Dynamic Spectrum Management for Mixtures of ...

44th Annual Conference on Information Sciences and Systems

1

Dynamic Spectrum Management for Mixtures of Vectored and Non-vectored DSL Systems Mehdi Mohseni, Member, IEEE, Georgios Ginis, Senior Member, IEEE, and John M. Cioffi, Fellow, IEEE

Abstract—A number of Vectored DSL system prototypes have confirmed the substantial performance gains from Far-End Crosstalk (FEXT) cancellation. With field trials and initial deployments expected to materialize in the next two years, a very important question is how such Vectored DSL systems can coexist with non-vectored DSL systems sharing the same cable. This paper shows that the performance gains for Vectored DSL systems can be maintained in a mixture of vectored and nonvectored lines if a Spectrum Management Center (SMC) is assigned to control the impact of crosstalk from the non-vectored to the vectored lines. Simulation results are presented for certain scenarios, first, to illustrate that an Iterative Waterfilling strategy can achieve performance near the optimal, and second, to explore the trade-off between limiting the rates of the non-vectored lines and boosting the rates of the Vectored DSL systems. Index Terms—Digital subscriber line, dynamic spectrum management, power management, vectoring, iterative waterfilling.

I. INTRODUCTION

V

ectored DSL systems are capable of reducing the crosstalk effects that limit data rate performance in dense deployment of DSL lines operating in the very-high-speed region (above 15 Mbps) [1][2][3]. Vectored DSL system prototypes have already confirmed the very significant performance gains predicted by earlier theory [4]. Recently, a standard for Vectored DSL was published [5], and field trials and initial deployments are expected to happen within the next two years. The major breakthrough of Vectored DSL technology is that it pushes the data rates achievable on a single copper twisted-pair to the region of 100 Mbps. This is dedicated bandwidth delivered to each customer, as opposed to shared bandwidth available through other broadband access systems using coaxial cable or Passive Optical Networks (PON). To realize the full benefits from vectoring, there are three important problems that need to be solved via management of the physical layer [4]. First, computational resources may not be sufficient to apply vectoring across all lines served from a given node. Especially for nodes serving a number of lines of M. Mohseni, and G. Ginis, are with ASSIA, Inc, Redwood City, California (phone: 650-654-3400; fax: 650-654-3404; e-mails: [email protected], [email protected],). J. M. Cioffi is with ASSIA, Inc, and Stanford University.

100 or more, cancellation of crosstalk occurring from any pair to any pair can be computationally prohibitive. In such cases, a Spectrum Management Center (SMC) [6] guides the assignment of computational resources to the pairs that have the highest service requirements, and which can indeed improve their data rate performance through vectoring. Second, the advantages of higher data rates with Vectored DSL make sense only if lines can meet quality of service requirements, such as very low counts of transmission errors and very rare occurrences of reinitializations. Events of errors and reinitializations are typically associated with either impulse noise or with other time-varying noise sources. Preventing these events requires tuning of the impulse noise protection parameters or of the available SNR margin of the DSL link by an SMC. Third, Vectored DSL lines are expected to share cable with existing non-vectored DSL systems (most commonly VDSL2 [7]). This can be the case for two reasons: At least in the initial phases of Vectored DSL deployment, operators will be unwilling to replace existing DSL equipment. A probable scenario is that VDSL2 and Vectored DSL systems will be deployed from the same node and will be sharing cable binders. Additionally, in geographies with loop unbundling, an operator with Vectored DSL systems may share binders with an operator that is using (non-vectored) VDSL2 systems, but with no plan for upgrading to Vectored DSL. In these scenarios, a very important task of the SMC is to prevent the data rates of Vectored DSL from degrading because of crosstalk induced by the non-vectored lines. This paper studies this third class of management problems for Vectored DSL. Dynamic Spectrum Management (DSM) [6] includes techniques for managing the transmitted Power Spectral Density (PSD) of lines to mitigate the crosstalk effects. Several DSM strategies have been proposed for mitigating crosstalk in non-vectored DSL systems [8] [9][10][11], but no results were publicly available until recently for mixtures of Vectored DSL and VDSL2 systems. An investigation of practical DSM techniques for such mixtures was very recently presented in [12], specifically focusing on upstream transmission. This paper investigates DSM for downstream transmission with mixtures of Vectored DSL and VDSL2 systems. Simulation results are presented to first compare a practical Iterative Waterfilling [8] management strategy with the

44th Annual Conference on Information Sciences and Systems theoretically optimal solution, obtained through Optimal Spectrum Balancing (OSB) [9]. Further simulation results are then presented for various scenarios of service requirements for the non-vectored lines. Finally, a few comments are made on practical aspects of operating an SMC with a mixture of Vectored and non-vectored DSL.

