The Digital Future Of The Telephone Network
A Study of Evolving Technology
By Lee Goeller
Originally published by Probe Research Inc. 1979. Reprinted by permission

Chapter 3
Fundamentals Of Speech Transmission

Introduction

The advantage of digital transmission in general, and pulse code modulation (PCM) in particular, is that, once a signal is coded, it does not change. The amplitude does not change, the phase does not change, and no noise is added, as long as the error rate in the pulses transmitted is kept low. All this comes about because PCM signals are not amplified linearly as are analog signals: they are regenerated.

Regeneration is a technique known since the early days of the telegraph. Each telegraph circuit would operate a relay at its far end; the relay would key a new circuit, reproducing a clear, sharp pulse. No noise would pass on to the second (or third or fourth) channel. With voice encoded into pulses, an electronic circuit (located, typically, every 6000 feet along a T span-line) looks for the presence or absence of pulses at carefully predetermined times. If it finds a pulse, no matter how straggly, distorted or dispersed, it sends forward a new, sharp, clean pulse to once again do battle with the dragons thoughtfully supplied by a provident nature to keep communications from being a trivial exercise. Further, cross-talk and noise pick-up are minimized because such spurious signals must either be above a given threshold to create a false pulse, or must drive an existing pulse below a slightly different threshold to make the regenerator miss it. With proper system design, these errors can be minimized.

Like everything else in this world, however, digital transmission has some disadvantages. Initially, in the vacuum tube world, cost was the biggie. Today, however, LSI has changed all that. Now the major problem appears to be the bandwidth required. To code speech for digital transmission using PCM, 64,000 bits per second has been accepted as the standard. With other forms of modulation, notably delta-modulation, the bit rate can be cut in half at the very least, but there is already so much PCM in the plant that a change would not be practicable.

To get a feel for what 64 kb/s means, a broadband data channel today is 50 kb/s, and the telephone industry implements such data channels on an analog carrier "group" of 12 voice channels. On the other hand, 12 voice channels in analog form require only about 48 kHz, and two groups of 12, comparable to the "digroup" of 24 lines on one T span-line, need less than 100 kHz (hundred thousand cycles per second) of bandwidth. The T-carrier di-group, however, requires 1.544 million bits per second. Naturally, bits per second and cycles per second do not equate on a one-to-one basis, but it should be evident that T-carrier has to take a lot more bandwidth for voice signals, even though each of its voice channels can (potentially) handle about the same digital data that 12 analog voice channels can.

Economics of Digital Modulation

It is fortunate that, with satellites, fiber optics, millimeter wave guides, etc., etc., bandwidth is getting less and less expensive all the time. But in the immediate future, it will be a major problem. The whole point with T-carrier, the American telephone industry's PCM digital carrier, was that it could be put on existing "exchange cable," the pairs of wires already in place and used individually as trunks. T-carrier puts 24 voice channels on two pairs of wires (one pair for each direction), increasing for a modest cost in electronics the number of trunks by more than an order of magnitude. In Europe, CCITT standards call for 30 voice channels plus two signaling channels on two pairs; transmission advances now allow American T to run 48 channels on two pairs.

The economics of T-carrier, as developed, depended on having pairs of wires used as single trunks to begin with. If two pairs could handle 12 times as many trunks by using the available bandwidth above the normal voice range (4 kHz), a swap of unused bandwidth for economy of manufacture and use would be a good trade. With fiber optics, where the bandwidth per physical channel extends far beyond that available on pairs of wires designed for voice transmission, the trade-off is even better. But the catch comes when we turn to long-haul trunks. These, again for obvious economic reasons, were the first to be put on analog carrier systems. Thus no "empty" bandwidth exists; where less than 4 kHz per trunk is available, digital modulation simply hasn't room to operate.

