T1E1.4/99-414 Project: Title: Source: VDSL The Impact of Upstream Power Back-Off on VDSL Frequency Planning Presenter: Krista S. Jacobsen Author: K.S. Jacobsen Texas Instruments 243 Samaritan Drive San Jose, CA 95124 Phone: 48.879.239 FAX: 48.879.2912 Location/Date: Baltimore, MD, 23-27 August 1999 Distribution: T1E1.4 Status: For information Abstract This contribution points out the need to consider power back-off in the upstream direction when VDSL frequency plan alternatives are evaluated. It is shown that upstream power back-off, which is necessary in distributed topologies, can affect the bit rates achievable with any chosen frequency plan. Also shown is that certain power back-off methods result in achievable bit rates that are very close to those computed assuming self-fext, which enables simplification of simulations to evaluate VDSL frequency plan alternatives. NOTICE This contribution has been prepared to assist ANSI Standards Committee T1 - Telecommunications. This document is offered to the Committee as a basis for discussion and is not a binding proposal on Texas Instruments. The requirements are subject to change after further study. Texas Instruments specifically reserves the right to add to, ammend, or withdraw the statements contained herein.
August 16, 1999 Page 1 of 8 The Impact of Upstream Power Back-off on VDSL Frequency Planning Krista S. Jacobsen, Texas Instruments Broadband Access Group Abstract This contribution points out the need to consider power back-off in the upstream direction when VDSL frequency plan alternatives are evaluated. It is shown that upstream power back-off, which is necessary in distributed topologies, can affect the bit rates achievable with any chosen frequency plan. Also shown is that certain power backoff methods result in achievable bit rates that are very close to those computed assuming self-fext, which enables simplification of simulations to evaluate VDSL frequency plan alternatives. 1. Introduction: the near-far problem Simulations to project VDSL performance generally assume that any far-end crosstalk (FEXT) at the VDSL receiver is due to disturbers that are identical to the line under consideration. In other words, it is assumed that all disturbers span the same distance and transmit the same power spectrum as the line under consideration, and self- FEXT results. In reality, lines emanating from a single ONU, CO or LEx may span a variety of distances in a distributed topology, as shown by the example in Figure 1. d N L N d N-1 L N-1 ONU/CO... d N-2 L N-2 d 1 L 1 Figure 1. Example distributed topology Projections of downstream VDSL performance that assume self-fext are accurate or slightly pessimistic when lines of varying lengths reside in the same binder as in Figure 1. The assumption of self-fext is accurate for short lines, on which FEXT levels are highest. For long lines, the assumption of self-fext is slightly pessimistic because the shorter lines do not couple over the entire length of the long lines. However, this effect is not likely to change performance projections appreciably because FEXT is not the dominant impairment on lines longer than several hundred meters. Instead, line attenuation limits the performance. In the upstream direction, however, the assumption of self-fext is optimistic for longer lines. If all VTU-Rs transmit at their maximum PSD levels, signals on shorter lines will detrimentally affect the upstream performance on longer lines. To illustrate, assume the maximum VDSL transmit PSD is -6 dbm/hz. Referring to Figure 1, the signal transmitted by the VTU-R on L N at this level will be attenuated significantly by the time it travels a distance (d N -d 1 ). At this point, transmissions from the VTU-R on L 1 may begin to couple into L N. If the transmit PSD of the VTU-R on L 1 is -6 dbm/hz, then it is significantly higher than the attenuated level of the desired signal on L N. The result of the (relatively) high-power interference is a degradation in achievable upstream rate on L N. Figure 2 shows the achievable bit rate in the upstream direction 1 as a function of range with two FEXT assumptions. The dashed curve illustrates achievable upstream bit rates assuming 1 self-fext disturbers. The solid curve is the upstream performance when the 1 disturbers consist of the following: 4 disturbers on loops 2 meters in length, 3 disturbers on loops 5 meters in length, and 3 disturbers on loops 9 meters in length. Referring to
