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
10
VDSL: REQUIREMENTS AND IMPLEMENTATION
VDSL is not as mature as ADSL, so in this chapter I can only de®ne the system requirements, and describe and compare proposals for the duplexing and modulation. The ITUwill probably choose one (or a combination) sometime after I submit the manuscript of this book (March, 1999) and before its publication.1The process by which the choice will be made is very controversial. For ADSL the choice was based on the results of competitive tests, but sentiment is running against these for VDSL. The ideal alternative would be complete disclosure by all proponents of both the theory and practice of their method followed by objective analysis and discussion. I know that to hope for this would be naively unrealistic; corporate alliances, investment in committed silicon, intellectual hubris, managerial incompetence, backroom cajoling and armtwisting, shortsightedness, and a difference of interest between the telcos and the modem manufacturers will all play a part.
There are three candidates for the duplexing method: FDD, an interesting technique called Zipper, which is a way of doing FDD with a minimum of ®ltering, and S(ynchronized)TDD; these are discussed in Sections 10.3 through 10.5. It is unlikely that STDD will be chosen because of the telcos’ fear that it will be dif®cult (perhaps occasionally impossible) to synchronize all the VTUs in a cable.2Nevertheless, some form of synchronized DMT will be needed for a VDSL system that has to be binder-group compatible with TDD ISDN (see Section 9.3); a form of Zipper has been proposed for this also. In the last section of this chapter I will try to be unbiased3and provide an objective comparison between the two pairs of candidates: FDD and Zipper for “general-purpose” VDSL, and SDMT and Zipper for ISDN-synchronized VDSL.
There are also three candidates for the modulation method: single-carrier QAM and CAP, and DMT, but I think readers will empathize if I discuss only the last one. All three are discussed in [Ciof® et al., 1999]: a good survey paper. This
1 The worst possible timing: the reader will know more than the author!
2More technical discussion of this in Sections 10.5.3 and 10.7.12, and a personal opinion in Section 10.8.
3I originally proposed it for VDSL, and I still believe it would be the best, so that will be hard!
201
chapter is much less detailed than Chapter 8 on ADSL for two reasons: many of the implementation details are similar to ADSL and need not be repeated, and many other details have not yet been worked out.
10.1 SYSTEM REQUIREMENTS AND CONSEQUENCES THEREOF
The requirements are completely de®ned4in [Ciof®, 1998] and [ETSI, 1998]. For our purposes the six most important are (1) the services/ranges/rates combinations; (2) transmit PSDs; (3) compatibility with wireless systems, particularly AM and amateur (“ham”) radio; (4) coexistence with ADSL; (5) operation on the same pair with BRI; and (6) position of the network termination (NT); these are discussed in the following subsections. Some of the consequences of these requirements are common to any method of implementation; these are discussed in this section. Some are speci®c to either SDMT or Zipper and are discussed in Section 10.4 or 10.5.
TABLE 10.1 VDSL Services, Ranges, and Rates Asymmetric Long “Extended” Symmetric Notes: 1 (0.3 km) 52 6.4 note 5 34 34 4 Two important early inputs to the process of developing systems requirements were a conference of mostly European and Asian telcos [FSAN, 1996] and [Foster et al., 1997], in which the “al.” was many of the participants in that conference.
Service Mode and Loop
Short
Medium
Very long
Very short
Short
Medium
Long
Very long
Extra long
1. These conversions between kft and km are about 9% inaccurate; it is not clear whether the kft or the km should be paramount. We consider the (shorter) kft ranges in this chapter.
2. The very long 6.5/0.8-Mbit /s service thus overlaps the ADSL service (but with a much shorter range!) I would not dare to predict what this portends for the future coexistence of ADSL and VDSL modems! 3. Intermediate asymmetries of down/up ˆ 4:1 and 2:1 have not been speci®ed, but may be useful later. 4. Interest has been expressed in some countries for a service on a 0.9-mm (approximately 19-AWG) loop as long as 3.4 km. The ITU may have to take this into account in their deliberations.