For the use of digital modulation on radio (as in microwave carrier systems, in particular), the excessively wide bandwidth required, as compared with analog modulation, squanders the scarce bandwidth available. However, the arguments get more and more complex. Because of the ability of a digital signal to hold its shape as long as the bit error rate is low and the signal at the receiver is above the threshold, less power is required and radio beams can be placed closer together. Further, "cross-polarization," permitting two independent signals to be transmitted on the same frequency but with their electromagnetic components at right angles to, and thus independent of, each other, is possible.

Thus, advocates of digital radio suggest that, in a given volume of space, as many digital radio voice channels can be transmitted as analog. Although the argument seems more complex than believable, and the game of "bits per baud" is slowly being won, the existing analog systems would have to be replaced to use PCM on long-haul trunks. Digital modulation should certainly be considered with satellites and fiber optics, if and when such facilities are planned on a large scale. But for the present, long-haul voice circuits have very large numbers of analog trunks in place and economics, both in terms of money and radio spectrum, dictates that these trunks will be there for quite a while.

Quantizing Noise

The next major problem with digital modulation is quantizing noise. Each voice signal is "sampled" 8000 times a second, and the height of each sample is coded into a (binary) number. That is, if the voltage amplitude of the sample is between two different thresholds, it is assigned one number. If it is between another pair of thresholds, it is given another number. The number is sent to the distant end where it is converted back to a voltage amplitude. However, it is given the value half-way between the two original thresholds, and not the exact value of the original signal. Thus the largest error in the reproduced signal is half the distance between the thresholds, and the average error is about half of the maximum error.

It can be seen that quantizing noise can be minimized if the thresholds can be kept very close together. In the original T-carrier, there were 128 levels into which a continuous analog signal could be coded. Since seven binary digits can represent 128 different things, seven bits were used to code the voice signal, and an additional bit was used for "supervision," or transmitting the on-hook/off-hook signal from one office to the other. This inexpensive built-in signaling, independent of voice signals and not subject to distortion (as was the case with SF signals) was another strong argument in favor of T-carrier.

As more and more T-carrier systems were installed, the probability of getting two or more T trunks in a given connection increased rapidly. One injection of quantizing noise was tolerable, but two or more added up to make trouble. Thus T-carrier was redesigned to add another 128 levels by using 8-bit coding rather than 7. To handle supervision, one bit was stolen every six frames, leaving, in those frames, only 7 bits for speech, and reducing the number of times the distant office was given on-hook/off-hook information from 8000 per second to 1333 times.

With this improvement, a number of conversions from A to D and back become possible and tandem connections through several T-trunks used as analog facilities can be carried out without noticeable impact of quantizing noise. Indeed, the only time the telephone company brings up the problem is when a business customer threatens to use a PCM PBX or, worse, a digitally modulated microwave system. Under such circumstances, quantizing noise (along with other T-carrier effects such as strobing on fax and modem-connected data) is trotted out like veritable boogey men.

Eight-bit coding five sixths of the time is not, however, without its ironies in the context of future digital switching. When the signal is not reconverted to analog but, rather, is passed through the switch in digital form, there is no law that says the 7-bit frame on the outgoing trunk will be the same as the 7-bit frame on the incoming trunk. Indeed, to force such an alignment (over a "superframe" which, for reasons related to the Panel System's obsolete signaling system, consists of 12 frames rather than 6), a 2-millisecond delay would have to be introduced at each switch. Since this much extra delay cannot be tolerated, as will be discussed below, the odds are good that each digital switch traversed by a digital signal will take away the 8th bit in a different frame. Even though the effect accumulates, "The impact of signaling frame realignment (called digit robbing) on SDN transmission performance is not expected to degrade service." It should be evident, however, that improving the capabilities of T-carrier for analog switching has not been much help for digital switching.