II. SYSTEM MODEL The system model used in this paper for investigating the co-existence of Vectored and non-vectored DSL systems is shown in Fig. 1. There are N 1 lines originating at the nonvectored DSL access node, and N 2 lines originating at the Vectored DSL access node. The non-vectored lines always use VDSL2 [7] technology, and the two nodes are “colocated”. The N 1 +N 2 pairs are assumed to be within the same 25-pair binder, and are thus experiencing crosstalk. To simplify this initial study of the problem, all pairs are assumed to have equal length, and to be within the same binder for their entire length. The access nodes are connected to an SMC, and thus, the connected lines have their physical layer control parameters programmed by the SMC [6].

Fig. 1. Mixed deployment of Vectored and Non-Vectored DSL, with N 1 lines using non-vectored DSL (VDSL) and N 2 lines using Vectored DSL (VDSL). All lines are assumed to have equal length.

The multi-pair channel model for downstream transmission is as defined in [2], where for each DMT tone, the output samples can be expressed as

Y = H⋅X+N,

(1)

where Y is the vector of received samples (each element corresponding to a different line), X is the vector of transmitted samples, N is the vector of received noise samples, and H is the channel transfer matrix at the specified tone. The diagonal elements of the channel transfer matrix are defined using the transmission line model specified in [14]. The off-diagonal elements of the channel transfer matrix are as specified for the first 25-pair binder defined in [15].

2

III. COMPARISON OF ITERATIVE WATERFILLING WITH OPTIMUM SPECTRUM BALANCING A. Algorithm Description The presence of even one non-vectored line contributes to crosstalk for the vectored lines, which cannot be eliminated through vectoring. This leads to data rate degradation compared to the case when all lines belong to a single vectored group. In the opposite direction, non-vectored DSL systems are affected by the crosstalk created by Vectored DSL systems, but in the same way as crosstalk created by other nonvectored systems (also known as “self-crosstalk”). Nonvectored DSL systems are typically already limited by selfcrosstalk regardless of the presence of Vectored DSL systems, but are often transmitting at maximum power, regardless of the lines’ service requirements. These observations lead to the proposal of reducing the transmitted power spectral density of the non-vectored DSL lines, while preserving the lines’ service requirements. This reduction limits the crosstalk induced on the Vectored DSL systems, and is not affecting the service of the non-vectored lines. This proposal is first investigated by comparing the solutions obtained by the two best-known algorithms for DSM: Iterative Waterfilling [2], and Optimum Spectrum Balancing (OSB) [9]. Iterative Waterfilling is a practical power control algorithm, which on the one hand has a robust distributed implementation, but on the other hand is known to be suboptimal in certain “near-far” scenarios. OSB is an algorithm for obtaining the optimum solution to the power control problem, but which cannot be implemented in practical DSL systems, and which becomes computationally very expensive for even a moderate number of lines. The intention of this investigation is to understand how different are the solutions obtained by these two diverse algorithms, at least for scenarios of co-located nodes and equal (or near-equal) loop lengths. The formulation of the optimization problem for OSB is next described. The optimization objective can be stated as maximize the minimum downstream rate among all vectored lines. This maximization is performed over all allowed power assignments for the vectored and non-vectored lines. The following constraints apply: 1) The downstream rate of every non-vectored line must be at least equal to the target data rate for non-vectored lines. 2) The aggregate transmit power limits must not be exceeded for any line. 3) The PSD mask must not be exceeded for any line. The above maxmin formulation can be easily transformed into the OSB problem described in [9] using standard methods known from convex optimization theory. The strategy with Iterative Waterfilling is to obtain the power assignments for the vectored and non-vectored lines following the following iterative approach: 1) Derive the waterfilling power allocation for the target data rate for each of the non-vectored lines, taking into account crosstalk from all other lines, and requiring a given


