DMT modems operate differently on each line by partitioning the spectrum into narrow sub-carriers. Each sub-carrier, or frequency bin, is modulated separately with QAM, and can contain from 1 to 15 data bits. The number of bits in each frequency bin, along with its power level, are determined adaptively on a continuous basis by the DMT modem to correspond to loop attenuation and the noise environment.

In contrast, SCM modems produce a fixed spectrum using a single carrier, which is modulated with QAM. The shape of this spectrum is manipulated by changing the constellation size, carrier frequency, and symbol rate, and by performing additional filtering after modulation. Due to limited granularity in parameter control, the SCM schemes offer less flexibility than a DMT modem when allocating data rate and providing power spectral density shaping.

2. Better Resilience to RF Interference
RF interference from external radio bands is often noisier than the VDSL signal itself. It is thus desirable to filter out these spectral areas before trying to receive the VDSL signal. This filtering operation needs to move dynamically, and may need to remove multiple frequency regions simultaneously.

DMT uses a windowed FFT to constrain the interference to as narrow a region as possible, and then separate and receive the other sub-channels. This FFT is effectively a large programmable filter that can selectively reject frequencies that contain interference. A DMT VDSL solution also uses cancellation techniques to remove more interference tones in the frequency domain. In contrast, SCM modems rely on its adaptive equalizer to provide a notch at the interfering frequencies.

3. Easily Programmed to Meet VDSL FDD Requirements
Duplexing is a necessary function for sending and receiving signals simultaneously on a single twisted pair. Duplexing can be achieved using time division duplexing (TDD), FDD, or full duplex through echo cancellation.

Standards bodies have chosen FDD as the duplexing method for VDSL, allowing two separate frequency bands for downstream traffic, and two for upstream traffic. There are currently three widely accepted band plans--Plan998, Plan997, and the FX plan--and more are likely to emerge.

To achieve FDD, receive signals must be filtered from reflecting signals emanating from the transmitter. This filtering can be done with analog filters in the line interface, or with digital filters after the signal has passed through the analog-to-digital converter (ADC).

Digital duplexing provides the flexibility required to design a widely applicable, future-proof modem with fewer analog components and complex circuit variants. Digital duplexing reduces the need for complex analog interface circuits, and eliminates the need to use different front-end analog filters for every FDD band plan. Digital duplexing does, however, require more performance from the ADC because the echo reflected from the transmitter may be much larger than the receive signal.

DMT modems are naturally suited to digital duplexing because the central element of these modems--the FFT--handles duplex filtering very well. QAM modems can in theory, perform digital duplexing, but to date have not, because this capability comes at the cost of additional DSP circuitry. Digital duplexing is more easily incorporated into a DMT modem than an SCM modem

Additionally, new frequency plans will likely be oriented to symmetrical operation for private networks, or for country-specific plans. DMT modems can be reprogrammed to meet new band plans; SCM modems currently cannot.

Provisioning a DMT modem to fit a new band plan is easy; the upstream and downstream bands are simply programmed into the modem. SCM requires more filtering to split the bands, and these analog band split filters cannot be programmed to different frequencies. Additionally, these filters require a larger transition band between frequencies than DMT modems, about 1 to 1.5 MHz versus only about 4.3 KHz (Figure 1). This is another very significant advantage of DMT modems.

4. Low-Power Operation
Contrary to some claims, DMT solutions are both cost and power efficient. DMT VDSL has already achieved power consumption of less than 1.5 W/port with a standards-compliant solution.

When SCM advocates claim DMT is less power efficient, they usually point first to the FFT. As discussed, the FFT is effectively a programmable filter that separates frequencies. SCM solutions also require filtering (usually in the analog domain), and need additional transmitters/receivers to support multiple frequency bands. Deep sub-micron digital IC technology provides low-power digital circuits for today's FFTs, and future technologies are expected to deliver large reductions in power consumption for multi-channel DMT-based devices.

A detailed analysis of the components of the two types of modems shows that power consumption is very similar for each component. One influential study, by Burton Saltzberg¹, formerly a Bell Labs research manager, compared power consumption for DMT vs. QAM modems, and concluded that a QAM digital datapump actually requires more digital processing complexity than a DMT datapump.

The differences in power consumption are largely due to expected differences in the analog front end and the line driver. "For a given speed of operation, the cost increases rapidly with the number of bits of conversion," Saltzberg stated. In fact, the relative increment in cost (i.e. silicon area) and power consumption of an ADC to deliver the required higher performance is negligible.

Additionally, today's DMT VDSL modems use peak-to-average ratio (PAR) reduction techniques (more to follow), that eliminate the need for higher-cost, higher-power line drivers, receiver amps, and hybrid circuits.

Saltzberg perceived the cost of digital echo cancellation to be another main contributor to the supposedly higher cost of DMT. However, echo cancellation is not used for DMT VDSL.

