View source for Talk:Digital subscriber line (DSL)

== 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.