Getting rid of the signaling bit altogether with CCIS is recognized as the way out of this absurd situation, but CCIS leads to another obscure problem. There are occasions where transmission of supervisory information from one end of a connection to the other is desirable. In PBXs, the switch-hook flash (momentary depression and release of the switch-hook) calls in the control system so that special features can be requested. There are occasions when a switch-hook flash in one PBX should call in features in another PBX via a tie line. If the supervisory bit could pass through one or more tie lines in tandem along with the voice bits, the distant PBX could detect the flash immediately without waiting for the cumulative delays required by timing to detect a flash rather than a hang up, then passing the flash signal along from switch to switch via CCIS, and then recreating it at the far end. Note, too, that hang up supervision on regular long distance connections could be detected and trunks freed at the end of the call appreciably more quickly if hang up timing could be started at all switches simultaneously.

Companding

There is another aspect of quantizing noise that is even more important. We only get the 256 (or 128) levels if a signal is large enough to swing between the two extremes that the system can handle. If a person speaks softly, or if loss has been working on a signal before it gets to the T-carrier terminal, the signal to be coded may be able to swing only over two or three levels of the 256. This make the magnitude of the quantizing noise huge compared with the actual signal to be encoded rather than the maximum possible signal.

To get around this, "companding" is used. In analog carrier systems (and, indeed, in the early versions of T-carrier), a large signal was COMpressed and a small signal was exPANDed to make all signals to be transmitted more or less the same size. (It should be noted that, in one-way media, companding is common and simple: broadcasting stations use it, tape recorders use it, etc. What makes companding difficult in telephony is that, at the far end of the channel, the signal must be restored to its proper size. That is, the COMpandor at one end must track with the comPANDOR at the other end.)

To do companding in digital systems, it is now standard to have the thresholds spaced differently at different signal levels. They are very close together for small amplitudes, and get farther and farther apart as the analog amplitude to be encoded gets bigger. With this approach, the quantizing noise has about the same ratio for small signals as for large ones. Unfortunately, standards for companding in America differ from those used in Europe (CCITT).

Companding now introduces a problem of its own. In a well-designed communication system, noise is small compared with the signal to be transmitted. And, when people are talking, this is true in T-carrier. However, one person is usually quiet while the other speaks, so each direction of transmission has only noise on it more than half the time. Companding, with small steps near 0 level, tends to accentuate the noise entering the system in the absence of voice; thus great care must be taken to keep 60 Hz hum from power lines from being picked up and giving other noise a free ride into audibility when speech is not present to mask the noise out.

Amplitude Control

Suppose we want, for some reason, to change the amplitude of a digital signal. We have two choices. We can convert it back to analog, put in an amplifier or attenuator, and recode it, or we can find some means for operating on the signal in its digital form. There are several ways of doing the latter. Automatic Electric, TRW and others are using Read Only Memories. You use the incoming coded signal as the address to the memory, and the signal you get back, stored in that address, is the equivalent signal 2 dB lower in level (if you are using a 2 dB pad). Alternatively, you can use some sort of a circuit that multiplies the input signal by a factor to get an output signal that has the desired relationship.

There are several problems in all this, however. Suppose, for instance, our digital signal is not voice, but is data that somehow has gotten directly into the system, bypassing the A/D converters. Translating it by some fixed amount will convert it to garbage. Another problem, dealing exclusively with voice signals, is more subtle. If our quantizing thresholds were evenly spaced, direct conversion for an analog signal would be the same for loud signals and quiet signals, and for the peak of loud signals as for their parts near the zero-crossing level. With companding, it can be seen that a 2 dB loss is somewhat different for the low-level portion of a signal compared with the high level portion. This makes for complications, to say the least.

In any event, we have been leading up to one point: not only does a digitally coded signal hold its amplitude and phase from input to output, but it is darn hard to change, even if you wanted to. As long as one is considering a single trunk between two analog switches, where the signal must be brought back to analog to be switched on a per channel analog basis, there is no problem. A digital carrier system works just like an analog carrier system, and level adjustments can be made in standard and well understood ways, always on the analog side. But when digital signals are to be switched by digital switches, the level-adjust problem must be faced.

(On first reading, the material between here and the Summary at the end of this section, which is highly technical, may be omitted.)

