the call is in progress, so someone else trying to call either A or Вwill get a busy signal.

The phone call progresses as each end sends packets of voice signals to the other end. The voice signals are usually embedded in IP packets using a suitable layer 4 protocol, such as User Datagram Protocol (UDP) (see Chapter 5). The phone call packets are multiplexed along with all other packets to and from a number of customers, using the same multiplex structure as for all other data. The router is already needed to serve data purposes, and the telephone call is simply a few additional packets that must be transported.

The softswitch is nothing more than a big database, which must have up-to-date information on the phone number and identification of all subscribers. Whereas a class 5 switch is located in a neighborhood and typically serves a few thousand subscribers, the softswitch, depending on its design capacity, can serve many more subscribers and may be located at any convenient point in the network. Once the call is established, the softswitch’s job is done, and it can go on to set up another call. Call packets do not flow through it; rather, they flow through the router, as do all other packets. There is no traditional switching fabric; instead, the switching fabric is replaced by routing within the network of routers that is provided for other data services anyway.

Of course, real systems are somewhat more complex than this simplified picture. Most calls don’t originate and terminate on the same switch (switched circuit), nor are they carried through only one router (VoIP). The VoIP system must interface to the PSTN to transfer calls not destined to another telephone on the same system, and both must connect to long-distance switched circuit trunks today in order to complete long-distance calls. While it is possible to complete long-distance calls via the public Internet, this is not a common practice, due to quality of service (QoS) issues, which are discussed in Chapter 5.

6.6.2 A Practical VoIP System

Figure 6.16 illustrates a practical VoIP system based on the Multimedia Gateway Control Protocol (MGCP).⁸ Where several names are in common use for the same block, we have shown all of them. The term frequently used in the PacketCable specifications is italicized (there is some inconsistency in the terminology used in the PacketCable specifications). MGCP is a popular protocol for VoIP systems. CableLabs has defined a subset (plus extensions and modifications) to be the standard PacketCable protocol for signaling between network platforms, called the Trunking Gateway Control Protocol (TGCP). Another subset of MGCP, that part used for the signaling between the subscriber’s phone and the softswitch, was adapted and named network call signaling (NCS). NCS is used with TGCP.

Figure 6.16 Practical VoIP system based on MGCP.

Referring to Figure 6.16, start with the telephone in the subscriber’s house. It communicates with a customer gateway, the client, which contains the conversion equipment to translate signals to the NCS format used on the cable plant. Customer gateways and other points at which the voice packets can end are known as endpoints. The gateway may be built into an internal modem, or it may reside on the outside of the house, with a DOCSIS 1.1 or higher cable modem (if used). The telephone uses the common analog interfaces described earlier in this chapter, including analog voice, DTMF tones, switch hook status, and caller ID functions. Among other things, the customer gateway will contain interface circuits, often called a subscriber line interface circuit (SLIC) and a subscriber line audio-processing circuit (SLAC). The SLIC provides the battery feed, overvoltage protection, ringing, supervision, coding, hybrid transformer, and testing (BORSCHT) functions for the telephone. The SLAC processes voice-band analog signals into pulse code modulated (PCM) outputs and converts the PCM inputs into analog outputs.

The plant-side interface from the gateway in the example is an Ethernet interface bearing IP packets, with MGCP or TGCP as the call management protocol.

(Ethernet is used to connect to the modem. If the customer gateway were integrated with the modem or other network interface, Ethernet would not be needed.) The customer gateway may employ voice compression to reduce the bandwidth, but compression is not advisable from a quality perspective. (See the later section on compression in voice circuits). The cable modem connects to a CMTS at the headend, and in turn the CMTS interfaces with the router facility. One of the devices connected to the router is a server called a softswitch, or, in PacketCable parlance, a media gateway controller. This is the device described earlier, which contains the intelligence of the VoIP system. It has a database associating phone numbers with endpoint names for subscribers on the network. Endpoint names are constructed in the same manner as are email addresses. Of course, there must ultimately be a translation to an IP address to permit delivery of packets. It is possible to associate phone numbers with IP addresses, but using endpoint names enhances the reliability of the network by allowing multiple softswitches to be available and to concentrate the IP address database in one place (the DHCP server — see Chapter 5).

