uses advanced signal-processing concepts that force mathematician Claude Shannon's 1948 theory of channel-information capacity to its ultimate level, pushing data onto the local loop at several-megabit-per-second rates. But making xDSL a reality takes more than sophisticated algorithms running on DSPs. The analog-front-end (AFE) circuitry that sits between the telephone company loop and the DSP plays a key role in successful and practical implementation of any xDSL variation. This AFE must digitize or re-create analog signals with excellent dynamic-range performance, drive fast-slewing signals into lossy copper lines, and capture the weak incoming analog signals that far stronger outbound signals overshadow—all while using little power at each end of the link. Basic structure has complexity Many alphabet-soup variations of xDSL exist, each having a unique combination of data rates, upstream/downstream data-rate asymmetries, and distance reach. However, all xDSL designs need to pump significant signal levels onto the line to ensure that the signal power at the received end is sufficient for data recovery. Typical power levels are 100 to 400 mW, but transient peak levels can reach 5 to 10W. The path between the local loop and the digital-processing circuitry consists of a line-transformer interface, analog filters, line drivers and receivers, and A/D and D/A converters. The circuitry may also include a splitter circuit, which has several filters to separate the conventional POTS signals from the xDSL signal. The unpredictable static and dynamic variations of the local loop make xDSL an analog challenge. In the United States, the loop's length between the CO and the customer-premises equipment (CPE) can be 100 to 20,000 ft. Studies indicate that, although 60% of US lines are shorter than 16,000 ft, you need to cover loops as long as 18,000 ft to extend access to 95% of residential lines. This significant span of lengths defines the need for wide-dynamic-range components in your xDSL system. A signal that is optimized for the shorter length may have insufficient power for the longer one; a signal that is optimized for the full length is likely to overload shorter loops. Although the local loop in the United States is normally unshielded twisted-pair cabling, it does not even come close to resembling a classic transmission line. Each local loop has its own personality, and length is only one contributing quality. Some lines are relatively direct; others run in circuitous paths near noise sources. Some lines are clean, point-to-point runs; others show the scars of a long history of use. These older lines may have bridged taps, which are un-terminated stubs of the twisted-pair cable left over from previous cable use. In other cases, twisted-pair segments are spliced to make a longer loop or a loop to a new location. The situation in Europe adds another layer of complexity: European loops are generally shorter, but they vary from country to country. Further, European loop wire gauges vary considerably, and some loops use aluminum wire or spliced aluminum and copper. Even if, once you know the length and its static condition, you adjust your xDSL system parameters to best meet the requirements of a given local loop, you still need to accommodate some difficult and changing line conditions. There's noise, of course, as the loop picks up all sorts of transients and EMI from electrical discharges and adjacent lines. There are also the noises that the loop's users generate as part of conventional POTS. These noises include sudden transitions due to on-/off-hook state changes, dial tone, the ringing voltage that the CO sends to the phone, call-waiting interruptions, and dialing signals that the phone generates when a user places a call. To further complicate things, phone systems maintain legacy considerations and backward compatibility, so your line may also include old-fashioned rotary-phone (pulse) dialing clicks. These noise factors and transients stress the xDSL circuitry and system in two ways. First, the hardware must have sufficient head room and dynamic range to accommodate the noise factor and transients and be resilient enough to work with the drastic impedance changes that often accompany them. (After all, when you pick up the POTS phone and take the line from on- to off-hook, the line's dc termination goes from an open- circuit to a closed-circuit condition.) Second, these disruptions often cause the system algorithms to go to a retraining mode as they learn to adapt to the new signal