5. These ranges have not yet been de®ned.
Range Downstream Rate Upstream Rate (kft) (Mbit/s) (Mbit/s)
3 (1 km) 26 3.2
(note 1)
4.5 (1.5 km) 13 1.6
(note 1)
6 6.5 1.6 or 0.8 (note 2) (note 2) (note 3) (note 4)
1 26 26
3 13 13
4.5 6.5 6.5 note 5 4.3 4.3 note 5 2.3 2.3
Both asymmetric (with a down/up ratio of 8 : 1) and symmetric services should be supported: preferably by a simple recon®guration of one versatile modem. The various combinations of services, ranges, and rates are de®ned in Table 10.1. How many of these combinations will have to coexist in the same binder group will depend on the telco/LEC. ADSL loop lengths within one distribution binder group typically do not vary by more than about Æ 30% about some average, and offering every customer in a binder group the same rate is not very suboptimal. The variance of the shorter VDSL loops, however, will probably be much greater, and service providers will have to decide whether to “preserve the convoy”5or to allow different speeds (with all the dangers of collision). One particular symm/asymm combination that was predicted in [Foster et al., 1997] is short-range symmetric (for business customers) and long-range asymmetric (for domestic customers).
The services in a binder group can be summarized:
(a) Homogeneous binder group (all asymmetric or all symmetric) (b) Mixed binder group (long asymmetric and short symmetric) (c) Mixed with similar ranges (and therefore aggregate rates) for
asymmetric and symmetric
and the expectation is that prob(a) > prob(b) > prob(c). It is desirable (though not essential) that the range/rate for any particular service should not be reduced by the need to provide for a service with a lower probability (i.e., homogeneous gets top priority, etc.).
10.1.2 Transmit PSDs and Bit Loading
The ®rst proposal was for a maximum PSD of À60 dBm/HzÐmodi®ed for compatibility with ham radio (see Section 10.1.5)Ðacross the entire used band. The lower limit of this band was set at 0.3 MHz (see Section 10.1.4), and the upper limit of the usable band for very short loops is about 15 MHz. At À60 dBm/Hz the total power in this band (excluding the ham bands) would be about 11.0 dBm; the maximum total transmit power is speci®ed as 11.5 dBm.
PSD “boosts” to as high as À50 dBm/Hz outside the ADSL band will probably be allowed, but the total power would still be limited to 11.5 dBm. A mask as shown in [Ciof®, 1998] would allow up to 17.5 dBm if the maximum were transmitted at every frequency, or looked at another way, would limit the bandwidth used to about 1.4 MHz if the maximum allowed PSD were used. Clearly, these two inconsistent speci®cations will allow a lot of ¯exibility in the choice of PSD and bit loading.
5“The speed of a convoy is that of the slowest ship.”
If the PSD is limited to À60 dBm/Hz, the bit loading procedure is very simple: the usable frequency bands and/or time slots for any session are de®ned by the higher-level system management, and the procedure (for both down and up) to meet a request for a pair of data rates is essentially the same as used for ADSL, which is described in Section 8.5.2. If, however, power boosts are allowed and the dominant constraint becomes the 11.5-dBm total power, then the optimum procedure will be complicated. Over a signi®cant part of the band the receiver noise is dominated by kindred FEXT, so if all VTUs in the binder group obeyed the same rule, the PSD in that part of the band could be reduced below À60 dBm/Hz and the savings in total power devoted to the higher frequencies where the noise is AWGN-limited. That is, however, a big “if”; if the ®nal standard allows power boosts to much higher than À60 dBm/Hz at any frequency, a unilateral reduction of PSD (i.e., without any assurance that the other VTUs will do the same) in the FEXT-dominated parts of the band would be foolish! If a PSD strategy is not coordinated at a higher level, however, then in the PMD layer each VTUmust consider the measured noise as unalterable, and optimize its own transmission under the mixedÐtotal power and PSDÐ constraint using an algorithm like that in Section 5.3.2.