For subscribers residing on the VoIP network, the softswitch instructs them to send voice packets to each other; the softswitch then has nothing further to do with the call except to be informed when it ends. The packets carrying voice communications between callers are collectively known as the bearer channel. Contrasted to the bearer channel is the signaling channel, or signaling plane, which is the collection of packets used to set up and tear down a call.

For subscribers not on this network, the softswitch knows the endpoint name or IP address of at least one voice gateway, the device that interfaces between the VoIP system and the PSTN’s class 5 switch and with an inter-exchange carrier (IXC — a long-distance company). As described earlier, when a phone on the system goes off-hook, the softswitch is notified and instructs the customer gateway to send a dial tone to the telephone. The softswitch receives the dialed number and associates it with either an endpoint name on the VoIP network or, if the called party is not on the network, refers the call to the voice gateway, which converts it to a standard switch interface, such as TR-303 (described earlier in this chapter). At the same time, the softswitch provides calling instructions to either the PSTN class 5 switch or the IXC class 4 switch, giving it the called party’s number. The signaling gateway may be a separate unit, but often it is incorporated into the softswitch and provides translation to the SS7 signaling protocol used by the switches. Frequently, the functions of the softswitch, signaling gateway, and voice gateway will be incorporated into one device, also called a softswitch. Sometimes the functions are built into a class 4 or class 5 switch, providing an integrated system for smaller operators.

A gateway may be used to interface to a class 5 switch, with the class 5 switch performing the functions of the softswitch. The gateway in this case must be able to translate SS7 signaling into the appropriate VoIP commands. A soft-switch would not be used in that configuration. More capability is demanded of the class 5 switch than if it were used just to handle calls originating or terminating off the VoIP network.

6.6.3 Key Concepts of MGCP and TGCP

The material presented here applies generally to both MGCP, the IETF standard, and to NCS/TGCP, the PacketCable standard. Although we shall normally refer only to MGCP, we really mean both. Differences are noted where they exist.

MGCP assumes a connection model where the basic constructs (concepts) are endpoints and connections. Examples of endpoints include customer gateways, voice gateways, and softswitches. Note that several endpoints can exist in one product, such as a softswitch combined with a voice gateway, which likely has software supporting other services, such as interactive voice response. Connections connect these endpoints as required for a call. Connections are grouped to form a call. One or more connections can belong to one call. Connections and calls are set up at the initiative of one or several call agents, the softswitch software that sets up and manages the call. If the same softswitch handles both voice gateways involved in the call, there is usually only one call agent at work. If the two voice gateways are handled by different softswitches, then the two call agents communicate between themselves to set up the call.

A key feature of MGCP, as opposed to certain other VoIP protocols, is that the customer gateways have relatively little intelligence. The intelligence resides at the softswitch as much as possible. This has several benefits. The first is that the cost of the customer gateway can be lower, since it does less. Also, features may be added to the system by merely adding software to the softswitch and not doing anything to the customer gateways.

For example, the softswitch instructs the customer gateway to notify it when a phone goes off-hook. When so notified, the softswitch instructs the customer gateway to collect dialed digits. In MGCP, the customer gateway may be instructed to send each digit to the softswitch as it is sent by the telephone. But this generates a lot of short messages on the network, and many short messages are inefficient in their use of the network. A single long message is less wasteful. The softswitch may instead instruct the customer gateway to collect all dialed digits before sending them to the softswitch. This creates a complication for the customer gateway, though, in that different numbers of digits may have to be dialed, depending on what the subscriber is dialing. A local number in North America is dialed with either seven or 10 digits, depending on the location, whereas a domestic long distance number is dialed with 11 digits, and a call to outside of North America is dialed with 13 or more digits. This can be resolved by having the softswitch supply the customer gateway with a digit map describing how many digits are expected, depending on the first digits dialed. Digit maps are not a part of the present NCS/TGCP specification: Each digit is forwarded to the softswitch as it is dialed, in short packets. This simplifies the operation of the network, at a slight expense of inefficiency.

