Computer Networks 4th Ed Andrew S. Tanenbaum [Electronic resources] نسخه متنی

2.5 The Public Switched Telephone Network

When two computers owned by the same company or organization and located close to each other need to communicate, it is often easiest just to run a cable between them. LANs work this way. However, when the distances are large or there are many computers or the cables have to pass through a public road or other public right of way, the costs of running private cables are usually prohibitive. Furthermore, in just about every country in the world, stringing private transmission lines across (or underneath) public property is also illegal. Consequently, the network designers must rely on the existing telecommunication facilities.

These facilities, especially the PSTN (Public Switched Telephone Network), were usually designed many years ago, with a completely different goal in mind: transmitting the human voice in a more-or-less recognizable form. Their suitability for use in computer-computer communication is often marginal at best, but the situation is rapidly changing with the introduction of fiber optics and digital technology. In any event, the telephone system is so tightly intertwined with (wide area) computer networks, that it is worth devoting some time to studying it.

To see the order of magnitude of the problem, let us make a rough but illustrative comparison of the properties of a typical computer-computer connection via a local cable and via a dial-up telephone line. A cable running between two computers can transfer data at 10⁹ bps, maybe more. In contrast, a dial-up line has a maximum data rate of 56 kbps, a difference of a factor of almost 20,000. That is the difference between a duck waddling leisurely through the grass and a rocket to the moon. If the dial-up line is replaced by an ADSL connection, there is still a factor of 10002000 difference.

The trouble, of course, is that computer systems designers are used to working with computer systems and when suddenly confronted with another system whose performance (from their point of view) is 3 or 4 orders of magnitude worse, they, not surprising, devoted much time and effort to trying to figure out how to use it efficiently. In the following sections we will describe the telephone system and show how it works. For additional information about the innards of the telephone system see (Bellamy, 2000).

2.5.1 Structure of the Telephone System

Soon after Alexander Graham Bell patented the telephone in 1876 (just a few hours ahead of his rival, Elisha Gray), there was an enormous demand for his new invention. The initial market was for the sale of telephones, which came in pairs. It was up to the customer to string a single wire between them. The electrons returned through the earth. If a telephone owner wanted to talk to n other telephone owners, separate wires had to be strung to all n houses. Within a year, the cities were covered with wires passing over houses and trees in a wild jumble. It became immediately obvious that the model of connecting every telephone to every other telephone, as shown in Fig. 2-20(a), was not going to work.

Figure 2-20. (a) Fully-interconnected network. (b) Centralized switch. (c) Two-level hierarchy.

To his credit, Bell saw this and formed the Bell Telephone Company, which opened its first switching office (in New Haven, Connecticut) in 1878. The company ran a wire to each customer's house or office. To make a call, the customer would crank the phone to make a ringing sound in the telephone company office to attract the attention of an operator, who would then manually connect the caller to the callee by using a jumper cable. The model of a single switching office is illustrated in Fig. 2-20(b).

Pretty soon, Bell System switching offices were springing up everywhere and people wanted to make long-distance calls between cities, so the Bell system began to connect the switching offices. The original problem soon returned: to connect every switching office to every other switching office by means of a wire between them quickly became unmanageable, so second-level switching offices were invented. After a while, multiple second-level offices were needed, as illustrated in Fig. 2-20(c). Eventually, the hierarchy grew to five levels.

By 1890, the three major parts of the telephone system were in place: the switching offices, the wires between the customers and the switching offices (by now balanced, insulated, twisted pairs instead of open wires with an earth return), and the long-distance connections between the switching offices. While there have been improvements in all three areas since then, the basic Bell System model has remained essentially intact for over 100 years. For a short technical history of the telephone system, see (Hawley, 1991).

Prior to the 1984 breakup of AT&T, the telephone system was organized as a highly-redundant, multilevel hierarchy. The following description is highly simplified but gives the essential flavor nevertheless. Each telephone has two copper wires coming out of it that go directly to the telephone company's nearest end office (also called a local central office). The distance is typically 1 to 10 km, being shorter in cities than in rural areas. In the United States alone there are about 22,000 end offices. The two-wire connections between each subscriber's telephone and the end office are known in the trade as the local loop. If the world's local loops were stretched out end to end, they would extend to the moon and back 1000 times.

At one time, 80 percent of AT&T's capital value was the copper in the local loops. AT&T was then, in effect, the world's largest copper mine. Fortunately, this fact was not widely known in the investment community. Had it been known, some corporate raider might have bought AT&T, terminated all telephone service in the United States, ripped out all the wire, and sold the wire to a copper refiner to get a quick payback.

If a subscriber attached to a given end office calls another subscriber attached to the same end office, the switching mechanism within the office sets up a direct electrical connection between the two local loops. This connection remains intact for the duration of the call.

If the called telephone is attached to another end office, a different procedure has to be used. Each end office has a number of outgoing lines to one or more nearby switching centers, called toll offices (or if they are within the same local area, tandem offices). These lines are called toll connecting trunks. If both the caller's and callee's end offices happen to have a toll connecting trunk to the same toll office (a likely occurrence if they are relatively close by), the connection may be established within the toll office. A telephone network consisting only of telephones (the small dots), end offices (the large dots), and toll offices (the squares) is shown in Fig. 2-20(c).

If the caller and callee do not have a toll office in common, the path will have to be established somewhere higher up in the hierarchy. Primary, sectional, and regional offices form a network by which the toll offices are connected. The toll, primary, sectional, and regional exchanges communicate with each other via high-bandwidth intertoll trunks (also called interoffice trunks). The number of different kinds of switching centers and their topology (e.g., can two sectional offices have a direct connection or must they go through a regional office?) varies from country to country depending on the country's telephone density. Figure 2-21 shows how a medium-distance connection might be routed.

Figure 2-21. A typical circuit route for a medium-distance call.

A variety of transmission media are used for telecommunication. Local loops consist of category 3 twisted pairs nowadays, although in the early days of telephony, uninsulated wires spaced 25 cm apart on telephone poles were common. Between switching offices, coaxial cables, microwaves, and especially fiber optics are widely used.

In the past, transmission throughout the telephone system was analog, with the actual voice signal being transmitted as an electrical voltage from source to destination. With the advent of fiber optics, digital electronics, and computers, all the trunks and switches are now digital, leaving the local loop as the last piece of analog technology in the system. Digital transmission is preferred because it is not necessary to accurately reproduce an analog waveform after it has passed through many amplifiers on a long call. Being able to correctly distinguish a 0 from a 1 is enough. This property makes digital transmission more reliable than analog. It is also cheaper and easier to maintain.

In summary, the telephone system consists of three major components:

Local loops (analog twisted pairs going into houses and businesses).

Trunks (digital fiber optics connecting the switching offices).

Switching offices (where calls are moved from one trunk to another).

After a short digression on the politics of telephones, we will come back to each of these three components in some detail. The local loops provide everyone access to the whole system, so they are critical. Unfortunately, they are also the weakest link in the system. For the long-haul trunks, the main issue is how to collect multiple calls together and send them out over the same fiber. This subject is called multiplexing, and we will study three different ways to do it. Finally, there are two fundamentally different ways of doing switching; we will look at both.

2.5.2 The Politics of Telephones

For decades prior to 1984, the Bell System provided both local and long distance service throughout most of the United States. In the 1970s, the U.S. Federal Government came to believe that this was an illegal monopoly and sued to break it up. The government won, and on January 1, 1984, AT&T was broken up into AT&T Long Lines, 23 BOCs (Bell Operating Companies), and a few other pieces. The 23 BOCs were grouped into seven regional BOCs (RBOCs) to make them economically viable. The entire nature of telecommunication in the United States was changed overnight by court order (not by an act of Congress).