FEXT free vectored lines rate Min of vectored lines rate for OSB, non-vectored capped at 25Mbps Min of vectored lines rate for IW, non-vectored capped at 25Mbps Non-vectored line rate for both OSB and IW Min of vectored lines rate, non-vectored line rate is uncapped Non-vectored line rate, uncapped

180

160

140

120

Data rates (Mbps)

target SNR margin (also known as “fixed-margin waterfilling”). 2) Derive the waterfilling power allocation for maximizing the rate of each vectored line, taking into account crosstalk only from non-vectored lines (also known as “rate-adaptive waterfilling). 3) Repeat the above steps until the data rates have converged. The next subsection compares the results obtained with the the OSB and Iterative Waterfilling methods.

3

100

80

60

40

B. Simulation Results The simulation parameters are listed in Table I.

20

0 500

1000

1500

2000

2500

3000

3500

Loop length (ft)

TABLE I SIMULATION PARAMETERS Parameter

Value

VDSL2 profile Upstream PSD limit mask Downstream PSD limit mask Background noise Target SNR margin Net coding gain SNR gap for uncoded QAM Max number of bits per tone

17a EU-32 D-32 AWGN, -140 dBm/Hz 6 dB 4.2 dB 9.8 dB 15

To reduce the computational burden of OSB, N 1 =1 nonvectored DSL lines, and N 2 =2 Vectored DSL lines were considered. These are only 3 pairs from the 25-pair binder, specifically chosen because of their strong crosstalk coupling. Pair number 24 of the binder model [15] was used for the nonvectored line, and pair numbers 8 and 25 were used for the Vectored DSL lines. The bit loading per bin was performed using the “gap” approximation with the parameters listed in Table I. Fig. 2 presents downstream data rates obtained at different loop lengths for the vectored and non-vectored lines, with either OSB or with Iterative Waterfilling. On this figure, the “FEXT-free” vectored lines rate is the rate that would be achieved if all lines were vectored (see top curve). This represents the level of performance that reached if perfect binder management were applied to eliminate any nonvectored lines from the binder. For example, at 1500 feet, the downstream data rate achievable is 124 Mbps. Unfortunately, such practices are rejected by most service providers as inappropriate for large-scale networks. Next, on the same figure, the data rates are considered corresponding to the case when no DSM is applied: The data rate of the worstperforming vectored line is shown with square data-points, and the data rate of the non-vectored line is shown with cross datapoints. Clearly, the data rate of the vectored lines degrades significantly if no attempt is made to control the tranmit PSD of the non-vectored line. For example, at 1500 feet, the downstream data rate of the worst-performing vectored line is 75 Mbps, while the rate of the non-vectored line is 65 Mbps.

Fig. 2. Downstream data rates of vectored and non-vectored lines with OSB and with Iterative Waterfilling (IW) and 25 Mbps data rate target for the non-vectored line. Binder includes 2 vectored lines and 1 non-vectored line.

The next set of curves show the data rates of the worstperforming vectored line assuming that the non-vectored line has its rate limited to no more than 25 Mbps. Two curves are shown, corresponding to the OSB (diamond data-points) and to the Iterative Waterfilling (triangle data-points) algorithms. The two curves are really close at almost all loop lengths. For example, at 1500 feet, the OSB solution provides a rate of 117 Mbps, while the Iterative Waterfilling solution provides 114 Mbps. Both of these curves compare favorably with the “FEXT-free” data rate curve. To better illustrate the benefit from the DSM methods, Table II presents in an alternate way some of the results of Fig. 2. The table shows the downstream data rate of the worstperforming vectored line at 1000, 1500 and 2000 feet. TABLE II DOWNSTREAM DATA RATE OF WORST-PERFORMING VECTORED LINE Method

1000 feet

1500 feet

2000 feet

No DSM With Iter. Waterf. With OSB FEXT-free

83 Mbps 150 Mbps 151 Mbps 158 Mbps



There are two important observations in this section: First, power optimization for the non-vectored line can greatly reduce the impact of crosstalk from the non-vectored line to the vectored lines. Second, for this particular scenario of colocated nodes and equal loop lengths, management of the power level is nearly as effective as management of the exact PSD shape. This last observation contrasts with the results presented in [12] for upstream transmission, where it is shown that multi-level PSD shaping can achieve performance that is significantly better than “flat” power adjustment. Therefore, the conclusion drawn here about the effectiveness of managing the power level must be viewed only in the context of the assumed scenario.


