July 2, 1998

High costs and confusion over industrywide standards have stalled consumers' ability to gain access to the Internet via asymmetrical-digital-subscriber-line (ADSL) technology. As a result, cable modems have surged ahead of ADSL in providing home access to the information superhighway. Now, that situation may change: The Universal ADSL Working Group (UAWG) is rallying around the International Telecommunications Union's (ITU's) ADSL G.lite specification for a lower cost, longer reaching ADSL standard that employs no splitters. By solidifying a global standard, G.lite enables ADSL chip vendors to optimize their chips and software to produce consumer-grade modems; nevertheless, questions still remain about the widespread deployment of ADSL.

ADSL is not out of the Internet-access race, however. For example, Incumbent Local Exchange Carriers (ILECs), such as US West (www.uswest.com) and GTE (www.gte.com), and Competitive LECs (CLECs) are now deploying ADSL services for Internet access. In Europe, Video on Demand, the original driver for ADSL, remains a viable business (see sidebar "Europe checks out ADSL"). However, questions remain on whether a single worldwide standard for ADSL is possible. These questions revolve around the myriad regional telephone systems and consumer handsets that such a standard would have to accommodate. Even so, ADSL boosters are urgently rallying to G.lite because they see cable-TV operators pushing ahead in broadband access. They may be right: Analysts estimate that, by 2000, vendors will sell 3.7 million standards-compliant cable modems (Reference 1).

In contrast, analysts predicted in December 1996 that 5 million ADSL connections would exist by 2000, but this year analysts place that number closer to 1.4 million. This significant drop in forecasted connections, although not unusual for an overhyped technology, results from several problems. For example, the lack of standard end-to-end network protocols and the quandary over whether carrier-amplitude-phase (CAP) or discrete-multitone (DMT) line coding would prevail caused access-network and service providers to be hesitant about rolling out products. In addition, the much-publicized ADSL field trials proved that widespread deployment was more expensive than anticipated because of problems qualifying the twisted-pair cable plant for ADSL service and spectrum compatibility between signals in the cable plant.

On the bright side, an apparent end to the ADSL line-code battle is the broad acceptance of DMT as the standard. However, numerous trials have shown not only DMT's technical merits, but also its need to mature as a reliable standard line code. Also, in February, the G.lite working group agreed to use DMT line coding. However, the ITU will not complete the G.lite specification until late this year. Even then, it is uncertain whether G.lite will decrease the modem and deployment cost, and it will not provide end-to-end protocols.

End-to-end compatibility is essential for consumer applications to communicate seamlessly with Internet service providers, content providers, and corporate networks. The consensus is that asynchronous-transfer-mode (ATM) transport protocols can provide such network compatibility. Both the proposed ADSL Forum's Point-to-Point Protocol (PPP)-over-ATM-over-ADSL recommendation and the ANSI T1.413 Issue 2 with its ATM transconvergence-layer specification form the framework for end-to-end compatibility over an ATM network. ATM transconvergence is optional in the ANSI Issue 2 specification, but it is a good fit for systemwide interoperability. ATM's ability to scale bandwidth matches well with the rate-adaptive features of ADSL. Also, ATM supports guaranteed bit rates and latencies that provide the quality-of-service (QoS) and class-of-service features that ADSL network providers can price according to services supplied to consumers and businesses.

After establishing the ATM connection, PPP--a de facto standard for host dial-up connections--authenticates and configures the session-setup and-release phases. In addition, PPP performs multiple operational functions, including encryption, compression, and billing. Companies such as Harris & Jeffries, Virata, and Trillium are exploring opportunities to provide the connecting software needed for PPP and ATM end-to-end compatibility.

Besides the interoperability issues, expensive installation procedures and costly modems also hampered ADSL deployment. Installing full-rate ADSL required technicians to remove load coils and bridge taps from local loops between central offices (COs) and customer premises and to inspect and repair wiring inside the premises. Inadequate loop-condition records--placement of coils and bridge taps, length of loop, and gauge of wire--made loop qualification even more difficult.

