12/04/99 1 T1E1.4/99-560 Project: T1E1.4: VDSL Title: Revisiting Bridged Tap and Spectrum Issue for VDSL Performance (560) Contact: J. Cioffi, W. Yu, and G. Ginis Dept of EE, Stanford U., Stanford, CA 94305 Cioffi@stanford.edu, 1-650-723-2150, F: 1-650-724-3652 Date: December 5, 1999 Dist'n: T1E1.4 Abstract: This paper addresses the issue of robust VDSL transmission in the presence of bridged taps in the local loop. The impact of bridged taps on transmission performance is described and evaluated for various frequency-division duplexing plans. It is shown that using only 4 bands will under several realistic circumstances result in very poor performance. The conclusion is that a robust duplexing plan will need at least 7 bands (4 upstream and 3 downstream). An action for T1E1.4 is that should this group try to reach consensus on a spectrum plan that such a plan must include 7 bands, 4 up and 3 down, in order to ensure that VDSL system performance requirements are met in the presence of bridged-taps. NOTICE This contribution has been prepared to assist Standards Committee T1 - Telecommunications. This document is offered to the Committee as a basis for discussion and is not a binding on Stanford University. The requirements are subject to change after further study. The authors specifically reserve the right to add to, amend, or withdraw the statements contained herein.
12/04/99 2 T1E1.4/99-560 Revisiting Bridged Tap and Spectrum Issue for VDSL Performance (99-560) J. Cioffi, W. Yu, and G. Ginis Department of Electrical Engineering Stanford University, Stanford, CA 94305 Phone: 650-723-2525 ; Fax: 650-724-3652 ABSTRACT This paper addresses the issue of robust VDSL transmission in the presence of bridged taps in the local loop. The impact of bridged taps on transmission performance is described and evaluated for various frequency-division duplexing plans. It is shown that using only 4 bands will under several realistic circumstances result in very poor performance. The conclusion is that a robust duplexing plan will need at least 7 bands (4 upstream and 3 downstream). An action for T1E1.4 is that should this group try to reach consensus on a spectrum plan that such a plan must include 7 bands, 4 up and 3 down, in order to ensure that VDSL system performance requirements are met in the presence of bridged-taps. 1. Introduction: VDSL service promises to provide data transmission at high-rates on copper twisted-pair wire reaching lengths up to 4500 ft. The commercial success of the service will largely depend on the soundness of the system design. This paper investigates the benefit of 7-bands over 4-bands both without and with bridged taps. Gains in range of as much as 30% without bridged taps and 40% with bridged taps are illustrated. This contribution, unlike an earlier version in the ITU (NT-39, Nashville, 11/99), allows all methods to use 0% excess bandwidth - the conclusions are identical though to previous conclusions. This paper is a second update of an earlier paper provided to the FSAN group of service providers. It appears accepted that VDSL transmission will use (at most) 20 MHz of bandwidth. Frequency-division duplexing (FDD) will be employed to separate the 2 directions of transmission (upstream and downstream)[1, Issue 2.5]. The FDD plan must be identical for both symmetric and asymmetric services, although it may differ between wide geographical regions (e.g. Europe and US). A consensus for the exact spectrum plan has not been reached [1, Issue 2.10]. The final spectrum plan(s) will constitute a major engineering decision, which affects enormously the whole design procedure, and any mistake at this stage may be difficult to correct later. The derivation of the duplexing plan needs to take into account several factors, which may at first seem to have low visibility. VDSL must implement a range of data rates, symmetries, and performance objectives. Among the most important effects which have to be considered are frequency-selective line disturbances, such as bridged-taps, and mobile radio ingress. These disturbances may cause frequency bands on the order of a few hundred kilohertz to become completely unusable for data transmission. Moreover, those frequency bands cannot be known in advance, because they will depend on the actual bridged-tap length, or on the spectrum of the radio-noise source. This means that the duplexing plan will have to be provisioned for different conditions to be robust. This paper mainly studies the effect of bridged-taps on VDSL transmission, and shows that any robust duplexing plan must consist of at least 7 bands (4 upstream and 3 downstream). We note that mobile radio
