Chuong 4.html

ADSL, VDSL, and Multicarrier Modulation . John A. C. Bingham Copyright # 2000 John Wiley & Sons, Inc. Print ISBN 0-471-29099-8 Electronic ISBN 0-471-20072-7

4

DSL SYSTEMS: CAPACITY, DUPLEXING, SPECTRAL COMPATIBILITY, AND SYSTEM MANAGEMENT

Capacity and methods of duplexing [i.e., echo cancellation, frequency-division duplexing (FDD), and time-division duplexing (TDD)]are discussed together here because for a crosstalk-dominated medium such as DSL, the capacity in one direction is very dependent on what is being transmitted in the other direction. The sum of downstream plus upstream maximum data rates is probably the best measure of capacity; in Sections 4.3 and 4.4 we describe how to maximize this and partition it between downstream and upstream.

In Section 4.5 we discuss spectral compatibility, which is essentially a way of controlling alien crosstalk. Without putting too much import on the meaning of compatible, two different systems A and B might be said to be fully compatible if A interferes with B no more than B does with itself, and vice versa. In Section 4.6 we discuss the impact of spectral compatibility and several other factors on management of a multipair physical layer xDSL system.

4.1 CAPACITY

The term capacity as used in this book is a measure of the data rate that can be transmitted through a channel, but it is not the theoretical limit developed by Shannon; it is a practical rate that depends not only on the signal/noise ratio, but also on the method of modulation and demodulation, the coding, the margin, and the required error rate. Each of the latter four factors deserves discussion.

4.1.1 Modulation and Demodulation

For any given channelÐand more speci®cally, for any given channel impulse response (IR)Ðthere are three different upper bounds for performance for single-carrier systems:

59

1. Matched ®lter bound. This is applicable if a symbol is received in isolation, uncontaminated by previous symbols.
2. Maximum likelihood bound. This is achieved if the receiver uses some type of maximum likelihood sequence detection (MLSD), usually a Viterbi detector, to take advantage of the interference from previous and subsequent symbols. If the channel is not too distorted and the IR is not too dispersed, the Viterbi detector can use all the energy in an IR for detection,¹ and the maximum likelihood bound is equal to the matched ®lter bound.
3. Decision feedback equalizer (DFE) bound. This is achieved when the receiver, after appropriate minimum mean squared error (MMSE) pre®ltering, subtracts the effects of previous decisions and bases its decision only on the main sample of the IR (de®ned as h₀).

Only the last of these, the DFE bound, concerns us here because (1) its performance can be directly related to that of a multicarrier system, and (2) the other two detection methods have not been considered for xDSL.²

4.1.2 Coding

Two methods of coding are typically used for DSL: Reed±Solomon forward error correction (R-S FEC) and trellis code modulation (TCM). A few details of the separate methods are given in Sections 8.2.4 and 8.2.7, but an analysis of the combination of the two would be extremely complex. For the moment we use crude estimates of the coding gains of the methods de®ned in T1.413: 2 dB (coding gain ˆ 1.58) for the FEC alone, and 4 dB (coding gain ˆ 2.51) for FEC plus TCM.

4.1.3 Margin

DSL service providers recognize that all measurements of loop impairments are merely samples, and the reported results are statistical. Therefore, to guarantee a particular service to their customers, they require that data and error rates be achieved with all anticipated crosstalk and noise levels increased by some margin m. For ADSL and VDSL 6 dB (m ˆ 4) is the accepted value.³

¹ In [Bingham, 1988]I coined the term compact to describe such an IR, but it did not seem to catch on.
²A lot of work has been done on precoding to make a multicarrier signal “®t” a distorted channel (see [Ruiz, 1989]and [Ruiz et al., 1992]and references therein). It is not clear (to me, anyway) whether such “vector coding” methods achieve the maximum likelihood bound of a channel.
³This is a very pessimistic strategy because, as we saw in Section 3.6, all crosstalk levels are already speci®ed as 1% worst case. There was much discussion in T1E1.4 as to whether all noise powers must be quadrupled. The “play-it-safe” conclusion was that they should be!

CAPACITY 61
4.1.4 Error Rate

Opinions about a tolerable bit error rate (BER) range from 10^À12 for highquality compressed video to 10^À4 for data transmission over ATM. For the present discussion we use the number speci®ed in T1.413: 10^À7.