levels, noise spectrum, and SNR parameters. Ultimately, this adaptation allows xDSL to work at high data rates, but the adaptation severely affects data throughput. Power transfer is critical Although each xDSL variation uses a different signaling pattern and format, you can get a good idea of signal levels by looking at the common asynchronous-digital-subscriber-line (ADSL) standard and its G.lite variation. Discrete-multitone (DMT) modulation is inefficient because it has a relatively high peak-to-rms ratio (PAR), or "crest factor," of 5.6 (15 dB). High-bit-rate digital-subscriber line (HDSL) has a more efficient PAR of 4 (12 dB). The line driver must deliver peak power to the line interface without clipping and distortion. The distortion specifications for xDSL are stringent, even by today's high-performance standards. Although your distortion goal depends on the xDSL variant, total harmonic distortion (THD) of better than –70 dBc is your nominal first-pass requirement. For comparison, a high-fidelity audio system with its tightly constrained line and load may boost superior THD of –90 to –100 dBc, but –70 dBc is still impressive. By contrast, many video systems operate with THDs of approximately –50 dBc. The system effects of distortion are significant. Excessive distortion means that the xDSL DSP must apply more sophisticated algorithms to implement error correction and, thus, may need more MIPS-processing and supply power. Conversely, distortion means that, for your design's given DSP/algorithm combination, you are restricted to lower data rates, range, or both than you would achieve with lower distortion. You can achieve low distortion in any design if you are willing to pay the price with reduced amplifier efficiency. However, a practical xDSL system is also power-constrained. To increase efficiency, you have several choices. First, consider amplifiers with low quiescent (static) efficiency. This low quiescent current can achieve only a small portion of the efficiency you need; the dynamics of the signal and, therefore, the amplifier's dynamic efficiency account for the bulk of the losses. The dynamic dissipation involves several factors. Most critical is the drive amplifier output's magnitude compared with the magnitude of its supply rail. The closer to the rail that the amplifier runs, the more efficient the amplifier (Figure 1). Extra head room between the peak signal and power supply wastes power. A line-driving amplifier that supports the peak-signal requirements while operating from lower power-supply rails is more efficient. This small voltage differential, in turn, requires that amplifier vendors exploit internal architectures that suit xDSL signals. For example, they can use an emitter-follower output stage rather than a Darlington stage to achieve the requisite gain. Another design element adds a degree a freedom to your distortion, voltage, and current trade-off. The transformer—that ancient electrical device—normally couples the xDSL driver/receiver to the line and allows you to select a voltage/current span for the power you need to put onto the line. All this concern about power and efficiency may seem excessive, given that the CO operates from a substantial ac line source and that the CPE xDSL termination is a single-channel device. In reality, though, both the CO and CPE are concerned with power dissipation—but for different reasons. At the CO, channel density—the number of customers that you can serve from a single cabinet—is critical, and power costs include the costs of obtaining and regulating power, providing battery backup if the ac line fails (telephone companies do worry about outages), and overall cooling. Further, if the telephone company powers the CPE, then the CO is concerned with both the power it can send down the line and the aggregate power it must deliver to all the CPE units it supports. The reasons for power concern at the end user CPE differ widely from those at the CO. Cooling and channel-density problems are secondary to the availability of supplies and their current capacity. Dual supplies, such as ±12, may be available, but the negative supply is often unregulated and needs further conditioning, or the supply has limited current-sourcing capability. In other cases, the system power is just a 3 or 5V rail, so any higher dc-rail voltage requires a step-up converter. Finally, although an xDSL link should be on all the time, its associated desktop or laptop PC may need to go into a sleep mode to conserve power, or the PC must meet Environmental