VNL

Let us now take a moment to understand why levels must be changed. It is obvious that analog carrier systems contain - amplifiers, and it would pose no problem, in principle, to have the signal at the output of a carrier channel be any level we like. Since analog carrier systems all have their inputs at -16 dB TLP* and their outputs at +7 dB TLP, this is just about what happens (what TLP and other exotic terms mean will come clear shortly). The important point here is that the output of an analog carrier system is 16 + 7 =23 dB louder than its input, or, to use other terms, the output is 200 times the amplitude of the input.

[*Footnote: What TLP and other exotic terms mean will come clear shortly.]

This 23 dB gain on all analog channels, regardless of length, is a standard. It allows a standard test signal to be inserted on all inputs, and a standard output to be observed on all outputs. This facilitates the set-up and testing of thousands of circuits. The fact that the gain is so large apparently comes from many years ago when the first carrier systems, implemented with vacuum tubes, were designed. Since transmission and switching were (and are) handled by separate divisions within the telephone company, and since the transmission equipment might well be in one building while the switching equipment was located in another, perhaps several city blocks away, it seemed reasonable to use the vacuum tube amplifiers required for the carrier system to compensate for the short-haul losses encountered between the carrier system and the switch. Of course, 23 dB difference between input and output levels meant that cables had to be segregated to eliminate cross-talk from the loud output to the sensitive input, but dealing with such problems was much simpler than adding additional expensive vacuum tube amplifiers with their low reliability.

Unfortunately, only the carrier channels have a standard input and output. The overall trunk, of which one or more carrier channels may be a part, must have very carefully prescribed loss introduced to minimize echo. Echo is the major problem in transmission, and if it can be held within limits, it just turns out that other problems, such as "singing,” will usually vanish.

Note that if there were two separate channels, one from transmitter A to receiver B and the other from transmitter B to receiver A, there would be no echo. That is, there would be no way for the signal from A’s transmitter to get back into A’s ear-piece, and all the following discussion would be meaningless. However, for the past 80 or 90 years, the transmitter and receiver in each telephone set have been connected to a single pair of wires to the central office through a "hybrid coil," and this pair to the CO carries voice signals in both directions. The Class 5 office operates on a two-wire basis so, for local calls, the entire path, telephone set to telephone set, is two wires. Clearly, this saves money. The "outside plant" amounts to about a third of the capital investment required to serve a customer, and doubling it to provide four-wire transmission would have been a burden on the telephone industry and the ratepayer in terms of pre-LSI technology. Whether or not this will remain true is one of the things we will explore in this paper.

In any event, Class 5 offices, including No. 1, No. 2 and No. 3 ESS, are two-wire switches matching two-wire customer plant. However, as of today, just about all trunk plant (circuits between switches) is four-wire. That is, most of it runs on carrier systems and, as a result, one direction of transmission must be separated from the other. Now, the bottom line here is that echoes come about when two-wire facilities interface four-wire facilities. The four-wire facilities, used for long haul, have amplification in them. And if the incoming signal "bounces off" the two-wire connecting circuit so that some of it goes into the outgoing trunk path, it will return to the far end as echo. If it bounces there, too, it will come back again, going round and round. Under proper (or improper) circumstances, a circulating signal can build up to a howl, just as a public address system, played back into its microphone, will scream. "Singing" of this sort is an extreme case; echo is more common and thus more annoying and, as has been mentioned, if it can be eliminated or reduced, singing isn't likely to occur.

Now, echo is a funny thing. When you speak, you hear what you have said in synchronism with your speech, and it doesn't bother you a bit. But if you hear it delayed slightly, it can cause troubles. Extensive tests conducted by Bell Labs specialists in human factors have shown that, within a considerable range, the greater the delay, the more annoying the echo. Or, to put it another way, the greater the delay, the more attenuation you need for the echo signal to keep it from driving the customer bananas.