4.1.5 The DFE Bound
The probability of a symbol error⁴ in the detector of a DFE for a squareconstellation QAM signal that conveys b bits (b even) is
r!
P
e
ˆ
4
kQ
3h₀=²
2^b À 1^…4:1†
where
₁^…I
Q…x†ˆ p e^Ày2dy …4:2† 2 x

b is the number of bits per symbol (i.e., 2^b is the number of points in the constellation), k an edge-effect correction factor that approaches unity for large b, h₀ the ®rst (main) sample of the IR, and ² the variance of the noise at the detector.

Equation (4.1) is only approximate for b odd, but it is an acceptable approximation for b55 and for calculations of overall capacity; for this purpose k ˆ 1 is also an acceptable approximation. For calculations of bit loading (see Section 5.3) the cases of b ˆ 1 and 3 must be treated separately. Then the simpli®ed form of (4.1) can be solved for b:

()
b
ˆ
log
2
1
‡
3h₀=
‰Q^À1…P_e=4†Š^{2 …4:3†}

[Price, 1972]showed how for a DFE, (4.3) can be modi®ed to be a function of the SNR at the input to the receiver. Then the data rate for a QAM signal in a Nyquist band from f₁ to f₂ with a margin mar and a coding gain cg is

()_{… f2}
^{R ˆ log 3SNR… f † df …4:4†2 1 ‡}…mar=cg†‰Q^À1…P_e=4†Š²_f1

where the limits of the band are such that SNR( f )51 for f₁ < f < f₂. We will see in Section 5.2 that this expression is applicable, with only minor modi®cation, to multicarrier as well, so from now on we use it for capacity calculations.

⁴When using a byte-organized R-S FEC, symbol errors containing only a few bit errors are corrected with the same ef®ciency as bit errors.
4.2 DUPLEXING METHODS
The ef®ciency of a duplexing scheme can be de®ned as
_{” ˆ} total…down ‡ up† data rate_{…4:5†capacity}
We consider three methods of duplexing and the ef®ciencies they can achieve.
4.2.1 Terminology

Before we consider echo canceling (EC) and frequency-division duplexing (FDD) we must clarify what these terms mean: as descriptions of both duplexing strategies and/or implementational tactics.

^* “Pure” EC (Section 4.2.2) means the simultaneous use of the entire band for transmission in both directions; it de®nes both the strategy and the tactics. It is used in HDSL.

^* “Pure” FDD (Section 4.2.3), strictly speaking, refers only to the strategy; it is one of the options for ADSL. It is also often used to describe the tactics of separating the bands by ®ltering; this, however, can be misleading because Zipper, proposed for VDSL (see Section 10.4) is FDD without ®ltering.

^* EC/FDD (Section 4.2.4) is a mixture: duplex transmission up to some frequency and simplex with FDD above that. It is used in HDSL2, which divides the band into three parts: both/downstream/upstream. The other option for ADSLÐEC up to 138 kHz and only downstream above thatÐ is a special case of EC / FDD.

^* FDD(EC)⁵ (Section 11.4) uses FDD as a duplexing strategy but uses EC instead of ®ltering to separate the signals.
4.2.2 Echo Canceling

The most ef®cient way to use any bidirectional medium would seem to be to use EC to remove the re¯ection of the locally transmitted transmit signal, and transmit in both directions simultaneously using the full available bandwidth for both. For voiceband modems where the noise is (or at least is assumed to be) independent of any other simultaneous transmissionsÐwhat we are calling alien noiseÐthis is the best strategy, and an ef®ciency of 100% can be achieved.

⁵This has been called pseudo-FDD, but that is misleading because it really is FDDuplexing.
DUPLEXING METHODS 63

Simultaneous transmission in both directions on a DSL, however, causes kindred NEXT; this may require a different approach to duplex transmission. Figure 3.13 shows the levels of signal and kindred NEXT for 12 kft of a 25-pair cable of 24 AWG loaded with 24 kindred systemsÐa typical medium ADSL loopÐand Figure 3.14 shows the same for 3 kftÐa typical VDSL loop. The SNR that is needed for transmission depends on the modulation method, the margin, and the coding gain, but a value of 10 dB at the band edge is a reasonable one to use for our ®rst discussion. It can be seen that the signal to NEXT ratios are 10 dB for the two loops at approximately 0.32 and 1.8 MHz. Clearly, simultaneous duplex transmission beyond these frequencies would be impossible; whether it would even be desirable below those frequencies remains to be seen.

4.2.3 Frequency-Division Duplexing
Frequency-division duplexing (FDD) is very well established and understood, so not much needs to be said. The frequency-domain ef®ciency is
_{“fd ˆ} downstream bandwidth ‡ upstream bandwidth_{…4:6†2 Â total channel bandwidth}

which, with practical ®lters, is typically limited to about 40%. The data ef®ciency, as de®ned by (4.5), may be slightly more or less than this, depending on the variation of SNR across the band.