The softswitch will also instruct a customer gateway to ring the phone, to provide a ring-back tone (telling the caller that the called phone is ringing), and the address(s) to which the customer gateway is to send packets. MGCP allows for a number of different networks over which packets could be sent, but NCS/TGCP assumes an IP network.

6.6.4 Delay

There are several sources of delay in a VoIP network. By delay, we mean the difference in time between when a word is spoken and when it is heard by the listener (or when the echo is heard by the talker, as described later). We already mentioned the delay due to compression. Delay is incurred both at compression and decompression. And there may be additional delay due to the processing circuitry and software. Then the signal must be put in packets. There is a tradeoff in the size of the packets. If you put fewer samples in a packet, there is less delay. But since each packet must have a header of 62 bytes (in the example system — see later), shorter packets are less efficient, in that they “burn” more bandwidth due to more headers. Longer packets have the same-length header; but since more information is transmitted per packet, there is less inefficiency. Longer packets introduce more delay because you cannot transmit the packet until all bits have been accumulated.

Think of the analogy of carrying water in a bucket that you fill from a faucet. The faucet supplies water at a fixed rate (corresponding to the rate at which packets are accumulated at the sending end of the call). If you have a bigger bucket (corresponding to larger packets), you can carry more water at a time (at least until the bucket gets too heavy), so you work more efficiently. But it takes more time (delay) to fill the bucket at the faucet.

Another significant source of delay in IP networks is the need to remove jitter from the packets arriving at the receiver. In order not to introduce uncomfortable interruptions in speech, it is necessary to make sure that each packet has arrived and is ready to be played by the time the previous packet is finished playing. However, IP networks by their nature cannot guarantee a constant arrival time of packets. Sometimes they will arrive after a short delay, and sometimes they will arrive after more delay, depending on network congestion. If you have tried to listen to an audio clip on the Internet, you have likely experienced an interruption in the audio when the next packet is delayed en route to your computer. To avoid any interruptions to the voice, the received packets are buffered, or held in queue, at the receiver for long enough to ensure that packets will always be available when needed. The amount of buffering needed depends on the jitter in the network. This dejitter buffer is another significant source of delay in a VoIP network.

Delay causes a couple of problems in telephony. If the delay gets too long (several hundred milliseconds), you begin to notice a delay in response from the other end. You ask a question, and the other party doesn’t answer as fast as you expect. If you have made overseas calls using satellite transmission, you can appreciate the problem. Some cell phone systems employ enough compression that they tend to approach the point where you would notice the delay. (Try carrying on a cell phone conversation with a colleague in the same room so that you hear each other over the cell phones and directly. You will see how much delay there is.)

One guideline to acceptable delay is found in ANSI T1.508-1998, American National Standard for Telecommunications — Loss Plan for Evolving Digital Networks.⁹ Section 6.2.1 of that document recommends that the round-trip delay between two end users be kept below 300 ms. To permit this in international circuits, each national segment is allocated 100 ms of round-trip delay. Satellite delays are of necessity longer.

6.6.5 Echo Cancellation

Even before the delay gets long enough to notice a response delay, echo can be a problem. For several reasons, when a signal crosses a boundary where it is converted between two-wire and four-wire transmission, some signal can be reflected to the talker. This produces an uncomfortable echo. The amount of delay that can be tolerated is subjective, so you might find different authorities using different numbers as the threshold over which echo cancellation (ECAN) must be used. ANSI T1.508-1998 suggests that when round-trip delay exceeds 5 ms, an echo canceler should be used. PacketCable suggests that the threshold is 20 ms. It is practically impossible for a VoIP system to introduce less delay, so echo cancellation should be used. The current PacketCable specification requires that all endpoints have ECANs.

Echo cancelers are sometimes as simple as devices that attenuate the far-end received signal, but these are not particularly satisfactory. Most modern echo cancelers measure the round-trip delay and construct adaptive filters that mimic that delay, subtracting the originated signal from the returned signal. Very likely at some point you have started a long-distance or cell phone call and noticed a bad echo, which goes away after a second or two of conversation. This is your echo canceler measuring the echo and adapting itself to cancel it. Each link of a telephone system that has the potential to introduce echo is responsible for canceling the echo it introduces.

