Talk:Digital subscriber line (DSL)
The topic is very important, and the authors are certainly among the best possible ones for this topic. The article is very well written, and will certainly be useful to the reader of Scholarpedia.
In the following, I offer few comments to the authors.
Author's response) The authors thank the anonymous reviewer for comments and suggestions. Our response is below.
Section "Key Technologies and Features"
Reviewer's question) The authors mention only DMT, but I believe that a least a reference to CAP should be given, after all theory predicts comparable performance under ideal implementations. Now we know enough about these modulation schemes to compare their inherent similarities and differences in performance and cost, for example DMT having some advantage in digital processing, but requiring greater cost in analog circuitry.
Author's response) We have added a small subsection entitled “Selection of DMT,” which reads
Selection of DMT
Some readers of this article have asked about the selection of DMT for DSL, so this section is added to explain it. While it was initially controversial, several early manufacturers were misled by supposed theoretical equivalences between single-carrier DSL proposals and DMT. These proposals concluded that the two would perform about the same if implemented correctly. The theoretical equivalence of multi-tone transmission and single-carrier only holds under strict conditions that are almost never met in DSL channels. The lack of equivalent is why DMT outperformed (by more than a factor of 10 in noise immunity, thus an SNR advantage of more than 10 dB) single-carrier methods in two separate DSL competitions held by neutral test laboratories for ADSL in 1993 and again for VDSL in 2003. DMT has a highly efficient implementation that benefits from Fast Fourier Transform (FFT)s that are often used in sophisticated digital-signal-processing systems to reduce the complexity. Even with infinite complexity and infinitely long “equalizer” filters, single carrier methods will not perform as well as DMT on DSL connections – single-carrier methods can always perform as well as DMT on voiceband modem channels below 4 kHz of bandwidth, which lead to the confusion of many in proposing the single-carrier methods for DSL.
Section "Discrete Multitone"
Reviewer's question) The word "narrow" in the following sentence is misleading.
"Instead, DMT partitions the transmit spectra into **narrow** subchannels, and only those subchannels that pass through the channel are loaded with information (or bits) via the process of water-filling and bit-loading."
Author's response) Ok, we see the point – let us try the adjusted text below instead. "A heuristic view of DMT is that it partitions the transmit spectra into narrow subchannels, and only those subchannels that pass through the channel are loaded with information (or bits) via the process of water-filling and bit-loading. In implementation, the word ‘narrow’ is relative to a time-domain guard period placed between transmission of information on each successive use of the tones. When the number of tones is large, so that over any fixed bandwidth the tones are correspondingly `narrow,’ the overhead of the guard period becomes negligible"
Reviewer's question) It is not necessary that the subchannels be "narrow", I suggest replacing "narrow" with "multiple".
I believe the following sentence is technically imprecise, and should be deleted.
"The subchannels are chosen to be sufficiently narrow, so an equalizer is not required.
Author's reponse) We have adjusted the sentence.
"When the number of subchannels is sufficiently large, then an equalizer is not required.
Reviewer's question) The width of the sub-channel is irrelevant with respect to equalization. Only the Guard Interval (GI) has a role in equalization and a GI that is longer than the maximum duration of the imulse response completely eliminates Inter-Symbol Interference (ISI) irrespectively of the subchannel width (or of the number of carriers given the total bandwidth). The only role of subchannel width (or, equivalently, the number of carriers given the total bandwidth) is to vary the transmission efficiency and, therefore, compensate for the presence of the GI. Smaller subchannel width (more carriers for same total bandwifth) causes the symbol duration to be longer and, therefore, the ratio of the GI duration to the total symbol duration (overhead) goes to zero asymptotically.
Author's response) We believe we have addressed the above paragraph with the suggested corrections.