After qualifying the loop, the technician installed a splitter to separate the analog voice signal from the ADSL signal. Another technician would then install a network-interface card inside the customer's computer and finally configure the software. Add all this labor expense to the fact that ADSL standard-compliant modems were expensive and immature--partly because of the complexity of DMT line coding and first-generation chip sets to implement it (Reference 2).

Another problem that sidetracked deployment, though not a showstopper, was the spectral compatibility with other signals in the cable plant. Even though the ADSL standards group analyzed DMT signals with a maximum number of disturbers, such as analog T1, integrated-services digital network (ISDN), and high-bit-rate digital subscriber line (HDSL), real-world evidence pointed to problems. Putting too many digital-subscriber-line (DSL) cables in one bundle reduces the reach for all of them. In addition, ISDN compatibility is important in Europe, where one twisted-pair wire carries both services (see sidebar "ADSL complements ISDN"). Spectrum incompatibility results in crosstalk between the signals.

Crosstalk problems surface when twisted-pair cables with ADSL lines, analog T1 lines with repeaters, and HDSL cables reside in the same cable binder. These binders group as many as 50 cable pairs into one bundle. In these binders, high-power T1 repeaters boost the T1 signal at 3000- to 6000-ft intervals. Unfortunately, the repeaters cause crosstalk that "steps on" the ADSL signals in the binder group. Note that ADSL's asymmetry depends on the fact that remote receivers need not encounter near-end crosstalk (NEXT) from adjacent high-power transmitters. Without NEXT, the remote receivers can pick many bits out of weak signals arriving in the downstream direction. However, in CO equipment that closely groups many transmitters, NEXT poses a problem for upstream receivers trying to pick bits out of weak signals. As a result, the upstream data rate is lower. With T1 repeaters and symmetrical technologies, such as HDSL, high-power transmitters at both ends and in the middle of the cables step on the ADSL signals at the CO and near the remote receivers. The interference is enough to reduce the range of the ADSL signals.

More spectral compatibility problems arise from placing multiple full-rate ADSL lines that use echo cancellation in the same binder groups. To maximize downstream data rates, the T1.413 Category 2 specification allows the downstream modulation to overlap the upstream. CO modems use echo cancellation to subtract their transmitted signal from the arriving upstream signal. Echo cancellation works on a transceiver's signal but cannot cancel the NEXT from adjacent ADSL transmitters that overlap the upstream signals. This NEXT problem for echo cancellation has prompted all ADSL providers to use only frequency-division-multiplexing ADSL because it has no overlapping signals.

Overcoming problems

To overcome ADSL's problems, several companies have introduced proprietary DSL technologies for home use. For example, Rockwell has introduced Consumer DSL as a low-cost alternative to ADSL. Consumer DSL lacks splitters and uses quadrature-amplitude-modulation (QAM) instead of DMT line coding to achieve 1-Mbps downstream rates. Another example is Lucent's WildWire DSL technology, which uses Aware's DSL Lite. DSL Lite delivers 1.5-Mbps downstream rates using DMT. WildWire also has no splitters, offers low cost, and is easy to use. Other proprietary implementations for splitterless ADSL, such as Universal DSL from Centillium, and Consumer-installed DSL from Globespan, will also lay the groundwork for a universal standard.

The proprietary versions of splitterless ADSL are good experiments, but computer manufacturers, access-network operators, and regional-network operators believe that widespread deployment demands the development of a standard. The ADSL Forum, ANSI T1, the European Telecommunications Standards Institute (ETSI), and the ITU all started to work on splitterless versions of ADSL, but progress was slow. To try to speed the process, a subgroup of the ADSL Forum led by Intel (www. intel.com), Microsoft (www.microsoft. com), and Compaq (www.compaq.com) formed the UAWG. All the US ILECs and the leading European and Asian telephone companies have joined as well. These core promoters and a group of technical contributor companies are developing a proposal to submit to the ITU G.lite Working Group to create an international standard for G.lite.

