Design Feature: August 1, 1996

Line drivers and receivers represent a potential bottleneck in communications systems. Though the systems these devices support usually contain sophisticated and multilayered signal processing, the signal path eventually funnels down to this single, functionally simple point of contact between system and cable. If the driver or receiver does not suit the application or if you improperly specify or apply these items, then the intended signal becomes corrupted or lost, thus negating all those complex layers above it.

Increasingly, line and cable drivers and receivers are supporting greater bandwidths and distances. This situation is true for digital signals representing data and for analog signals, such as RGB (component) video for high-resolution monitors. Even the digital signals, however, must handle the analog reality of driving any signal—whether it represents information in digital or analog format—through a wire.

In addition, applications such as asynchronous digital-subscriber line (ADSL)/high-bit-rate digital-subscriber line (HDSL) and cable modems require relatively high voltage and current drive. You must supply this drive, match line impedance, minimize the effects of common-mode voltage, and provide isolation in some situations. Like the unity- or low-gain buffer within a circuit, the line driver usually provides no amplification. Unlike those internal buffers, it must also handle a more varied load situation, as well as more arduous EMI/RFI, crosstalk, and fault conditions due to line length and unfriendly surroundings.

Start with cable basics

The most common choice in electrical cable is between coaxial cable and shielded or unshielded twisted-pair (UTP) cable (Reference 1). An application may require you to use existing cable and, thus, you have no choice of what type to use. In more fortunate circumstances, however, you can select the kind that gives the performance and price combination you need or that a LAN standard you are implementing specifies.

Coaxial cable offers the advantages over twisted pair of greater bandwidth and inherent shielding, but coaxial cable costs more and is slightly more difficult to terminate. Twisted pair is available in single-pair or multipair bundles with as many as 25 pairs and, thus, can support more than one signal path in one cable. Basic UTP has a 100(ohm) impedance, and the shielded version typically has 150(ohm) impedance. The most common impedances for coaxial cables are 50 and 75(ohm), although other values are available.

A coaxial cable supports single-ended (grounded) connections, and twisted pair normally supports balanced, differential terminations. To get the benefits of shielded coaxial cable in a balanced configuration, you need to use two coax cables or switch to triaxial cable, which has two parallel conductors within a single shield. Be careful, though, with twisted-pair lines: At high data rates, the slight but unavoidable asymmetry between conductors, caused by imperfections in their twist, causes subtle timing-skew problems due to differences in propagation times, as well as less-than-ideal common-mode rejection. Special twisted-pair cables and coaxial cable pairs minimize these problems.

In directly coupled connections, your equipment most likely must interface with a 50 or 75(ohm) line. The type of cable your application has historically used primarily determines this choice. In most situations, you must include a line-matching resistor at the driver (Figure 1). This resistor matches the source impedance to the line impedance and minimizes reflections from the source end. These reflections occur when the far end of the line reflects back transmitted signals; mismatch at the source then again reflects this echo. The double echo appears as a delayed, attenuated version of the original signal and causes distortion and errors at the receiver where it adds to the original signal (Figure 2 and Reference 2).

Many terms exist for this matching-resistor function, including "back-termination," "back-matching," "reverse-termination," and "double-terminated." Re-gardless of its name, however, the resistor has another critical effect on your signal: It causes the receiver to see the signal at one-half its original amplitude, due to the voltage divider that the resistor and the line form. You can use 2-times gain at the driver to maintain nominal signal levels and S/N ratio.

To reduce the number of external discrete components, some vendors incorporate the gain-setting/feedback resistors on-board. Harris, for example, offers the HFA1112, a 550-MHz cable driver that provides gain of 2, 1, or -1 by pin-strap selection; dual (HFA1212) and quad (HFA1412) versions are also available. For closer matching of the internal device count to the application, the HA5013 for RGB video incorporates just three 125-MHz drivers.

Direct coupling to the line is satisfactory in many, but not all, applications. Transformer coupling is a beneficial and often necessary alternative (see box, "Transformers solve coupling problems").

Despite all the interest in single-supply ICs, you may have to use a bipolar supply and driver IC. A single-supply device may be unable to slew as fast or as wide as your requirements dictate or to provide the needed output current. A single supply provides a virtual—not real—ac ground, which is difficult to isolate and also contributes to ground noise.

A bipolar signal operating from a split supply not only automatically doubles the dynamic range and, thus, S/N ratio for a given maximum amplitude excursion, but also eases the design and increases the performance of the drive circuitry within the output stages of the line-driver IC. The choice, in fact, may not be yours. Many communications applications require signals that are symmetrical with respect to ground to minimize ground differences and offsets. To drive a given amount of power injected onto the cable, a bipolar supply also requires one-half the current of a unipolar supply.