The exact details of the divestiture were described in the so-called MFJ (Modified Final Judgment, an oxymoron if ever there was oneif the judgment could be modified, it clearly was not final). This event led to increased competition, better service, and lower long distance prices to consumers and businesses. However, prices for local service rose as the cross subsidies from long-distance calling were eliminated and local service had to become self supporting. Many other countries have now introduced competition along similar lines.

To make it clear who could do what, the United States was divided up into 164 LATAs (Local Access and Transport Areas). Very roughly, a LATA is about as big as the area covered by one area code. Within a LATA, there was one LEC (Local Exchange Carrier) that had a monopoly on traditional telephone service within its area. The most important LECs were the BOCs, although some LATAs contained one or more of the 1500 independent telephone companies operating as LECs.

All inter-LATA traffic was handled by a different kind of company, an IXC (IntereXchange Carrier). Originally, AT&T Long Lines was the only serious IXC, but now WorldCom and Sprint are well-established competitors in the IXC business. One of the concerns at the breakup was to ensure that all the IXCs would be treated equally in terms of line quality, tariffs, and the number of digits their customers would have to dial to use them. The way this is handled is illustrated in Fig. 2-22. Here we see three example LATAs, each with several end offices. LATAs 2 and 3 also have a small hierarchy with tandem offices (intra-LATA toll offices).

Figure 2-22. The relationship of LATAs, LECs, and IXCs. All the circles are LEC switching offices. Each hexagon belongs to the IXC whose number is in it.

Any IXC that wishes to handle calls originating in a LATA can build a switching office called a POP (Point of Presence) there. The LEC is required to connect each IXC to every end office, either directly, as in LATAs 1 and 3, or indirectly, as in LATA 2. Furthermore, the terms of the connection, both technical and financial, must be identical for all IXCs. In this way, a subscriber in, say, LATA 1, can choose which IXC to use for calling subscribers in LATA 3.

As part of the MFJ, the IXCs were forbidden to offer local telephone service and the LECs were forbidden to offer inter-LATA telephone service, although both were free to enter any other business, such as operating fried chicken restaurants. In 1984, that was a fairly unambiguous statement. Unfortunately, technology has a funny way of making the law obsolete. Neither cable television nor mobile phones were covered by the agreement. As cable television went from one way to two way and mobile phones exploded in popularity, both LECs and IXCs began buying up or merging with cable and mobile operators.

By 1995, Congress saw that trying to maintain a distinction between the various kinds of companies was no longer tenable and drafted a bill to allow cable TV companies, local telephone companies, long-distance carriers, and mobile operators to enter one another's businesses. The idea was that any company could then offer its customers a single integrated package containing cable TV, telephone, and information services and that different companies would compete on service and price. The bill was enacted into law in February 1996. As a result, some BOCs became IXCs and some other companies, such as cable television operators, began offering local telephone service in competition with the LECs.

One interesting property of the 1996 law is the requirement that LECs implement local number portability. This means that a customer can change local telephone companies without having to get a new telephone number. This provision removes a huge hurdle for many people and makes them much more inclined to switch LECs, thus increasing competition. As a result, the U.S. telecommunications landscape is currently undergoing a radical restructuring. Again, many other countries are starting to follow suit. Often other countries wait to see how this kind of experiment works out in the U.S. If it works well, they do the same thing; if it works badly, they try something else.

2.5.3 The Local Loop: Modems, ADSL, and Wireless

It is now time to start our detailed study of how the telephone system works. The main parts of the system are illustrated in Fig. 2-23. Here we see the local loops, the trunks, and the toll offices and end offices, both of which contain switching equipment that switches calls. An end office has up to 10,000 local loops (in the U.S. and other large countries). In fact, until recently, the area code + exchange indicated the end office, so (212) 601-xxxx was a specific end office with 10,000 subscribers, numbered 0000 through 9999. With the advent of competition for local service, this system was no longer tenable because multiple companies wanted to own the end office code. Also, the number of codes was basically used up, so complex mapping schemes had to be introduced.

Figure 2-23. The use of both analog and digital transmission for a computer to computer call. Conversion is done by the modems and codecs.

Let us begin with the part that most people are familiar with: the two-wire local loop coming from a telephone company end office into houses and small businesses. The local loop is also frequently referred to as the ''last mile,'' although the length can be up to several miles. It has used analog signaling for over 100 years and is likely to continue doing so for some years to come, due to the high cost of converting to digital. Nevertheless, even in this last bastion of analog transmission, change is taking place. In this section we will study the traditional local loop and the new developments taking place here, with particular emphasis on data communication from home computers.

When a computer wishes to send digital data over an analog dial-up line, the data must first be converted to analog form for transmission over the local loop. This conversion is done by a device called a modem, something we will study shortly. At the telephone company end office the data are converted to digital form for transmission over the long-haul trunks.

If the other end is a computer with a modem, the reverse conversiondigital to analogis needed to traverse the local loop at the destination. This arrangement is shown in Fig. 2-23 for ISP 1 (Internet Service Provider), which has a bank of modems, each connected to a different local loop. This ISP can handle as many connections as it has modems (assuming its server or servers have enough computing power). This arrangement was the normal one until 56-kbps modems appeared, for reasons that will become apparent shortly.

Analog signaling consists of varying a voltage with time to represent an information stream. If transmission media were perfect, the receiver would receive exactly the same signal that the transmitter sent. Unfortunately, media are not perfect, so the received signal is not the same as the transmitted signal. For digital data, this difference can lead to errors.

Transmission lines suffer from three major problems: attenuation, delay distortion, and noise. Attenuation is the loss of energy as the signal propagates outward. The loss is expressed in decibels per kilometer. The amount of energy lost depends on the frequency. To see the effect of this frequency dependence, imagine a signal not as a simple waveform, but as a series of Fourier components. Each component is attenuated by a different amount, which results in a different Fourier spectrum at the receiver.

To make things worse, the different Fourier components also propagate at different speeds in the wire. This speed difference leads to distortion of the signal received at the other end.

Another problem is noise, which is unwanted energy from sources other than the transmitter. Thermal noise is caused by the random motion of the electrons in a wire and is unavoidable. Crosstalk is caused by inductive coupling between two wires that are close to each other. Sometimes when talking on the telephone, you can hear another conversation in the background. That is crosstalk. Finally, there is impulse noise, caused by spikes on the power line or other causes. For digital data, impulse noise can wipe out one or more bits.

Modems

Due to the problems just discussed, especially the fact that both attenuation and propagation speed are frequency dependent, it is undesirable to have a wide range of frequencies in the signal. Unfortunately, the square waves used in digital signals have a wide frequency spectrum and thus are subject to strong attenuation and delay distortion. These effects make baseband (DC) signaling unsuitable except at slow speeds and over short distances.

To get around the problems associated with DC signaling, especially on telephone lines, AC signaling is used. A continuous tone in the 1000 to 2000-Hz range, called a sine wave carrier, is introduced. Its amplitude, frequency, or phase can be modulated to transmit information. In amplitude modulation, two different amplitudes are used to represent 0 and 1, respectively. In frequency modulation, also known as frequency shift keying, two (or more) different tones are used. (The term keying is also widely used in the industry as a synonym for modulation.) In the simplest form of phase modulation, the carrier wave is systematically shifted 0 or 180 degrees at uniformly spaced intervals. A better scheme is to use shifts of 45, 135, 225, or 315 degrees to transmit 2 bits of information per time interval. Also, always requiring a phase shift at the end of every time interval, makes it is easier for the receiver to recognize the boundaries of the time intervals.