Protection Agency Energy Star or similar European Community Blue Angel guidelines for standby power consumption (approximately 1W). Both give and receive Although efficiency and delivering low-distortion power to the load are critical to the transmitting side of the xDSL link, there are also important issues regarding the complementary receiving end of the data path (Figure 2). Depending on loop length, receiver input range, and other factors, the receiver may typically need to attenuate the incoming signal by about 13 dB or provide 40-dB gain to compensate for the signal's average loss on long loops. Therefore, the receiver front end needs typical gain as large as –13+40=27 dB. In itself, this value is not a lot of gain, and you would have little difficulty implementing it in a well-designed circuit. As usual, though, noise—the pervasive nemesis of ideal circuits—enters the picture. To make sure that your front end does not significantly add to the noise level and degrade SNR, keep added noise at less than one-half those noise-density values. An xDSL AFE is a lot more than just a line driver and receiver. The receiver-path A/D converter and corresponding transmitter-path D/A converter (or pulse former) must also provide very-low-distortion operation, or their deficiencies overshadow the virtues of the line driver and receivers. You want converters (codecs) with at least 12 (and, more likely, 14) bits of resolution, although resolution is only one measure of the converter's suitability (Figure 3). Because the xDSL signal is spread over a wide band, the converter's frequency-domain performance is also critical. Further, sampling rates are 2 to 4M samples/sec or faster, yielding a resolution/speed point that is fairly difficult to achieve with high quality and low spurious-free dynamic range (SFDR). An attractive market opportunity for xDSL AFE vendors that have strong linear and mixed-signal expertise, and the plethora of component and chip-set offerings clearly reflects this diversity. But beware: Not every part that appears suitable for xDSL applications targets the unique requirements of the application. Vendors have "repurposed" some of these devices from other communications, video, or more general applications. Repurposing is not necessarily bad, but you need to carefully read data sheets to see if the vendor assessed the part's specifications using the various xDSL standards as its context. Consider the basic line-driving function. For many amplifiers, vendors normally specify the maximum output current as the maximum current that the amplifier can deliver into a short circuit (or nearly short-circuit load of a few ohms), with V_OUT=0 and R_LOAD=0. This specification is not meaningful for an xDSL driver. With an xDSL driver, you look for substantial current output at a known worst-case distortion level. A more meaningful specification defines the maximum current under application-specific conditions, such as I_OUT with SFDR of –60 dB at 1 MHz into 25W. If the driver or receiver IC is a dual-channel device, make sure that the specifications are for both channels in the typical differential configuration that xDSL drivers and receivers use, rather than the one-channel-active/second-channel-quiescent mode. You also need to compare components from different vendors with an xDSL chip set from a single supplier. There's no best approach here. Considering components from disparate vendors allows you to pick the best of what's available or, even better, allows you to match components to your own unique requirements. At the same time, vendors who offer chip sets often compensate for weakness in one component by offering strength in another. Perhaps most important, the chip-set vendor can assess the combined set of ICs at a higher level of the appropriate xDSL standard, taking from you more of the integration and test burden. Filters are an important part of the xDSL front end. Most vendors incorporate filters on the xDSL ICs to the extent possible to both save OEM cost and space and to match filter performance to driver, receiver, and converter performance. Switched-capacitor filters are easier to fabricate on the IC than continuous-time designs, but switched-capacitor filters have higher noise. However, when the vendor designs the filter switch as part of a larger IC or chip set, it can synchronize filter switching and converter and amplifier operation. This system wide context minimizes noise degradation. Analog Devices has individual components, including AFE chip sets as well as