Section "Adaptation"
Reviewer's question) Is also the GI of ADSL variable depending on channel condition? The delay spread of a 6 kft loop is certainly smaller than the delay spread of a 12 kft loop, thus imposing different requirements on the GI length. If the GI and the number of carriers are fixed, then transmission efficiency may vary depending on the loop topology. The reader would be interested to know if also this kind of adaption has been included in the ITU standards and, if not, why.
Author's response) The following sentence has been added to this section.
This process of bit-swapping is independent of any variation in the underlying exact symbol lengths and inter-symbol guard intervals, the latter of which may vary slightly from channel to channel in some of the advanced ITU transmission standards’ details.
Section "Dynamic Spectrum Management"
Reviewer's question) "INP and delay parameters of each DSL" are not defined anywhere.
Author's response) We proposed to add INP = Impulse Noise Protection is a parameter in standards that measures the length of noise bursts (in units of 0.25 milliseconds). Delay = the delay caused by the combination of transmitter interleaving and receiver de-interleaving in a DSL connection, measured in milliseconds
Reviewer's question) A major component of advanced DSL management is a database of loop information for DSL provisioning and maintenance. This database is envisioned as having a variety of information on loops, noise, and the histories of deployed DSLs extending far beyond existing loop databases. It would store loop makeups or loop responses, data on deployed DSLs, binder information, measured noise, information on crosstalk between lines, and so on. From this point of view, techniques for loop make-up identification as well as crosstalk identification become important enablers for DSM.
The authors could spend a couple of paragraph on these aspects of DSM.
Author's response) We agree this is a good addition and we like your text as a good start with the additions you request also in red below.
A major component of advanced DSL management is a database of loop information for DSL provisioning and maintenance. This database is envisioned as having a variety of information on loops, noise, and the histories of deployed DSLs extending far beyond existing loop databases. It would store loop makeups or loop responses, data on deployed DSLs, binder information, measured noise, information on crosstalk between lines, and so on. From this point of view, techniques for loop make-up identification as well as crosstalk identification become important enablers for DSM. Information gathered from DSL equipment often is collected through an SNMP (Simple Network Management Protocol) line at the SMC and processed to identify line transfers, crosstalk levels and transfers between lines, as well as to calculate and characterize other line noises. Older systems may instead collect such information through element management systems (EMS) over a protocol known as TL1. Such interfaces are standardized in the American DSM Technical Report as well as the International Telecommunications Union G.997.1 standard, and are sometimes known as maintenance information bases (MIBs).
Such information can be used to characterize DSLs as being significantly affected by the presence of bridged taps (unused branches of the telephone line), bad connections, poor balance with respect to ground and types of noise (like AM radio, crosstalk, impulse). When significant, the DSL operator may elect to use this information to organize maintenance activities to remove the problems
External links
Reviewer's question) I suggest adding the following one managed by Ken Kerpez (Telcordia) http://net3.argreenhouse.com:8080/dsl-test/index.htm
Author's response) We will certainly add the links.
Reviewer A: Additional minor comments on first revision
I thank the authors for the accurate revision of the paper, which very well addresses my comments. I now offer only very minor suggestions.
Author's response) The authors thank again the anonymous reviewer for comments and suggestions.
Section "Discrete Multitone"
Reviewer's question) The quotation mark before "When the number of subchannels..." should be deleted.
Author's response) We removed the quotation mark.
Reviewer's question) The sentence below
"In implementation, the word ‘narrow’ is relative to a time-domain guard period placed between transmission of information on each successive use of the tones"
could probably be re-phrased as:
"In implementation, the word ‘narrow’ **is intended in comparison to the reciprocal of the duration of the guard interval** placed between transmission of **consecutive DMT symbols**"
Author's response) We adjusted the sentence as follows: "In implementation, the word ‘narrow’ is intended in comparison to the reciprocal of the time-domain guard interval placed between transmission of consecutive DMT symbols"
Section "Dynamic Spectrum Management"
Reviewer's question) when talking about loop make-up and crosstak ID, the authors may mention and reference the efforts in G.selt. I do not know what the status of these activieties are but, if these efforts are still alive, they should probably be mentioned.