Figure 2-24 illustrates the three forms of modulation. In Fig. 2-24(a) one of the amplitudes is nonzero and one is zero. In Fig. 2-24(b) two frequencies are used. In Fig. 2-24(c) a phase shift is either present or absent at each bit boundary. A device that accepts a serial stream of bits as input and produces a carrier modulated by one (or more) of these methods (or vice versa) is called a modem (for modulator-demodulator). The modem is inserted between the (digital) computer and the (analog) telephone system.

Figure 2-24. (a) A binary signal. (b) Amplitude modulation. (c) Frequency modulation. (d) Phase modulation.

To go to higher and higher speeds, it is not possible to just keep increasing the sampling rate. The Nyquist theorem says that even with a perfect 3000-Hz line (which a dial-up telephone is decidedly not), there is no point in sampling faster than 6000 Hz. In practice, most modems sample 2400 times/sec and focus on getting more bits per sample.

The number of samples per second is measured in baud. During each baud, one symbol is sent. Thus, an n-baud line transmits n symbols/sec. For example, a 2400-baud line sends one symbol about every 416.667 µsec. If the symbol consists of 0 volts for a logical 0 and 1 volt for a logical 1, the bit rate is 2400 bps. If, however, the voltages 0, 1, 2, and 3 volts are used, every symbol consists of 2 bits, so a 2400-baud line can transmit 2400 symbols/sec at a data rate of 4800 bps. Similarly, with four possible phase shifts, there are also 2 bits/symbol, so again here the bit rate is twice the baud rate. The latter technique is widely used and called QPSK (Quadrature Phase Shift Keying).

The concepts of bandwidth, baud, symbol, and bit rate are commonly confused, so let us restate them here. The bandwidth of a medium is the range of frequencies that pass through it with minimum attenuation. It is a physical property of the medium (usually from 0 to some maximum frequency) and measured in Hz. The baud rate is the number of samples/sec made. Each sample sends one piece of information, that is, one symbol. The baud rate and symbol rate are thus the same. The modulation technique (e.g., QPSK) determines the number of bits/symbol. The bit rate is the amount of information sent over the channel and is equal to the number of symbols/sec times the number of bits/symbol.

All advanced modems use a combination of modulation techniques to transmit multiple bits per baud. Often multiple amplitudes and multiple phase shifts are combined to transmit several bits/symbol. In Fig. 2-25(a), we see dots at 45, 135, 225, and 315 degrees with constant amplitude (distance from the origin). The phase of a dot is indicated by the angle a line from it to the origin makes with the positive x-axis. Fig. 2-25(a) has four valid combinations and can be used to transmit 2 bits per symbol. It is QPSK.

Figure 2-25. (a) QPSK. (b) QAM-16. (c) QAM-64.

In Fig. 2-25(b) we see a different modulation scheme, in which four amplitudes and four phases are used, for a total of 16 different combinations. This modulation scheme can be used to transmit 4 bits per symbol. It is called QAM-16 (Quadrature Amplitude Modulation). Sometimes the term 16-QAM is used instead. QAM-16 can be used, for example, to transmit 9600 bps over a 2400-baud line.

Figure 2-25(c) is yet another modulation scheme involving amplitude and phase. It allows 64 different combinations, so 6 bits can be transmitted per symbol. It is called QAM-64. Higher-order QAMs also are used.

Diagrams such as those of Fig. 2-25, which show the legal combinations of amplitude and phase, are called constellation diagrams. Each high-speed modem standard has its own constellation pattern and can talk only to other modems that use the same one (although most modems can emulate all the slower ones).

With many points in the constellation pattern, even a small amount of noise in the detected amplitude or phase can result in an error and, potentially, many bad bits. To reduce the chance of an error, standards for the higher speeds modems do error correction by adding extra bits to each sample. The schemes are known as TCM (Trellis Coded Modulation). Thus, for example, the V.32 modem standard uses 32 constellation points to transmit 4 data bits and 1 parity bit per symbol at 2400 baud to achieve 9600 bps with error correction. Its constellation pattern is shown in Fig. 2-26(a). The decision to ''rotate'' around the origin by 45 degrees was done for engineering reasons; the rotated and unrotated constellations have the same information capacity.

Figure 2-26. (a) V.32 for 9600 bps. (b) V32 bis for 14,400 bps.

The next step above 9600 bps is 14,400 bps. It is called V.32 bis. This speed is achieved by transmitting 6 data bits and 1 parity bit per sample at 2400 baud. Its constellation pattern has 128 points when QAM-128 is used and is shown in Fig. 2-26(b). Fax modems use this speed to transmit pages that have been scanned in as bit maps. QAM-256 is not used in any standard telephone modems, but it is used on cable networks, as we shall see.

The next telephone modem after V.32 bis is V.34, which runs at 28,800 bps at 2400 baud with 12 data bits/symbol. The final modem in this series is V.34 bis which uses 14 data bits/symbol at 2400 baud to achieve 33,600 bps.

To increase the effective data rate further, many modems compress the data before transmitting it, to get an effective data rate higher than 33,600 bps. On the other hand, nearly all modems test the line before starting to transmit user data, and if they find the quality lacking, cut back to a speed lower than the rated maximum. Thus, the effective modem speed observed by the user can be lower, equal to, or higher than the official rating.

All modern modems allow traffic in both directions at the same time (by using different frequencies for different directions). A connection that allows traffic in both directions simultaneously is called full duplex. A two-lane road is full duplex. A connection that allows traffic either way, but only one way at a time is called half duplex. A single railroad track is half duplex. A connection that allows traffic only one way is called simplex. A one-way street is simplex. Another example of a simplex connection is an optical fiber with a laser on one end and a light detector on the other end.

The reason that standard modems stop at 33,600 is that the Shannon limit for the telephone system is about 35 kbps, so going faster than this would violate the laws of physics (department of thermodynamics). To find out whether 56-kbps modems are theoretically possible, stay tuned.

But why is the theoretical limit 35 kbps? It has to do with the average length of the local loops and the quality of these lines. The 35 kbps is determined by the average length of the local loops. In Fig. 2-23, a call originating at the computer on the left and terminating at ISP 1 goes over two local loops as an analog signal, once at the source and once at the destination. Each of these adds noise to the signal. If we could get rid of one of these local loops, the maximum rate would be doubled.

ISP 2 does precisely that. It has a pure digital feed from the nearest end office. The digital signal used on the trunks is fed directly to ISP 2, eliminating the codecs, modems, and analog transmission on its end. Thus, when one end of the connection is purely digital, as it is with most ISPs now, the maximum data rate can be as high as 70 kbps. Between two home users with modems and analog lines, the maximum is 33.6 kbps.

The reason that 56 kbps modems are in use has to do with the Nyquist theorem. The telephone channel is about 4000 Hz wide (including the guard bands). The maximum number of independent samples per second is thus 8000. The number of bits per sample in the U.S. is 8, one of which is used for control purposes, allowing 56,000 bit/sec of user data. In Europe, all 8 bits are available to users, so 64,000-bit/sec modems could have been used, but to get international agreement on a standard, 56,000 was chosen.

This modem standard is called V.90. It provides for a 33.6-kbps upstream channel (user to ISP), but a 56 kbps downstream channel (ISP to user) because there is usually more data transport from the ISP to the user than the other way (e.g., requesting a Web page takes only a few bytes, but the actual page could be megabytes). In theory, an upstream channel wider than 33.6 kbps would have been possible, but since many local loops are too noisy for even 33.6 kbps, it was decided to allocate more of the bandwidth to the downstream channel to increase the chances of it actually working at 56 kbps.