It takes finite time for a signal to travel from one end of a trunk to the other, bounce, and return to the sender. In carrier systems, the speed of transmission is slightly less than the speed of light in vacuum, or 186 miles per millisecond. Thus the delay of the echo can be calculated from knowing the length of the circuit in miles. Through use of quantitative data measured in the above-mentioned human factors experiments, delay can be related to the minimum loss that must be in a circuit to make echo tolerable, assuming the reflection at the 4/2 wire interface is held within reasonable limits. The minimum loss is used, since, with automatic alternate routing, a path between two points may traverse built-up connections of considerably different lengths, and the subjective difference encountered by the callers must be kept to a minimum.

In any event, any given toll connection between two Class 5 offices should have loss that is approximated by 4 dB plus Via Net Loss, or VNL, which is dependent on distance. The 4 dB is divided up between the two ends of the connection, and is put in the "toll-connecting trunks" between Class 5 and Class 4 offices. The distance-based VNL is included in "Intertoll Trunks." Ideally, toll-connecting trunks have 2 dB plus VNL, but their length is usually so short that VNL is negligible.

To calculate VNL, a "via net loss factor" is provided for various different facilities. For all types of carrier systems, it is 0.0015. Multiply this by the length of the facility in miles, and the loss in dB to insert in each direction of transmission is determined. Note that the echo passes through the loss twice, while the signal, only once.

To allow for the natural variation in loss in each trunk in a built-up connection, VNL is actually increased by .4 dB. Thus after multiplying the distance by the VNLF, one adds .4 dB to the loss obtained. A further refinement in recent years adds the loss in the Class 5 office, assumed to be .5 dB, to each end of the connection. Thus the actual loss from the line side of one Class 5 office to the line side of the other is 5 dB plus VNL plus .4 dB times the number of trunks in the connection.

This is a fair amount of loss. If we go from New York to San Francisco, 2500 miles would require 3.85 dB from the VNLF alone, and if four trunks were in the connection (two toll connecting and two intertoll), we'd get another 1.6 dB, for a grand total of 10.45 dB. (Actually, distance related loss is rounded off so this isn't quite the right number, but the idea is clear). For the complete connection, loss from the customer locations to the Class 5 offices must be added.

To limit the maximum loss, trunks longer than 1850 miles, and thus requiring more than 2.78 dB of distance-related loss, are NOT operated VNL. Rather, they are operated at 0 loss, and have echo suppressors built in. This makes conversation easier on long trunks, and simplifies the administration of echo suppressors and their inclusion in built-up connections. Since the distance is relatively great, echo suppressors only show up in trunks from one major region of the country to another. Thus they tend to be between switches high in the DDD hierarchy, and they are accessed by short-haul trunks on both ends. Thus minimal precautions are required to prevent two echo-suppressors from being inserted in one connection. This is important, because each echo suppressor, acting independently, can lock out the other party and both callers can be shouting at each other simultaneously without a sound getting through. (Speaker-phones on each end don't do much to help things, either).

It is interesting to note that, at the short end, of the distribution of trunk lengths, VNL is 3 dB or more above what is required for "optimal" control of echoes (based on the above-mentioned human factors experiments), while in the range of distances beyond 1000 miles, it tracks very well. Since there are many more short-haul than long-haul trunks, it would appear that 2 or 3 dB loss in short-haul connections is of little importance. Indeed, AT&T's Notes on Distance Dialing, 1975, says, "Although the VNL plan provides more loss than optimum for short connections, the difference is not sufficiently great to have any appreciable effect on grade of service."