August 16, 1999 Page 2 of 8 Figure 1, the configuration corresponds N=4,with d i = 2 m, d i+1 = 5 m, and d i+2 = 9 m. The loop of interest varies in length from 1 to 2 meters in 1-meter increments. All disturbers are assumed to transmit upstream at -6 dbm/hz. 12 Accounting for distribution of loops Assuming self FEXT 1 Upstream bit rate (Mbps) 8 6 4 2 2 4 6 8 1 12 14 16 18 2 Figure 2. Impact of the near-far effect on upstream VDSL bit rates Figure 2 shows the dramatic effect of the incorrect assumption of self-fext on projected upstream performance when all loops transmit at the maximum PSD level. On most loops the actual achievable rates are less than half the values projected assuming self-fext. Note that when the line of interest is shorter than about 2 meters, the assumption of self-fext is pessimistic because signals on 9 of the 1 disturbers are attenuated by loops longer than the line of interest. 2. Improving performance using upstream power back-off The near-far problem is most severe when a very long line is degraded by FEXT caused by shorter lines on which too high an upstream transmit PSD is used. 2 Thus, it is necessary to decrease the transmit PSDs on shorter lines. Several upstream power back-off methods have been identified. This section summarizes three methods and provides simulations to illustrate their achievable performances. To begin, an example loop configuration is established. Figure 3 illustrates a distributed topology with 1 nodes. At the end of each node are 2 VTU-Rs. The loops range in length from 15 meters to 15 meters, in increments of 15 meters. This configuration is used in all simulations. 1. To illustrate the need for upstream power back-off, the bit rate achievable in bandwidth from dc to 1 MHz is considered. In reality, of course, not all this bandwidth could be allocated to the upstream direction in a frequency-division duplexed system. 2. Loosely speaking, the transmit PSD on a short line is too high if its level is significantly higher than the level of the signal on the long line at the point at which the shorter line begins to couple into the longer line.
August 16, 1999 Page 3 of 8 ONU/CO... d 1 = 15 m d 9 = 135 m d 8 = 12 m L 8 L 9 L 1 L d 1 1 = 15 m Figure 3. Example distributed topology: 1 nodes, 2 VTU-Rs per node The VTU-Rs transmit upstream in the frequency bands from 1.1-2.5 MHz, 4.-5. MHz and 1.15-17.7 MHz. 2.1 Constant power back-off In the constant power back-off method, the upstream transmit PSDs on shorter lines are reduced equally across the upstream frequency band. Thus, the transmit PSD is flat, at some level less than or equal to the maximum allowed VDSL level, within the frequency band(s) allocated for upstream transmission. The amount by which the PSD is reduced is computed using a desired receive PSD at a reference frequency and either a model or measurements of the actual loop transfer function, which provides the loop attenuation at the reference frequency. The required transmit PSD level is then simply the desired received PSD at the reference frequency divided by the attenuation of the loop at the reference frequency. Using this method, the upstream PSD level is set so the received PSD level at the reference frequency is the same at the ONU for every loop. The desired receive PSD at the reference frequency can be computed using some reference loop, typically the longest loop in the binder. Alternatively, an arbitrary desired receive PSD level can be selected. Figure 4 illustrates the achievable upstream bit rates with the constant power back-off method on the example distributed topology. The bit rates with three reference frequencies (1., 2. and 3. MHz) are shown along with the achievable bit rates without power back-off. The desired receive PSD level at the specified reference frequency is the same as would be received at that frequency on a 15-meter loop (i.e., the longest loop in the binder). The performance of the constant power back-off method can be improved by using different desired receive PSD values and different reference frequencies in each upstream band. Note that when multiple reference frequencies are used the overall back-off is no longer constant over the entire frequency range but rather varies from band to band (but is constant within each band). Figure 5 shows the achievable upstream bit rates when the reference frequencies are the midpoints of the three upstream bands. The desired receive PSD levels at these reference frequencies were computed using a different reference length for each band. For the lowest band, the reference loop was 15 meters. For the middle and upper bands, the reference loops were 9 meters and 45 meters, respectively. Obviously, different choices for the desired receive PSDs at the reference frequencies will yield different results. Figure 5 reveals that using a different reference frequency and desired receive PSD level in each band generally results in higher overall upstream bit rates.