The next step beyond V.90 is V.92. These modems are capable of 48 kbps on the upstream channel if the line can handle it. They also determine the appropriate speed to use in about half of the usual 30 seconds required by older modems. Finally, they allow an incoming telephone call to interrupt an Internet session, provided that the line has call waiting service.

Digital Subscriber Lines

When the telephone industry finally got to 56 kbps, it patted itself on the back for a job well done. Meanwhile, the cable TV industry was offering speeds up to 10 Mbps on shared cables, and satellite companies were planning to offer upward of 50 Mbps. As Internet access became an increasingly important part of their business, the telephone companies (LECs) began to realize they needed a more competitive product. Their answer was to start offering new digital services over the local loop. Services with more bandwidth than standard telephone service are sometimes called broadband, although the term really is more of a marketing concept than a specific technical concept.

Initially, there were many overlapping offerings, all under the general name of xDSL (Digital Subscriber Line), for various x. Below we will discuss these but primarily focus on what is probably going to become the most popular of these services, ADSL (Asymmetric DSL). Since ADSL is still being developed and not all the standards are fully in place, some of the details given below may change in time, but the basic picture should remain valid. For more information about ADSL, see (Summers, 1999; and Vetter et al., 2000).

The reason that modems are so slow is that telephones were invented for carrying the human voice and the entire system has been carefully optimized for this purpose. Data have always been stepchildren. At the point where each local loop terminates in the end office, the wire runs through a filter that attenuates all frequencies below 300 Hz and above 3400 Hz. The cutoff is not sharp300 Hz and 3400 Hz are the 3 dB pointsso the bandwidth is usually quoted as 4000 Hz even though the distance between the 3 dB points is 3100 Hz. Data are thus also restricted to this narrow band.

The trick that makes xDSL work is that when a customer subscribes to it, the incoming line is connected to a different kind of switch, one that does not have this filter, thus making the entire capacity of the local loop available. The limiting factor then becomes the physics of the local loop, not the artificial 3100 Hz bandwidth created by the filter.

Unfortunately, the capacity of the local loop depends on several factors, including its length, thickness, and general quality. A plot of the potential bandwidth as a function of distance is given in Fig. 2-27. This figure assumes that all the other factors are optimal (new wires, modest bundles, etc.).

Figure 2-27. Bandwidth versus distance over category 3 UTP for DSL.

The implication of this figure creates a problem for the telephone company. When it picks a speed to offer, it is simultaneously picking a radius from its end offices beyond which the service cannot be offered. This means that when distant customers try to sign up for the service, they may be told ''Thanks a lot for your interest, but you live 100 meters too far from the nearest end office to get the service. Could you please move?'' The lower the chosen speed, the larger the radius and the more customers covered. But the lower the speed, the less attractive the service and the fewer the people who will be willing to pay for it. This is where business meets technology. (One potential solution is building mini end offices out in the neighborhoods, but that is an expensive proposition.)

The xDSL services have all been designed with certain goals in mind. First, the services must work over the existing category 3 twisted pair local loops. Second, they must not affect customers' existing telephones and fax machines. Third, they must be much faster than 56 kbps. Fourth, they should be always on, with just a monthly charge but no per-minute charge.

The initial ADSL offering was from AT&T and worked by dividing the spectrum available on the local loop, which is about 1.1 MHz, into three frequency bands: POTS (Plain Old Telephone Service) upstream (user to end office) and downstream (end office to user). The technique of having multiple frequency bands is called frequency division multiplexing; we will study it in detail in a later section. Subsequent offerings from other providers have taken a different approach, and it appears this one is likely to win out, so we will describe it below.

The alternative approach, called DMT (Discrete MultiTone), is illustrated in Fig. 2-28. In effect, what it does is divide the available 1.1 MHz spectrum on the local loop into 256 independent channels of 4312.5 Hz each. Channel 0 is used for POTS. Channels 15 are not used, to keep the voice signal and data signals from interfering with each other. Of the remaining 250 channels, one is used for upstream control and one is used for downstream control. The rest are available for user data.

Figure 2-28. Operation of ADSL using discrete multitone modulation.

In principle, each of the remaining channels can be used for a full-duplex data stream, but harmonics, crosstalk, and other effects keep practical systems well below the theoretical limit. It is up to the provider to determine how many channels are used for upstream and how many for downstream. A 5050 mix of upstream and downstream is technically possible, but most providers allocate something like 80%90% of the bandwidth to the downstream channel since most users download more data than they upload. This choice gives rise to the ''A'' in ADSL. A common split is 32 channels for upstream and the rest downstream. It is also possible to have a few of the highest upstream channels be bidirectional for increased bandwidth, although making this optimization requires adding a special circuit to cancel echoes.

The ADSL standard (ANSI T1.413 and ITU G.992.1) allows speeds of as much as 8 Mbps downstream and 1 Mbps upstream. However, few providers offer this speed. Typically, providers offer 512 kbps downstream and 64 kbps upstream (standard service) and 1 Mbps downstream and 256 kbps upstream (premium service).

Within each channel, a modulation scheme similar to V.34 is used, although the sampling rate is 4000 baud instead of 2400 baud. The line quality in each channel is constantly monitored and the data rate adjusted continuously as needed, so different channels may have different data rates. The actual data are sent with QAM modulation, with up to 15 bits per baud, using a constellation diagram analogous to that of Fig. 2-25(b). With, for example, 224 downstream channels and 15 bits/baud at 4000 baud, the downstream bandwidth is 13.44 Mbps. In practice, the signal-to-noise ratio is never good enough to achieve this rate, but 8 Mbps is possible on short runs over high-quality loops, which is why the standard goes up this far.

A typical ADSL arrangement is shown in Fig. 2-29. In this scheme, a telephone company technician must install a NID (Network Interface Device) on the customer's premises. This small plastic box marks the end of the telephone company's property and the start of the customer's property. Close to the NID (or sometimes combined with it) is a splitter, an analog filter that separates the 0-4000 Hz band used by POTS from the data. The POTS signal is routed to the existing telephone or fax machine, and the data signal is routed to an ADSL modem. The ADSL modem is actually a digital signal processor that has been set up to act as 250 QAM modems operating in parallel at different frequencies. Since most current ADSL modems are external, the computer must be connected to it at high speed. Usually, this is done by putting an Ethernet card in the computer and operating a very short two-node Ethernet containing only the computer and ADSL modem. Occasionally the USB port is used instead of Ethernet. In the future, internal ADSL modem cards will no doubt become available.

Figure 2-29. A typical ADSL equipment configuration.

At the other end of the wire, on the end office side, a corresponding splitter is installed. Here the voice portion of the signal is filtered out and sent to the normal voice switch. The signal above 26 kHz is routed to a new kind of device called a DSLAM (Digital Subscriber Line Access Multiplexer), which contains the same kind of digital signal processor as the ADSL modem. Once the digital signal has been recovered into a bit stream, packets are formed and sent off to the ISP.

This complete separation between the voice system and ADSL makes it relatively easy for a telephone company to deploy ADSL. All that is needed is buying a DSLAM and splitter and attaching the ADSL subscribers to the splitter. Other high-bandwidth services (e.g., ISDN) require much greater changes to the existing switching equipment.