4

IV. PERFORMANCE OF ITERATIVE WATERFILLING WITH RANDOM PAIR SELECTION

180 FEXT free data rate Vectored: mean rate over 100 pair selections Non-vectored: mean rate over 100 pair selections Vectored: max rate over 100 pair selections Non-vectored: max rate over 100 pair selections Vectored: min rate over 100 pair selections Non-vectored: min rate over 100 pair selections

160

140

120

Data rate (Mbps)

A. Discussion Based on the findings of the previous section, a further investigation was undertaken in this section to understand the trade-off between reducing the transmitted power of nonvectored lines and improving the data rate performance of the vectored lines, and only assuming the use of the Iterative Waterfilling algorithm. Further in the analysis of this section, the data rate variation resulting from different pair selections within the binder is evaluated.

100

80

60

40

20

0 500

1000

1500

2000

2500

3000

3500

Loop length (ft)

Fig. 3. Downstream data rates of vectored and non-vectored lines without DSM, and downstream data rate assuming all lines are vectored (“FEXTfree”). Binder includes 20 vectored lines and 5 non-vectored lines.

The second scenario assumes that DSM is applied, and that using Iterative Waterfilling, the transmitted power of the nonvectored lines is limited to the minimum power necessary to achieve the desired data rate target. Fig. 4 shows the results for a downstream data rate target of 25 Mbps for the nonvectored lines. In this case, all non-vectored lines are forced to operate at 25 Mbps, and as a result are able to conserve their transmitted power. The vectored lines are then able to achieve much higher rates than when non-vectored lines transmitted with full power. At 1500 feet, the vectored lines operate within a narrow band of 108 to 124 Mbps with a mean rate of 116 Mbps. That compares favorably with the FEXTfree rate of 124 Mbps. 180 Vectored: mean rate over 100 pair selections Non-vectored: mean rate over 100 pair selections Vectored: max rate over 100 pair selections Non-vectored: max rate over 100 pair selections Vectored: min rate over 100 pair selections Non-vectored: min rate over 100 pair selections

160 140 120

Data rate (Mbps)

B. Simulation Results The simulation parameters listed in Table I are also used here. It was assumed that there are N 1 =5 non-vectored lines and N 2 =20 vectored lines. To evaluate the statistical variation of data rates resulting from pair placement within the binder, a Monte Carlo method was used, where in each “experiment”, a different pair selection was made. A total of 100 experiments were performed for each simulation scenario. For each of the experiments, data rates were recorded for the worstperforming pairs, both within the vectored group and within the non-vectored group. After completing all 100 experiments, the minimum, mean and maximum data rates of the worstperforming pairs within the vectored and within the nonvectored groups were computed. The first scenario assumes that no DSM is applied, and that all lines are allowed to transmit at their maximum transmit power. This is shown in Fig. 3, where the FEXT-free data rate is also presented for comparison. The curves on the plot show the minimum, mean and maximum rates of the worstperforming pairs within the vectored and non-vectored groups. It is very clear that there is an immediate dramatic effect on data rate from including even a small number of non-vectored lines in the same binder where vectored lines are operating. As an example, the downstream data rate of a vectored line at 1500 feet is 124 Mbps. When the 5 non-vectored pairs are included, this data rate drops, and varies between 64 Mbps and 87 Mbps, with an average of 73 Mbps. The data rate performance of the vectored lines is better than the data rate performance of non-vectored lines, but that it is much lower than the “FEXT-free” data rate.

100 80 60 40 20

0 500

1000

1500

2000

2500

3000

3500

Loop length (ft)

Fig. 4. Downstream data rates of vectored and non-vectored lines with Iterative Waterfilling and 25 Mbps data rate target for non-vectored lines. Binder includes 20 vectored lines and 5 non-vectored lines.

Similar results can be found in Figures 5 to 7, when the target rate is respectively limited to 30 Mbps, 35 Mbps and 45 Mbps. As the target data rate increases two effects are observed for the Vectored DSL lines: a) The data rates decrease on average. b) The variation of data rates increases.