4.2.4 EC/FDD

Figures 3.13 and 3.14 suggest that a mixed strategyÐduplex transmission using EC up to some crossover frequency, and simplex transmission using FDD above thatÐmight be best. Figure 4.1 shows the aggregate (downstream plus upstream) data rates as a function of this frequency that are achievable (assuming perfect band-separating and perfect EC⁶) on 12 kft of 24 AWG under two different 1% worst-case noise/crosstalk conditions:

1. With 20 ADSL systems as crosstalkers
2. With 10 ADSL systems and 10 HDSL systems⁷ as crosstalkers

It can be seen that if there is a mixture of ADSL and HDSL in the binder group, the band up to about 200 kHz is already degraded by the full-duplex HDSL, so the ADSL might as well use it similarly: that is, EC up to about 200

⁶ Neither is realistic of course, but we are interested here only in comparisons, not in absolute performance.
⁷HDSL is an average interferer: not as bad as T1 and worse than BR ISDN; see Annex B.2 of T1.413 for a de®nition of its PSD.

Figure 4.1 Total (downstream plus upstream) data rates on 12 kft of 26 AWG.

kHz and FDD or TDD beyond. This is the strategy that was adopted (somewhat serendipitously!) for ADSL in T1.413: EC up to a nominal 134 kHz, and only downstream beyond that. If, on the other hand, there are only kindred systems in the cable, the total capacity is maximized by using echo cancellation up to only about 60 kHz; in fact, the maximum is so broad that the bene®ts of any EC at all are very small. This is an important point; if there is to be mass deployment of ADSL, it will soon greatly outnumber older systems, and EC will become almost useless.

4.2.5 Time-Division Duplexing⁸

Time-division duplexing (TDD) is, in an imprecise way, the dual of FDD, but it is simpler in that the system design does not depend on complicated ®lter calculations. A superframe is de®ned comprising, in sequence, downstream transmission, a quiet or guard period, upstream transmission, and a second quiet period: d/q/u/q. Then the data rate ef®ciency is equal to the time-domain

⁸TDD has also been called time compression multiplexing, but TCM is now the abbrebiation for trellis code modulation. Only TDD should be used from now on.
CAPACITY REVISITED 65
ef®ciency:
tdown ‡ tup
^{” ˆ “td ˆ} 2Â…t_down ‡ t_up ‡ 2t_guard†^…4:7†

One system constraint is that the guard period must be greater than the one-way propagation delay, t_prop, so it might seem that nearly 50% ef®ciency could be achieved by making the superframe period (t_down ‡ t_up ‡ 2t_guard) very long. There is usually, however, a system constraint on the maximum latency,⁹t_lat ; the superframe period must be less than this. Therefore, the maximum ef®ciency

“_max ˆ “_td ˆ ^{tlat À 2tprop} …4:8†_tlat

For both ADSL and VDSL systems the speci®cations of t_prop and t_lat would allow an implementation of TDD that achieves about 45% ef®ciency, but nevertheless, EC/FDD was chosen for ADSL.

Early Use of TDD. TDD was selected as the multiplexing technique for BRI in Japan (see Annex III of ITU-G.961 and Section 9.3) and was also considered for Europe.

Synchronized TDD. Because TDD uses the full band for transmission in each direction, kindred NEXT would be fatal; it must be avoided by synchronizing the frames in all the systems in a binder group. This is done by using the same frame clock for all transmitters in a CO or ONU, and loop timing (i.e., frequency and time locking to the downstream signal) at all the remotes (see Section 10.2). This synchronization may be more dif®cult now that unbundling, the leasing of pairs in a cable (probably without the associated frame clock) to a competitive local exchange carrier (CLEC) and/or an Internet service provider (ISP), is becoming common. The pros and cons of synchronized DMT (SDMT) for VDSL are discussed in detail in Chapter 10.

4.3 CAPACITY REVISITED

With the general principles of duplexing established above, it will be useful to calculate downstream and upstream capacities with 10 ADSL and 10 HDSL crosstalkers under three different sets of conditions:
1.3. Margin ˆ6dB
1.4. Total coding gain ˆ 4 dB (i.e., R-S FEC and TCM)

1. T1.413 guaranteed performance: that is,
1.1. All crosstalks at 1% worst-case level
1.2. EC up to 130 kHz; only downstream above that

⁹This is a ridiculous name for a system processing delay, but I have given up ®ghting!