Author's response) We have included SELT as recommended.
From this point of view, techniques for loop make-up identification as well as crosstalk identification become important enablers for DSM One good example of these techniques is SELT (Single-ended loop testing), which enables loop make-up and crosstalk identification without installing any device at the customers’ side, and is standardized as G.selt by ITU.
Section "Applications"
Reviewer's question) The word "plethora" is used, but in English this word is used with a negative connotation. I suggest substituing with multitude, abundance, variety, etc.
Author's response) We updates the sentence as recommended.
In addition, future vectored-DSL systems as well as CuPON systems would enable a variety of new applications to be supported, for example, next-generation home networks.
Section "References"
Reviewer's question) I suggest that the authors consider adding/selecting a few papers on loop make up and xtalk ID. Below, a list of relevant (and I also believe, exhaustive) journal papers in chronological order.
Author's response) We will certainly add these papers.
Boets, P.; Bostoen, T.; Van Biesen, L.; Pollet, T.; "Preprocessing of Signals for Single-Ended Subscriber Line Testing," IEEE Transactions on Instrumentation and Measurement, Volume 55, Issue 5, Oct. 2006 Page(s):1509 - 1518
Shi, Y.; Ding, F.; Chen, T., "Multirate Crosstalk Identification in xDSL Systems," IEEE Transactions on Communications, Volume 54, Issue 10, Oct. 2006 Page(s):1878 - 1886
S. Galli and K. J. Kerpez, “Single-ended loop make-up identification — Part I: A method of analyzing TDR measurements,” IEEE Transactions on Instrumentation and Measurement, vol. 55, no. 2, pp. 528–537, Apr. 2006
Papandreou, N.; Antonakopoulos, T., "Far-end crosstalk identification method based on channel training sequences," IEEE Transactions on Instrumentation and Measurement, Volume 54, Issue 6, Dec. 2005 Page(s):2204 - 2212
T. Bostoen, P. Boets, M. Zekri, L. Van Biesen, T. Pollet, D. Rabijns, “Estimation of the transfer function of a subscriber loop by means of a one-port scattering parameter measurement at the central office,” IEEE J. Sel. Areas Commun., vol. 20, no. 5, pp. 936–948, Jun. 2002
Salvekar, A.A.; Louveaux, J.; Aldana, C.; Fang, J.L.; de Carvalho, E.; Cioffi, J.M., "Profile detection in multiuser digital subscriber line systems," IEEE Journal on Selected Areas in Communications, Volume 20, Issue 5, June 2002 Page(s):1116 - 1125
S. Galli and D. L. Waring, “Loop makeup identification via single ended testing: Beyond mere loop qualification,” IEEE J. Sel. Areas Commun., vol. 20, no. 5, pp. 923–935, Jun. 2002.
S. Galli, C. Valenti, K. Kerpez, "A Frequency-Domain Approach to Crosstalk Identification in DSL Systems," IEEE Journal on Selected Areas in Communications, vol.19, no.8, August 2001.
C. Zeng, C. Aldana, A. Salvekar, and J. M. Cioffi, “Crosstalk Identification in xDSL systems,” IEEE J. Sel. Areas Commun., vol. 19, no. 8, pp. 1488–1496, Aug. 2001.
Reviewer B
Nice small overview article about DSL! It probably fits the audience of Scholarpedia quite well.
Author's response) The authors thank the anonymous reviewer for comments and suggestions. Our response is below.
A couple of suggestions for improvements to the authors:
Reviewer's question) In the Overview, ISDN is stated to have little success with low rate and strong crosstalk as the major reason. This does not give a clear and fair picture, since ISDN was quite speedy at the time introduced long time before ADSL. The major reason it was not so successful seem rather to be that operators did not invest in it broadly and that the price for the service was often set way to high. Germany and India are two markets which were quite successful for ISDN, and before ADSL services were available. Also, to remember ISDN was introduced at a time before the Internet had a deep impact on society.