One disadvantage of the design of Fig. 2-29 is the presence of the NID and splitter on the customer premises. Installing these can only be done by a telephone company technician, necessitating an expensive ''truck roll'' (i.e., sending a technician to the customer's premises). Therefore, an alternative splitterless design has also been standardized. It is informally called G.lite but the ITU standard number is G.992.2. It is the same as Fig. 2-29 but without the splitter. The existing telephone line is used as is. The only difference is that a microfilter has to be inserted into each telephone jack between the telephone or ADSL modem and the wire. The microfilter for the telephone is a low-pass filter eliminating frequencies above 3400 Hz; the microfilter for the ADSL modem is a high-pass filter eliminating frequencies below 26 kHz. However this system is not as reliable as having a splitter, so G.lite can be used only up to 1.5 Mbps (versus 8 Mbps for ADSL with a splitter). G.lite still requires a splitter in the end office, however, but that installation does not require thousands of truck rolls.

ADSL is just a physical layer standard. What runs on top of it depends on the carrier. Often the choice is ATM due to ATM's ability to manage quality of service and the fact that many telephone companies run ATM in the core network.

Wireless Local Loops

Since 1996 in the U.S. and a bit later in other countries, companies that wish to compete with the entrenched local telephone company (the former monopolist), called an ILEC (Incumbent LEC), are free to do so. The most likely candidates are long-distance telephone companies (IXCs). Any IXC wishing to get into the local phone business in some city must do the following things. First, it must buy or lease a building for its first end office in that city. Second, it must fill the end office with telephone switches and other equipment, all of which are available as off-the-shelf products from various vendors. Third, it must run a fiber between the end office and its nearest toll office so the new local customers will have access to its national network. Fourth, it must acquire customers, typically by advertising better service or lower prices than those of the ILEC.

Then the hard part begins. Suppose that some customers actually show up. How is the new local phone company, called a CLEC (Competitive LEC) going to connect customer telephones and computers to its shiny new end office? Buying the necessary rights of way and stringing wires or fibers is prohibitively expensive. Many CLECs have discovered a cheaper alternative to the traditional twisted-pair local loop: the WLL (Wireless Local Loop).

In a certain sense, a fixed telephone using a wireless local loop is a bit like a mobile phone, but there are three crucial technical differences. First, the wireless local loop customer often wants high-speed Internet connectivity, often at speeds at least equal to ADSL. Second, the new customer probably does not mind having a CLEC technician install a large directional antenna on his roof pointed at the CLEC's end office. Third, the user does not move, eliminating all the problems with mobility and cell handoff that we will study later in this chapter. And thus a new industry is born: fixed wireless (local telephone and Internet service run by CLECs over wireless local loops).

Although WLLs began serious operation in 1998, we first have to go back to 1969 to see the origin. In that year the FCC allocated two television channels (at 6 MHz each) for instructional television at 2.1 GHz. In subsequent years, 31 more channels were added at 2.5 GHz for a total of 198 MHz.

Instructional television never took off and in 1998, the FCC took the frequencies back and allocated them to two-way radio. They were immediately seized upon for wireless local loops. At these frequencies, the microwaves are 1012 cm long. They have a range of about 50 km and can penetrate vegetation and rain moderately well. The 198 MHz of new spectrum was immediately put to use for wireless local loops as a service called MMDS (Multichannel Multipoint Distribution Service). MMDS can be regarded as a MAN (Metropolitan Area Network), as can its cousin LMDS (discussed below).

The big advantage of this service is that the technology is well established and the equipment is readily available. The disadvantage is that the total bandwidth available is modest and must be shared by many users over a fairly large geographic area.

The low bandwidth of MMDS led to interest in millimeter waves as an alternative. At frequencies of 2831 GHz in the U.S. and 40 GHz in Europe, no frequencies were allocated because it is difficult to build silicon integrated circuits that operate so fast. That problem was solved with the invention of gallium arsenide integrated circuits, opening up millimeter bands for radio communication. The FCC responded to the demand by allocating 1.3 GHz to a new wireless local loop service called LMDS (Local Multipoint Distribution Service). This allocation is the single largest chunk of bandwidth ever allocated by the FCC for any one use. A similar chunk is being allocated in Europe, but at 40 GHz.

The operation of LMDS is shown in Fig. 2-30. Here a tower is shown with multiple antennas on it, each pointing in a different direction. Since millimeter waves are highly directional, each antenna defines a sector, independent of the other ones. At this frequency, the range is 25 km, which means that many towers are needed to cover a city.

Figure 2-30. Architecture of an LMDS system.

Like ADSL, LMDS uses an asymmetric bandwidth allocation favoring the downstream channel. With current technology, each sector can have 36 Gbps downstream and 1 Mbps upstream, shared among all the users in that sector. If each active user downloads three 5-KB pages per minute, the user is occupying an average of 2000 bps of spectrum, which allows a maximum of 18,000 active users per sector. To keep the delay reasonable, no more than 9000 active users should be supported, though. With four sectors, as shown in Chap. 4.

2.5.4 Trunks and Multiplexing

Economies of scale play an important role in the telephone system. It costs essentially the same amount of money to install and maintain a high-bandwidth trunk as a low-bandwidth trunk between two switching offices (i.e., the costs come from having to dig the trench and not from the copper wire or optical fiber). Consequently, telephone companies have developed elaborate schemes for multiplexing many conversations over a single physical trunk. These multiplexing schemes can be divided into two basic categories: FDM (Frequency Division Multiplexing) and TDM (Time Division Multiplexing). In FDM, the frequency spectrum is divided into frequency bands, with each user having exclusive possession of some band. In TDM, the users take turns (in a round-robin fashion), each one periodically getting the entire bandwidth for a little burst of time.

AM radio broadcasting provides illustrations of both kinds of multiplexing. The allocated spectrum is about 1 MHz, roughly 500 to 1500 kHz. Different frequencies are allocated to different logical channels (stations), each operating in a portion of the spectrum, with the interchannel separation great enough to prevent interference. This system is an example of frequency division multiplexing. In addition (in some countries), the individual stations have two logical subchannels: music and advertising. These two alternate in time on the same frequency, first a burst of music, then a burst of advertising, then more music, and so on. This situation is time division multiplexing.

Below we will examine frequency division multiplexing. After that we will see how FDM can be applied to fiber optics (wavelength division multiplexing). Then we will turn to TDM, and end with an advanced TDM system used for fiber optics (SONET).

Frequency Division Multiplexing

Figure 2-31 shows how three voice-grade telephone channels are multiplexed using FDM. Filters limit the usable bandwidth to about 3100 Hz per voice-grade channel. When many channels are multiplexed together, 4000 Hz is allocated to each channel to keep them well separated. First the voice channels are raised in frequency, each by a different amount. Then they can be combined because no two channels now occupy the same portion of the spectrum. Notice that even though there are gaps (guard bands) between the channels, there is some overlap between adjacent channels because the filters do not have sharp edges. This overlap means that a strong spike at the edge of one channel will be felt in the adjacent one as nonthermal noise.

Figure 2-31. Frequency division multiplexing. (a) The original bandwidths. (b) The bandwidths raised in frequency. (c) The multiplexed channel.

The FDM schemes used around the world are to some degree standardized. A widespread standard is twelve 4000-Hz voice channels multiplexed into the 60 to 108 kHz band. This unit is called a group. The 12-kHz to 60-kHz band is sometimes used for another group. Many carriers offer a 48- to 56-kbps leased line service to customers, based on the group. Five groups (60 voice channels) can be multiplexed to form a supergroup. The next unit is the mastergroup, which is five supergroups (CCITT standard) or ten supergroups (Bell system). Other standards of up to 230,000 voice channels also exist.