The UAWG focuses on the physical layer but maintains sensitivity to the link- and network-layer protocols. A completed draft version circulated within the UAWG in June, but the UAWG is working under strict nondisclosure guidelines, so the details are not public. The UAWG is assuming that the ITU will ratify the G.lite standard in October. After ITU ratification of a standard, the UAWG will dissolve, its mission accomplished.

The UAWG's goals for G.lite align with those of the ADSL Forum, ANSI, ETSI, and the ITU: simplifying the customer premise modem, the ATU-R (ADSL transceiver unit-remote), and maintaining compliance with the full-rate ADSL ANSI standard. Compliance with ANSI-specified ADSL determines that DMT line coding is necessary for G.lite. By maintaining compliance, ADSL service providers can deploy full-rate ADSL equipment, including splitters, as the ADSL transceiver unit-central office (ATU-C). The full-rate modem at the CO provides an upgrade path to full-rate ADSL for consumers when the applications require it and the price is right.

Simplifying the customer-premise modem means that a consumer should be able to buy a retail modem, plug it in at home, and have it work as simply as an analog modem. The modem must be less sensitive to line-qualification problems. To achieve this goal, the UAWG developers have to trade off high bit rates for longer reach and less sensitivity to loop-condition problems. The trade-off reduces the maximum bit rate to 1.5 Mbps for the downstream and 384 kbps for the upstream data. Rate-adaptive bit rates below these maximums allow modems on local loops with impedance-discontinuity problems to fall back to a rate the loop can support. Reducing the maximum bit rate also allows limiting of the working frequency that DMT line coding uses to less than 500 kHz, which makes the signals less susceptible to attenuation at long distances. Lower frequencies are also less susceptible to impedance discontinuities, such as bridge taps and wire-gauge changes. Using lower frequencies should help with the line-qualification problems that have plagued deployment so far.

Simplifying installation also means either entirely eliminating the splitter or putting a filter at each phone (Figure 2). Eliminating the splitter is a controversial issue because of the problems that arise from noisy phones and customer-premise wiring. A splitter isolates phones and faulty house wiring from the local loop. However, installing a splitter adds the cost of sending a technician to consumers' homes. Some local-exchange owners, especially in Europe, view splitters as necessary to demarcate their local-loop wiring from the consumer's wiring. The G.lite standard may have multiple annexes to address the splitter issue regionally.

When a splitter does not isolate the phone from the digital signal, the ADSL transceiver chip must compensate. DSP algorithms must recognize transient noise coming from such phone-set activities as off-hook and dialing. To avoid losing data, the DSP then needs to reassign in real time the data in the DMT carriers that noise affects. The problem becomes even more complex because of the variety of handsets, each with their unique noise profiles, that consumers may have. The scale of this problem has left many critics doubting whether a splitterless technology will work in the real world of discount-department-store phones and do-it-yourself home-phone wiring. The critics believe that a more workable solution would be for consumers to install inline microfilters at each phone to isolate high-frequency noise from the rest of the phone line. But these inline-filter modules change the line characteristics and could cause impedance-matching problems for high-frequency signals.

Simply removing splitters and reducing the bit rate of G.lite modems may not translate to reducing the cost of consumer modems. Instead, the signal-processing requirements increase because the ATU-R modems have to detect and retrain for the transient line noise. Previously, the splitter effectively isolated the high-speed data, but DSP would now have to compensate for noise introduced by consumers' phone appliances and house wiring. However, chip vendors are exploring other ways to reduce cost. For example, Alcatel and Motorola believe that G.lite's low bit rate will enable the technology to screen the analog-front-end (AFE) device to relaxed data-sheet requirements. Testing to relaxed AFE specifications should result in higher chip yields and lower cost.

The ADSL chip sets

ADSL chip vendors offer a variety of products to reduce cost and maintain performance (Table 1). The vendors base their chip sets either on general-purpose, high-performance DSPs or on lower performing DSPs surrounded with specialized coprocessor blocks. The coprocessors accelerate computationally intensive functions, such as Reed-Solomon and Viterbi coding. High-performance DSPs offer advantages for flexible programmability. However, the DSPs must be very high performance to keep up with the real-time processing G.lite requires or the full-rate speeds of ADSL. On the other hand, the special coprocessors perform parallel tasks, such as trellis coding and echo cancellation in fast and power-miser algorithms. The coprocessors take the burden of simultaneously or rapidly performing many functions off the general-purpose DSP.

