Read 9e145590.PDF text version

12/04/99

1

T1E1.4/99-559

___________________________________________________

Project: T1E1.4: VDSL __________________________________________________________ Title: Vectored VDSL (99-559) ____________________________________________________________ Contact:

J. Cioffi, G. Ginis, W. Yu Dept of EE, Stanford U., Stanford, CA 94305 [email protected] , 1-650-723-2150, F: 1-650-724-3652

______________________________________________________ Date: December 5, 1999__________________________________ Dist'n: T1E1.4

Abstract:

ONU vectoring is introduced in this contribution to reduce the effect of both NEXT and FEXT self-crosstalk to allow VDSL with universal band allocation to meet all service requirements, even when asymmetric and symmetric services are mixed. Both bundled and unbundled situations are investigated with the consequence that a recent universal band allocation proposal [1] allows all service requirements to be met, and in many cases exceeded significantly.

__________________________________________________________________________________

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

T1E1.4/99-559

Vectored VDSL (99-559)

J. Cioffi, G. Ginis, and W. Yu

Department of Electrical Engineering Stanford University, Stanford, CA 94305

Phone: 650-723-2525 ; Fax: 650-724-3652 ABSTRACT

ONU vectoring is introduced in this contribution to reduce the effect of both NEXT and FEXT self-crosstalk to allow VDSL with universal band allocation to meet all service requirements, even when asymmetric and symmetric services are mixed. Both bundled and unbundled situations are investigated with the consequence that a recent universal band allocation proposal [1] allows all service requirements to be met, and in many cases exceeded significantly.

1.

Introduction:

Unbundling of telephone line cables is rapidly becoming a worldwide reality with the consequence that the 1 individual phone lines within a single cable may use transmission systems of different service providers. Such unbundled use by multiple service providers increases the importance of standardization of various phone-line transmission methods to ensure good mutual spectrum use and benefit in the presence of the inevitable crosstalk between lines. Standardization can delineate the bands to be used nationally and consistently for upstream and downstream transmission thus ensuring through frequency division duplexing a spectrum compatibility between the different transmission systems of different service providers in the same cable. Standardization can also leverage the growing knowledge of digital signal processing and coding methods, often assuming the use of sufficient signal processing sophistication, to avert crosstalking problems and effects. This paper investigates one such method known as vectored VDSL that can significantly reduce the effect of self-crosstalk if the pairs belonging to any one service provider are coordinated. The motivation for such coordination will be significantly better performance, in data rate, in range, and in particular here, allowing a universal band allocation to meet both ETSI and ANSI symmetric and asymmetric performance objectives in the same binder. It is important to note that the FEXT-reduction methods presented here are discretionary to the vendor and/or service provider and are provided for information only. The current spectral masks remain the same and this method is not a form of coordinated power back-off (but does not prevent power back-off nor alter it if used), but rather an independent use of coordination to improve rate/range trade-offs. Section 2 introduces the basic vector concept and studies how crosstalk is generally mitigated via vectoring, while Section 3 progresses to evaluation of the improvement for VDSL with universal band allocation in particular for VDSL requirements in ETSI and in North America. In particular, this contribution shows that all ETSI and North American VDSL requirements for asymmetric and symmetric can be met with a single universal plan [1] if sufficient vectoring is used.

2.

1

Basic Vectoring

This phenomenon is sometimes heuristically referred to as "dark copper."

12/04/99

3

T1E1.4/99-559

The concept of vectored or matrix channels has been studied in other contexts for multiuser transmission [2], disk recording with multiple heads [3], and wireless transmission with multiple antennae [4-5] with enormous increases in achievable data rate in the last two cases (between a factor of 10 and 100 increase). In DSL, the possible improvement is not quite as large, but still sufficiently large to merit serious study. The matrix channel in DSL encompasses the adaptively identified crosstalk-coupling transfer functions as well as the line insertion loss. When the crosstalk is significant, substantial improvement is possible through service-provider-discretionary coordination of a single service provider's transmit signals.

fiber

. Vector Mod. .

. . .