Author's response) This was addressed by simply saying “a reason” instead of “most important reason.” This article does not attempt to critique or make excuses for ISDN’s failure, but rather to credit it for origins of today’s DSL. The reason the sentence appears is that DSL is faster than ISDN, not to critique ISDN – hopefully this minor change helps with this comment.
Reviewer's response) OK!
Reviewer's question) The last sentence under the "Extended Bandwidth over Analog Telephones": "Typically, passive microfilter devices can be placed in series with any analog telephone to prevent higher-frequency "ring-trip" voice-band signals from disrupting the DSL signals." Comment: Equally important is to protect the telephone voice-band signals from the interference coming from the DSL signals.
Author's response) We added a sentence to say that the microfilter also protects voice signals.
Reviewer's response) Good!
Reviewer's question) Comments to the explanation of DMT: DMT should also be referred to as "baseband OFDM with bitloading". DMT evolved from OFDM due to the characteristics of the telephone twisted pair cables - baseband transmission without fading. See further comment below.
Author's response) Sorry, this is not the evolution path of DMT – it actually evolved from attempts in voice band modem area to use OFDM (wireless came after DSL – in fact the first successful wireless broadcast leading to OFDM use there was done with a DMT modem with a fixed bit distribution and a wireless front-end replacing the line driver circuits in the early 1990’s. It was only at this point that the coding aspect (the DMT modem that was used in the US Government/EIA tests for early OFDM broadcast actually used a Reed Solomon code with 60% parity) became fully appreciated as an alternative to trying to track a time-varying frequency-selective channel with an equalizer for a single-carrier system. DAB, DVB, and Hyperlan/Wifi followed those test results. Rather than claim credit for WiFi here, which DMT could do, we thought a more humble stance is the wording currently present.).
Reviewer's response) I stand corrected! You if someone should know the history of DMT. I like more the new explanation. Baseband and associated double FFT-size explanation etc is fine! Nevertheless, I find it a bit strong and slightly misleading to say the other (or first) OFDM-DMT difference is "adaptive bit-loading". Just "bit-loading" would be more descriptive and fair in my opinion.
Why? My reasoning: Since the essential bit-loading is performed before show time any small adaption thereafter is essentially due to a (hopefully) small mismatch from off-line training information to the actual occurring noise environment (and usually a reason of concern for a potentially troublesome line). As you correctly point out, the rate typically remains constant, which severely limits the adaptiveness since its not constructed to adapt to wildly changing noise environments or instantaneous rate needs by the end users. What I would like to make clear is that DSL is constructed from ground up essentially for a very time-invariant channel and noise environment to yield constant BER and rate for every user. The constant rate, sometimes good, can also be a severe disadvantage. Energy is largely wasted since users in practice very seldom utilize their connection to their "max" rate (in either or both directions and simultaneously, with or without DSM). Its like always leaving the lights on... ADSL2 low power modes (L2 & L3) is a "quick fix" attempt trying to address this problem coarsely for the network operators (they are getting more and more concerned about their electricity bills). However, the switching between low power modes and normal full L0 mode can be quite hazardous for the stability of a network (therefore not used by most operators today, despite their increasing electricity costs...). This is largely due to the fact that all current DSLs are constructed ground up for a constant rate and BER (with or without DSM), and to yield essentially constant crosstalk levels during show time. I assume that at the time when DSL was first constructed, energy consumption and associated costs was of a very low concern, since power was easily available (no battaries needed) and electricity quite cheap... but unfortunately, these costs and our environment has changed since then. Still missing in existing crosstalk limited DSLs (and in current DSM's, up to 'Level 2'): Wouldn't it be nice with an elegant time-adaptive solution (truly on the fly, "one-sided" if possible, time-adaptive bit-loading or time adaptive coding and modulation) which rapidly vary rate & power consumption (to follow the instantaneous traffic need of the users - to "drive" quicker and slower and with a troubleless "quick acceleration") ground-up constructed for the "new" scenario - time varying crosstalk levels?