August 16, 1999 Page 4 of 8 7 6 1.MHz reference frequency 2.MHz reference frequency 3.MHz reference frequency No power back off 5 4 3 2 1 5 1 15 Figure 4. Upstream bit rates using constant power back-off with three reference frequency values and using a 15-meter loop to derive desired receive PSD values at reference frequencies. Also shown are bit rates without power back-off. 7 6 With power back off No power back off 5 4 3 2 1 5 1 15 Figure 5. Upstream bit rates using a different constant power back-off level in each upstream band. Desired receive PSD levels were computed using a different reference loop for each upstream band. Also shown are bit rates without power back-off.
August 16, 1999 Page 5 of 8 2.2 Reference length method The reference length method is a generalization of the constant power back-off method. In contrast to the constant power back-off method, where a desired receive PSD is established at only a single frequency, the reference length method sets the upstream transmit PSD so that the received PSD at all frequencies is equal to the PSD that would be received if a constant (that is, flat) PSD were transmitted on a loop of length L reference. To compute the required transmit PSD, the desired receive PSD and the actual loop attenuation within the upstream frequency bands must be known to the VTU-R. The required transmit PSD is then simply the desired received PSD divided by the loop attenuation. Thus, use of the reference length method requires a frequency-dependent reduction of the upstream transmit PSD. Applied strictly as defined above, the reference length method backs off the upstream transmit PSDs on all loops without taking into account the data-carrying capabilities of those bands on different length loops. The performance of the method can be improved substantially if it is recognized that although systems on short loops will transmit data in most or all of the available upstream bandwidth, systems on long loops use only the lowest bands. Consequently, the reference length method can provide higher upstream bit rates on shorter loops by increasing their transmit PSDs in higher frequency ranges. Because these frequency bands are too attenuated to support data on longer loops that are most susceptible to near-far FEXT, the higher transmit PSD levels will not affect upstream bit rates on longer loops. The transmit PSDs can be increased using a number of techniques. One method is to compute, using a model, the level of FEXT the computed transmit PSD would inject into other lines. In frequency bands in which the FEXT level is less than some maximum level (for example, the noise floor), the PSD can be increased so that the FEXT level is equal to the maximum. Figure 6 plots the performance of the reference length method when the PSDs are increased, if possible, using this method. The maximum acceptable FEXT level is the AWGN floor of -14 dbm/hz. 25 2 45m reference length 9m reference length 135m reference length No power back off 15 1 5 5 1 15 Figure 6. Upstream bit rates using the reference length method with various reference lengths. Loops are allowed to cause FEXT at the AWGN floor of -14 dbm/hz. Also shown are achievable bit rates without power back-off. Another implementation of the reference length method provides increased PSDs by using a different reference length in each upstream band. (This approach is similar to using a different reference frequency and desired PSD level in each band when the constant power back-off method is used.) For example, in the lowest band, a reference length of 15 meters might be used, while in the highest band a reference length of 3 meters might be chosen.
August 16, 1999 Page 6 of 8 2.3 Equalized-FEXT method The equalized-fext method attempts to equalize, using a known FEXT model, the level of FEXT received on each line at the ONU/CO. The upstream transmit PSDs are adjusted so that all lines affect each other approximately equally. The method is similar in approach to the reference length method; however, the equalized FEXT method allows higher transmit PSDs at lower frequencies because FEXT coupling at these frequencies is less than at higher frequencies. As in the reference length method, the transmit PSDs on shorter loops can be increased at higher frequencies without affecting performance on longer loops. Figure 7 plots the achievable upstream bit rates using the equalized-fext method and applying the same PSD-increasing procedure described in the previous section with a maximum level of -14 dbm/hz. Note that, as expected, the performance is similar to when the reference length method is used. 25 2 45m reference length 9m reference length 135m reference length No power back off 15 1 5 5 1 15 Figure 7. Upstream bit rates using equalized-fext method with three reference length values. Also shown are bit rates without power back-off. 3. Impact of upstream power back-off on simulations As the previous section showed, the achievable bit rates in the upstream direction can be affected dramatically by the chosen power back-off method. This effect must be considered when VDSL frequency plan alternatives are evaluated; the assumption of simple self-fext can result in optimistic projections of upstream bit rates. This section compares the results of simulations in which (a) the upstream bit rates in a distributed topology are computed assuming self-fext (i.e., what might be done in computer simulations); (b) the actual upstream bit rates are computed accounting for the distributed topology but not using any power back-off (i.e., the bit rates that would result if transceivers were deployed without power back-off capabilities in a distributed topology); and (c) the actual upstream bit rates are computed accounting for the distributed topology and applying a particular power back-off method (i.e., the bit rates that would result is transceivers with power back-off capabilities were deployed in a distributed topology). For consistency, the same example loop topology and upstream bands established in the previous section are used again. Figure 8 shows the achievable upstream bit rates when the constant power back-off method is applied in its most basic form with a single reference frequency of 1. MHz. The plot illustrates clearly that simulations assuming self-fext are generally highly optimistic relative to the rates sustainable with constant power back-off with a 1.- MHz reference frequency.