5

For example, at 1500 feet and for a target rate of 40 Mbps for the non-vectored lines, the vectored lines have a data rate range between 90 and 117 Mbps, with a mean of 100 Mbps. Still, these data rate speeds are comparable to the FEXT-free rate of 124 Mbps. Vectored: mean rate over 100 pair selections Non-vectored: mean rate over 100 pair selections Vectored: max rate over 100 pair selections Non-vectored: max rate over 100 pair selections Vectored: min rate over 100 pair selections Non-vectored: min rate over 100 pair selections

160

Data rate (Mbps)

140

120

100 80

60

40

100

20

0 500

20

1000

1500

2000

2500

3000

2000

2500

3000

3500

3500

Loop length (ft)

V. CONCLUSION


180 Vectored: mean rate over 100 pair selections Non-vectored: mean rate over 100 pair selections Vectored: max rate over 100 pair selections Non-vectored: max rate over 100 pair selections Vectored: min rate over 100 pair selections Non-vectored: min rate over 100 pair selections

160 140 120 100 80 60 40 20 0 500

1500


40

0 500

1000

Loop length (ft)

60

Data rate (Mbps)

140

120

80

Vectored: mean rate over 100 pair selections Non-vectored: mean rate over 100 pair selections Vectored: max rate over 100 pair selections Non-vectored: max rate over 100 pair selections Vectored: min rate over 100 pair selections Non-vectored: min rate over 100 pair selections

160

Data rate (Mbps)

180

180

1000

1500

2000

2500

3000

3500

Loop length (ft)


This paper showed that Vectored and non-vectored DSL systems can co-exist in the same binder, when a SMC is able to apply DSM methods on the lines to limit the crosstalk induced from the non-vectored lines to the vectored lines. A few practical aspects of DSM for mixtures of Vectored and non-vectored DSL systems are here discussed as closing remarks. In actual field deployment, it is not possible to apply a single set of configuration parameters to all non-vectored lines. This is because a) service providers offer a diverse set of service products, each with different data rate and quality of service requirements; and, b) different lines require different configurations for protection from impulse noise (e.g. through stronger coding-interleaving or retransmission), or for improved resiliency to time-varying noise (e.g. through provisioning to maintain adequate SNR margin at all times). The best practice is for an SMC to monitor all lines individually for sufficiently long periods of time. Subsequently, the SMC configures each of the non-vectored lines by taking into account both the service requirements and the historical conditions of each line. In order to achieve the transmitted power reductions described in this paper, any of several control parameters can be configured, such as the maximum SNR margin, the maximum aggregate transmit power, or the PSD mask, depending on the capabilities of the DSL equipment. This practice leads to the largest possible benefits from deploying Vectored DSL systems.


REFERENCES [1]

[2]

[3]

[4]

[5] [6] [7] [8]

[9]

[10]

[11]

[12]

[13]

[14] [15]