D A C

VDSL 1 Vector channel H=QR y . . . VDSL L

x

D A C

Figure 1 - Illustration of Vectored VDSL Concept. Each tone in DMT has all VDSL transmitters in ONU coordinated. Figure 1 illustrates the basic concept. The set of channel inputs is viewed as an entity over several VDSL systems of the same service provider. In vectored VDSL, a single vector x (for each tone in DMT) models all the VDSL inputs at the common ONU. Each VDSL transmission system is presumed to use digital duplexing [6] and each is also synchronized to the same frame/symbol clock. Other service providers may not be sychronized to (nor coordinated with) the first service provider, and so the case of partial vectoring is also studied later in the results of Section 3. The channel can be described in terms of a "matrix" insertion loss H. The receivers are not coordinated because they will be in physically distinct locations -however, this will be of little consequence in practical VDSL as long as the ONU side can be coordinated. One can think of the channel output as a single discrete-time vector y = Hx + n , even though the elements of that output vector may be in physically distinct and uncoordinated locations.2 With FDM multiplexing, only FEXT is of concern for vectoring and for the model of the matrix H. The vector x would include one element for each VDSL transmitter that is coordinated (per DMT tone). Many entries in the matrix H will be zero or nearly zero, but significant crosstalkers will create nonzero entries in H. The matrix H is assumed known to the ONU-vectored VDSL transmitter (and identified by an appropriate channel identification procedure). With digitally duplexed DMT, there is no cross-tone crosstalk, allowing for the vector model above to be simple and of small dimension with maximum equal to the number of VDSL transmission systems of one provider in the same ONU, and typically much less. The matrix H has a well-known and readily computed decomposition in digital signal processing called "QR" factorization, specifically H=RQ* , where R is triangular and Q* is a "unitary matrix" (QQ*=Q*Q=I) and a superscript of * means conjugate transpose. By setting

x = QX

the noise-free channel output becomes y = RX , which is triangular. An optimum receiver for this signal if all the elements of y were in same place would be a simple decision feedback scheme that exploits the

2

This model applies to each tone and at each DMT symbol period. Thus, the H is both a function of tone index and possibly time as well. We avoid such a preponderance of notation here to illustrate concepts.

12/04/99

4

T1E1.4/99-559

triangular structure of R=SG where S is a diagonal matrix and G remains triangular but monic (ones along diagonal). However, the elements are not co-located, so instead a "block flexible" precoder is used at the

X X' + mod Q'

x

I-R

Figure 2 - Illustration of vector modulator for any DMT tone (see text). transmitter as shown in Figure 2. The mod box simply means the constellation is mapped modulo the boundary of the original constellation, just like in Tomlison precoding [11], for each tone. The outputs of the vector y elements at each receiver will then be free of interference from the other inputs, namely FEXT is eliminated, albeit at some overall loss in energy to the receivers because each of the y values has a scaling by the corresponding component of S. In VDSL, this loss is very small as Section 3 shows. Since S is diagonal, the individual receivers can independently implement

Y = S -1 y = X + N ,

which is a single multiply on each subchannel/tone per receiver. The receiver processing is thus relatively small. The transmitter processing involves implementation of the precoder and of the matrix Q, but these are relatively small since they are independent for each tone. For instance a well-designed system might see the combined complexity of these operations at about 20-30 multiplies per tone, which is well within the complexity of VDSL transceiver implementation. The transmission then has a set of parallel channels for each user where multiuser FEXT has been prequalized to be of smaller consequence on each tone of a DMT system. The performance improvement can be large. In fact, the authors could find no practical situations where the result was discernable from having no FEXT at all. In the upstream direction, the dual processing can be applied in that the upstream matrix can be factored as

* H u = Qu Ru and the upstream receiver forms Yu = Qu y u = Ru x + N u

which can be optimally decoded by decision feedback or soft cancellation for each tone [5], [8].3 VDSL Self-NEXT: The reader may note that the methods presented here also allow a mechanism for reduction of self-NEXT at both ends, but this additional advantage was not exploited because only FDM with nonoverlapping bands was considered. At the ONU, it is relatively trivial to eliminate self NEXT from a coordinated ONU with overlapping spectra (should it ever be used), amounting to one complex coefficient in an adaptive filter per tone per significant crosstalker [10]. Self-NEXT at the LT would have to be handled by the methods in [8], preferably exploiting soft information in the forward error correction syndromes. However, since NEXT is not an issue in FDM VDSL, we have not pursued these topics further here. However, overlapping bands with such NEXT cancellation would most certainly further improve performance.

3

Error propagation can be an issue and may require special attention to mitigate in the upstream receiver, but we do not address that here. Questions can be addressed to first author on this subject.

12/04/99

5

T1E1.4/99-559

3.

Simulation Results:

This section provides some data rate and range results for the use of both fully vectored (single service provider) and partially vectored (multiple service providers) within the same binder and compares the improvement with a system that does not use vectoring.