5. Supporting More Services on the Copper Line
In many countries, it will be necessary to provide VDSL on top of other services--such as POTS, ISDN, and ADSL--that are being delivered over the same copper line. With DMT, these services can be avoided by adjusting power spectral density to simply turn off the carriers within these frequency regions. Single carrier solutions cannot do this as easily.

6. EMC Friendly
In the real world, suppression of RF egress into amateur radio and other bands is an important issue. Copper telephone wires are twisted to provide a balanced cancellation of opposite polarity signals (differential drive). This cancellation effect was intended to prevent crosstalk.

A VDSL transmission on a typical telephone wire can still produce RF egress at high frequencies because copper wire has progressively worse balance as frequency increases. High quality category 5 copper pair provides very good cancellation. However, in most copper access networks, and particularly in apartment buildings, the twisted pair may be category 3 or worse. These types of lines not only suffer from RF ingress but also cause RF egress that could upset amateur radio.

To suppress interference, a DMT VDSL transmitter simply turns off certain sub-channels in the spectral region used by amateur radio. SCM modems have limited capability in this area, but typically can provide at least one frequency notch by changing their transmit filter characteristic.

7. Improved Impulse Noise Handling
Impulse noise, a common problem in copper access networks, is generally caused by switching transients from ring generator relays in central offices or electric motors on the customer premise.

DMT helps diminish impulse noise. An impulse may wipe out the receive signal for 5 μs (microseconds) or more, but a DMT symbol spans 250 μs. An impulse will therefore destroy only some of the samples used for a FFT. The FFT will spread the time domain pulse over the frequency bins, and the noise margin will, to some extent, absorb the impulse energy.

Saltzberg also analyzed the sensitivity of DMT and QAM modems to impulse noise. Although he gave an advantage to DMT for low-level impulse noise, he said this advantage vanishes for higher-level impulse noise if a restriction is placed on interleaving depth due to latency (delay) requirements.

In fact, the T1E1.4 VDSL trial-use standard allows for large interleaving depth, and dual latency "fast" and "slow" channels. These fast and slow channels separate traffic (like voice) that is sensitive to delay but not as sensitive to bit error rate (BER), and data traffic, which is sensitive to BER but not to delay. The delay of the fast channel in DMT VDSL is within the 1-ms T1E1.4 requirement. Thus, DMT modulation, along with interleaving and forward error correction (FEC), provides a very robust method for dealing with impulse noise.

8. Scalability--An Unexpected Advantage
Until now, the question of what happens when the number of VDSL lines begins to scale has been largely absent from the DMT vs. CAP/QAM debate. As VDSL scales, many lines will be active at one time in a single binder cable. Simulations used to predict VDSL performance typically model the transmission of signals in the presence of crosstalk noise with the average power spectral density (PSD) of other services in the binder cable. However, in the real world, much of this noise is due to self far-end crosstalk (FEXT) due to modems of the same type.

When a group of DMT modems initialize themselves, a sort of competitive environment ensues where only a few modems, or perhaps only one, dominate the use of sub-carrier waves with low SNRs (signal-to-noise ratio). The other DMT modems quickly abandon these low SNR sub-carriers, because they cannot allocate even a single bit to their frequency bins. This effect most often occurs in the higher-frequency sub-carriers, which typically experience more crosstalk. As a result, groups of DMT modems tend to optimize the available bandwidth in the twisted copper pair, in turn maximizing throughput in large-scale deployments.

9. Meets the PAR Demands
PAR is the measure of peak-to-average signal voltage transmitted to the line. Different modulation techniques have different theoretical PAR values. PAR is an important parameter because the higher its value, the higher the voltage swing that must be accommodated by the analog circuits, which typically translates into higher device costs and more power consumption.

It's commonly said that DMT is inferior to CAP/QAM because DMT has a higher PAR. On the surface, this appears to be the case since DMT modulation typically offers a PAR of about 15 dB while CAP/QAM deliver a PAR of roughly 12 dB. In reality, however, DMT VDSL modems use built-in DSP algorithms to reduce PAR. These algorithms require very little power and cost, resulting in a typical reduction of PAR by 3 dB. This places the PAR for DMT and CAP/QAM VDSL at roughly equal ratios.

10. Similar Start-Up Time, Acquisition, and Handshake
QAM modems do not require a handshake to start, and QAM proponents claim this leads to faster start-up times. In fact, a DMT VDSL modem starts up within about the same time period as a DMT ADSL modem -- a time that is well within the expectations of DSL users. DMT VDSL modems also use standardized handshake signaling identical to that of ADSL.

Wrap Up
From the analysis above, it can be seen that DMT provides some nice benefits to designers developing VDSL architectures. Table 1 provides you with a recap of the advantages.

While DMT may appear to be a better choice, the ultimate winner in this battle will be determined by the standards bodies. Currently, the standards bodies are wrestling with this issue.

Reference
1. B. Saltzberg, "Comparison of Single-Carrier and Multitone Modulation for ADSL Applications". IEEE Communications Magazine, Nov. 1998.