12/04/99 3 T1E1.4/99-560 ingress may be thought of as having an impact similar to bridged-taps. A combination of the two disturbances is not an unlikely situation, and it only furthers this paper's argument that at least 7 bands are absolutely necessary. When power control and G.pnt compatibility are further considered, this number is certain to grow. 2. Robustness VDSL must be able to accommodate bridged-taps of different lengths. This is here called "robustness." Although it is impossible to avoid bridged-tap effects completely, it is highly desirable for performance to degrade somewhat gracefully. For a symmetric service this can be interpreted as maintaining the ratio of upstream rate to downstream rate close to 1. In this case it would be highly undesirable to have a huge data rate loss in only one direction. In asymmetric transmission, it is desirable to maintain the ratio of asymmetry under different bridged-tap configurations. First, this section illustrates the adverse effects of bridged-taps on transmission performance. Figure 1 shows the transfer function (in db) of a 4050 ft loop, with bridged-taps (66, 56, 46, and 36 ft long) and without bridged-taps. The bridged-taps cause the transfer function to exhibit notches periodically in frequency. As the bridged-taps get shorter, the notches become more deep and they move to higher frequencies. The existence of such notches (10-20 db deep) can seriously harm transmission. Below the graph, two different 4-band frequency plans are drawn. Note that this specific loop has very large attenuation at frequencies above 7 MHz, so the spectrum above 7 MHz is unsuitable for data transmission. Therefore, only the lower 2 bands would actually be used. Each frequency plan copes differently with this kind of disturbance. If plan A were used in the presence of a bridged-tap 36 to 66 ft long, then upstream transmission performance would be degraded significantly, although the downstream direction would not be affected. If plan B were used, then the downstream transmission performance would be degraded. Both plans fail to be robust. One might argue that some other 4-band plan would actually show more immunity to such situations. We explain why this is false: The bridged-tap length is not determined, so it may vary from 10 to more than 100 ft. This means that the notches may actually occur in almost any frequency of the VDSL spectrum. For any 4-band plan, there will always be a bridged-tap with such a length, that performance will solely be degraded in one direction. As [2] correctly notes, this direction is often the upstream direction using conventional models as are used in this paper. This intuitive argument is supported by the simulations of Section 3. With real crosstalk in the filed, which is highly frequency selective, it is harder to say that one direction is degraded more often and indeed more robustness in both directions are required. However, we have not pursued this topic further here, but such pursuit would only increase the number of bands required, whence the conclusion later of at least 7 bands. However, there is one optimal frequency-division duplexing scheme, which one can prove attains the maximum possible robustness. The solution is to partition the spectrum into infinitesimally small bands and alternatively assign them to upstream and downstream transmission. Then, any frequency-selective distubance (such as a bridged-tap) will have an equal impact on both directions of transmission. Figure 2 shows such a frequency plan, and illustrates why symmetric service is maintained. The implementation of this optimal scheme may prove too complex 1, so suboptimal schemes with adequate robustness may have to be used instead. By interpolating between the 4-band plan and the optimal plan, we deduce that a number of bands as large as possible is highly desirable. As the number of bands increases, the data rate loss caused by a bridged-tap will be distributed more evenly between the two directions of transmission. The simulations in Section 3 demonstrate this fact, and it is also shown that 7 is the least acceptable number of bands. 1 Recent demonstrations of full zippering have been able to suggest that at least in some situations, large numbers of alternating up/down bands are indeed feasible with acceptable implementation.
12/04/99 4 T1E1.4/99-560 Plan A f f Plan B Figure 1 - Illustration of robustness with bridged taps. The graph shows the insertion loss (in db) of a 4050' loop with bridged-taps of length 66', 56', 46', and 36' (20m, 17m, 14m, and 11m respectively). Below the graph, two different frequency plans are shown. When plan A is used, only upstream transmission is affected. When plan B is used, only downstream transmission is affected. In both cases symmetric service is disabled.