3.1 Performance versus ETSI requirements

Crosstalk coupling functions were individually modeled as those in the ETSI VDSL sytem requirements [6] where the phase was modeled as linear somewhat arbitrarily. We found the results insensitive to phase assumption, so this linear-phase assumption was not of signficant consequence. In practice, the actual coupling function, magnitude and phase of each crosstalker needs to be known to the ONU for implementation, but said known values can span a significant range without altering achievableperformance results. Channel models were also taken from [6] for the 6 noises (A-F). The same universal band allocation from [1] was used for all channels (whether symmetric, asymmetric, FTTCab or FTTex). To assist the reader's understanding, Table 1 summarizes the various rate/range requirements for ETSI. Table 1 - European Service Types Designation Upstream Data Rate (Mbps) Dnstrm Data Rate (Mbps) Range (meters) loop1-4 A1 (long asym - 1) 2.1 6.4 1600-1800 900 (C) A2 (long asym - 2) 2.1 8.6 1500-1700 900 (C) A3 (medium asym) 3.1 14.5 1200-1400 900 (C) A4 (short asym) 4.1 23.2 850-1000 S1 (long symmetric - 1) 6.4 6.4 900-1500 S2 (long symmetric - 2) 8.6 8.6 600-1250 S3 (medium symmetric) 14.5 14.5 600-800 S4 (short symmetric 1) 23.2 23.2 250-400 S5 (short symmetric 2) 28.3 28.3 200-300 Figures 3a, 3b, and 3c illustrate successively the achievable data rate versus service class for ETSI Loop 1 (we found this loop to be worst case and so only report for it, others have better performance) when no vectoring, partial vectoring, and full vectoring are used. In Figure 3a, one sees that ETSI requirements are not met, which is indicated by any of the service classes being below the "goal" line in Figure 3a - in this case 20 VDSL self-FEXT were used. Partial vectoring where only 5 VDSL self-FEXT are present from another service provider in the same cable is better in Figure 3b and just barely meets requirements. Full vectoring in Figure 3c easily meets all requirements. Full digital duplexed DMT as exactly proposed in [1] is used in all curves. The vectoring allows all the requirements to be met.

12/04/99

6

T1E1.4/99-559

DMT with No Vectoring: ETSI VDSL

40 35 Data Rate in Mbps 30 25 20 15 10 5 A1-UP A2-UP A3-UP A4-UP 0 A1-DOWN A2-DOWN A3-DOWN A4-DOWN S1 S2 S3 S4 S5 Noise A Noise B Noise C Noise D Noise E Noise F GOAL

Service Type

Figure 3a - Illustration of VDSL DMT with no vectoring and 20 VDSL self-FEXT. Note downstream asymmetric performance is below the requirement goal of ETSI in [6].

Partial Vectored DMT: ETSI VDSL

45 40 Data Rate in Mbps 35 30 25 20 15 10 5

A1-DOWN A2-DOWN A3-DOWN A4-DOWN A1-UP A2-UP A3-UP A4-UP S1 S2 S3 S4 S5

Noise A Noise B Noise C Noise D Noise E Noise F GOAL

0

Service Type

Figure 3(b) - Illustration of VDSL DMT with partial vectoring and 5 VDSL self-FEXT from another service provider. Note downstream asymmetric performance is very close to the requirement of ETSI in [6].

12/04/99

7

T1E1.4/99-559

Fully Vectored DMT: ETSI VDSL

90 80 Data Rate in Mbps 70 60 50 40 30 20 10 0

A1-UP A2-UP A3-UP A4-UP A1-DOWN A2-DOWN A3-DOWN A4-DOWN S1 S2 S3 S4 S5

Noise A Noise B Noise C Noise D Noise E Noise F GOAL

Service Type

Figure 3c - Illustration of VDSL DMT with full vectoring. Note downstream asymmetric performance easily exceeds the requirement of ETSI in [6], using a single universal band allocation for all types of service.

3.2 Performance versus ANSI requirements

North American results are listed here. These results use the same spectrum plan [1] as in the ETSI simulations of this paper. North American Noise D is essentially the same as North American Noise A, except that 2 T1 crosstalkers are included [9]. T1 noise is hostile to VDSL and prevents requirements from being met in any circumstances (even without unbundling). However, contribution NT-043 [8] shows how to mitigiate T1 crosstalk if necessary. Thus in Figures 4a-4c for American requirements, the only situations that vectoring of VDSL does not meet requirements are because of the T1 noise, which can be eliminated through other means. There are no noises B,C nor E,F in the ANSI VDSL requirements.

12/04/99

8

T1E1.4/99-559

American VDSL Performance - 90% range requirement with noises A and D, and 20 self-FEXT

40 Data Rate in Mbps 35 30 25 20 15 10 5 MA-UP MA-DN 0 LA-UP SA-UP LA-DN SA-DN MS LS SS Noise A - 24-gauge Noise A - 26-gauge Noise D - 24-gauge Noise D - 26-gauge GOAL

Service Type

Figure 4a - American requirements versus VDSL with no vectoring. Note asymmetric downstream requirements are not met, but symmetric are met.