PSD Cutbacks to Reduce ULFEXT. In Section 4.6.3 we saw if there is a (CO) mix of short and long loops in a binder-group, UL FEXT from a close-in upstream transmitter would be the dominant crosstalk into the upstream on the long loop. This would be particularly troublesome for VDSL if a LEC offered both long asymmetric and short symmetric: clearly the upstream PSD must be reduced.
T1.413 de®nes a procedure for physical-layer-control of the downstream ADSL PSD, but this mix of VDSL services is much more complicated; it requires system management to decide on an acceptable trade-off between rates, and then control (via a downstream message from the VTU-C) the upstream PSD. At the time of going to press this is un®nished business.
10.1.3 Coexistence with ADSL
VDSL and ADSL systems may use pairs in the same binder group in the two ways shown in Figure 10.1: FTTE may result in an exchange mix and FTTC may result in a remote mix. The potential for crosstalk is different in the two cases.
Exchange Mix. With an exchange mix the ADSL downstream signal may NEXT into the VDSL upstream and/or FEXT into the VDSL downstream. The expected levels for a 4.5-kft VDSL loop are shown in Figure 10.2 over somewhat more than the ADSL frequency band. It can be seen that the (VDSL) signal to (ADSL) NEXT ratio varies between ‡ 12 dB at 0.3 MHz and approximately À11 dB at 1 MHz; clearly, the entire ADSL band would be useless for upstream VDSL. For the downstream signal the VDSL signal to ADSL FEXT ratio varies between 30 and 18 dB. This is not negligible, but as can be seen from
Figure 10.1 VDSL and ADSL in the same binder group: (a) ®ber to the exchange: exchange mix; (b) ®ber to the cabinet: remote mix. 1.0 3.9 58.1
Figure 10.2 VDSL signal and ADSL NEXT and FEXT for 4.5 kft with exchange mix. TABLE 10.2 Downstream Capacities VDSL Downstream Capacity (Mbits/s) ÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐ Loop Length (kft) In ADSL Band In Rest of Band
3.0 2.6 22.9
4.5 2.3 7.1
Remote Mix. With a remote mix, even though VDSL’s PSD is 20 dB lower than ADSL’s, NEXT from the upstream and/or ULFEXT from the downstream may be the dominant crosstalkers. Figure 10.3 shows the receive levels of downstream ADSL on a 9-kft 26-AWG loop compared to the two standardized crosstalks, ADSL FEXT and HDSL NEXT, and the two new crosstalks, NEXT from upstream VDSL and UL FEXT from 3 kft of down stream VDSL. It can be seen that beyond about 0.3 MHz, both directions of VDSL intefere very badly with ADSL. The downstream ADSL data rates achievable are shown in Table 10.3.
Figure 10.3 ADSL received signal on 9 kft of 26 AWG, with standardized ADSL FEXT and HDSL NEXT, and new VDSL NEXT and UL FEXT from a 3-kft remotely mixed loop. TABLE 10.3 Downstream ADSL on 9 kft of 26 AWG with XT from Remote Mix VDSL D-s ADSL (Mbits/s) ÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐÐ Interferers 1-kft VDSL Loop 3-kft VDSL Loop HDSL ‡ ADSL 6.9 (note) 6.9 Note: This is approximately the rate speci®ed in T1.413. Exchange Ð (see note) Ð
HDSL ‡ ADSL ‡ VFEXT 3.1 4.9
HDSL ‡ ADSL ‡ VNEXT 2.3 2.3
TABLE 10.4 Crude Quantitative Summary of XT Between ADSL and VDSL VDSL Down into ADSL Up Mix (see note)
Remote FEXT VDSL Up into ADSL Down
NEXT
ADSL Down into ADSL Down into VDSL Down VDSL Up FEXT NEXT FEXT Ð
Note: These are XTs from a new service into an existing service; they are therefore completely unacceptable.
Summary.The preceding results are summarized in Table 10.4, where the size of the XTer indicates the seriousness of the interference.