12/04/99 5 T1E1.4/99-560 Bridged-tap equally affects down and up Plan Z f Figure 2 - Illustration of optimal robust duplexing solution - effect of bridged-tap is shared between upstream and downstream. 3. Simulations The simulation results that are shown below were obtained using a popular telco simulation tool. The 4 different frequency plans that were evaluated are shown below (numbers refer to MHz): SCM 4-band plan - provided by Alcatel/Telia ([3]) up = (3.25 4.65, 10.9 17.6) down = (0.4 2.75, 5.35 9.9) SCM 4-band plan - provided by Savan/Infineon in ETSI TD65 ([4]) Up = (4.3-7.9, 14.85-20) Down=(.35-3.4, 8.92=13.65) Digital Duplexing 5-band plan up = (0.03-0.138, 3.08-4.78, 10.242-17.66) down = complement of up Digital Duplexing 7-band plan up = (0.03-0.138, 2.5-3.5, 4.5-5.5, 11-17.66) down = complement of up Digital Duplexing 15-band plan up = (0.03-0.138, 2.1-2.5, 2.75 3, 3.25-3.5, 4-4.25, 4.5-4.75, 5-5.5, 10.5-17.66) down = complement of up The SCM simulations assumed a 0% efficiency loss, while the Digital Duplexing simulations assumed a 10% efficiency loss, thereby providing a rather optimistic view of SCM. Efficiency loss is a time-domain overhead. The analog-duplexed 4-band SCM system also used approximately 20% excess bandwidth. The bridged-tap lengths ranged from 10 to 20 meters. The other parameters in the popular FSAN telco tool were set as follows: SNRmax = 45 db Gap = 12 db (with 3 db of coding gain and 6 db of margin, the gap is usually 9 db, but we did not use coding in these simulations, so all would improve slightly) Cable : TP1
12/04/99 6 T1E1.4/99-560 HAM bands were not notched The services evaluated were medium symmetric, long symmetric, extra long symmetric, medium asymmetric, and long asymmetric in a noise A and noise D environment. For each service the reach in meters was computed both with and without bridged-taps. The following tables show the resulting reaches, and the percentage gains achieved by the plans using more than 4 bands in comparison to the SCM plan. The services types were recently ranked by some incumbent local exchange carriers in [5] and those weightings helped develop the results here. These results may appear at first inconsistent with some in a recent contribution to another group [6]. Such is not the case - this contribution here focuses on calculation of reach and percentage reach loss at a given data rate, while the results in [6] evaluate percentage data rate loss at a given reach. The percentage improvements of multiband plan differs under the two measures (but in both cases is substantial). ANSI Noise- A Infineon Savan TD65 Alcatel Telia Alcatel Telia Alcatel Telia S_med(13/13) 867 713-17.76 872 0.58 870 0.35 S_long(6.5/6.5) 1085 1147 5.71 1262 16.31 1270 17.05 S_xlong(2/2) 1291 1531 18.59 1688 30.75 1773 37.34 A_med(26/3.2) 735 869 18.23 763 3.81 758 3.13 A_long(13/1.6) 1324 1466 10.73 1329 0.38 1328 0.30 S_med(13/13) 507 513 1.18 594 17.16 603 18.93 S_long(6.5/6.5) 835 695-16.77 986 18.08 1046 25.27 S_xlong(2/2) 1099 1284 16.83 1428 29.94 1491 35.67 A_med(26/3.2) 686 818 19.24 712 3.79 713 3.94 A_long(13/1.6) 1134 1295 14.20 1210 6.70 1213 6.97 ANSI Noise- D (T1 noise) S_med(13/13) 658 623-5.32 669 1.67 661 0.46 S_long(6.5/6.5) 783 753-3.83 849 8.43 812 3.70 S_xlong(2/2) 911 1081 18.66 1183 29.86 1058 16.14 A_med(26/3.2) 565 607 7.43 590 4.42 589 4.25 A_long(13/1.6) 768 800 4.17 755-1.69 768 0.00 S_med(13/13) 452 466 3.10 498 10.18 493 9.07 S_long(6.5/6.5) 634 620-2.21 734 15.77 726 14.51 S_xlong(2/2) 790 861 8.99 948 20.00 945 19.62 A_med(26/3.2) 496 551 11.09 537 8.27 539 8.67
12/04/99 7 T1E1.4/99-560 A_long(13/1.6) 681 753 10.57 693 1.76 707 3.82 We immediately see that using a larger number of bands always improves performance. The advantage of more bands is even more clear when bridged-taps are included. Significant reach gains are achieved, which are directly translated to increased revenue potential for service providers adopting a multi-band implementation. As explained in Section 2, using a large number of bands provides increased robustness to bridged-taps. It is worth noting that the reach of the extra long symmetric service is improved by more than 35% (300-500 meters), when more than 7 bands are used. This service represents an important market segment, and might be the first VDSL service to be deployed. The asymmetric results were all limited in range by downstream while all the symmetric results were limited by upstream range, with one exception: the Infineon long-range asymmetric at 13/1.6 was limited in range by upstream. 4. Conclusion This paper examined the problem of robust VDSL transmission in the presence of bridged-taps. It was shown that bridged-taps have the potential to disable symmetric service in several realistic scenarios. In order to combat this impairment the duplexing plan must consist of as many bands as possible. Simulation tests for typical scenarios indicate that no fewer than 7 bands (4 upstream and 3 downstream) must be employed. 5. References [1] S. Palm, " G.vdsl: Updated Issues List for G.vdsl", ITU SG15/Q4 Temporary Document NT-U11, November 1, 1999, Nashville, TN. [2] V. Oksman, "G.vdsl: Impact of Bridged Taps on VDSL Performance," ITU SG15/Q4 Temporary Document NT-091, November 1, 1999, Nashville, TN. [3] M. Mielants, Alcatel Bell, Belgium, use of program described in ITU SG15/Q4 Temporary Document NG-075, August, 1999, Nuremberg, Germany. [4] G. Porat, "A Study of Band Allocations for VDSL," ETSI TM6 Contribution TD65, Edinburgh, Scotland, September 1999. [5] T. Starr et al, "G.vdsl: Ranking of VDSL Service Types for Determining Band Allocation," ITU SG15/Q4 Temporary Document NT-027, Nashville, TN, November 1, 1999. [6] J. Cioffi, "Robust Duplexing in VDSL," ETSI TM6 Contribution TD14, Edinburgh, Scotland, September 1999.