The maximum distance you can drive a signal through a cable is a function of many interrelated factors: data rate and bandwidth, signal swing, ambient noise, and cable impedance and type. Typically, a short-haul cable run is approximately 1000m or less, and a long run is 4000 to 5000m. Maximum data rates range from several megabits to several hundred megabits/second, but there are so many variables that it is impractical to give precise values. As a general guideline, the achievable data rate falls in inverse proportion to the square of the distance: You get one-fourth the data rate at twice the distance, assuming that all other factors remain unchanged.

You may not have all the flexibility with these factors that you would like. Some applications have de facto or official standards that define or limit variables. UTP applications, such as ADSL, limit the power output to 26 dBm, to minimize both near-end crosstalk (when the transmitted signal interferes with a receiver at the same end of the cable) and far-end crosstalk (when the transmitted signal interferes with a receiver at the opposite end) (Reference 3).

Video pushes speeds

It's not hard to support composite-video applications, such as NTSC (National TV Systems Committee), PAL (phase-alternation line), or SECAM (Systeme Electronique Couleur avec Memoire), with their relatively low bandwidths of less than 10 MHz. You can choose among many 50-MHz op amps with differential gain and phase errors less than 0.01% and 0.018, respectively.

With high-performance RGB video, the speeds are much higher. Although needs vary, a typical application requires 0.1-dB gain flatness to at least 30 MHz. For broadcast and high-definition-TV vendors are pushing performance and relatively low power. Because the application inherently needs power to drive the line, however, the vendors emphasize quiescent and standby power consumption, rather than active power consumption.

Maxim's unity-gain-stable MAX4102, for example, has a 250-MHz ,-3-dB bandwidth; 130-MHz, -0.1-dB gain flatness; 350V/msec slew rate; and 0.002% and 0.002° differential gain and phase errors, respectively, at 3.58 MHz (Figure 3). Its companion MAX4103 is compensated for gains of 2V/V or greater, with 180-MHz bandwidth, 80-MHz gain flatness, and 0.008% and 0.003° gain and phase errors, respectively.

In video distribution and switching systems, such as studios or multiple-monitor setups, you typically feed the signal to multiple receivers. Although you can use a driver that directly sources multiple receivers, you risk losing all output if any receiver fails or if a user accidentally shorts the driver or connects another output to it. Unless your design has no user-supplied connections or can tolerate such catastrophic failures, use the common but more costly technique of providing one driver per channel as a buffer. This approach provides a level of protection and eases troubleshooting.

Not all applications are point-to-point. The loop-through, or bridging, amplifier (Figure 4), lets you tap a signal voltage between the coaxial-cable conductor and its shield. The input impedance of the amplifier must be high enough to make cable loading negligible. Also, ensure that the amplifier, even when powered down, maintains this high-impedance input, because the taps often are independently controlled.

Increasingly, serial digital video (SDV) is driving professional and studio production and editing. ("SDV" also stands for switched digital video (Reference 4).) The Society of Motion Picture and TV Engineers (SMPTE) standard 259M allows digitized video at rates of 143 Mbps (NTSC), 177 Mbps (PAL), 270 Mbps (Serial Digital Component) and 360 Mbps (Widescreen Component), with drive currents of 26 mA per 75(ohm) load. Technical similarity exists between parts for the SMPTE rates and the DS-3 and synchronous-optical-network/synchronous-digital-hierarchy rates of 44.7, 51.84, 155.52, and 311.04 Mbps. Each application benefits from devices designed initially for another. At these high rates, the cable's dispersive loss characteristics increasingly affect received-signal integrity, and you have to consider compensation and equalization techniques.

To meet the SDV needs, a driver such as National/Comlinear's CLC007 accepts single-ended or differential ECL input signals and drives four ac-coupled, 75(ohm), back-matched loads with single-ended ECL levels of 800 mV p-p. The 400-Mbps device can drive more than 300m of cable, such as Belden 1505 or 8281, with 650-psec rise and fall times. The CLC006 driver is similar to the CLC007, but the CLC006 lets you adjust output amplitude via an external resistor and drive two loads.

ADSL, HDSL, and very high-speed ADSL (VDSL) demand drivers that not only provide bandwidth and fidelity, but also put substantial power onto the line. The amount of power needed depends on the distance and the line characteristics. Also important for this class of application are the driver-distortion specifications, because these multitone signals are especially susceptible to THD and intermodulation distortion (IMD). Typical distortion figures are -80 to -90 dBc.

Analog Devices' AD815 drives such links; three resistors set its gain (Figure 5). With a ±15V supply, it delivers 400 mA rms (1A pk) with differential voltages as high as to 40V p-p; it has 50-MHz, -3-dB bandwidth; 680V/µsec slew rate; and 60-nsec settling time to 0.01%.

Keep thermal design at the top of your design list for ADSL and HDSL applications, because these technologies' power dissipation is much greater than that in video and LAN applications. Packages with heat sinks and power surface-mount packages with heat spreaders let you dissipate the heat. Be sure to do the thermal calculations and allow for necessary circuit board copper as needed.