G. Ginis, and J. M. Cioffi, “Vectored transmission for digital subscriber line systems,” IEEE J. Select. Areas Comm., vol. 20, no. 5, pp. 10851104, June 2002. R. Cendrillon, G. Ginis, E. Van den Bogaert, and M. Moonen, “A nearoptimal linear crosstalk canceler for upstream VDSL,” IEEE Trans. Signal Proc., vol. 54, no. 8, pp. 3136-3146, Aug. 2006. R. Cendrillon, G. Ginis, E. Van den Bogaert, and M. Moonen, “A nearoptimal linear crosstalk precoder for downstream VDSL,” IEEE Trans. Comm., vol. 55, no. 5, pp. 860-863, May 2007. J. M. Cioffi, K. Fisher, A. Clausen, M. Peeters, P.-E. Eriksson, and G. Ginis, The Path to 100 Mbps DSL Services, IEEE Globecom 2009, Access Forum, session 203, December 2003. Self-FEXT Cancellation (Vectoring) for Use with VDSL2 Transceivers, ITU-T Recommendation G.993.5 (g.vector), January 2010. Dynamic Spectrum Management, ATIS NIPP-NAI Technical Report, ATIS-PP-0600007. Very High Speed Digital Subscriber Line Transceivers 2, ITU-T Recommendation G.993.2, January 2010. W. Yu, G. Ginis and J.M. Cioffi, “Distributed multiuser power control for digital subscriber lines,” IEEE J. Select. Areas Comm., vol. 20, no. 5, pp. 1105-1115, June 2002. R. Cendrillon, W. Yu, M. Moonen, J. Verlinden, and T. Bostoen, “Optimal multiuser spectrum balancing for digital subscriber lines,” IEEE Trans. Comm., vol. 54, no. 5, pp. 922 – 933, May 2006. R. Cendrillon, J. Huang, M. Chiang, and M. Moonen, “Autonomous Spectrum Balancing for Digital Subscriber Lines,” IEEE Trans. Signal Processing, vol. 55, no. 8, pp. 4241–4257, Aug. 2007. S. Jagannathan, and J. M. Cioffi, “Distributed, Adaptive Bit-loading for Spectrum Optimization in Multi-user Multicarrier Systems,” IEEE ICC 2008, pp. 625-630, May 2008. J. M. Cioffi, A. Chowdhery, H. Zou, and P. J. Silverman, “Mixed Vectored and Non-Vectored VDSL2,” ATIS Committee COAST-NAI, Working Group NAI (DSL Access), contribution COAST-NAI-2010005, Dallas, TX, January 27, 2010. Physical Layer Management for Digital Subscriber Line (DSL) Transceivers, ITU-T Recommendation G.997.1 (g.ploam), January 2010. Spectrum Management for Loop Transmission Systems, ATIS Technical Report 0600417. Multiple-Input Multiple-Output Crosstalk Channel Model, ATIS NIPPNAI Technical Report, ATIS-PP-0600024.

6 Mehdi Mohseni (S’02–M’06) received a B.S degree in electrical engineering from Sharif University of Technology, Tehran, Iran in 2001, and M.S. and Ph.D. degrees in electrical engineering from Stanford University, Stanford, California, in 2003 and 2006, respectively. He joined ASSIA, Inc, Redwood City, California in 2006. He is currently holding the position of Systems Architect and is involved in research and development efforts for software modules implementing dynamic spectrum management for DSL. His research interests include multi-user information theory, convex optimization techniques, and their application to wireless and broadband communications. Dr Mohseni received the IEEE Communications Society Award for Outstanding Paper on New Communications Topics in 2008, and the Award for Outstanding Paper in ISSLS 2004. Georgios Ginis (S’97–M’02–SM’08) received the Diploma in electrical and computer engineering from the National Technical University of Athens, Athens, Greece, in 1997, and the M.S. and Ph.D. degrees in electrical engineering from Stanford University, Stanford, California, in 1998 and 2002, respectively. Between 2002 and 2005 he was a Systems Engineer, Member Group Technical Staff, at the Broadband Communications Group of Texas Instruments, San Jose, California, where he was involved in the design of DSL chipsets for central office equipment and residential gateways. In 2005 he joined ASSIA, Inc in Redwood City, California as Director of Technology, responsible for technology development, intellectual property and standards. He is currently Vice President of Expresse Software with ASSIA, Inc, overseeing development and deployment of the DSL Expresse management product. His research interests include dynamic spectrum management for DSL applications. Dr Ginis is also serving as an associate editor for IEEE Communications Letters. John M. Cioffi (S’77–M’78–SM’90–F’96) received the B.S.E.E degree from the University of Illinois, Urbana-Champaign, in 1978 and the Ph.D.E.E. degree from Stanford University, Stanford, CA, in 1984. He was with Bell Laboratories, Holmdel, NJ, from 1978 to 1984 and IBM Research, San Jose, CA from 1984 to 1986. He has been with Stanford Universityas an Electrical Engineering Professor from 1986 to the present. He founded Amati Communications Corporation, Palo Alto, CA, in 1991 (it was purchased by Texas Instruments, Incorporated in 1997), and was Officer/Director from 1991 to 1997. He presently serves as CEO and Chairman of the Board at ASSIA, Inc, Redwood City, California. His specific interests are in the area of high performance digital transmission. Dr Cioffi is a Marconi Fellow (2006) and a member of the National Academy of Engineering (2001). He has received the IEEE Kobayashi Medal (2001), the IEEE Millennium Medal (2000), and the IEE JJ Tomson Medal (2000). He has published over 250 papers with several winning best paper awards and holds over 100 patents.