Bottom line. (Do you see what I mean and do you share this opinion?) During show time - current DSLs, even with DSM, are all constructed ground up to be essentially static systems, and not from the beginning to be very time adaptive.
Author's response) We understand. We changed "adaptive" to "variable" and also added a sentence referring forward to DSM as a way to introduce more smoothly rapid response to channel noise variation.
Reviewer's question) The DSM Levels 1 and 2, as detailed in Figure 6, are a bit unclear defined. Typically DSM Levels 1 and Level 2 can be more clearly distinguished by:
Level 1 - Single line spectral optimization. Uncoordinated, user by user, optimization of users' spectra performed locally for each line.
Level 2 - Multiple line spectral optimization. Coordinated multiuser optimization of users' spectra by a central cable bundle entity (a SMC).
Author's response) We have edited the figure to add single-line spectrum control to Level 1. We’ve not yet made a change to Level 2 because we want to avoid a 2nd confusion that might be introduced instead by your correction. We believe that adding the word “coordinated” or “uncoordinated” or giving anywhere the impression that central control of all lines is necessary, which is of course an incorrect characterization of DSM. (For instance, DSM easily works at all levels in unbundled environments, but often this is confused in the minds of those who do not understand it and criticize it as requiring a “central controller,” which is not true – each service provider can have their own controller for only their lines.) Thus, we tried to avoid promoting that misconception in this Scholarpedia article on DSL, where we try to keep mention of DSM brief. If you have wording that avoids that issue, and is better than the existing wording, we’d be happy to see it.
Reviewer's response) I accept this and share your concern. But according to your current DSM Level definitions its a bit unclear which DSM Levels are based on the assistance from a SMC and for which Levels are not. Spectral (PSD) assistance or full signal (phase-amplitude, full vectoring) assistance ?
For me the table is fine except that it would help to specify which Levels are based on (one or several) SMCs and if it is about spectral coordination or "full" signal coordination. As I prefer (but you might have a different opinion), extending the table (in Figure 6) with something like:
Level 0 - no DSM, (no SMC).
Level 1 - Single line DSM, no SMC.
Level 2 - One or several SMC(s). Multi-user spectral coordination - PSDs
Level 3 - SMC (one only, in DSLAM!). Multi-user signal coordination - phases & amplitudes.
Then also adding a vertical downward arrow (similar to the existing, or better use the same) with "Multi-user coordination" to also show the increased coordination required.
Author's response) We have edited Figure 6 to add "Multi-user Coordination" in the arrow. Regarding the SMC's use, we have changed the second sentence of the second paragraph in the DSM section to "Based on the available data about the lines and the controls at DSM levels 1, 2, or 3, the SMC optimizes ..." The SMC is required at all levels other than level 0. Although SMC has the expansion of Spectrum Management Center, its role includes any type of optimization of the DSL parameters (of individual lines or jointly for multiple lines). This role of the SMC is defined in the American DSM Technical Report, which is cited in the article. The use of one or multiple SMCs is dependent on whether the loop deployment is unbundled or not. In Level-3 DSM, the SMC need not be placed in the DSLAM. It simply communicates the line-grouping, tone-selection, optimized spectra, etc. to the xTU-C, which then performs the signal-level coordinated transmission or reception.
Reviewer's question) Note: both Level 1 and Level 2 DSM aim to be polite - using this strategy it is understood that politeness is favorable for the all wires in the cable, rather than just for a few lines. UPBO is a good example of DSM Level 1 - uncoordinated spectral politeness. DSM Level 2 has better possibility to achieve higher politeness since all crosstalk information between the lines is gathered at one place for processing, the SMC.
Author's response) We added the sentence “Simple variation of the upstream transmit spectrum according to measured line attenuation (or thus effectively length) could be viewed as a form of level 1 DSM and is sometimes also known as upstream power back-off or UPBO.”
