Because VDSL is not a simple technological upgrade to ADSL, it is not really valid to argue that interoperability with the ADSL systems in place today should be the paramount driver of modulation technique decisions. As noted above, it is application requirements and the connectivity required on each side (fiber and on-site at the customer) that dictates the choice. Thus, in comparing the two approaches, system maturity, technical requirements, intended applications and environmental conditions should be given equal weight.

VDSL Operation
How VDSL serves as a fiber extension technology in fiber-to-the-curb or fiber-to-the-building applications is shown in Figure 1. The VDSL lines bridge the gap between very high-speed backbone infrastructures and customer premises equipment and/or integrated access devices. Furthermore, with the ability to provide Ethernet over VDSL, telecom operators can now connect a business LAN or an entire campus to a 10/40-Gbit Ethernet wide-area network on the way to achieving an all-Internet Protocol infrastructure.

In new deployments and new networks, which represent a large portion of the systems installed since QAM VDSL technology became available in 1998, equipment and service providers are free to choose the transport protocol. Ethernet is the frequent choice, providing connectivity economies of scale for end users, and creating a requirement for a simple, robust, plug-and-play VDSL solution. On a per-link basis, Ethernet-over-VDSL implementations have reached pricing parity with ATM-over-ADSL, making the technology very attractive for new deployments. Alternatively, when VDSL is deployed as an upgrade to existing DSLAM-based systems, equipment providers and carriers find the higher per-link costs for ATM transport acceptable. Thus, the choice of transport protocol is driven by application requirements, not technology.

The VDSL standard, as defined in ETSI's TS 10127002, ANSI's T1.424 and the ITU's G.993.1, defines two possible modulation technologies: single-carrier modulation, also referred to as QAM, and multicarrier modulation, also referred to as DMT. The QAM line code uses the four-band division structure illustrated in Figure 2. For service flexibility, the standardization bodies have defined band allocation plans. Every country can have different service needs, but within each region, spectral compatibility and band plan conformance must be kept in order to commit to the service offered.

DMT divides an entire channel, or band, into subchannels positioned 4.3125 kHz apart, where each subchannel carries only a fraction of the total payload. It uses techniques such as the fast Fourier transform (FFT) to modulate data on up to 4,096 subcarriers in a full-rate VDSL implementation, creating a large number of transmissions in which the subchannels have to be processed in parallel.

QAM is the approach used in all currently deployed VDSL solutions, with the number of ports deployed worldwide now approaching 2 million. Third- and fourth-generation QAM-VDSL chip sets that support the four-band standard are now on the market. The field experience gained in the four-year deployment history of QAM VDSL shows excellent performance and cost-effectiveness in terms of power supply and bandwidth use.

QAM/DMT Trade-offs
QAM and DMT differ not only with regard to the implementation of the four-band scheme, but also in how they deal with other critical performance aspects. The QAM approach focuses on time-domain processing of the VDSL signal, taking into account the serial and analog nature of the signal on the wire and optimizing digital and analog components of the system. The DMT frequency-domain processing of the signal requires a time-to-frequency (IFFT/FFT) and serial-to-parallel data stream conversion at each end of the line before any measures can be applied to counter the analog noise and attenuation inherent in the copper POTS wiring. QAM VDSL enjoys the benefits of technology maturity, low power consumption and plug-and-play features for powerful and simple link configuration. DMT was also the modulation technology of choice for ADSL solutions. For providers that view VDSL only as an upgrade to highly asymmetric data transmission services, or that have legacy issues with existing ADSL installations, applying DMT to very high-bit-rate systems may be a logical approach. As noted earlier, Ethernet-over-VDSL link costs also are at parity with the price of ATM-over-ADSL links, making them economically attractive for all types of system implementations.

Clearly, both QAM and DMT deliver pluses and minuses, with the choice being dictated by analysis of the real-world conditions and other practical matters applying to a specific application and installation environment. Trade-offs exist in six major areas affecting performance: burst noise, outside-noise interference, plug-and-play functionality, power consumption, bandwidth utilization and interoperability.

Noise Handling
Bursts of noise impact any xDSL technology. They erase the signal over a period of time, and how this is handled by the modulation scheme is of extreme importance. The combination of a data interleaver and Reed-Solomon error correction is used by both QAM and DMT. The interleaver used by QAM introduces less delay (about 10 milliseconds less) into the transmission path, and uses less memory than a DMT interleaver.

A major concern in a DMT-based system is that impulse noise with a duration of tens of microseconds can cause errors in all the used subchannels at the receiver and erase a complete FFT symbol. A DMT system must always program the impulse noise protection for whole FFT symbol lengths, not just for the length of impulse found in the network. The defined VDSL standards require impulse-noise protection of 500 microseconds. As illustrated in Figure 3, a 250-μs-long DMT symbol dictates impulse noise protection of three DMT symbols, or 750 μs, to meet the standards requirement. This introduces extra delay (about 10 ms) and extra interleaver memory. Thus, while even a short burst of noise may erase all of the information in one DMT frame, the effect of such a burst in a QAM modem will be smaller due to the short duration of QAM symbols, which usually last less than a microsecond.

