Status and Issues in xDSL

The original DSL was of course ISDN, developed in the early 80s; it squeezes 160 Kbps into an 80 khz bandwidth of local loop. ISDN uses a simple four level PAM modulation (2B1Q), to reach a standard range of 18,000 feet, although more recent products use smarter signal processing to achieve longer range. ISDN has not been a great success in the U.S., although it has done very well in other countries (such as Germany).

The follow-on was HDSL (High bit rate DSL), which used the same 2B1Q modulation but on a larger bandwidth and with a lot more sophisticated DSP, to deliver much faster rates, over CSA range (12,000 ft. of 24 AWG). HDSL most commonly operates with two pairs to deliver symmetric T1 or E1 rates (1.544 Mbps or 2.048 Mbps), which it does by sending half the data on one pair, and half on the other, both operating as full-duplex echo-cancelled links (with either 768 Kbps for T1 and either 1168 or 1024 Kbps for E1).

HDSL is a lot simpler and more robust than the old T1 service, which required repeaters every few hundred yards and was consequently difficult to install or maintain. As a result, it has essentially replaced T1 and the odds are good that if you have recently got a "T1 line", it was actually HDSL. (To be strictly accurate, you should refer to DS1—Digital Subscriber 1—for rates or services, while "T1" refers to the older physical implementation using AMI—Alternate Mark Inversion—line code on two simplex connections). As well as high speed data connections, HDSL is popular for connecting wireless basestations into the PSTN (multiple connections, a lot cheaper than fibre) and for 'pairgain' applications—squeezing many voice channels onto one piece of copper.

In Europe, the standards body (ETSI) defined in their HDSL technical report what was essentially a familly of 'HDSLs', all delivering E1 rates using the same 2B1Q line code, and operating on four, three (old but still used), two (nowadays the usual), or one pair (there are also definitions for one and two pair CAP systems). This has the advantage of being quite straightforward and simple, but each implementation would have a different reach, with single pair falling perhaps 10-20% short of CSA range. In Europe where loops tend to be much shorter than in the U.S. (especially in Italy or Germany), this is acceptable.

ANSI is leading the discussions on next generation HDSL - HDSL2 (although ETSI has opened discussion on an enhanced HDSL, which perhaps will support some rate adaptive approach, they are most probably probably going to wait and look at ANSI's proposal before making too many decisions). This has the target of being a technology that will last, delivering T1 rates over a full CSA range with the reliability that has been proven on HDSL (it is possible that this should support rate adaptive services too, allowing lower-speed access even on longer loops).

The difficulty is in meeting the range and strict latency requirements while still maintaining spectral compatibility and coping with real-world noise and interference. Latency of less than 500 ms is mandatory for many existing services—this is a tough requirement. CSA range is necessary for operational reasons—the telco's customer databases often only classify distances as "within CSA? yes/no"; if a customer requersts a service, that is the only test if they can get it or not.

The expectation is that agreement will be reached over the next few meetings, with Adtran, Pairgain, Level 1, Globespan, and others close to a consensus. Due to latency, a multi-carrier system is unlikely, and the discussion on line code seems to have converged in favour of a single carrier technology (such as coded 64CAP or QAM). The uncertainty is more over error-correction and coding techniques, but all of the proposals exhibit impressive performance, with demonstrated achieved coding gains of some 5 to 6 dB (at least one vendor is using Turbo-codes—probably the sexiest idea in information theory today).

While the above are all standards (or quasi-standards), several companies have lower-speed/lower price variants that are being proposed. Many simply consist of using half an HDSL chipset on a single line (for instance, to give 768 Kbps), whilst others are new developments, targetting slower applications (such as 512 or 384 Kbps). These are discussed for Internet access or low cost pairgain—especially in less developed countries—and examples include MDSL (Moderate speed), PCM-n, or Brooktree's DSL/384.

Importantly, all of these technologies are loop-powered (in other words, they are powered from the central office), and all use the POTS band for data (while you can have voice, it is needs to be incorporated digitally, and is not the not the transparent/backwards compatible approach of ADSL).

Although some people use SDSL (Symmetric DSL) to describe single pair HDSLit better describes the (somewhat oxymoronic) Symmetric ADSL (for instance, by changing the up/down allocation of normal ADSL chips to give the same rate in each direction). Significantly, this interpretation means it operates over POTS and need not be loop-powered (both unlike HDSL). Depending on crosstalk (a VERY big "it depends"! crosstalk more than any other factor limits how well a service will operate, and whether things interfere with one another), this could deliver 1+ Mbps in each direction, and operate over the full CSA distance; however, it has not yet been standardized or widely available, and there are important issues in spectral compatibility that would need to be resolved. The standards bodies have not decided on this, although at least one supplier is marketing such a product.

ADSL has been standardized by both ANSI and ETSI—the T1.413 was published in late 1995 and is a very comprehensive specification, describing all manner of issues with ADSL from physical modulation, to coding, framing, and management type operations. Discussions are now very well advanced on Issue 2 of this specification, and all the substantive items are likely to be frozen in the next few weeks, leaving editing and organizing before the new edition of the standard is published next year. Most of the changes have been in updating the standard with the benefit of experience of the last two years experience, and updating it to reflect changes in the market—for example, incorporating more support for data mode services and Internet, rather than the video-focus of the first version (despite some comments, the standard is extremely well suited to data services without major change, but the document will be updated to include more details).

Another change is to describe rate adaptive ADSL (in a marketing coup, one manufacturer appropriated the acronym 'RADSL' for this; a bit of a surprise for the rest of the industry, which had always had a rate adaptive ADSL!). The omission is not in the technology, as T1.413 has a good description of how DMT adjusts rate in 32 Kbps steps, but rather in standardizing the training and management protocols to ensure interoperability between different rate adaptive modems.