Spectrum Management. It is clear from Tables 10.2 and 10.4 that if there is an exchange mix of ADSL and VDSL, then VDSL up should not use the ADSL band (0.3 to 1.1 MHz); VDSL down may use that band with a small to moderate bene®t depending on the length of the VDSL loop. From Tables 10.3 and 10.4 it can be seen that if there is a remote mix, VDSL must not use the ADSL band for either direction. The extent of the control required from the LEC depends on the duplexing method used.
NOTE: Early versions of the requirements document do not recognize any difference between exchange and remote mixes; the presence of any ADSL in any binder group puts the ADSL band off-limits. This seems unnecessarily restrictive; it may be relaxed.
10.1.4 Coexistence with Echo-Canceled BRI
It was agreed from the early days of discussion of ADSL that there would be (at least) two versions: one to operate on the same pair with POTS (to work “over” POTS in the frequency domain), and the other to work over BRI6; the extra 140 or so kiloHertz of bandwidth gained if the loop did not have to carry ISDN made the “POTS” version worthwhile.
6These are de®ned in Annexes A and B of G.992.
For VDSL, however, this extra bandwidth is insigni®cant compared to the total bandwidth (about 12 MHz on a short loop), so it was decided from the start that there need be only one version of VDSL7with a usable band starting at about 300 kHz. This had the advantage of simplifying all interference calculations because above 300 kHz HDSL NEXT falls off very rapidly, and the only signi®cant alien interferer is ADSL.
10.1.5 Compatibility with Amateur (Ham) and AM Radio
This requires the ability to deal with interference from nearby transmitters (ingress), and in the case of ham radio only, also requires the prevention of interference with nearby receivers (egress).8Both ingress and egress are more serious at the remote end, where the coupling is via the (relatively) unbalanced drop wire; an unbalanced and unshielded pair out of a CO or an ONU(a “head end”) is less common. Nevertheless, some operators use aerial cable all the way from a head end to the customer premises, so ingress and egress at both ends must be considered.
Ingress.The levels are de®ned in Section 3.7; methods of dealing with the ingress are described in Sections 10.4 and 10.5.
Egress. The four factors controlling VDSL interference with ham receivers are the VDSL (differential mode) transmit PSD, the differential-mode-to-commonmode balance of the loop near the transmitter, the separation of loop and ham antenna (controlling the path loss), and the sensitivity of the ham receiver. These were all analyzed in [Bingham et al., 1996a], and the conclusion of T1E1.4 was that in the ham bands (de®ned in Section 3.7) the transmit PSDÐboth upstream and downstreamÐshould be no higher than À80 dBm/Hz.
10.1.6 The Network Termination
It is intended that early versions of VDSL service should terminate at a network termination (NT) at the entrance to the customer premises, and that some other method of distributionÐas yet unspeci®edÐwould be used within the premises. This means that the in-house wiring, particularly the short bridge taps, need not be considered as part of any loop. Whether VDSL will later follow the example of ADSL and be required to operateÐat reduced rates if necessaryÐthrough existing house wiring remains to be seen.
7 Another difference between the situations for the two systems was that ADSL was developed originally as a standard for North America, where ADSL over ISDN is of little interest. In the development of VDSL, on the other hand, Europe has been the leading partner, and there a mix of ISDN and VDSL on the same pair is important. DUPLEXING
8Interference by VDSL with AM receivers is not considered to be a problem.
10.2 DUPLEXING
10.2.1 Echo Cancellation?
The ®rst question when considering a duplexing method for VDSL is whether echo cancellation should be used. A strategy for ADSL that is discussed in Section 4.2 is EC up to some changeover frequency and FDD thereafter; for VDSL we can generalize this to consider either FDD or TDD above some frequency. The calculations of Section 4.3 were repeated for the short, medium, and long VDSL ranges (1, 3, and 4.5 kft), but now with only kindred crosstalk as an impairment.9
Figure 10.4 shows the aggregate (downstream and upstream) data rates (without any consideration of the methods or ef®ciency of partitioning) as a function of the crossover frequency. It can be seen that on a short loop, EC up to about 3 MHz would be slightly bene®cial (approximately 10% increase over pure xDD), but beyond that the bene®ts are insigni®cant. Clearly, EC for VDSL would not be worth the added complexity.