2. Same as condition set 1 except that only FDD is used (upstream up to 120 kHz, downstream from 155 kHz)
3. Average performance: that is,
3.1. Crosstalk at median level
3.2. FDD as in condition set 2
3.3. Margin ˆ0dB
3.4. Coding gain ˆ 2 dB (i.e., no TCM)

The downstream and upstream rates are plotted as a function of length for 24 AWG in Figure 4.2(a) and (b). It must be emphasized that these ®gures are for idealized conditions: no bridge taps, front-end noise À140 dBm/Hz, no VDSL crosstalk. As be®ts such conditions, the “T1.413” rates at 12 kft are about 15% higher than speci®ed for CSA loop 8.

The new and enticing part of these results is what can be achieved “on average.” The high rates at 10 and 12 kft are not particularly interesting, but the more than threefold increase over the conventionally speci®ed system beyond 18 kft is very intriguing. It must be emphasized, however, that for these very long loops and with kindred FEXT as the only signi®cant crosstalk, the total “noise” is dominated by the so-called AWGN, which was optimisticallyÐand many would argue, unrealisticallyÐset at À140 dBm/Hz. Achieving this level of front-end noise will be a severe and crucial challenge for the analog designer, because with a 20-kft loop, each decibel above À140 dBm / Hz costs approximately 150 kbit / s of downstream data rate. One consolation to be derived from the fact that for the long loops the performance is noise limited is that the available data rate does not change much with the number of interferers, thus simplifying the task of dynamic rate adaptation (see Section 4.6.2).

4.4 A DECISION: EC OR NOT?

As we have seen, compared to conventional FDD (downstream above upstream), EC increases the downstream rate by using the low band for transmission but decreases all upstream rates by subjecting signals received at the CO on that pair to uncanceled echo, and on all other pairs to kindred NEXT. The frequency range over which the increase exceeds the decrease (for a net increase in duplex capacity) decreases with loop length.

For ADSL applications within the CSA¹⁰for which (1) the downstream rate is paramount and (2), a lot of alien NEXT is expected, EC in the band up to 138 kHz, as de®ned by T1.413, was a good choice. On the other hand, for longer

¹⁰For example, video on demand (VoD). A DECISION: EC OR NOT? 67
Figure 4.2 Data rates for 24 AWG as a function of loop length: (a) downstream; (b) upstream.

loops and ADSL¹¹ applications in which (1) a ratio of upstream to downstream rates of at least 1:8 is needed and (2) ADSL is the majority service in the cable, the disadvantages of ECÐreduced upstream rate, much greater complexity, longer training time, and greater dif®culty of spectrum managementÐoutweigh its advantages.

Furthermore, as discussed in Section 9.1.3, if the customer-premises unit is operated without the low-pass ®lter part of the splitter (“splitterless”), it may be necessary to reduce the upstream transmit power in order to avoid noise in the telephone speaker. In this case the signal received at the CO would be even more vulnerable to kindred NEXT and uncanceled echo. EC is perhaps not obsolete, but for all the reasons discussed above it has to be concluded that for most of the emerging ADSL market it is obsolescent.

Loops with Very High Noise. On loops with noise according to model B in Annex H of T1.413¹²the usable band for ADSL downstream extends up to only about 400 kHz. Use of the band up to 150 kHz for downstreamÐmade possible by ECÐis highly desirable, perhaps even essential.

4.5 SPECTRAL COMPATIBILITY

The (physical layer) system that we should consider here is the combination of a large feeder cable, an FDI, and many distribution cables, shown in Figure 3.1. These together have to deliver a variety of xDSL services to customers over a wide range of distances from the CO (some perhaps < 5 kft, some > 20 kft).

One basic principle of design for xDSL systems is that each new xDSL system should¹³ meet its rate/range requirements with crosstalk from kindred systems or any and all previously standardized (“old”) systems in the same binder group. I know of only two exceptions to this:

1. T1.413 did not de®ne any tests with ADSL and T1 systems in the same binder group; crosstalk from an adjacent binder group is assumed to be reduced by 10 dB. The implication was that local exchange carriers (LECs) would keep ADSL and T1 separate all the way out to the customer premises.

2. The speci®cations of ADSL out-of-band PSDs were tightened in Issue 2 of T1.413 to reduce crosstalk into VDSL, a newer system.

¹¹ Only ADSL is considered in this section because the TDD system proposed for VDSL avoids echoes and NEXT completely.
¹²This model is for European loops; it de®nes the noise as À100 dBm/Hz from 10 to 300 kHz and thus avoids all problems of de®ning crosstalk. I do not know how realistic it is.
¹³I use “should” rather than the “shall” of standards because the entire subject is still too vague for imperatives.