Although drivers get the most attention, don't forget the role of the line receiver. It has the unenviable role of extracting a noisy, distorted signal from the line, often with common-mode voltage as an additional challenge. Although receiving on a simplex or half-duplex path connection is difficult, a full-duplex system adds another problem for the receiver: The relatively high-level transmission signal at the receiver input threatens to overwhelm the lower level received signal.

By providing transmitting and receiving amplifier pairs within a single IC, you can gain satisfactory performance with minimum package count. For example, Elantec provides two power line drivers and two receiver line amplifiers in the EL1501C subscriber line interface for ADSL (Figure 6). From a single-ended source, the drivers produce a differential output as high as 45V p-p into 200(ohm), letting you achieve data rates as high as 2 Mbps at 5 km. The receiver amplifiers cancel the transmitting signal, passing the received signal from the line, and presenting it as a differential signal. (You can convert the differential signal back to single-ended, if necessary, using a conventional instrumentation-amplifier configuration.)

The 80-MHz bandwidth drivers in the EL1501C slew as fast as 1000V/µsec, with output distortion lower than -60 dB and receive distortion better than -73 dB. As with all differential receivers, the ability of the receiver pair to reject the adjacent driver's output signal depends heavily on matching between the line impedance and resistors.

Compensated drivers eliminate the need for external components to maintain stability at the amplifiers' load and gain setting. These drivers are easier to use than are uncompensated devices and provide predictable performance.

However, you should also consider using an uncompensated amplifier with external compensation to maximize performance. You can develop non-unity-gain amplifiers that provide very high bandwidth and slew rates and adjust the bandwidth to yield lower overall noise. To optimize the compensation network, you must understand your driver and load, including the roll-off characteristics of the cable itself. Don't assume that less-than-expected bandwidth through the driver/cable combination is the fault of a driver that cannot handle the capacitive load of the cable, for example (Reference 5). By adjusting the driver-frequency response, you can counteract the cable roll-off and extend the effective system bandwidth to greater than that of the cable itself.

Some drivers have a built-in compensation network that you can include as you need it, tailoring it for capacitive or other loads. The LT1207 from Linear Technology Corp, a dual 250-mA, 60-MHz current-feedback amplifier for ADSL, HDSL, RGB, and video applications, includes an optional network for correcting the output-stage peaking of capacitive loads (Figure 7). Driving a 200-pF load and no compensation, the capacitance causes a 5-dB peak. With the internal compensation network active, and set by a 0.01 µF capacitor and lower feedback-resistor value, the response is flat to 0.35 dB and to 35 MHz.

Beyond the driver or receiver themselves, look at adaptive equalization as another compensating technique for correcting system deficiencies (see box, "Adapting is not just for Darwin").

You must match signal fidelity and the effects of driver, cable, receiver, and system compensation to your application. Analog signals require more accurate recovery than do digital signals, which can be retimed and regenerated. Regardless of the signal type, an eye pattern lets you qualitatively and quantitatively assess how well your system is transferring the signal before you resort to spectrum analyzers or bit-error-rate testers (Figure 8).

The eye pattern overlays all repetitive waveforms and bits onto each other on a synchronized display. Besides providing a clear summary of signal-timing jitter, noise, distortion, and margin, the eye pattern lets you instantly see the impact of changes you make in any circuitry or any equalization schemes you implement. The eye pattern provides a real-time window into the signals your equipment sends and receives and how any aspect of the link affects those signals.

Table 1 shows just a sample of available line-driver and -receiver ICs. Don't let functional descriptions, such as "video driver" or "ADSL driver," constrain your choice. Depending on your application, devices from other categories will also likely meet your combination of requirements. Also, because the IC architectures are similar and the various applications have much commonality, devices often satisfy more than one application.

You can reach Technical Editor Bill Schweber at (617) 558-4484, fax (617) 558-4470, email bill.schweber@cahners.com

References

1. Albrecht, Alan, and Patricia Thaler, "Introduction to 100VG-AnyLAN and the IEEE 802.12 Local Area Network Standard," Hewlett-Packard Journal, August 1995, pg 6.
2. Kester, Walt, Editor, Practical Analog Design Techniques, Analog Devices Inc, 1995, Chapter 2.
3. Coles, Alastair, et al, "Physical Signaling in 100VG-AnyLAN," Hewlett-Packard Journal, August 1995, pg 18.
4. Wright, Maury, "Delivering digital video," EDN, March 14, 1996, pg 39.
5. High Speed Signal Processing Seminar, Harris Semiconductor Corp, 1994, Chapter 2.

Acknowledgment

Thanks to Ken Fields of Elantec Inc, Mark Gordon and Mike Sedayo of Linear Technology Corp, and Charles Allen and Steve Pratt of Maxim Integrated Products for their insight and comments.