The varied approaches to chip-set architecture make it difficult to make a straightforward comparison between chip sets. Some, such as those from Analog Devices, include all the chips you need for a modem. Others include transceivers and AFEs but not controllers and line drivers. Ask each vendor whether its chip set includes all the components you need to complete a modem design or whether you need to buy other components.

Various categories and issues exist in the ADSL standard, and most vendors claim compliance with one or some of these categories. However, this compliance does not necessarily mean the vendors' chip sets interoperate with each other. ANSI ratified the T1.413 Issue 1 ADSL standard in late 1995. Category 1 of this standard is the base performance specification. Category 2 is the high-performance option that includes trellis coding and echo cancellation for higher bit rates over longer distances. The follow-on T1.413 Issue 2 is complete and awaiting final ratification. Issue 2 clears up ambiguities in Issue 1 and adds rate-adaptive operating mode to the standard. Also, Issue 2 adds the optional ATM transconvergence layer. Categories 1 and 2 retain the same distinctions in the Issue 2 of T1.413.

Motorola's CopperGold ADSL transceiver is the only Issue 1, Category 2-compliant and the most highly integrated chip set, which centers on the MC145650 DMT transceiver. This device integrates a general-purpose DSP, a T1.413-compliant data interface, and an AFE. In addition to the CopperGold transceiver, Motorola offers the line driver, the ATM cell processor, and a choice of host-controller options to complete the ADSL modem. The evaluation board, application-programming interfaces, and a system analyzer are also available.

The CopperGold transceiver blends DSPs with coprocessors to accelerate performance. The embedded 563xx-series 24-bit DSP core efficiently processes the 20-bit precision of DMT algorithms, and hardware accelerators process error-correction and echo-cancellation algorithms. The accelerators also allow the DSP to work at lower clock speeds and therefore save power: CopperGold typically dissipates less than 1W of power.

Texas Instruments provides a good example of using a general-purpose, high-performance DSP as the core of an ADSL chip set. TI offers three chip sets--two for client-side modems and one for network-side modems. All the TI chip sets comply with ANSI T1.413 Issue 2, Category 2 and support 8-Mbps-downstream and 800-kbps-upstream data rates. The one-chip TNETD2000C transceiver for CO equipment offers two-line support. The multiple lines allow increased density, lower power consumption, and reduced cost per line. The two full-rate ADSL lines will work in DSL access multiplexers (DSLAMs) and ADSL line cards for G.lite and are programmable for upgrades to evolving standards. TI offers a $5000 multiline evaluation system for CO applications (Figure 3).

TI's TNETD2000R chip set works in client-side external and network-termination-box modems, and the TNETD2000P works in internal PCI-adapter-card modems. The 2000P includes an ATM host-interface controller and Microsoft Windows drivers. A production-worthy adapter-card reference design for using the 2000P will be available in the third quarter of 1998. A client-modem evaluation kit costs $2500.

Meanwhile, Alcatel, which pioneered the use of ATM as the transport protocol for ADSL services, offers the third-generation DynaMiTe ADSL chip set, which supports the Issue 2 option for ATM transconvergence (Figure 4). The DMT modem chip integrates the ATM framing function but does not include the ATM segmentation-and-reassembly (SAR) function, because Alcatel assumes that only 2% of a host Pentium's processing power can perform the SAR function at ADSL bit rates. DynaMiTe also includes a real-time controller that eliminates any real-time constraints on the modem or host mP. A reference design is available.