SPECTRAL COMPATIBILITY 69

This principle might be called spectral optimization; it is what a lot of the rest of this book is about. The complementary principle of spectral compatibility is that each new system should cause no more degradation (via crosstalk) to existing systems than they do to themselves. “No more degradation” is, however, controversial, and three interpretations have been proposed:

1. The PSD of the new system must fall below that of at least one old system at all frequencies. This would guarantee that the new system would cause less crosstalk than would the established system.

2. The PSD of the new system should fall below a superset of old systems; that is, at any frequency a new PSD may be as high as the highest old one.
3. The capacity of the old systems with crosstalk from the new system must be no less than with any mix of old systems [Zimmerman, 1997a and 1997c].

It has been recognized by the spectral compatibility group of T1E1.4 that rule 1 is too stringent; it would unnecessarily inhibit the future development of systems. Indeed, neither ADSL nor the emerging VDSL meets this requirement; their bandwidths are much greater than anything previously used. On the other hand, rule 2 is too lax; Figure 4.3 shows the PSDs of BRI, HDSL, and upstream ADSL, and the maxima thereof. A new system could, in aggregate, be a more severe disturber than any old one.

Figure 4.3 Standardized PSDs.

Rule 3 requires the calculation of capacity according to either equation (4.3) for DMT or similar equations given in [Zimmerman, 1998]for PAM, QAM, and CAP systems. It allows the trade-off of higher PSD (and therefore higher crosstalk) in some frequency regions (preferably those where the SNR is already low) for lower in other regions. The ®rst application of this rule was in the PSD of HDSL2,¹⁴which is shown in Figure 4.4, and it is likely that T1E1.4 will issue a standard based on rule 3 by mid-1999.

Figure 4.4 PSD of HDSL2. ¹⁴Such a widely varying PSD in the passband would be a natural for DMT, which can achieve such variations without causing signal distortion. I regret that DMT was not pushed harder for HDSL2.
4.6 SYSTEM MANAGEMENT
Management of the xDSL system shown in Figure 3.1 requires many decisions by a LEC; some of them are:

1. How to offer unbundled pairs to a competitive LEC while maintaining spectral compatibility and other necessities for service quality
2. What mix of data rates and services can be offered, and whether these should be adaptive
3. Whether to impose controls on some transmit PSDs beyond those de®ned by the standard
4. Whether to enable some of the options de®ned in the standard (e.g., EC in ADSL)
5. In which binder group to put the pairs carrying different services

These problems are discussed in Sections 4.6.1 through 4.6.5.
4.6.1 Local Exchange Carriers: Incumbent and Competitive

Traditionally, the loop plant has been owned and operated by LECs; these were originally part of Bell Telephone and then of the regional bell operating companies (RBOCs). The LECs owned the plant and provided the services thereon, so the only criterion by which they were judged was the quality of those services. The Telecommunications Act of 1996, however, required that in exchange for permission to compete in the long-distance transmission market, the incumbent LECs (ILECs) must “unbundle” their cables and lease some of their loops to competitive LECs (CLECs), which could provide some DSL services. Section 271 of the act de®ned the criteria by which it would be judged whether the local market had indeed become competitive.

One of the ways in which compliance with section 271 could be demonstrated is by providing to the CLECs more information about the loops and their datacarrying potential.¹⁵ In ascending order of sophistication, the details of each leased loop that are needed are:

1. Assurance that loading coils and repeaters have been removed.
2. Length and dc resistance: usually readily available.
3. DSL services presently provided and contemplated for all other pairs in both the feeder and distribution binder groups. This is discussed in more detail in Section 4.6.5.
4. The ADSL capacity of that loop with some agreed-upon set of interferers. In Section 4.3 we showed how to analyze a loop given its length and gauge and the position, length, and gauge of all bridge taps, and in Section 4.1.5 we showed how to calculate the capacity. It has been claimed, however [Sapphyre, 1998], that such detailed information of the frequency response is not needed and that capacity can be calculated approximately just from measurements made in the voiceband using a V.34 modem. Whether the method can (and will) be widely used remains to be seen.

¹⁵Capacity is de®ned in Chapter 4, but it is too precise a word for here.