Reviewer's response) I like it!
Reviewer's question) I strongly suggest to further explain why some duplex method, in particular FDD is used, i.e. to (efficiently and cost effectively) avoid the strong NEXT and Near Echo present in DSL. Without the NEXT (true) full duplexing would make most sense. Furthermore, I propose to shortly explain the fundamental difference between the 'analog FDD' with DMT used for ADSL, and the digital FDD, or Zipper DMT, used for VDSL (1&2). Particularly the flexibility and efficiency advantages with the latter. Please refer to the publications Zipper: a duplex method for VDSL based on DMT (IEEE-TCOM, aug 1999) and Asynchronous Zipper (IEEE- ICC, 1999).
Author's response) Very good. We have added the new paragraph in the DMT section and the reference “Zipper: a duplex method for VDSL based on DMT (IEEE-TCOM, aug 1999) and Asynchronous Zipper (IEEE- ICC, 1999)” at the end.
The duplexing method used for ADSL and VDSL DMT is frequency-division as shown in Figure 2 above. This type of duplexing avoids what is called "near-end crosstalk," which is interference between DSL signals traversing opposite directions on the loop. Such crosstalk can be a large from a transmitter in one direction placed closed to a receiver for the opposite direction. This duplexing is achieved with analog filtering in ADSL. However, VDSL progressed to a digital-signal processing method known as digital duplexing or "zippering," introduced by (Isaksson, 1999).
Reviewer's response) Thanks!
Reviewer's question) I suggest to further mention the main differences between widely known OFDM used for radio communication and DMT used for twisted-pair copper. That is 1) the DSL baseband communication effectively requires a real valued baseband signal, which leads to twice as large FFT, 2N, as the number of 'uniquely used' subcarriers, N, since the baseband spectrum must be complex symmetric (thereby forcing complex conjuated carriers representing the 'negative frequencies' x_{2N-k} = x_k^*, which isn't necessary for radio OFDM). 2) Bitloading is possible due to the almost stationary DSL channels.
Author's response) Good. We have added the paragraph in the DMT section.
For DSL, DMT differs from the OFDM used in wireless systems in two fundamental ways. The first is the adaptive bit-loading on each carrier that is used for DMT. Second is the DMT system is baseband in that it uses an FFT of size two times the number of carriers to ensure the actual transmitted and received signal is one-dimensional and real. (Wireless systems instead use a size N complex-to-complex FFT that produces a quadrature or two-dimensional signal.)
Reviewer's response) Fine! (With the remarks above)
Reviewer's question) Figure 8: I suggest to provide a higher resolution version of the figure. Its difficult to see the details in the current low-res version.
Author's response) We provided a link for a larger version below the figure.
Reviewer's response) Excellent!
Reviewer's question) Please mention that bit-swapping is intended to handle typcally small changes in the noise spectrum, but not designed to handle larger noise changes after the bitloading, e.g. due to RFI or impulse noise (where bitswapping effectively fails, not seldom requiring a re-synch and troubled customers, at least with 'triple-play'...). As far I know in the DSL standards, bit-swapping does not change the bit-rate, which remains constant. So with overall increasing noise occurring after the bitloading, bitswapping won't help.
Author's response) Thank you – we’ve added the sentences “Typically bit-swapping retains the same data rate but redistributes energy so that best (and lowest necessary power) bit distribution is maintained. Bit swapping and gain swapping allow up to at least a 14 dB range of noise variation (a factor of 60 and large enough to handle many noises). However, some noises may exceed this range and then re-initialization is necessary.”
Reviewer's response) Bravo!
Reviewer's question) Referring to the "narrow subchannels", GI etc and previous reviewer comments (I fully agree with his GI length statment); I further suggest to mention that a time equalizer (TEQ) placed before of the GI removal at the receiver is often used to "shorten" the overall impulse response length (memory) to within the GI length, (a beneficial technique used for ADSL and often necessary especially for long loops which have longer channel memory.)