Figure 10.4 Aggregate (down plus up) rates as a function of maximum EC frequency. 9The VDSL bandwidth is such that HDSL and ISDN can be ignored. The only potentially serious interferer is ADSL downstream, which we consider shortly.
10.2.2 FDD or TDD?
With EC eliminated, the duplexing choice is between FDD and TDD. Special features and capabilities of FDD, Zipper, and STDD are discussed in Sections 10.3 to 10.5, but motivation for the three proposals, comparison between them, and possible reasons for choosing one or the other are postponed to Section 10.7.
10.2.3 Mixed Services
The basic problem with mixing 8:1 and 1:1 services in the same binder group is that if they are independently optimized, part of the asymmetric’s downstream and the symmetric’s upstream must share either bandwidth (if using FDD) or time (if using TDD), and will unavoidably NEXT into each other. An extension of the result in Section 10.2.1 is that ifÐand only if10Ðsuch a mix is needed, then above some frequency that depends on the loop length both downstream asymm and upstream symm must be constrained so that they do not overlap and do not NEXT into each other. This is fairly easy to do with TDD (see Section 10.6.2) because the capacities of all the symbols in a superframe are approximately equal. It is very dif®cult to do ef®ciently with FDD because the capacities of the frequency bands differ widely.
10.3 FDD
The basic principles of DMT FDD have already been discussed under ADSL, and we need only to consider three important differences between VDSL and ADSL:
1. ADSL uses a ®xed crossover between downstream and upstream, but it can get away with it because only an 8:1 asymmetry is needed. On long loops the crossover frequency is too high, and the downstream rate is the limiting one, but not seriously so. For VDSL, however, the effect is more severe: for the 1:1 symmetrical service the optimum crossover frequency varies much more with loop length. Flexible crossover frequency(ies) are needed, and modems must be able to change the passbands of all the ®lters.
2. The requirements for symmetrical operation and for ¯exibility in choice of downstream and upstream subcarriers mean that both directions must use FFTs spanning the full frequency range; the system considered here uses the 512-pt FFT of ADSL and SDMT.11
10 The “only if” follows from the principle established in Section 10.1.1: a homogeneous service must be con®gured for maximum performance.
11For FDD there is no great advantage to be gained from using the very large FFTs required by Zipper.
3. For better spectral compatibility with ADSL, VDSL should not transmit upstream below 1.1 MHz; that is, the downstream bandÐor at least pat of itÐmust be below the upstream. Most binder groups will probably not carry both ADSL and VDSL, so this rule would not apply to them, but if the rule can be obeyed without any loss of performance, then for the sake of simplicity, it should be obeyed on all loops.
10.3.1 Mixture of Symmetric and Asymmetric Services
If a mix of 8:1 and 1:1 is thought of as a 16:2 and 9:9 mix, it can be seen that if all tones were used for both services, 7/16 of the asymmetric downstream tones would suffer kindred12NEXT from symmetric upstream; similarly, 7/9 of the symmetric upstream tones would suffer kindred NEXT from asymmetric downstream. This would cause a very severe decrease in the data rates, particularly for the symmetric service.
Just as for SDMT (see Section 10.5.4), for any but the shortest loops, it is a nearly optimum strategy to avoid NEXT altogether by assigning tones exclusively to downstream or upstream.
10.4 ZIPPER
Zipper was described in a series of ANSI and ETSI contributions, [Isaksson et al., 1997], [Bengtsson et al., 1997], [Olsson et al., 1997], and the more accessible [Isaksson et al., 1998b]. The name was derived from the original version of the method, shown in Figure 10.5, in which downstream subcarriers alternate with upstream subcarrriers as in Lewis Walker’s invention of 1913.13Separation of the transmit and receive signals is achieved by the orthogonality of the DMT subbands.