There are some conditions, though, under which echo cancellation must be turned off, so all systems have the ability to detect those conditions and to instruct the endpoint to turn off echo cancellation. For example, when a telephone link is handling fax or modem traffic, the type of traffic is detected and the echo cancellation is turned off. Note that even with data and VoIP available, there will still be times when people need to use the VoIP circuit to transmit these older data communications protocols. There is an alternative way to transmit faxes, in which the fax signal is demodulated in the network near the point of origination, the data is transmitted, and the fax signal is reconstructed near the user. This alternative is called fax relay, described in ITU T.38. These facilities, not in widespread use, are listed as optional in PacketCable.

6.6.6 Real-Time Protocol

Real-Time Protocol (RTP) is really two related protocols found in many VoIP systems. RTP is used to transmit bearer packets, that is, the packets containing the voice information. Real-Time Control Protocol (RTCP) is the related protocol used to carry data relating to setting up and tearing down the RTP session. These two protocols are maintained by the Internet Engineering Task Force (IETF).

Thus, we find two fundamental types of packets in use in the example VoIP system. Bearer channels, containing packets of voice signals, are carried in RTP. Signaling packets are carried in RTCP. The two types of packets do not identify themselves as to which type they are. Rather, this is done through the use of different port numbers at layer 4, where UDP is employed. RTP and RTCP are often, but not always, carried in UDP packets.

6.6.7 User Datagram Protocol

Chapter 5 includes a brief description of UDP. It is an alternate to TCP, used when the packets being sent are not to be acknowledged by the receiver. Acknowledgment is not feasible when a packet is intended for multiple receivers or in cases, such as voice transmission, when retransmission of a missed packet is not workable — by the time the lost packet is identified and a retransmission is received, it is too late to use the packet. There are other methods of covering lost packets, such as repeating the previous packet. UDP packets are usually put in IP packets for transmission on the network.

6.6.8 Makeup of VoIP Packets

Figure 6.17 illustrates the makeup of the IP packets containing VoIP bearer packets according to the protocol suite being discussed, which is used in Packet Cable and other VoIP applications. See Chapter 5 for information concerning the makeup of the UDP and IP packets. RTP uses application layer framing, meaning that the application can interpret what is in the packet, so there could be interpretations of the packet other than what we show here.

Figure 6.17 RTP in UDP in IP in Ethernet.

RTP uses a compact header to keep from adding a lot of overhead. The version number defines the version of RTP; currently there is only one version. The padding bit indicates whether extra bytes have been added to pad the packet or whether all are valid data. The extension bit will indicate if there is any additional header information other than what is shown here; so far, no extensions have been defined. The contributor count tells how many contributing source identifiers the message contains. Normally in point-to-point VoIP calls, there will be only one contributing source. However, the contributor count could be used for conference calling, as described later. Next is the marker bit, available to the application for its use. Finally, the payload type describes the data to follow. A number of different formats of both audio and video may be carried in RTP packets. We are dealing with only a small subset in the VoIP application.

The synchronization source identifier, abbreviated SSRC, identifies the original sender of the message. The original sender identifies the sequence number and the time stamp for the data. Contributing sources are defined when more than one endpoint is originating data. There is only one contributing source in a point-to-point voice call, but if you are holding a conference call, you will have more than one contributing source.

6.6.9 Conference Calling

In order to accommodate multiple sources, RTP uses the concept of mixers. A mixer serves the same function that an audio mixer performs in audio applications, but it does it in a different way. Applied to the VoIP application (which uses a subset of what RTP is capable of doing), a mixer receives inputs from all the participants on a conference call, adds them, and sends out a new RTP packet to all participants so that they can hear each other at the same time. Note that in applications other than VoIP, a mixer might also combine images or do other operations where it is desired to combine elements from different sources.