chip sets that include DSPs. The $2.78 (1000) AD8017 dual high-current amplifier provides a 1600V/µsec slew rate that can pump 270 mA into a 10W load and has an SFDR of –58 dBc at 1 MHz. Quiescent current is 7 mA for each amplifier in the dual package, and the amplifier aids dynamic efficiency by sourcing voltage to within 1V of a 12V supply rail. At the other end of the integration spectrum, Analog offers the $50 (OEM) AD20msp918 chip set for ADSL. This five-IC chip set supports Internet access with 4.5-Mbps downstream and 450-kbps upstream data rates and includes an asynchronous-transfer-mode modem interface, a DMT coprocessor, analog/digital converters, line driver/receivers, a DSP, and management software. All features comply with relevant ANSI, ETSI, and ITU standards. Burr-Brown is also active in the line-driver area. The company offers the $2.95 (1000) DRV1101, a 5V-supply, 230-mA-pk current driver for ADSL G.lite systems features THD of less than –81 dBc at that full voltage span, into a 100W load. The eight-pin IC supplies as much as 10-dBm average line power with a crest factor of 5.3 for a peak delivered power of 25 dBm. Moving up the integration chain, Burr-Brown also offers a series of AFEs, such as the AFE1203 and similar $14.40 (1000) AFE1205 for HDSL. These devices allow a single-pair HDSL connection to run at twice its previous speed, to 2.3 Mbps, and feature a scalable data rate that adjusts operation, power dissipation, and filter responses according to clock frequency and data rate—to as little as 160 kbps. Within the 48-pin SSOP, a transmitting section generates, filters, and buffers output data in 2B1Q format. The transmitting section's internal differential line driver uses a composite stage output configuration; Class B operation efficiently drives large signals, and Class AB operation minimizes crossover distortion. The receiver channel, with its fourth order delta-sigma A/D-converter, programmable gain amplifier (0 to 9 dB), and 24X over sampling ratio, filters and digitizes the symbol data received at symbol rates as high as 1168 kHz (for 2.3-Mbps operation). Although Texas Instruments is sometimes primarily known as a DSP supplier, the company has developed a broad xDSL analog portfolio. For example, it offers the $4.80 (1000), 400-mA dual-differential-line-driver THS6012 for the CO side, and the similar 200-mA, $3.71 (1000) dual-driver THS6022 for the CPE side. The THS6012 offers –72 dB third-order harmonic distortion at 1 MHz with a 25W load and a 20V p-p signal. Its current-feedback topology provides slew rates as high as 1300V/µsec at gain of 5. The company targets the CPE device at lighter loads and lower current levels. The device's corresponding distortion figures are –69-dB third-order harmonic distortion into a 50W load at 20V p-p. TI also has codecs for either link end. For the CPE side, consider the TLV320AD11APZ codec for ADSL, which supports 8-Mbps downstream and 640-kbps upstream operation. This 3.3V, 100-pin IC contains 14-bit transmitter and receiver circuitry (including gain blocks and filters), a clock, a reference, and a host interface (parallel for data and serial for control), as well as POTS filters. TI's complementary codec for the CO side is the TLV320AD12APZ . STM-i-croelectronics divides its functions into the STLC60135 DMT transceiver and the $50 (10,000) STLC60134 AFE. This front end has a 12-bit, 8.8M-sample/sec A/D converter and two D/A converters. It handles a 16- to 640-kbps upstream channel and a 1.536- to 8.192-Mbps downstream channels. The 64-pin device, which consumes 0.5W at 3.3V, also includes third- and fourth-order, tunable continuous-time filters; transmit gain control in 1-dB steps over a 15-dB range; and receive gain control over a 31-dB range, also in 1-dB steps. Alcatel also partitions its $50 (OEM) MTK-20131 ADSL modem chip set into digital and analog functions. The MTC-21034 AFE has 12-bit A/D and D/A converters at 8.8M samples/sec supporting the 640-kbps upstream and 8.8M-sample/sec downstream data. Tunable, 1-dB- maximum ripple analog filters let you set cutoff frequencies to 138 kHz for the upstream channel and to 1.1 MHz for the downstream channel. The AFE includes a linear predriver that drives the external line driver. You need to separately add the line driver. Some less-well-known vendors are also putting ADSL devices onto the market. Datapath Systems Inc is building on mixed-signal experience in disk-storage components. The company has developed the DSP8000 for CPE and DSP8001 for CO AFEs. They include 