Noise sensitivity and, even more important, noise resistance determine the performance of a chip solution or an access system. This is a common issue for all copper-wire xDSL technologies, and its impact increases as frequency rises. VDSL uses the high end of the spectrum (as high as 12 MHz), therefore making noise interference an issue. By its nature, a DMT signal has a higher crest factor than a QAM signal, requiring typically at least 13 bits for the analog-to-digital conversion (compared with 10 to 11 bits in a QAM signal). This causes analog-to-digital converters to clip the signal at the higher levels. This clipping produces errors at the receiving end, forcing the receiver to use the Reed-Solomon algorithm to try and correct these errors. As the "budget" of Reed-Solomon code is limited, the DMT signal consumes a substantial portion of this resource by using it to correct errors deliberately created at the transmitting side. This leaves a smaller margin of Reed-Solomon code to correct errors created by noise.

Comparatively, QAM does not induce errors through designed-in clipping, and has a stronger protection against potential external noises induced on the transmission link. QAM uses equalization as a strong protection against potential noises induced on the transmission link.

Modern signal-processing techniques and advances in silicon technology provide the ability to develop very robust equalizers that allow a QAM-VDSL modem to work in harsh environments, including bridged taps and high self-crosstalk levels from additional VDSL systems sharing the same cable binder. Five years ago, when DMT was accepted as the standard modulation for ADSL, proponents were able to show that the technology uses fewer Mips (millions of instructions per second) of processing power for similar equalization. Today, Mips budgets are not as restrictive, overcoming this now-outdated objection to QAM implementation.

QAM uses blind equalization, which means that there is no need for a training sequence or handshake. A QAM-based system therefore takes about 100 ms to start operating. In contrast, DMT has to send a training sequence across the line. The receiver adjusts itself to the received sequence and then sends the information to the transmitter for distribution among all the subchannels. So a DMT-based system may require several seconds to start operating. Although this is generally well within the expectations of DSL users surfing the Internet, it may not be satisfactory for consumers accustomed to "instant-on" television tuners.

QAM-based VDSL includes integrated communication channels that allow simple plug-and-play configuration of the link to any service required. QAM allows the user to simply choose a profile that matches the service required, and the link automatically configures. DMT VDSL must exchange parameters for all the 4,096 tones used and, in a high-crosstalk environment (VDSL frequencies), one DMT line could compete with its neighbor for usage of a certain tone.

Power Consumption
Power consumption is extremely critical for systems deployed on telephone poles, in street cabinets or in the wiring room of a multitenant/multidwelling unit. In these deployment scenarios, size and power consumption budgets are limited and strongly influence the feasibility of providing the service. DMT modulation requires a large analog/digital converter scheme (A/D and D/A), typically at least 1 to 2 bits more than QAM. (As noted above, even a 13-bit A/D D/A increases error risk due to clipping.) Each additional bit in the A/D can increase power consumption by a factor of 2.

DMT also requires more digital signal processing, memory and data buffer resources to cope with 4,096 subchannels, all of which consume power. The four-band QAM VDSL has a per-port power consumption of 1.5 watts. It achieves this level by handling signal conversions, without clipping, using only a 12-bit A/D D/A. Effective-bandwidth QAM uses decision feedback equalization, which averages the input signal-to-noise ratio (S/N) over the whole frequency band, while DMT averages the bits per subchannel. In cases where the minimal S/N level goes below the signal, DMT cannot use the signal in this subchannel, as illustrated in Figure 4. On the other hand, a QAM modem can easily use frequencies with levels of S/N far below the levels required for adequate performance of a DMT device.

Of course, a critical factor in realizing market potential is system interoperability. In this respect, QAM VDSL is well ahead of the game. Vendors have already achieved interoperability of their respective chip set solutions, including live demos of video-over-VDSL at this year's Supercomm exhibition in Atlanta.

DMT VDSL interoperability, on the other hand, is an unknown, since the technology is not yet available from multiple suppliers. It is true that because DMT is the standard line code for ADSL, an existing ATM-based ADSL modem at the customer-premises end of a line will be able to communicate with an ATM-based DMT-VDSL modem at the other end.

The deployment scenarios seen today, however, are ones in which Ethernet-over-VDSL links to fiber, not to ATM-equipped DSLAM devices. Equipment providers and service providers make decisions based on a range of technological and commercial factors. In this respect, QAM's proven functionality, market history, large and growing installed base, and demonstrated interoperability qualify it as the technology of choice for true broadband connectivity over copper wire.

Additional Articles
For more on this subject, dive into:

1. "10 Reasons to Choose DMT for VDSL Designs," at www.commsdesign.com/story/ OEG20020514S0009
2. "Broadband Access: G.SHDSL—Reaching the Access Network," at www.commsdesign.com/story/OEG20010109S1078