Figure 6.18 illustrates use of a mixer to facilitate a conference call between four subscribers. Rather than route packets directly from one subscriber to another, in a conference call all packets are routed to the mixer, which combines them and sends the combined packets to all participants. The mixer may be a separate function, or it may be implemented in the softswitch. It is necessary in a VoIP system only if conference calling is to be supported. Though not supported in early versions of the PacketCable standard, some VoIP systems offer three-way calling without use of a mixer. In these systems, each endpoint is capable of accepting two independent audio streams and mixing them, in essence performing the function of a two-input mixer. In such systems, the mixer would be needed only for conferences of more than three endpoints.

Figure 6.18 Conference calling using a mixer.

Of course, not all of the call sources will necessarily be on the VoIP network as shown. Some could be on the PSTN and be routed through the media gateway.

While not necessarily useful in typical VoIP networks, we’ll mention one other function defined in the RTP protocol, the translator. A translator is used to connect a subscriber who has different network capabilities than do other subscribers. For example, suppose one subscriber to a VoIP network was connected using a 64-kb/s ISDN connection, which is not fast enough to support uncompressed voice packets. The subscriber to whom the first subscriber is talking is connected to a cable modem and is not using compression. The translator would receive packets from the cable modem end, compress them so that they fit in the 64-kb/s channel, and send them to the subscriber on the ISDN connection. For packets in the other direction, the translator would decompress the packets and send them on to the cable modem subscriber.

6.6.10 Bandwidth Required for a VoIP Signal

The bandwidth required for a switched circuit signal is normally 64 kb/s, with very little overhead. However, when voice is transmitted in packets, we have to add overhead in order to manage the packets. The required bandwidth thus is the sum of that required to send the voice data and the packet overhead. There is a trade-off between the number of voice samples sent in the packet and the total amount of overhead. The amount of overhead is fixed per packet and is dictated by the protocols being used. A short packet will demand this overhead for fewer voice packets, so the resultant data rate will increase. A longer packet will demand this same overhead for more voice packets, so the total data rate will decrease. The penalty is that more delay is introduced.

Refer to Figure 6.17, which shows the makeup of a typical VoIP bearer (voice) packet carried on Ethernet. The application data is the actual voice data, which in an uncompressed (G.711) system occupies 64 kb/s of bandwidth. It is placed in an RTP packet, which adds a 12-byte header. We are assuming a simple call between two points, so there are no contributing source identifiers. The RTP packet is put in a UDP packet, which adds yet another 8 bytes of header. This UDP packet is placed in an IP packet, which adds 20 bytes (IPv4), and finally the IP packet is put in an Ethernet packet, which adds 18 bytes of header and four bytes of frame check, for a total addition of 22 bytes. Thus, for a single block of application data, we add a total of 12 + 8 + 20 + 22 = 62 extra bytes of overhead. These 62 bytes of overhead are added to every block of application data, so the amount of overhead depends on the size of the data block.

As shown in Table 6.3, there is a trade-off between the delay a voice packet experiences and the amount of overhead added. Short delays are good, but they add more overhead by virtue of the fact that more packets must be transmitted to convey the information. In order to mitigate the effect of the added header information, DOCSIS 1.1 and later versions offer a way of replacing the packet header with an abbreviated header, which reduces the overhead. Called the payload header suppression (PHS), it is used to suppress packet headers with a short identifier. Rather than transmit the entire packet, a short identifier replaces the header for transmission between the CMTS and the cable modem, which replaces the original header before sending the packet to a non-DOCSIS system.

Table 6.3 VoIP Block Size and Data Rate

Because of the time-critical nature of VoIP traffic, it is necessary to provide for good quality of service (QoS) parameters for transmission of voice packets. This may be done in several different ways, as described in Chapter 5. VoIP packets can be transmitted on the public Internet, and this has been done. The problem is that the Internet is not guaranteed to support the necessary QoS, and there is no guarantee that packets will be delivered to the receiver in a timely fashion and in the correct order. In order to handle voice reliably and with quality competitive with the PSTN, it is necessary to make sure that the entire network handling the call has the requisite QoS parameters.

6.6.11 Other VoIP Control Protocols

We have been describing MGCP and NCS/TGCP, common control protocols for VoIP transport in cable TV plants. But there are other protocols you may encounter from time to time.