In FDD the ratio of downstream/upstream data rates is controlled by the frequency bands (both positions and widths) of the ®lters. Optimizing these ®lters for a variety of ranges and ratios is a dif®cult task off line; it is even more dif®cult on line. Some versions of Zipper control the ratios by using small groups
Figure 10.5 Pure Zipper. 12Cousins rather than siblings.
13The patent on that has expired, so there is no problem with IP.
of very narrow subbands in which the number used for downstream and upstream is determined by the downstream/upstream ratio; the prototype system is for symmetrical VDSL, in which alternating tones are used for down and up. This type of Zipper system applied to a channel in which the SNR changes with frequency is an approximate dual of a TDD system applied to a channel in which the SNR changes slowly with time: down/up ratios are controlled by the assignment of successive tones (symbols).
It was recognized, however, that tones near the edges of the bands may suffer from both echo and NEXT (see Section 10.5.3), and changing direction many times across the frequency band increases the number of such tones; more recent versions of Zipper therefore group the down and up tones in large blocks. With this arrangement the name is no longer appropriate, so because it is basically an FDD system implemented without ®lters the name digital duplexing (DD) has been proposed.
The main ways in which Zipper/DD differs from DMT are that it uses:
1. A much longer cyclic extension (a cyclic pre®x plus a cyclic suf®x) to allow for both the receive transient (just as with conventional DMT) and the propagation delay of the loop.
2. A much longer symbol (typically, using a 4096-pt FFT) to maintain ef®ciency with the longer cyclic extension.
3. Ranging on each UTP in a binder group to ensure that both VTU-O and VTU-R transmitters on any given loop start their symbols at the same time. A method of doing this has not been described, but it is not hard to devise one.14
10.4.1 Basic Zipper/DD System
Conventional DMT de®nes Tsymband Tcp, the (data) symbol and cyclic pre®x durations, and Trtran, the time for the receive transient (after equalization) to decay so that distortion does not add signi®cantly to the noise; the requirement is that Tcp>Trtran. In addition, Zipper/DD must consider Tp, the one-way propagation delay, Tetran, the time for the echo of the transient on another subcarrier on the same UTP to decay so that it does not add signi®cantly to the noise, and TNtran, the time for the NEXT transient from another UTP to similarly decay. These are shown in Figure 10.6, which shows the timing of transmit and receive signals.
Important points to note are:
* Ranging ensures that transmission starts at the same time at both ends, so the timing diagram (with only the one-way propagation delay accounted for) is the same at both ends.
14 For example, during training the VTU-O could measure 2Tpby sending a probing signal, which the VTU-R would immediately return. It would then instruct the VTU-R to start its symbols Tp before the end of the received symbols. Figure 10.6 Zipper timing diagram on longest loop.
* Echoes need be considered only as “synchronous” transients because transmit and receive use different subcarriers.
* NEXT is also transient, but if symbols on all the UTPs in a binder group are not synchronized,15Zipper operates in the asynchronous mode, and the transient may occur at any time in the receive symbol. The sidelobes of the full NEXTing signal then impinge on the receive subcarriers, and subcarriers must be assigned in large blocks to minimize edge effects (see Section 10.5.4).
* If all the UTPs in a binder group can be synchronizedÐwhen, for example, all the VTU-C/Os are colocatedÐthe NEXT transients occur during the cyclic extension, and their effect is much reduced. Zipper can then operate in a synchronous mode, and a more ¯exible method of subcarrier assignment can be used.
It can be seen that the requirements on the cyclic extension are
Tce > Tp ‡ Trtran for no distortion …10:1a† Tce > Tetran for no echo …10:1b†
Equation (10.1 a) is just an extension of the Tcp > Trtran requirement of conventional DMT, but it adds a little ¯exibility: if Tce is chosen for a marginally acceptable amount of distortion on the longest loop (the worst case), distortion will be less on all shorter loops. This will turn out to be very useful when we consider the equalizer in Section 10.4.8. The requirement of equation (10.1b) is considered in the next section.