Pad Switching

For Intertoll trunks, VNL or echo suppressors are provided. For toll connecting trunks, loss is VNL +2 dB. In dial tandem networks used by large business customers, tie-trunks connect directly to PBXs. They are switched through for trunk-to-trunk connections, and to local station users for trunk-to-line connections. In the first case, tie-trunks should be connected together on a "Via" basis, without a 2 dB pad. In the second instance, however, where a "terminal" connection is made, the 2 dB pad must be inserted. This operation is called pad switching, and used to be common in the public network many years ago. However, when Class 5 offices were arranged to make line-to-line and line-to-trunk connections their main function in life, leaving trunk-to-trunk connections to Class 4 and higher offices, pad switching vanished in favor of 2 dB permanently associated with toll connecting trunks. It is vital for PBX designers to remember the lost art of "pad switching," since combined tandem and PBX switches are highly cost-effective. In particular, they require no "access lines" (the private network equivalent of toll connecting trunks) between local and long-haul switching, since the whole job is done in one machine.

TLP

In the public telephone network, trunks have been measured, traditionally, from outgoing switch to outgoing switch. This made a lot of sense in the days of manual and SXS switching, since a trunk circuit, interfacing the switching equipment to the transmission facility, was more nearly part of the latter than the former. The trunk circuit would provide for an appearance in front of the operators or the selectors, and continue this appearance on to the test desk. Thus the outgoing switch (or jack) was a logical point of access for everybody. It was defined as the 0 transmission level point, or 0 TLP, since a 1 milliwatt test tone at 1000 Hz is standard and, when loss (or gain) is measured in decibels (dB), a 1 mw tone is at the 0 or reference level.

In later systems, particularly Crossbar and ESS, the trunk circuit became more closely related to the switch than the transmission facility. Access to the trunk circuit could only be achieved via a path through the switch itself; in ESS, there is no direct path to the test desk. Thus the outgoing switch is not directly accessible as a test point. In No. 4 ESS, and in other digital switching systems where the boundary between trunk and switch has vanished completely, the "outgoing switch" concept has no meaning at all. It is likely, however, that half way into the next century, trunks will still be measured from outgoing switch to outgoing switch, whatever that will mean.

We have already seen that carrier systems accept inputs at -16 TLP, and provide outputs at +7 TLP. This shows how levels at different points in the office vary in the natural course of equipment usage. For testing and other purposes, it is desirable to know the deviation from standard rather than the absolute level at any given point. Thus if all measurements are referred to 0 TLP, the net behavior of the trunk can be discussed, independent of the point where the measurement was made. Or, to put it another way, a 1-mw tone at 0 TLP is equivalent to a .02512-mw tone at -16 TLP (.02512 is 16 dB below 1 mw) or a 5.0119-mw tone at +7 TLP. Measurements are made in dBm rather than milliwatts, and all are related to the 1 milliwatt reference level at 0 TLP.

Since 2 dB of loss is carefully placed in all toll-connecting trunks (to insure VNL + 4 dB loss on all toll connections), Class 4 offices and CSPs (control switching points—Class 3, 2 and 1 offices) are assumed to have their outgoing switches at -2 TLP. The idea here is that, if a 0 mw tone is applied at the outgoing switch of a Class 5 office, it should arrive at the outgoing switch of its nearby Class 4 offices 2 dB down. In any event, -2 TLP is the reference level for toll switches.

There are two objectives to all this, remember. First, test and line-up procedures require identical outputs for all facilities to simplify testing, and second, the actual desired loss in a facility must be easily measured and interpreted, regardless of the signal level at the point of measurement. This can best be seen with an example. Figure 3 shows a traditional analog trunk between two four-wire toll offices. Note that the test tone is assumed to be applied at the outgoing switch of one office, and the detector or meter is dialed up through the distant office and is, once again, assumed to be at the outgoing switch. When measurements are made in the opposite direction, the other switching matrix is included.

Figure 3 shows that both directions of carrier circuit have 23 dB of gain, and can be measured with a -16 dBm signal at the input and a meter intending to read +7 dBm at the output. Any deviation from one channel to the next can be easily spotted. Second, the loss from the outgoing switch to the carrier terminal can also be measured as a standard 14 dB value. Keep in mind that there are jacks for test access at most of these points in a crossbar office and also that such test access is a major investment.