H.323 is the original VoIP protocol. It and Session Initiation Protocol (SIP) differ fundamentally from MGCP in that they place the network intelligence at the endpoints. No softswitches are required, though they can be used. Rather, each endpoint is expected to know the network well enough that it can establish and manage any possible call on its own. Obviously this demands a lot more intelligence in the endpoints, increasing their cost.

MEGACO is a superset of MGCP; it adds new services beyond voice transport management. The fundamental concepts of MEGACO are terminations and contexts. Terminations are streams of data (voice, video, or whatever). They are grouped into contexts, which define two or more terminations that are in communication with each other.

6.8.1 Timing Recovery

A few methods have been developed to permit timing recovery at the user end of a DS1 network. Here we describe adaptive clocking, a method that allows DS1 circuits to be transmitted over IP networks that do not maintain precise end-to-end timing. Figure 6.19 illustrates the use of adaptive clocking in timing recovery. The 1.544-Mb/s datastream is delivered to the headend from an outside network (for example, from a telephone switch connecting to a PBX at a business). This is shown in the upper left of the figure. A clock recovery circuit supplies timing to a phase locked loop (PLL), which in turn controls a clock for the recovered upstream information, as we describe shortly.

Figure 6.19 Timing recovery in a TDMolP system using adaptive clocking.

The incoming data is transmitted in the IP network, such as via a cable modem system, to a system that recovers the DS1 at the user end. At the user end, as packets arrive (not on a regular basis, but rather with jitter added by the IP network), they are clocked (loaded) into a first-in-first-out register (FIFO). This is a special-purpose shift register that has independent input and output clocks. As a packet arrives, it is clocked into the input of the FIFO, and it immediately moves as far toward the output as it can, getting in line behind packets that got there earlier. A separate output clock shifts data out of the FIFO to the using system, possibly via an ECAN, described earlier.

The output clock must run at exactly the same speed as the clock that controlled entry of the downstream DS1 data into the network, but it has no direct knowledge of that clock rate. However, the user-end recovery system can infer the clock rate by monitoring how full the FIFO is. A nominal fill point is established in the FIFO. That is, the output clock will attempt to adjust itself to the frequency that keeps the FIFO filled to the nominal fill. However, since data arrives in packets, it is inevitable that sometimes the FIFO will fill or empty beyond this nominal fill. Thus, two thresholds are set. If the output clock is running faster than the controlling system clock, then the FIFO will empty below the “—” end of the jitter range shown. When this happens, the output clock is slowed down (that is, its frequency is reduced slightly), slowing the speed of data being shifted out. Similarly, if the output clock is running too slowly, the FIFO will begin filling more, and when it fills to the “+” limit, the output clock will speed up. Thus, over time the output clock will either speed up or slow down until it is running at the same frequency as the clock that controls delivery of the DS1 to the IP network.

If the DS1 is carrying telephone calls, it is important to provide for echo cancellation if the delay exceeds a small amount, as described earlier. This is the job of the echo canceler (the ECAN) shown at the output. Some DS1 streams carry phone calls in some time slots and other data in other time slots. The ECAN must be able to recognize what kind of data is in each time slot, and either cancel echo or not.

The downstream output clock establishes timing (which is carried in the DS1 datastream) for the user-end system. The user-end system delivers the upstream DS-1 traffic to the IP network in the lower right of Figure 6.19. At the headend, as upstream packets arrive, they are loaded into the upstream FIFO register, where they immediately move as far to the left in the upstream FIFO as possible. They are then clocked out by the upstream output clock, which is the one locked to the downstream DS-1, as shown earlier. Since the user-end clock is locked to the downstream DS-1 and the upstream output clock is locked to this same downstream DS-1, we are assured that we will not encounter a slip, in which either there is not enough data to fill the DS-1 due to late arrival or there is too much data, resulting in something having to be discarded.

It is extremely important that DS-1 circuits exhibit reliable service. If a DS-1 packet is dropped, the receiving circuit should be able to detect that fact and insert null data into the FIFO where the dropped packet should have been. This will not help with the problem of important data being dropped, but at least it will maintain timing, which is essential to the operation of DS-1 interfaces. When DS-1 packets are being carried, they must receive very high transport priority.