Reverse ADSL. If space for its equipment is not available in the CO, a CLEC may use a nearby building, and access the CO via other pairs that are crossconnected in the CO to the pairs going to its customers. This reverse ADSL (transmitting a downstream signal upstream, and vice versa) con®guration is shown in Figure 4.5, where the CLEC customer is shown closer to the CO to emphasize that the range is reduced by l_B, the distance of the CLEC’s of®ce from the CO. Such an arrangement would, strictly speaking, violate many of the spectral rules discussed in the preceding section, but a pragmatic relaxing of the rules is needed; it can be shown as follows that the crosstalk into orthodox ADSL pairs can be held to an acceptable level.

Figure 4.5 shows two other types of crosstalk that will be generated from the CLEC’s “ATU-C” on pair B¹⁶ into pair A: ANEXT into the downstream and, if the CLEC uses EC, UL FEXT into the upstream. A reasonable requirement is that both of these should be less than what conventional ADSL systems to do themselves: that is, kindred FEXT and kinderd NEXT, respectively. The limitations on l_B can therefore be calculated as follows.

ANEXT. In the downstream band
p

15 log… f†À …l^p …4:9†_A À l_B† <20 log… f†‡ 10 log…l_A†À l_A f Figure 4.5 Reverse ADSL to access ATU-C outside the CO. ¹⁶ The crosstalk from the CLEC’s upstream signal going downstream on pair B will be attenuated and inconsequential.
That is, _{lB <} 5 log… f†‡ 10 log…l_A†_{…4:10a† }
f

For 26 AWG, the most common gauge out of the CO, %7.8 dB/kft, and the CSA limit for l_A is 9 kft, so that at f ˆ 0.3 MHz (the “sweet” part of the downstream band) l_B < 1:63 kft, and at f ˆ 1.0 MHz l_B <1.22 kft. By averaging the crosstalks across the whole downstream band a conservative constraint is that

l_B < 1:5 kft …4:10b†

ULFEXT. At f ˆ 0.12 MHz, the upper limit of the upstream band, ULFEXT is greatest at approximately 1.5 kft, so it should be compared with kindred NEXT at this frequency and range. Downstream and upstream PSDs are À40 and À38 dBm/Hz, respectively, so

NEXTˆÀ40 À 51 ‡ 15 log…0:12†
%À105 dBm=Hz …4:11†
^{p}ULFEXTˆÀ38 À 51 ‡ 20 log…0:12†‡ 10 log…1:5†À 6:2  1:5  0:12 %À109 dBm=Hz …4:12†

Therefore, the ULFEXT constraint on l_B is looser than the ANEXT constraint. Hence, if the reverse distance is limited to less than 1.5 kft, reverse ADSL will present no problems to any conventional systems. Since it is in the CLEC’s interest to minimize this distance because it subtracts from the achievable range to their customers, the proposed limit of 1.5 kft should be acceptable to everybody. If the distribution binder-group already contains T1, reverse ADSL from a 1.5 kft distance would generate much less NEXT than would the T1, and the 1.5 kft limit could be stretched signi®cantly if the ILEC and the CLEC could agree on a rule.

Extended Reverse ADSL. Another type of reverse ADSL does not use a crossconnect to a conventional pair, but terminates the loop with an ATU-R in the CO. The services offered in this case are totally different, but the NEXT dependence on l_B would be exactly the same as in the preceding case. Conservatively, x should be limited to 1.5 kft, but could be stretched depending on what else is in the binder group.

4.6.2 Mix of Data Rates and Rate Adaptation

One of the big advantages of DMT is its rate adaptability, and basic rate adaptation in the PMD layerЮnding the maximum data rate that a particular line at a particular time (startup or any time thereafter) can transportÐis relatively straightforward; adaptation at startup and on-line are discussed in Sections 5.3 and 8.5.4, respectively. Rate adaptation in the higher layers is, however, much more complicated. Section 4.3 shows that there is a big difference between maximum guaranteeable and average data rates. This difference gets even bigger if one considers loops that are worse than the sets de®ned in T1.413. I believe that the biggest challenge in the wide deployment and acceptance of ADSL is at the higher layers; can they deal with (and get the most out of) a medium that may offer a range of data rates as high as 6: 1? Some of the questions that must be considered are:

^* Can dynamic rate adaptation (DRA) be used effectively at the higher layers?
^* Can all applications in the higher layers accept a short break in transmission during a rate change, or must DRA be restricted for some applications?
^* Would a “seamless” method of DRA (i.e., with no errors or interruption of service during a rate change) be more acceptable?
^* Even if a high data rate is possible upon startup (because there are no interferers), should the rate be limited to that which can be achieved on that loop under some “almost-worst-case” scenario?
^* How much information can the LEC provide about how many potential interferers there are in the binder group: that is, about the worst case for that loop?
^* How can that information be passed to the ATU-C or VTU-O so that it can be used during startup?