August 16, 1999 Page 7 of 8 7 6 With constant back off (1.MHz reference) Assuming self FEXT With no power back off 5 4 3 2 1 5 1 15 Figure 8. Comparison of upstream bit rates with and without constant power back-off. Also shown are the upstream rates achievable when self-fext is assumed in simulation. Figure 9 plots the upstream bit rates when the constant power back-off method is used with a different reference frequency and desired receive PSD in each of the three upstream bands. The reference frequencies are the same as in Section 2.1; the reference loops for the three bands are 12, 9 and 45 meters. Use of multiple reference frequencies clearly improves the overall performance and provides upstream bit rates much closer to those projected when self-fext is assumed in simulation. 7 6 With constant back off (multiple ref. freqs.) Assuming self FEXT With no power back off 5 4 3 2 1 5 1 15 Figure 9. Comparison of upstream bit rates with and without constant power back-off. Also shown are the upstream rates achievable when self-fext is assumed in simulation.
August 16, 1999 Page 8 of 8 Figure 1 shows the upstream bit rates achievable when the reference length method is applied with the PSD increasing technique described in Section 2.2. The reference length was chosen to be 15 meters. The figure shows clearly that use of the reference length method in conjunction with the PSD increasing technique results in a performance with power back-off that is very close to the performance when self-fext is assumed. This result is important: if this method of power back-off is presumed, simulations to project the performances of VDSL frequency plan alternatives can be simplified substantially by including only self-fext. Furthermore, consideration of the distributed topology is not necessary in evaluations of VDSL frequency plans. 7 6 With reference length back off Assuming self FEXT With no power back off 5 4 3 2 1 Figure 1. Comparison of upstream bit rates with and without reference length power back-off with the PSD increasing procedure. Also shown are the upstream rates achievable when self-fext is assumed in simulation. 4. Summary The simulation results presented in this document illustrate the importance of considering upstream power back-off when evaluating VDSL frequency plans. Power back-off is necessary in distributed topologies to provide reasonable upstream bit rates on longer loops. Simulations that assume a self-fext model can project overly-optimistic upstream VDSL data rates when the loop configuration is a distributed topology. Therefore, simulations of various VDSL frequency plans must include some method of power back-off. In addition, a specific configuration of loops must be assumed, as results vary for different configurations. An interesting result of this contribution is that if the reference length method with the PSD increasing technique is presumed, the actual upstream bit rates are very close to those computed using a self-fext model. Thus, if this method is presumed, simulations to project upstream bit rates can be simplified substantially by assuming self- FEXT and neglecting consideration of the distributed loop topology. 5. References 5 1 15 [1] FSAN VDSL Working Group. Power-Backoff methods for VDSL. ETSI TM6 contribution 983T17A. Lulea, Sweden. June 1998. [2] K.S. Jacobsen. The Equalized-FEXT Upstream Power Cutback Method to Mitigate the Near-Far FEXT Problem in VDSL. ETSI TM6 contribution 985t5r. Sophia Antipolis, France. November 1998.