Author's response) Ok, we’ve added the paragraph in the DMT section.
To ensure the independence of the tones in DMT, a guard interval is inserted between successive time-domain packets that contain roughly 8% of the packets time-domain samples repeated at the beginning of the block of samples. This is sometimes also known as a cyclic prefix. The cyclic prefix ensures a simple implementation of one transmit IFFT and one receive FFT. If the guard interval is not sufficiently long for the channel, then some equalization may be used to confine the span of the channel's impulse response to within a time period equal to the length of this guard interval. This technique has many equivalent implementations and sometimes called a "TEQ" or time-domain equalizer.
Reviewer's response) Looks good!
Reviewer's question) In future of generations of DSL, extending the bandwidth to far beyond 30 MHz on very short lines may be an economically viable solution in many countries to approach the Gigabit rates. That is, extending with fiber to the basement and similar near the customer drop-points.
Author's response) We agree, but did you want us to say this or are you just agreeing with what is there already says this?
Reviewer's response) Yes, please mention the extended bandwidth for even shorter wires if you also believe its a future direction (I don't find you explicitly pointing that out).
Author's response - ok we added the following sentences in the vectoring section.
Vectored DSLs may also offer the possibility of using more than 30 MHz of twisted-pair bandwidth. Some pairs (those involving alumninum of some type or other non-copper metals) may have linear bandwidth to only roughly 30 Mhz, but others may have a wider usable bandwidth such as evident for instance in category 5 wiring used in ethernet to 100's of MHz. Some future evolutions may thus exploit even wider bandwidths on the copper twisted pair than 30 MHz.
Finally, I would like to thank the authors for a good work and constructive revision of this article.
Reviewer's response) Last comments...
Since this is an overview DSL article I suggest you extend the reference list with a couple of important DSM papers (ranging from Level 1 to Level 3), as a guide to further reading. You don't need to refer in the text to each paper individually, just as a group and by listing the following papers in your reference list:
Wei Yu et. al. "Distributed Multiuser Power Control for Digital Subscriber Lines", IEEE Journal on Selected Areas in Communications, Special Issue on Twisted Pair Transmission, vol. 20, no.5, pp.1105-1115. June 2002. "Iterative Water-filling for Gaussian Vector Multiple Access Channels", IEEE T-IT, Jan. 2004. "Dual Methods for Nonconvex Spectrum Optimization of Multicarrier Systems", IEEE Transactions on Communications, July 2006.
Raphael Cendrillon et. al. "A Near-Optimal Linear Crosstalk Canceler for Upstream VDSL", IEEE Trans On. Signal Proc. Aug 2006. "A Near-Optimal Linear Crosstalk Precoder for Downstream VDSL", IEEE Trans on Commun., May 2007. "Autonomous Spectrum Balancing for Digital Subscriber Lines," in IEEE Transactions on Signal Processing, August 2007.
Driton Statovci, Tomas Nordström, and Rickard Nilsson, “The normalized-rate iterative algorithm: A practical dynamic spectrum management method for DSL,” EURASIP J. Appl. Signal Process., vol. 2006, pp. 1–17, 2006, Paper 95175.
Papandriopoulos, J.; Evans, J.S., "Low-Complexity Distributed Algorithms for Spectrum Balancing in Multi-User DSL Networks", ICC 2006.
Author's response) We have included the above references in the article except for the paper on IWF for Multiple Access Channels. There are many papers available on resource allocation in multiple-access and broadcast channels, which will make the references list extremely long. We can include all those if you would like.
In my previous review round, I had some further comments on the definition and explanation for the different DSM Levels (see text above; to clarify I referred to the "table" in Figure 6). However, you neither addressed it in the paper nor provided any "authors reply" in your last revision. Please address this issue.
Author's response) We have addressed that comment in the current revision.