4.6.3 PSD Controls

The usual interpretation of the roles of spectral compatibility and system management is that spectral compatibility should be ensured by standards and that system management should accept the spectra thus de®ned and work with them. There may, however, be some circumstances under which more systemspeci®c control over the PSDs is desirable. Examples of these are:

^* If at any time a receiver determines that the margin for the requested service is unnecessarily high, the performance of the system as a whole would be improved if the transmitter at the other end transmitted at a lower PSD level. Because the receiver sends PSD instructions to the transmitter via the g_i (see Section 8.5.1), it can, if told to do so by a higher layer, reduce the far-end’s transmit PSD.

^* In a VDSL system, unequal-level FEXT (ULFEXT) from a close upstream transmitter can be a serious impairment (see Section 10.1.2 for more details); if this is not controlled explicitly in the eventual VDSL standard, system management should do so.

4.6.4 Enabling or Disabling Options

There are two types of options that may be enabled or disabled during initialization of a pair of modems: those that can be negotiated by the modems themselves and do not require any systems management (e.g., a feature such as trellis coding would be used if both modems have the capability) and those that require a higher-layer decision (e.g., even if both modems have the capability, is this feature desirable for the system?). We need to consider here only the second type; of the following, the ®rst two are allowed for by T1.413, and the third will have to be anticipated in VDSL:

1. EC. The possibility of a mixture of EC and FDD ADSL systems¹⁷ in the same binder group raises the question of what the goal is for the overall system. Is it to maximize the sum of the data rates on all the loops, or to maximize the lowest data rate? There are certainly some cables for which these two goals would be opposing, but most service providers would probably choose the “maxmin” goal. The path to that goal, however, is tortuous. On very long loops (> 20 kft) the downstream rate might be signi®cantly increased by using EC, but the upstream rate could be maintained only if the other ADSL systems in the binder group did not use EC!

2. DRA. If the application layer cannot deal with a physical layer that changes data rate during a session, DRA must be disabled.
3. Upstream power control. This is particularly important in a TDD implementation of VDSL, where the upstream signal uses the full frequency band; ULFEXT from a full-power, close-in upstream transmitter would be lethal to other VTU-O receivers.

4.6.5 Binder-Group Management

LECs have very little freedom in populating the binder groups in the distribution (F2) cables going to the customer premises; the mix of services must be determined by customer requirements. On the other hand, they have a lot of freedom in the binder groups in the feeder (F1), cable emerging from the CO; the cross connections in the FDI can be used to transfer pairs as needed.

Segregation Out of the CO. An important question is: Can this freedom to con®gure the binder groups out of the CO be used advantageously? ¹⁷ Kindred certainly, but brothers and sisters often ®ght!

The choice, basically, is between binder-group continuity, in which, as far as possible, pairs leave the CO in what eventually becomes the distribution binder group (thereby minimizing cross connections), and binder-group homogeneity, in which the binder groups in the feeder cable contain, as far as possible, kindred¹⁸ systems. The choice determines the amount of alien NEXT (aNEXT) at the CO, and so affects the upstream performance.

In the band below 138 kHz the PSDs of ADSL downstream and HDSL are similar (ADSL is about 2 dB higher), so if echo-canceled ADSL systems are allowed, the three arrangements ÐADSL only, HDSL only, and mixedÐall generate similar levels of NEXT. There is little difference between binder group continuity and homogeneity; management cannot achieve much.

If, however, EC ADSL systems are not allowed, binder-group homogeneity would mean that the upstream ADSL experiences only kFEXT from pairs in the same binder group and aNEXT from pairs in other binder groups. This does not, of course, completely eliminate aNEXT, but reduces it by at least 10 dB, which is the number generally agreed upon for adjacent binder groups. The upstream capacity would be considerably increased, and this will become important as ADSL ranges are stretched and other quasi-symmetrical services are offered.¹⁹

Some LECs use homogeneous binder groups and, furthermore, try not to put different DSL systems in adjacent binder groups: thereby apparently reducing aNEXT even further. However, when the cable is spliced, which may be only 300 ft from the CO, there is no guarantee that the relative positions of the binder groups in the cable will be maintained; two separated binder groups may become adjacent. The conservative way to estimate upstream capacities is therefore to assume aNEXT at adjacent binder-group levels right at the CO.