Wavelength Division Multiplexing

For fiber optic channels, a variation of frequency division multiplexing is used. It is called WDM (Wavelength Division Multiplexing). The basic principle of WDM on fibers is depicted in Fig. 2-32. Here four fibers come together at an optical combiner, each with its energy present at a different wavelength. The four beams are combined onto a single shared fiber for transmission to a distant destination. At the far end, the beam is split up over as many fibers as there were on the input side. Each output fiber contains a short, specially-constructed core that filters out all but one wavelength. The resulting signals can be routed to their destination or recombined in different ways for additional multiplexed transport.

Figure 2-32. Wavelength division multiplexing.

There is really nothing new here. This is just frequency division multiplexing at very high frequencies. As long as each channel has its own frequency (i.e., wavelength) range and all the ranges are disjoint, they can be multiplexed together on the long-haul fiber. The only difference with electrical FDM is that an optical system using a diffraction grating is completely passive and thus highly reliable.

WDM technology has been progressing at a rate that puts computer technology to shame. WDM was invented around 1990. The first commercial systems had eight channels of 2.5 Gbps per channel. By 1998, systems with 40 channels of 2.5 Gbps were on the market. By 2001, there were products with 96 channels of 10 Gbps, for a total of 960 Gbps. This is enough bandwidth to transmit 30 full-length movies per second (in MPEG-2). Systems with 200 channels are already working in the laboratory. When the number of channels is very large and the wavelengths are spaced close together, for example, 0.1 nm, the system is often referred to as DWDM (Dense WDM).

It should be noted that the reason WDM is popular is that the energy on a single fiber is typically only a few gigahertz wide because it is currently impossible to convert between electrical and optical media any faster. By running many channels in parallel on different wavelengths, the aggregate bandwidth is increased linearly with the number of channels. Since the bandwidth of a single fiber band is about 25,000 GHz (see Fig. 2-6), there is theoretically room for 2500 10-Gbps channels even at 1 bit/Hz (and higher rates are also possible).

Another new development is all optical amplifiers. Previously, every 100 km it was necessary to split up all the channels and convert each one to an electrical signal for amplification separately before reconverting to optical and combining them. Nowadays, all optical amplifiers can regenerate the entire signal once every 1000 km without the need for multiple opto-electrical conversions.

In the example of Fig. 2-32, we have a fixed wavelength system. Bits from input fiber 1 go to output fiber 3, bits from input fiber 2 go to output fiber 1, etc. However, it is also possible to build WDM systems that are switched. In such a device, the output filters are tunable using Fabry-Perot or Mach-Zehnder interferometers. For more information about WDM and its application to Internet packet switching, see (Elmirghani and Mouftah, 2000; Hunter and Andonovic, 2000; and Listani et al., 2001).

Time Division Multiplexing

WDM technology is wonderful, but there is still a lot of copper wire in the telephone system, so let us turn back to it for a while. Although FDM is still used over copper wires or microwave channels, it requires analog circuitry and is not amenable to being done by a computer. In contrast, TDM can be handled entirely by digital electronics, so it has become far more widespread in recent years. Unfortunately, it can only be used for digital data. Since the local loops produce analog signals, a conversion is needed from analog to digital in the end office, where all the individual local loops come together to be combined onto outgoing trunks.

We will now look at how multiple analog voice signals are digitized and combined onto a single outgoing digital trunk. Computer data sent over a modem are also analog, so the following description also applies to them. The analog signals are digitized in the end office by a device called a codec (coder-decoder), producing a series of 8-bit numbers. The codec makes 8000 samples per second (125 µsec/sample) because the Nyquist theorem says that this is sufficient to capture all the information from the 4-kHz telephone channel bandwidth. At a lower sampling rate, information would be lost; at a higher one, no extra information would be gained. This technique is called PCM (Pulse Code Modulation). PCM forms the heart of the modern telephone system. As a consequence, virtually all time intervals within the telephone system are multiples of 125 µsec.

When digital transmission began emerging as a feasible technology, CCITT was unable to reach agreement on an international standard for PCM. Consequently, a variety of incompatible schemes are now in use in different countries around the world.

The method used in North America and Japan is the T1 carrier, depicted in Fig. 2-33. (Technically speaking, the format is called DS1 and the carrier is called T1, but following widespread industry tradition, we will not make that subtle distinction here.) The T1 carrier consists of 24 voice channels multiplexed together. Usually, the analog signals are sampled on a round-robin basis with the resulting analog stream being fed to the codec rather than having 24 separate codecs and then merging the digital output. Each of the 24 channels, in turn, gets to insert 8 bits into the output stream. Seven bits are data and one is for control, yielding 7 x 8000 = 56,000 bps of data, and 1 x 8000 = 8000 bps of signaling information per channel.

Figure 2-33. The T1 carrier (1.544 Mbps).

A frame consists of 24 x 8 = 192 bits plus one extra bit for framing, yielding 193 bits every 125 µsec. This gives a gross data rate of 1.544 Mbps. The 193rd bit is used for frame synchronization. It takes on the pattern 0101010101 . . . . Normally, the receiver keeps checking this bit to make sure that it has not lost synchronization. If it does get out of sync, the receiver can scan for this pattern to get resynchronized. Analog customers cannot generate the bit pattern at all because it corresponds to a sine wave at 4000 Hz, which would be filtered out. Digital customers can, of course, generate this pattern, but the odds are against its being present when the frame slips. When a T1 system is being used entirely for data, only 23 of the channels are used for data. The 24th one is used for a special synchronization pattern, to allow faster recovery in the event that the frame slips.

When CCITT finally did reach agreement, they felt that 8000 bps of signaling information was far too much, so its 1.544-Mbps standard is based on an 8- rather than a 7-bit data item; that is, the analog signal is quantized into 256 rather than 128 discrete levels. Two (incompatible) variations are provided. In common-channel signaling, the extra bit (which is attached onto the rear rather than the front of the 193-bit frame) takes on the values 10101010 . . . in the odd frames and contains signaling information for all the channels in the even frames.

In the other variation, channel-associated signaling, each channel has its own private signaling subchannel. A private subchannel is arranged by allocating one of the eight user bits in every sixth frame for signaling purposes, so five out of six samples are 8 bits wide, and the other one is only 7 bits wide. CCITT also recommended a PCM carrier at 2.048 Mbps called E1. This carrier has 32 8-bit data samples packed into the basic 125-µsec frame. Thirty of the channels are used for information and two are used for signaling. Each group of four frames provides 64 signaling bits, half of which are used for channel-associated signaling and half of which are used for frame synchronization or are reserved for each country to use as it wishes. Outside North America and Japan, the 2.048-Mbps E1 carrier is used instead of T1.

Once the voice signal has been digitized, it is tempting to try to use statistical techniques to reduce the number of bits needed per channel. These techniques are appropriate not only for encoding speech, but for the digitization of any analog signal. All of the compaction methods are based on the principle that the signal changes relatively slowly compared to the sampling frequency, so that much of the information in the 7- or 8-bit digital level is redundant.

One method, called differential pulse code modulation, consists of outputting not the digitized amplitude, but the difference between the current value and the previous one. Since jumps of ±16 or more on a scale of 128 are unlikely, 5 bits should suffice instead of 7. If the signal does occasionally jump wildly, the encoding logic may require several sampling periods to ''catch up.'' For speech, the error introduced can be ignored.