15The most important argument against STDD, which is described in Section 10.6, was the need for synchronizing the superframes on all the UTPs.
Ef®ciency. The formula for the ef®ciency of Zipper/DD is the same as for DMT: that is,
Tsymb
“Zipper ˆ2…Tsymb ‡ Tce† …10:2a†
Nsymb
ˆ 2…Nsymb ‡ Nce†…10:2b†
Tceis determined by the loop (length and IR) and the equalizer, so the ef®ciency is controllable only by Tsymb. The values proposed are Nsymbˆ 4096 and Nce ˆ 320, so that “Zipper ˆ 46.4%.
Shaping the Cyclic Extension. In Section 6.4 we discussed envelope-shaping of the cyclic pre®x. The advantage of shaping is that it reduces the sidelobes, and it can be performed wholly in the transmitter, wholly in the receiver, or partly in both, depending on where the reduction is needed. For Zipper (particularly in the asynchronous mode) the reduction is needed in both in order to reduce the effects of NEXT from nearby subcarriers, and one proposal for shaping was given in [Isaksson et al., 1998a]. The part of the cyclic extension that deals with the propagation delay can be shaped in the transmitter, and the part that deals with distortion can be shaped in the receiver. For the greatest overall reduction of NEXT with a given duration of the cyclic extension, it would be best to have these two parts of equal duration, but with the total cyclic extension proposed in Table 10.5 this is possible only on loops of 4.5 kft or less. The duration of the transmitter shaping can be further doubled by overlapping successive symbols as shown in Figure 10.7.
System Parameters. A preliminary set of parameters is given in Table 10.5. Figure 10.8 shows three plots of the end-to-end attenuation that results from either transmitter or receiver shaping, as a function of the number of tones away from a used bandÐwith no shaping, and with shaping over different numbers of samples: 100 (the maximum for the receiver on the longest loop), 160 (appropriate for both transmitter and receiver on all loops less than 4.5 kft),16and 220 (for the receiver on the longest loop). These are the attenuations that would be applied to both echo and ANEXT. Choosing the guard band (i.e., number of unused subcarriers) between a downstream and an upstream band to maximize
16 Isaksson and Mestdagh suggested shaping only 70 of the 220 available in the transmitter, and 70 of the 100 available in the receiver. This reduces the effects of distortion but also reduces the amount of sidelobe suppression. A careful study of the trade-offs is needed. TABLE 10.5 Preliminary Parameters for Zipper/DD fsamp (MHz) 22.08 MHz IFFT size (samples) 4096 Overhead for Tp and Trtran 320 (14.5 ms) (note 1) Data symbol rate 5.0 kHz Ef®ciency 46.4% Cyclic “suf®x” maximum 220 (note 2) (0.3±11 MHz) In-band transmit PSD without À60 dBm/Hz power boost Notes: Figure 10.7 Zipper transmitter (a) and receiver (b) shaping for sidelobe reduction. Figure 10.8 End-to-end sidelobe attenuation.
Cyclic extension (note 3) 320 ‡ 220 Subcarrier spacing 5.390625 kHz Used subcarriers 56±2047
1. An important point about this overheadÐnot mentioned in [Isaksson et al.,
1998a], but clearly premeditatedÐis that the total of 320 samples is proportionately the same as for ADSL (40/552). This augurs well for “scalable xDSL,” an idea that is being developed in SG 15. If the apportioning of samples between transmitter and receiver shaping is changed depending on the length (i.e., propagation delay) of the loop, it is important that this total and the resulting data symbol rate be preserved.