14-bit linear A/D and D/A converters operating at 4.416M samples/sec and receiving front ends with a noise floor that is –150 dBm/Hz higher than 300 kHz. These $22 (100), 128-pin ICs also include fourth-order lowpass filters with 5% cutoff-frequency accuracy, user-programmable gain in signal paths, and a 12-bit DAC for the voltage-controlled crystal oscillator. Other traditional linear vendors are contributing to the array of xDSL building blocks. Maxim's $39.96 (1000) MAX1201, a 5V, 14-bit A/D converter, provides conversion rates as high as 2.2M samples/sec with SFDR of 86 dB at 1.0021 MHz and SNR of 81 dB. For the xDSL receiver, the MAX414x series offers a choice of high-speed, low-distortion differential line receivers. Linear Technology Corp also has A/D converters that you can apply to xDSL applications. The $20 (1000) LTC1414, a 14-bit device with 2.2M-sample/sec capability, has SNR and distortion of 78 dB and SFDR of 86 dB at Nyquist input frequency of 1.1 MHz (Figure 5). This ±2.5V-input-range, 28-lead device also includes a sample/hold circuit with 40-MHz bandwidth. Elantec is also active in the line-driver area. Its $4.90 (10,000) EL1504C differential line driver supplies as much as 45V p-p into 200 W. Typical full-output distortion for this device is –60 dBc at 2 MHz. The 20-pin device is suitable for 6-Mbps, 18,000-ft ADSL applications. The 16-pin EL1505C may be suitable for reduced distances, speed, cost, and power dissipation. The $1.14 (1000) CLC5665 line driver from National Semiconductor may suit your needs if you are not running a full-rate or full-distance xDSL system. This 1800V/µsec device, which National originally targeted at video distribution, supplies a maximum output current of 85 mA. It has THD of –89 dBc at 1 MHz and develops 1V p-p into a 500W load. Providing a different mix of integration, the Fujitsu Microelectronics MB86626 ADSL front end integrates all active circuits (except the transmitting-side line drivers) into an 80-pin IC. It provides 15-bit A/D and D/A converters, active filters, and gain control with a 0- to 38-dB range. The device's architecture supports both analog and digital echo cancellation, and you can use the same IC in CO and CPE nodes, as well as in full-operation ADSL and its G.lite variation. Putting it together requires test, too Whatever ICs or chip set you decide you use, be sure to ask the vendor tough questions. ADSL devices, which exist in a 13-bit resolution, 2M-sample/sec environment, need test equipment with 16-bit, 6M- to 32M-sample/sec performance for meaningful performance data. Multi tone tests are critical to xDSL testing. If the vendor conceived and designed the part for xDSL from the start, make sure the company tested the part with multiple tones to meet the relevant specifications. If the vendor is refocusing the part from another initial target application toward xDSL, see how the device's specifications relate to xDSL regulatory requirements. If your vendor offers higher level or more highly integrated xDSL options, you have the advantage of tests that more closely relate to final performance in the system configuration. For example, an "uncancelled-echo" test, which Burr-Brown's AFE1205 data sheet describes, shows the noise and linearity errors that the transmitter and receiver generate independently of the far-end signal source or additive line noise. This test uses a known symbol sequence and passes it through the transmitter; a line-interface transformer; and a resistor, which simulates the line and then echoes the signal back into the receiver, which recovers the signal. A processor compares and subtracts the recovered data bits at the output of an adaptive filter, which also receives the original data stream. Be sure that the vendor also has data that compares AFE performance with parameters for the appropriate templates for the transmitting- and receiving-signal spectrums, power density, and time-domain that the relevant xDSL specifications explicitly call out or those parameters that you derive from these specifications. References

1. Kempainen, Stephen, "ADSL chip sets trim down with G.lite," EDN, July 2, 1998, pg 81.
2. Quarfoot, Jim, "Managing ADSL signals and contending with noise," Communication Systems Design, December 1998.
3. Chen, Walter Y, DSL Simulation Techniques and Standards Development for Digital Subscriber Line Systems, Macmillan Technical Publishing, 1998.
4. Ranschmayer, Dennis, ADSL/VDSL principles: Practical Implementation of DSL Networking, Macmillan Technical Publishing, 1998.