A variation of this compaction method requires each sampled value to differ from its predecessor by either +1 or -1. Under these conditions, a single bit can be transmitted, telling whether the new sample is above or below the previous one. This technique, called delta modulation, is illustrated in Fig. 2-34. Like all compaction techniques that assume small level changes between consecutive samples, delta encoding can get into trouble if the signal changes too fast, as shown in the figure. When this happens, information is lost.

Figure 2-34. Delta modulation.

An improvement to differential PCM is to extrapolate the previous few values to predict the next value and then to encode the difference between the actual signal and the predicted one. The transmitter and receiver must use the same prediction algorithm, of course. Such schemes are called predictive encoding. They are useful because they reduce the size of the numbers to be encoded, hence the number of bits to be sent.

Time division multiplexing allows multiple T1 carriers to be multiplexed into higher-order carriers. Figure 2-35 shows how this can be done. At the left we see four T1 channels being multiplexed onto one T2 channel. The multiplexing at T2 and above is done bit for bit, rather than byte for byte with the 24 voice channels that make up a T1 frame. Four T1 streams at 1.544 Mbps should generate 6.176 Mbps, but T2 is actually 6.312 Mbps. The extra bits are used for framing and recovery in case the carrier slips. T1 and T3 are widely used by customers, whereas T2 and T4 are only used within the telephone system itself, so they are not well known.

Figure 2-35. Multiplexing T1 streams onto higher carriers.

At the next level, seven T2 streams are combined bitwise to form a T3 stream. Then six T3 streams are joined to form a T4 stream. At each step a small amount of overhead is added for framing and recovery in case the synchronization between sender and receiver is lost.

Just as there is little agreement on the basic carrier between the United States and the rest of the world, there is equally little agreement on how it is to be multiplexed into higher-bandwidth carriers. The U.S. scheme of stepping up by 4, 7, and 6 did not strike everyone else as the way to go, so the CCITT standard calls for multiplexing four streams onto one stream at each level. Also, the framing and recovery data are different between the U.S. and CCITT standards. The CCITT hierarchy for 32, 128, 512, 2048, and 8192 channels runs at speeds of 2.048, 8.848, 34.304, 139.264, and 565.148 Mbps.

SONET/SDH

In the early days of fiber optics, every telephone company had its own proprietary optical TDM system. After AT&T was broken up in 1984, local telephone companies had to connect to multiple long-distance carriers, all with different optical TDM systems, so the need for standardization became obvious. In 1985, Bellcore, the RBOCs research arm, began working on a standard, called SONET (Synchronous Optical NETwork). Later, CCITT joined the effort, which resulted in a SONET standard and a set of parallel CCITT recommendations (G.707, G.708, and G.709) in 1989. The CCITT recommendations are called SDH (Synchronous Digital Hierarchy) but differ from SONET only in minor ways. Virtually all the long-distance telephone traffic in the United States, and much of it elsewhere, now uses trunks running SONET in the physical layer. For additional information about SONET, see (Bellamy, 2000; Goralski, 2000; and Shepard, 2001).

The SONET design had four major goals. First and foremost, SONET had to make it possible for different carriers to interwork. Achieving this goal required defining a common signaling standard with respect to wavelength, timing, framing structure, and other issues.

Second, some means was needed to unify the U.S., European, and Japanese digital systems, all of which were based on 64-kbps PCM channels, but all of which combined them in different (and incompatible) ways.

Third, SONET had to provide a way to multiplex multiple digital channels. At the time SONET was devised, the highest-speed digital carrier actually used widely in the United States was T3, at 44.736 Mbps. T4 was defined, but not used much, and nothing was even defined above T4 speed. Part of SONET's mission was to continue the hierarchy to gigabits/sec and beyond. A standard way to multiplex slower channels into one SONET channel was also needed.

Fourth, SONET had to provide support for operations, administration, and maintenance (OAM). Previous systems did not do this very well.

An early decision was to make SONET a traditional TDM system, with the entire bandwidth of the fiber devoted to one channel containing time slots for the various subchannels. As such, SONET is a synchronous system. It is controlled by a master clock with an accuracy of about 1 part in 10⁹. Bits on a SONET line are sent out at extremely precise intervals, controlled by the master clock. When cell switching was later proposed to be the basis of ATM, the fact that it permitted irregular cell arrivals got it labeled as Asynchronous Transfer Mode to contrast it to the synchronous operation of SONET. With SONET, the sender and receiver are tied to a common clock; with ATM they are not.

The basic SONET frame is a block of 810 bytes put out every 125 µsec. Since SONET is synchronous, frames are emitted whether or not there are any useful data to send. Having 8000 frames/sec exactly matches the sampling rate of the PCM channels used in all digital telephony systems.

The 810-byte SONET frames are best described as a rectangle of bytes, 90 columns wide by 9 rows high. Thus, 8 x 810 = 6480 bits are transmitted 8000 times per second, for a gross data rate of 51.84 Mbps. This is the basic SONET channel, called STS-1 (Synchronous Transport Signal-1). All SONET trunks are a multiple of STS-1.

The first three columns of each frame are reserved for system management information, as illustrated in Fig. 2-36. The first three rows contain the section overhead; the next six contain the line overhead. The section overhead is generated and checked at the start and end of each section, whereas the line overhead is generated and checked at the start and end of each line.

Figure 2-36. Two back-to-back SONET frames.

A SONET transmitter sends back-to-back 810-byte frames, without gaps between them, even when there are no data (in which case it sends dummy data). From the receiver's point of view, all it sees is a continuous bit stream, so how does it know where each frame begins? The answer is that the first two bytes of each frame contain a fixed pattern that the receiver searches for. If it finds this pattern in the same place in a large number of consecutive frames, it assumes that it is in sync with the sender. In theory, a user could insert this pattern into the payload in a regular way, but in practice it cannot be done due to the multiplexing of multiple users into the same frame and other reasons.

The remaining 87 columns hold 87 x 9 x 8 x 8000 = 50.112 Mbps of user data. However, the user data, called the SPE (Synchronous Payload Envelope), do not always begin in row 1, column 4. The SPE can begin anywhere within the frame. A pointer to the first byte is contained in the first row of the line overhead. The first column of the SPE is the path overhead (i.e., header for the end-to-end path sublayer protocol).

The ability to allow the SPE to begin anywhere within the SONET frame and even to span two frames, as shown in Fig. 2-36, gives added flexibility to the system. For example, if a payload arrives at the source while a dummy SONET frame is being constructed, it can be inserted into the current frame instead of being held until the start of the next one.

The SONET multiplexing hierarchy is shown in Fig. 2-37. Rates from STS-1 to STS-192 have been defined. The optical carrier corresponding to STS-n is called OC-n but is bit for bit the same except for a certain bit reordering needed for synchronization. The SDH names are different, and they start at OC-3 because CCITT-based systems do not have a rate near 51.84 Mbps. The OC-9 carrier is present because it closely matches the speed of a major high-speed trunk used in Japan. OC-18 and OC-36 are used in Japan. The gross data rate includes all the overhead. The SPE data rate excludes the line and section overhead. The user data rate excludes all overhead and counts only the 86 payload columns.

Figure 2-37. SONET and SDH multiplex rates.

As an aside, when a carrier, such as OC-3, is not multiplexed, but carries the data from only a single source, the letter c (for concatenated) is appended to the designation, so OC-3 indicates a 155.52-Mbps carrier consisting of three separate OC-1 carriers, but OC-3c indicates a data stream from a single source at 155.52 Mbps. The three OC-1 streams within an OC-3c stream are interleaved by column, first column 1 from stream 1, then column 1 from stream 2, then column 1 from stream 3, followed by column 2 from stream 1, and so on, leading to a frame 270 columns wide and 9 rows deep.