2. This allows for a propagation delay of 10 ms, which is plenty for the longest loop contemplated.
3. The terminology here is confusing, and it would certainly need to be clari®ed in a standard. The transmit signal comprises (4096 ‡ 320 ‡ 220) phasecontinuous samples, of which up to 220 can be used in a shaped pre®x and another 220 in a shaped suf®x, which is overlapped with the next symbol’s pre®x: leaving 100 or more to be treated by the receiver as a shapable pre®x (independent of the transmitter’s pre®x). These numbers are incorporated into Figure 10.7, which may help to reduce the confusion slightly.
the aggregate capacity would be very complicated, but probably at least 30 subcarriers will be needed. If two downstream and two upstream bands (the minimum for ¯exibility of services) are used, this would result in an overhead of 90/2048 % 4.5%; this is small compared to the guard bands of a ®ltered FDD system, but still not insigni®cant.
Clearly, the total cyclic extension would have to be de®ned in a standard, but how much more is a very vexed question. Since the shaping on one UTP affects NEXT into others, it is arguable that it should be de®ned, but more likely is a set of out-of-band PSD speci®cations just as for any conventional FDD system. How the transmitter and receiver would then agree on shapings needs further study.
10.4.2 Analog Front End and ADC
Much of the appeal of Zipper/DD lies in its purported ability to perform all the separation of transmit and receive signals with the IFFT and FFT; this brings a lot of ¯exibility in con®guring mixed systems (see Section 10.5.4). This assumes, however, that the full signal at the output of the 4W/2W hybrid (receive signal plus re¯ected transmit signal) is analog-to-digital converted; this assumption needs to be examined.
With a transmit PSD value of À60 dBm/Hz into a 4.5-kft loop of 26 AWG, the used band is approximately 0.3 to 3.0 MHz, and the transmit and receive powers in a symmetric system using “alternating” Zipper17are approximately ‡ 2.0 and À31.0 dBm. This means that to reduce the echoed signal just down to the level of the receive signalÐthereby doubling the power input to the ADC and requiring an extra half of a bit of quantization for the same level of quantizing noiseÐa trans-hybrid loss (THL) of 33 dB is needed.
Without bridge taps and line coupling transformers, and in a frequency region where the range of characteristic impedances of the cables used is small, this would not be dif®cult. The impedances of 24- and 26-AWG pairs, the most common UTPs in the United States, for example, change slowly and smoothly with frequency, and an RRC matching impedance that is a compromise for the two gauges can achieve a THL of better than 35 dB across the band. Many other countries, however, use a much wider variety of cables than this.
Transformers, however, typically have a Æ 10% tolerance on their inductance, and the worst-case THL achievable at the low end of the band is less than 26 dB. I have seen no analysis of the contribution of this high-level echo to the total echo power. Bridge taps present the worst problem. Figure 10.9(a) shows test loop VDSL4 de®ned in [Ciof®, 1998]; Figure 10.9(b) shows the return losses at both ends relative to the compromise impedance.18It is thus the highest RL that can be achieved without some adaptive hybrid, but at the RT it is abysmally low!
It is clear, therefore, that with the worst re¯ection and no pre®ltering, the number of ADC bits would have to be at least four greater than the number of DAC bits chosen19: probably at least 14 bits. It should be noted, however, that if large echoes are dealt with merely by increasing the number of ADC bits, all the responsibility for removing echoes and NEXT falls upon the FFT. The levels of asynchronous NEXT may be such that subcarriers must be assigned in large contiguous groups in order to reduce edge effects (see Section 10.5.3).
Another way of looking at this question was explained to me by Nick Sands.20 The peak voltage capability of an ADC is pretty much de®ned by the supply voltages. Therefore, increasing the level of the signal input to the receiver by x dB means that the front-end ampli®er gain must be reduced by x dB, thus moving the wanted signal x dB closer to the noise ¯oor (typically between À125 and À135 dBm/Hz at the present and foreseeable-future state of the art). Without the re¯ected transmit signal, the input noise would be dominant (as it should be); with x ˆ 20 or more the AFE noise ¯oor becomes the limiting factor over a large part of the band.
17 The calculation for a “block” DD system is much more complicated because of the different bands and bandwidths used for the two directions, so the Zipper system is used as a crude average.
1818
Không có nhận xét nào:
Đăng nhận xét