5. Rowe, Martin, "ADSL testing moves out of the labs," Test & Measurement World, April 1999.

Transformers: 150 years and still goingUsing a 1-to-1 ratio, you can reproduce your xDSL driver current and voltage output on the secondary line with the same ratio they have at the primary side. But engineers have known for many years that you can reduce IR line loss by stepping up the voltage on the line with a transformer (and thus reducing the current on the line) and then restoring the voltage at the far end with a step-down transformer ratio. (English physicist and chemist Sir Michael Faraday first explored the transformer in the 1840s!) Using a step-up transformer means that you do not have to develop high voltages with your active circuitry. This feature may be attractive, depending on your available supply rails, but you don't get something for nothing here. Lower voltage amplifiers tend to be less efficient than higher voltage ones, so you lose efficiency in the driver. Further, because you are looking to deliver power, the only way you can keep the VXI power product constant as you drop the internal voltage level is by increasing the current level in your driver. You need a driver that can deliver large amounts of current with low distortion into loads of 10 to 100W. Using a step-up transformer changes the problem of efficiently delivering voltage into one of efficiently delivering current, which may be a greater challenge for the driver. You also face a practical issue: Many line transformer ratios (1-to-1, 1-to-1.5, 1-to-2, 1-to-4) are available as standard vendor items. However, if you need a special ratio, such as 1-to-1.35, to precisely obtain your desired voltage and current range, you may find fewer sources, higher costs, and longer lead times. Don't postpone any investigation of the transformer until the end of your design cycle. The transformer you choose must work with your driver and your receiver and meet the numerous xDSL and telephone-company standards. Some IC vendors are working with transformer vendors to match these standards for you and give you some level of assurance that their ICs will have corresponding transformers to round out the component set. For example, Analog Devices has joined with Pulse Engineering Inc (www.pulseeng.com). Pulse targets its B2031 and B2032 line-interface transformers to work with Analog's AD20msp910 and AD20msp918 chip sets.

Shall we split up?An xDSL system must support both high-speed data and conventional plain-old telephone service (POTS). Because the POTS band is the lower end of the frequency spectrum—from a few hundred hertz to about 4 kHz—it overlaps with the much wider xDSL band. In standard asynchronous-digital-subscriber-line (ADSL), you use a splitter to separate the POTS bandwidth from the rest of the ADSL band. Although this physical filter provides potential for the best POTS and ADSL service, it is difficult to implement, and it complicates ADSL installation because a service technician must install it at the customer-premises equipment (CPE). A splitter-less ADSL installation, which is closely related to but not the same as G.lite, requires no filter inside the CPE. Instead, the ADSL CPE system performs more sophisticated, complex, and costly algorithms to separate POTS from ADSL service. Alternatively, the system may constrain you from simultaneously using POTS and ADSL—a situation that many end users would probably find unacceptable. A third alternative distributes the splitter function by placing the splitter as an inline low pass filter between the loop entrance to the CPE and any POTS phones that are on CPE premises. This distributed-splitter approach strikes a middle zone in system performance, installation difficulty, and technical challenge. Reference Warrier, Padmanand, "Universal DSL deployment of G.lite: issues and solutions," Texas Instruments Application Report SPAA007A, 1998, at www.ti.com.

Table 1—Vendors of xDSL Analog-Front-End components
Vendor	Circle No.	Line driver	Line receiver	A/D converter	D/A converter	Analog front end	Digital signal processing
Alcatel Microelectronics www.alcatel.com	301					x	x
Analog Devices Inc www.analog.com	302	x	x	x	x	x	x
Burr-Brown Corp www.burr-brown.com	303	x	x	x	x	x
DataPath Systems Inc www.datapathsystems.com	304					x
Elantec Semiconductor Inc www.elantec.com	305	x	x
Fujitsu Microelectronics Inc www.fujitsumicro.com	306					x	x
Linear Technology Corp www.linear-tech.com	307	x	x	x	x
Maxim Integrated Products www.maxim-ic.com	308	x	x	x	x
National Semiconductor Corp www.national.com	309	x	x	x	x
STMicroelectronics Inc www.st.com	310					x	x
Texas Instruments Inc www.ti.com	311	x	x	x	x	x	x

Tags: A/D converter, AD8017, AFE, AFE1205, Analog, analog-front-end, cable modems, CPE, D/A converter, differential line driver, Discrete-multitone modulation, DMT, DSP, EL1504C, HDSL, High-bit-rate digital-subscriber line, hybrid fiber-coax, ISDN, LTC1414, MTK-20131, ntegrated-services digital network, PAR, peak-to-rms ratio, plain-old-telephone-service, POTS, PSTN, satellite link, SFDR, spurious-free dynamic range, STLC60135 DMT, THD, THS6012, THS6022, TLV320AD12APZ, VXI, wireless local loops, xDSL, xDSL AFE

This entry was posted on Sunday, July 12th, 2009 at 7:52 am and is filed under Communication. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.