Figure 3. Amplitude Gain and Loss in an Intertoll Trunk (Simplified). Measurements are assumed to be made from outgoing switch to outgoing switch.

The final point is the measurement from the output of the carrier system to the outgoing switch in the called toll office. Here is where the VNL is inserted. A test tone of +7 dBm at the carrier output (or -2 dBm at the outgoing switch of the distant office) should come out to be below the -2 TLP by the value of the Via Net Loss. This makes the insertion and checking of VNL relatively simple.

The pads shown in the diagram have to be selected to augment the loss in the wiring, jack-fields, trunk circuits etc., and to bring the total loss to the required value. That is, to go from the outgoing switch to the carrier input, you don't use a 14 dB pad. You use a pad that is smaller than 14 dB by the amount of loss already present. This requires careful administration, to say the least. Keeping in mind that the diagram is greatly simplified (note, for instance, that I have not shown the SF signaling set), a feeling for the true administrative talents of the Bell System in this area can begin to be appreciated.

With regard to the signaling sets, they are often located fairly close to the carrier systems (being designed to work at the +7 and -16 levels), and they incorporate the pads, among other things. In a two-wire Class 5 office, they also usually include the hybrid circuit that converts from 4-wire to 2-wire. Figure 4 shows another simplified sketch of losses, this time covering a toll-connecting trunk.

Figure 4. Amplitude Gain and Loss in a Toll Connecting Trunk (Simplified).

When measurements are made, the procedure is somewhat different from that suggested above. However, the value of the TLP concept can, hopefully, be appreciated in terms of standard levels and expected deviations from a known norm.

Reflections

At the Class 5 Central Office, the four-wire trunk meets the two-wire customer loop. This meeting is effected through a transformer arrangement (there are modern electronic equivalents presently coming into favor) called a hybrid circuit. A hybrid circuit has four inputs of "ports." It is designed so that any signal entering one port will, under certain circumstances, divide equally between two of the remaining ports and be "balanced-out" of the final port. If the input port comes from the incoming side of the carrier system, and the port with the balanced-out signal goes to the outgoing side of the carrier system, we can see that there will be no echo. However, the balancing procedure is never perfect, and some of the incoming signal leaks through the hybrid and enters the outgoing side of the carrier system. The trick is to minimize this leakage.

The leakage can only be minimized if the customer line is matched by a "balancing network." Since a customer loop has a size (impedance) that varies depending on how long it is, how many telephone sets are scattered along its length, what frequency is being used for the measurement, whether loading coils or bridged taps are present, etc., a good matching network is hard to find. Some years ago, 900 ohms of resistance in series with 2.14 microfarads of capacitance was defined to be the best simple network. Since any trunk must be able to connect to any line on the CO, and, in analog systems, the hybrid and its matching network are part of the trunk, the difficulty in obtaining a good match at the trunk against all possible customer connections can be appreciated.

Note that if the Class 5 office were 4-wire rather than 2-wire, the hybrid would be on the line side, associated with each individual 2-wire line.* In principle, at least, it would be possible to adjust the balancing network to fit the customer loop with which it is associated. This is not supposed to be economically reasonable, however. But with the hybrid as part of the trunk, a good match is not even theoretically possible, although networks better than 900 ohms + 2.14 microfarads have been suggested.

[*Footnote: There would also be a lot more of them.]

Summary

In this section, we have seen that, by its nature, a digital signal coded in companded PCM has its amplitude and phase locked together, independent of distance or any other parameters (as long as the line bit-error-rate is reasonable). And, even if you want to change the amplitude using digital techniques, there are problems.

By contrast, the analog network as it presently exists is a maze of different amplitudes, some required by history, others required to limit echoes. Continual level variation throughout any given office is encountered as a matter of course, making a Transmission Level Point necessary to give meaning to a measurement made at any particular location.

[ Top ] [ Table of Contents ] [ Next Chapter ]

Copyright 2005 Lee Goeller. All Rights Reserved.