2.5.5 Switching

From the point of view of the average telephone engineer, the phone system is divided into two principal parts: outside plant (the local loops and trunks, since they are physically outside the switching offices) and inside plant (the switches), which are inside the switching offices. We have just looked at the outside plant. Now it is time to examine the inside plant.

Two different switching techniques are used nowadays: circuit switching and packet switching. We will give a brief introduction to each of them below. Then we will go into circuit switching in detail because that is how the telephone system works. We will study packet switching in detail in subsequent chapters.

Circuit Switching

When you or your computer places a telephone call, the switching equipment within the telephone system seeks out a physical path all the way from your telephone to the receiver's telephone. This technique is called circuit switching and is shown schematically in Fig. 2-38(a). Each of the six rectangles represents a carrier switching office (end office, toll office, etc.). In this example, each office has three incoming lines and three outgoing lines. When a call passes through a switching office, a physical connection is (conceptually) established between the line on which the call came in and one of the output lines, as shown by the dotted lines.

Figure 2-38. (a) Circuit switching. (b) Packet switching.

In the early days of the telephone, the connection was made by the operator plugging a jumper cable into the input and output sockets. In fact, a surprising little story is associated with the invention of automatic circuit switching equipment. It was invented by a 19th century Missouri undertaker named Almon B. Strowger. Shortly after the telephone was invented, when someone died, one of the survivors would call the town operator and say ''Please connect me to an undertaker.'' Unfortunately for Mr. Strowger, there were two undertakers in his town, and the other one's wife was the town telephone operator. He quickly saw that either he was going to have to invent automatic telephone switching equipment or he was going to go out of business. He chose the first option. For nearly 100 years, the circuit-switching equipment used worldwide was known as Strowger gear. (History does not record whether the now-unemployed switchboard operator got a job as an information operator, answering questions such as ''What is the phone number of an undertaker?'')

The model shown in Fig. 2-39(a) is highly simplified, of course, because parts of the physical path between the two telephones may, in fact, be microwave or fiber links onto which thousands of calls are multiplexed. Nevertheless, the basic idea is valid: once a call has been set up, a dedicated path between both ends exists and will continue to exist until the call is finished.

Figure 2-39. Timing of events in (a) circuit switching, (b) message switching, (c) packet switching.

The alternative to circuit switching is packet switching, shown in Fig. 2-38(b). With this technology, individual packets are sent as need be, with no dedicated path being set up in advance. It is up to each packet to find its way to the destination on its own.

An important property of circuit switching is the need to set up an end-to-end path before any data can be sent. The elapsed time between the end of dialing and the start of ringing can easily be 10 sec, more on long-distance or international calls. During this time interval, the telephone system is hunting for a path, as shown in Fig. 2-39(a). Note that before data transmission can even begin, the call request signal must propagate all the way to the destination and be acknowledged. For many computer applications (e.g., point-of-sale credit verification), long setup times are undesirable.

As a consequence of the reserved path between the calling parties, once the setup has been completed, the only delay for data is the propagation time for the electromagnetic signal, about 5 msec per 1000 km. Also as a consequence of the established path, there is no danger of congestionthat is, once the call has been put through, you never get busy signals. Of course, you might get one before the connection has been established due to lack of switching or trunk capacity.

Message Switching

An alternative switching strategy is message switching, illustrated in Chap. 1.

The first electromechanical telecommunication systems used message switching, namely, for telegrams. The message was punched on paper tape (off-line) at the sending office, and then read in and transmitted over a communication line to the next office along the way, where it was punched out on paper tape. An operator there tore the tape off and read it in on one of the many tape readers, one reader per outgoing trunk. Such a switching office was called a torn tape office. Paper tape is long gone and message switching is not used any more, so we will not discuss it further in this book.

Packet Switching

With message switching, there is no limit at all on block size, which means that routers (in a modern system) must have disks to buffer long blocks. It also means that a single block can tie up a router-router line for minutes, rendering message switching useless for interactive traffic. To get around these problems, packet switching was invented, as described in Chap. 1. Packet-switching networks place a tight upper limit on block size, allowing packets to be buffered in router main memory instead of on disk. By making sure that no user can monopolize any transmission line very long (milliseconds), packet-switching networks are well suited for handling interactive traffic. A further advantage of packet switching over message switching is shown in Fig. 2-39(b) and (c): the first packet of a multipacket message can be forwarded before the second one has fully arrived, reducing delay and improving throughput. For these reasons, computer networks are usually packet switched, occasionally circuit switched, but never message switched.

Circuit switching and packet switching differ in many respects. To start with, circuit switching requires that a circuit be set up end to end before communication begins. Packet switching does not require any advance setup. The first packet can just be sent as soon as it is available.

The result of the connection setup with circuit switching is the reservation of bandwidth all the way from the sender to the receiver. All packets follow this path. Among other properties, having all packets follow the same path means that they cannot arrive out of order. With packet switching there is no path, so different packets can follow different paths, depending on network conditions at the time they are sent. They may arrive out of order.

Packet switching is more fault tolerant than circuit switching. In fact, that is why it was invented. If a switch goes down, all of the circuits using it are terminated and no more traffic can be sent on any of them. With packet switching, packets can be routed around dead switches.

Setting up a path in advance also opens up the possibility of reserving bandwidth in advance. If bandwidth is reserved, then when a packet arrives, it can be sent out immediately over the reserved bandwidth. With packet switching, no bandwidth is reserved, so packets may have to wait their turn to be forwarded.

Having bandwidth reserved in advance means that no congestion can occur when a packet shows up (unless more packets show up than expected). On the other hand, when an attempt is made to establish a circuit, the attempt can fail due to congestion. Thus, congestion can occur at different times with circuit switching (at setup time) and packet switching (when packets are sent).

If a circuit has been reserved for a particular user and there is no traffic to send, the bandwidth of that circuit is wasted. It cannot be used for other traffic. Packet switching does not waste bandwidth and thus is more efficient from a system-wide perspective. Understanding this trade-off is crucial for comprehending the difference between circuit switching and packet switching. The trade-off is between guaranteed service and wasting resources versus not guaranteeing service and not wasting resources.

Packet switching uses store-and-forward transmission. A packet is accumulated in a router's memory, then sent on to the next router. With circuit switching, the bits just flow through the wire continuously. The store-and-forward technique adds delay.

Another difference is that circuit switching is completely transparent. The sender and receiver can use any bit rate, format, or framing method they want to. The carrier does not know or care. With packet switching, the carrier determines the basic parameters. A rough analogy is a road versus a railroad. In the former, the user determines the size, speed, and nature of the vehicle; in the latter, the carrier does. It is this transparency that allows voice, data, and fax to coexist within the phone system.

A final difference between circuit and packet switching is the charging algorithm. With circuit switching, charging has historically been based on distance and time. For mobile phones, distance usually does not play a role, except for international calls, and time plays only a minor role (e.g., a calling plan with 2000 free minutes costs more than one with 1000 free minutes and sometimes night or weekend calls are cheaper than normal). With packet switching, connect time is not an issue, but the volume of traffic sometimes is. For home users, ISPs usually charge a flat monthly rate because it is less work for them and their customers can understand this model easily, but backbone carriers charge regional networks based on the volume of their traffic. The differences are summarized in Fig. 2-40.

Figure 2-40. A comparison of circuit-switched and packet-switched networks.

Both circuit switching and packet switching are important enough that we will come back to them shortly and describe the various technologies used in detail.