Partial Vectoring, 5 VDSL FEXT vs American Requirement

50 45 40 35 30 25 20 15 10 5 0 LA- MA- SA- LA- MA- SA- LS MS SS UP UP UP DN DN DN

Figure 4b - American VDSL performance versus requirements for partially vectored DMT. Asymmetric requirements (outside of T1 noise in model D) are not met, but just barely.

Noise A - 24-gauge Noise A - 26-gauge Noise D - 24-gauge Noise D - 26-gauge GOAL

12/04/99

9

T1E1.4/99-559

Full Vectored VDSL vs American Requirement

80 70 Data Rate in Mbps 60 50 40 30 20 10 MA-UP MA-DN 0 LA-UP SA-UP LA-DN SA-DN MS LS SS Noise A - 24-gauge Noise A - 26-gauge Noise D - 24-gauge Noise D - 26-gauge GOAL

Service Type

Figure 4c - Full Vectoring in DMT VDSL versus American requirements. All requirements met easily (except T1 situation, and then on 26-gauge only). T1 noise needs handling as in [8].

4. Summary

Vectored VDSL enables a Universal Band Allocation for VDSL to meet all system requirements, even when asymmetric and symmetric services are mixed in the same cable. The complexity increase is relatively modest for DMT systems in terms of additional instructions per second because the precoding and pre-equalization can be independently executed for each tone. Coordination at the ONU side is necessary for vectoring, but such a site is the most likely site for coordination because many VDSL modems could be expected to reside in the same unit of a single manufacturer. Coordination of VDSL transceivers (as well as ADSL) is already used by a number of manufacturers for time-sharing of cost and power statistically, and the approach of this paper builds upon such structure to improve performance as well as to save power, size, and complexity in VDSL. This contribution proposes a sub-item for G.vdsl issues list 2.7 · Some nonzero TBD level of vectored VDSL may be assumed in assessing performance conformance under item 9.5 and in determining spectrum allocation in items under 2.10.x.

5. References

[1] Alcatel et al., "A Universal Spectrum Plan Proposal for DMT," ANSI T1E1.4 Contribution 99274R2, August 23, 1999. S. Verdu, "Adaptive Multiuser Detection," from Code Division Multiple Access Systems, (Editors, S.G. Glisic and P.A. Leppanen), Kluwer: Boston, 1995, pp. 97-116. P. Voois, "Two-Dimensional Signal Processing for Magnetic Storage Systems," Ph.D. Dissertation, Stanford University, December 1993.

[2]

[3]

12/04/99

10

T1E1.4/99-559

[4]

G. Raleigh and J. Cioffi, "Spatio-Temporal Coding for Wireless Communications," IEEE Transactions on Communications, Vol. 46, No. 3, March 1998. Cioffi, J.M. and Forney, G.D., Jr., "Generalized Decision-Feedback Equalization for Packet Transmission with ISI and Gaussian Noise," Chapter 4 of Communication, Computation, Control, and Signal Processing, (a tribute to Thomas Kailath), Kluwer: Boston, 1997, Ed: A. Paulraj, V. Roychowdhury, and C. Schaper. "Very high-speed Digital Subscriber Lines (VDSL) Part I: Functional Requirements," ETSI Technical Specification 101 270-1 V1.1.3 , Sophia-Antipolis, France, June 1999. Q. Wang, "VDSL Spectrum Plan Proposal," ANSI Contribution 99-345, Baltimore, MD, August 23, 1999. J.Cioffi, K. Cheong, W. Choi, and R. Negi, "G.vdsl: Fundamentals requirements induced by G.pnt/G.vdsl mixture," ITU SG15/Q4 Temporary Document NT-043, Nashville, TN, November 1, 1999. V. Oksman, S. Zeng, et al, "Noise Models for VDSL Performance," ANSI Contribution 99-438, Baltimore, MD, August 23, 1999. J. Cioffi, "Method for Crosstalk Cancellation," September 3, 1996 - available from author on request ([email protected]). T. Starr et al., Understanding Digital Subscriber Line Technology. Prentice-Hall: Upper Saddle River, NJ, 1999.

[5]

[6]

[7]

[8]

[9]

[10]

[11]

Information

9e145590.PDF

10 pages

Report File (DMCA)

Our content is added by our users. We aim to remove reported files within 1 working day. Please use this link to notify us:

Report this file as copyright or inappropriate

416237


You might also be interested in

BETA
Vectored transmission for digital subscriber line systems - Selected Areas in Communications, IEEE Journal on
9e145590.PDF
A01_RAO3334_01_SE_FM.QXD