Segregation of T1. Below about 200 kHz, NEXT from both downstream ADSL and HDSL is higher than from T1, so it is not necessary to segregate T1 in order to protect upstream ADSL. However, as discussed in Section 3.6.2, AmpFEXT from downstream T1 would be a signi®cant source²⁰of crosstalk into downstream ADSL if T1s were not segregated. It appears that T1 systems should be kept in separate binder groups wherever possible: that is, in both the feeder and distribution cables.

4.6.6 Rates, Ranges, or Numbers of Customers?

Modem engineers are typically concerned with rate as a function of loop length and amount of crosstalk, system engineers with range at a given rate, and LEC business managers with the percentage of their customers who can be offered a particular service. It is interesting to consider the relation between these three parameters.

¹⁸ The upstream capacity of pairs in a homogeneous HDSL binder group would be reduced slightly because of increased kNEXT, but HDSL systems should have been originally engineered to expect the worst-case XT anyway.
¹⁹For example, a videoconferencing requirement of 384-kbit/s symmetric plus whatever ABR is possible.
²⁰And, depending on the location of the ATU-R relative to the repeaters, might even occasionally be the dominant source.

HDSL in Same or Adjacent Distribution Binder Group. As an example, consider ADSL downstream transmission on long loops (about 18 kft) with HDSL in the same or in an adjacent binder group in the distribution cable.²¹²¹ AWG loops. It can be seen that, expressed as a percentage increase from the same to the adjacent binder group,

ÁRate…1707†) ÁRange…137† > ÁNumber of customes…57†…4:13†

A 5% increase in the number of potential customers does not increase LEC revenues by much, so by this criterion there would be little incentive to manage the distribution binder groups or to improve the equalizers.

Variation with Crosstalk Conditions. As another example, Table 4.2 shows the variation of achievable downstream rates (on a 16.5-kft loop of 24 AWG) with crosstalk conditions: from no crosstalk (someone who signs on at 3 A.M.?) to the

TABLE 4.1 Improvement in Rate, Range or Percentage of Customers Reached Rate at 18 kft Range at 1.544 Percentage of Customers (Mbit/s) Mbit/s (kft) Served (note 1)

HDSL in samebinder group 0.68 16.5 79
HDSL in adjacent binder group 1.84 18.6 83
(note2)

Notes :
1. From Figure 10.3 of [AT&T, 1982]: these percentages are national averages; they may vary signi®cantly from oneLEC to another.
2. With only ADSL FEXT, the performance becomes very dependent on front-end noise level and residual distortion. These numbers assume À135 dBm/Hz noise and nearly perfect equalization: both very optimistic!

TABLE 4.2 Variation of Achievable Data Rate with Crosstalk Conditions Rate(Mbit/s)

Noiseonly 3.18
1 average ADSL crosstalker 3.03
1 average HDSL crosstalker in same binder group 2.33
Full worst-case(10 ADSL ‡ 10 HDSL) 1.544

²¹In a new residential area an LEC might consider offering only ADSL in order to increase the downstream rate and/or range.
full 10 ADSL and 10 HDSL crosstalkers speci®ed in T1.413 (achieving basic 1.544-Mbit/s service).

The Lure of Higher Data Rates. In both cases above it can be seen that a more than 2:1 increase in data rate could frequently be offered if the TC and higher layers could be set up to take advantage of them.²²

4.7. SPECTRAL MANAGEMENT STANDARD: STATUS, FALL 1999
A draft standard was sent out for letter ballot [T1E1, 1999], but has provoked strong objections. The main controversies are:

^* Should this be a standard for spectral management, or should it merely de®ne spectral compatibility, and leave the management responsibility to the FCC?

^* Guarded systems are de®ned as those with which all new systems must be “spectrally compatible,”²³ but what systems are guarded: only those that have been standardized, or all systems that have so far been “widely deployed?” Does the category include, for example, repeatered HDSL (a favorite of ILECs) and/or various non-standard SDSLs (favorites of CLECs), both of which can interfere with ADSL more severely than any previously considered interferers?

The proposed standard was regarded by some CLECs as too protective of the status quo and restrictive on new and innovative services, and they feared that it would be used by ILECs as a reason (excuse?) for refusing to lease pairs to CLECs.²⁴ Conversely, it was considered by some manufacturers and operators to be too permissive concerning unstandardized (and therefore uncontrolled) systems. How this will be resolved remains to be seen.

²² And the front end and the equalizer can be designed to reduce added noise and distortion to below À135 dBm/Hz.
²³Recall that compatibility is bi-directional.
²⁴This has been thought and spoken, but perhaps never published before now!