The other change in the standard will be to follow the lead of the ATM Forum and to separate the standard into parts dealing with PMD (Physical media dependent) and TC (Transmission convergence) sub-layers. This may only be an editorial change in Issue 2, but it paves the way for more detailed specifications of ATM over ADSL or packet mode operations in future texts.

The big debate within ANSI is over the introduction of a second line code: whether the standards body should stay with DMT only, or should support two flavors of ADSL and document a version of the standard for CAP. The situation at the time of writing is that the main group of T1E1.4 decided to stick with the status quo, but that a parallel ad hoc group has been established and is working on standardizing CAP.

Some people have discussed 'reverse ADSL'—simply swapping two modems so that the high capacity direction is from the home to the CO. Unfortunately, in most cases this is not going to work. ADSL relies on all the 'loud' signals being located together (in other words, downstream sends are all at the CO), and all the weak received ones being located in a different frequency area, and physically separated. If you reverse this, then at the CO the loud 'send' of everyone else's downstream will be right where your reversed system is trying to listen to a very weak will be right where you are trying to receive the attenuated noisy weak high capacity 'upstream'—drowning it out. Conversely, your transmit signal will swamp everyone else. Given spectral compatibility constraints and 'good citizenship', this will limit reverse ADSL to perhaps 1000 ft. Of course, up and down are arbitrary—what matters is everyone has to operate in the same direction. It is a little like driving; in the U.S. people drive on the right; in the U.K. they drive on the left—either is fine, so long as you are consistent!

Finally, there is VDSL (Very high speed DSL). Whilst there is still much debate over the specifics, the gist is clear enough. VDSL is intended as the last drop, operating over copper, in applications such as Fibre-To-The-Curb (FTTC) or Fibre-To-The-Building (FTTB), where the head-end will be located in an ONU (Optical Network Unit) at the end of a length of fiber. It operates with very high data rates, but over short distances: 51 Mbps over 1000 ft. or 25 Mbps over 3000 ft. are typical. Intriguingly, even though the data rate is higher, it is likely that the shorter distance and more controlled environment may make this easier (and potentially cheaper) to implement than ADSL.

Various bodies are discussing this, including ANSI, ETSI, and DAVIC, and virtually every combination of line code, specification, and access method has been suggested. For example, while most suggest that an asymmetric system with perhaps 10:1 ratio is adequate, some prefer a fully symmetric system, and others are flexible. Within ANSI, there are a few main proposals.

Amati is championing "Synchronous DMT"—a ping-pong modulation method that uses Time Division Duplex, where the transmitter and receiver alternate roles. The attractions are that various asymmetries can easily be supported, simply by varying the duty cycle of send/receive. Secondly, the complexity is lower: little effort is needed to separate signals (in other words, no filters required), and hardware can be shared at each end—swapped between transmitting and receiving as required. On the other hand, DMT is perhaps more power-hungry than is desirable in the power-limitted environment of an ONU. Secondly, the ping-pong system must be very tightly controlled between all systems (if there is any difference or jitter, a transmitter will swamp a receiver); this might be difficult in a deregulated environment with competitive access. Finally, there is some concern that the proposed ping-pong frame rate (2 KHz) might be demodulated into an annoying audio-frequency signal (perhaps in an adjecent pair, which might not have a VDSL-splitter or filter)—this has been claimed to occur with TDMA digital cellphones, where the burst frequency is detected in hearing aids.

There is a large amount of support from a number of manufacturers are supporting a frequency division duplex system, with CAP for the downstream. This has the advantage that it is a low power solution, and can be optimized to give a simple transmitter. Being a high capacity broadband signal it is also resistant to RFI and can be placed in a region where there is less concern with impulse noise. Manufacturers including Analog Devices, Aware, Orckit, BBT, Globespan, Broadcom, and others all are developing a common draft standard based on this technology. (importantly, the first three have all developed DMT solutions for ADSL, showing that line code choices can be decided by technical criteria and are not unchanging).

However, the consensus breaks down for the upstream. Some manufactuers are proposing single carrier techniques (CAP or QPSK) which have the attractions of low power and simplicity. However, in the home (where the upstream signal starts), low power is not as critical as it is in the ONU, so this is of less use. Then there is the problem of noise, and how to cope with it. The low frequency bands can be very noisy, with a lot of impulse noise and ingress from sources like electric light dimmers, vacuum cleaners and the like—this can cause a lot of grief for a system, which is very hard to cope with. To cope with this, some systems (such as DAVIC 1.0's FTTC) place the upstream signal high in frequency—above the downstream. Here the system avoids the low-frequency wideband noise, and the filtering can be easier (reducing the waste from a large guard band). Unfortunately, this region has lots of attenuation in the copper, reducing the useful capacity of the system; the DAVIC system only has 1.6 Mbps upstream and it is doubtful if the same arguments would work for higher rates of say 3 Mbps. In addition, the (narrow) high frequency upstream is now very vulnerable to notches in the channel caused by bridge taps.

Recognizing the very different characteristics of downstream (power constraints, high speed / high bandwidth) and upstream (lots of impulse noise, concern on frequency plan), Analog Devices, Aware, and BBT have proposed a hybrid solution, that uses very different techniques for each direction—hopefully getting the best of both worlds. By adopting a CAP single carrier for the DS, the benefits listed above are achieved. The upstream is located at low frequencies, and uses a new muti-carrier technique—DWMT (Discrete Wavelet MultiTone). This copes extremely well with impulse noise, so the frequency plan can use the 'good' low frequency copper without worrying about noise. In addition, the technique easily lends itself to support multi-drop in-home wiring (multi-point to point / passive NT architectures).