In the interest of rapid ADSL adoption, Alcatel licensed the DMT/ATM transceiver and AFE to AMD, SGS-Thomson, and Integrated Telecom Express. The licensees must develop their own chips for the controller function. SGS-Thomson, with its Tosca chip set, is the first licensee to integrate the ATM transport and framing functions and a Utopia level 2 interface, which connects directly to ATM systems. The chip set can operate as an ATU-R or ATU-C by using the appropriate microcontroller code, which SGS-Thomson supplies. In addition, a "transmit-direction-politeness" feature lets you select signal attenuation to reduce crosstalk in short loops. In the receive direction, an option lets you control an external multiplexer to select external attenuation in short loops.

The second-generation AD20msp918 chip set from Analog Devices also adds the ATM framing options for Issue 2. The chip set complies with Issue 2, Category 1 plus trellis coding. Believing that reach is more important than data rate, the company's designers opted to omit adding echo cancellation to comply with Category 2. Using echo cancellation for higher data rates increases crosstalk, thereby making the device effective only over shorter distances. By adding trellis coding, however, the designers keep a low bit-error rate over long reaches. You can also program the frequency division between upstream, downstream, and ISDN to maximize the chip's use in applications in which ISDN operates over the same twisted pair.

Another new ADSL chip set, Lucent's WildWire, bypasses ANSI standards, instead targeting easy modem transitions for consumers by integrating analog and ADSL codecs (Figure 5). WildWire adds G.lite capability to the company's 56-kbps ITU V.90 chip set. The new chip set includes an autodetect feature that determines whether a CO has a DSL connection and allows the user's modem to transmit data at the highest rate available by using the analog or the ADSL codec. Software upgrades to G.lite compliance will become available when the standard is complete. The programmable DSP1690 couples two of Lucent's DSP1600 cores in one chip to provide 200-MIPS performance and typically consumes 1.5W of power.

Rather than claim compliance with ANSI standards, Globespan Semiconductor continues to offer ADSL chip sets based on the more mature CAP, which most ADSL deployments now use, instead of DMT line coding. Globespan intends to support the UAWG for G.lite but in the meantime is offering its version. The splitterless Consumer-installable DSL (CiDSL) chip set adds a micro data filter to the company's previous ADSL chip set. Consumers put the micro data filter, which costs less than $5, on their phone lines.

Not exactly an ADSL chip set but important in a modem is the network processor. Virata's Atom family of network processors uses the company's integrated, G.lite-compliant software-on-silicon technology for endpoint modems. The first chip in the family, the Hydrogen, facilitates ATM over any of the variations of xDSL and Ethernet over ATM over xDSL for designing customer-premise-equipment modems. It also enables basic bridging and routing for external DSL modems that connect to Ethernet hubs at the customer endpoints. The next chip in the family, Helium, enables multiclient CPE modems, such as hubs and switches with integrated routing, bridging, PPP, and ports that support six to 12 users. Helium also works in DSLAM-line-card applications.

The network processing software from Virata is available for licensing. One software package works for G.lite-to-VDSL bit rates because the processor chooses bit-rate performance on the fly. Modules include software for PPP over ATM and USB, PPP over ATM and Winsock 2.0, and ATM signaling and management. Software licenses cost $200,000 to $400,000 for access to all source code and $100,000 for focused designs.

All these compliant chip sets don't necessarily interoperate because the standards, such as ITU G.lite, are not yet final, so interoperability testing is critical. Chip vendors must adhere to the standards and cooperate with each other for testing at the physical layer, and modem vendors need to cooperate at the protocol level. The ADSL Forum is taking action on this subject by contracting with independent test labs to develop compliance-test specifications for ITU and ANSI standards. Also, individual companies are taking action. For example, Alcatel offers an open-implementation guide describing the choices, especially those for the ATM layer, it takes for its chips when interpreting the standards. In addition, Alcatel, TI, and Analog Devices plan to perform interoperability testing among themselves.

The interoperability testing should complete the consumer-grade modem products that are the UAWG's goal. The maturing of the DMT technology and the chip sets that make it work should be able to get ADSL back to the front of the race to pave the information superhighway to the home. Modem designers should soon see more highly integrated and application-targeted chip sets, because the chip vendors are confident the market will continue to expand.

ADSL Complements ISDN