October 23, 1997

RF-component diversity complicates design

Bill Schweber, Technical Editor

Designers of RF circuitry now have too much of a good thing. Discrete components, building blocks, system-level ICs, and even ASICs can meet your application requirements, but each choice carries important trade-offs in performance, cost, and flexibility.

RF design has always been a technical challenge, but your general strategy has been straightforward. You would knit together discrete components, such as transistors, supplemented by a few building blocks and a handful of passive components, to craft a clever configuration that would meet system objectives after much trial, error, and tweaking. "If only there were an easier, quicker way," was the common lament of the small core of capable RF designers.

These lamentations were before the explosion in wireless applications, especially high-volume ones, such as cellular phones. Today's situation brings to mind the old saying, "Careful what you wish for; you just might get it." The growth in applications has sparked a growth in RF-component choices. IC vendors previously provided either multifunction analog and mixed-signal devices that were suitable only at low to moderate frequencies or single-function components that worked for high frequencies. Now, these vendors apply their processes and design techniques with substantial success to the analog world of RF design. At the same time, traditional RF-component vendors have broadened their technologies and have a wider array of design choices.

Now, you begin your RF design by evaluating four fundamental approaches to the system design and deciding which one or combination best meets system requirements and trade-offs. Each choice brings trade-offs in performance, achievable specifications, cost, time to market, vendor selection, and many other factors. Inevitably, the answer to the question, "What's the best approach?" is familiar: It depends.

Range of choices

If your quantities are more than 100,000 to 500,000 and your needs are well-defined and fixed, you can consider working with an IC vendor that does as much of the circuitry as possible in a single custom device. This ASIC implements as much of the RF circuitry as possible with one IC process and may provide the lowest cost option in the long run.

That's the catch, though. In the fast-changing RF and wireless arenas, the "long run" may be too long to justify the up-front expense, time, and design risk of an ASIC. In addition, the single-process, single-IC constraint may limit the performance you can achieve and mandate unacceptable compromises to your target specifications. The ASIC option is usually not viable for most applications, especially complex ones, because of quantity, risk, or performance factors.

A step below the ASIC in functional completeness are the system-level ICs and chip sets. Here, a single large-scale, standard-product, mixed-signal IC or several acting in concert as a chip set comprise several major functional blocks of the overall system. Although you can use any one of the chip-set ICs individually, they normally function as part of the group of four to six ICs that nearly provide a complete system implementation.

The building-block IC, which performs one analog-RF function or implements a complete stage in the signal-flow path, offers low to medium integration. Typical examples of these building blocks are PLLs, IF stages, and variable-gain amplifiers. With these blocks, you gain flexibility by mixing processes, ICs, and components from various vendors to more closely match your target specifications. You also make performance trade-offs and optimizations that fit your situation. This approach, though, requires more ICs and passives than do the system-level-IC or chip-set options. You can implement the building blocks in a hybrid or multichip module (MCM), which provides some of the virtues of a single-chip ASIC but with lower risk and design time, using vendors such as Maxtek Components Corp; you can also consider the hybrid or MCM approach for the chip-set installation.

Finally, you can construct the signal-processing functions you need from discrete devices, such as basic transistors. This approach makes sense for functions such as RF front ends or power amplifiers and is often practical for mixers or other amplifier functions. Clearly, with the discrete approach, you have the ultimate in flexibility and in the ability to tailor the design to your needs. It's hard to compare final component count with the building-block-IC approach, because for some applications, the discrete design is a sparse and cost-effective implementation; for others, it is not.

Sources are critical

Whichever path you choose, you may need to deal with unfamiliar vendors. If your experience is in the RF world, you'll find respected IC vendors offering RF components; if you have mostly analog-IC exposure, you'll see that many established vendors have supplied the RF-design community for years.

But choosing a source is not just a matter of making new acquaintances. In general, philosophical, architectural, and terminological differences also exist between the vendor groups. For example, RF vendors tend to talk in terms of dBm, but IC vendors look at a design's voltages, impedances, and ability to drive loads at given points. If you mix components, you have to convert specifications and parameters at appropriate points.

One way to minimize your effort is to use a chip set from one vendor. This approach differs from using ICs from one vendor, because a chip set supposedly comprises ICs that are designed and tested to function as a group. If you use ICs but not a chip set from one vendor, you may find that the ICs have timing and interface incompatibilities. Some of these non-chip-set ICs may be 100%-tested for key specifications, and others may be mostly "typical"; again, you need to carefully study those data sheets.

Another issue is system architecture, which RF-design philosophy drives. Based on their experience; available technology; and what they see as the greatest obstacles to good system implementation, such as noise or insufficient isolation, traditional RF-component vendors have developed successful approaches to dividing the overall system into functional blocks. For example, they may prefer to separate blocks to allow measurement and adjustment of critical parameters that are hard to control. They also try to optimize blocks, because the signal-processing chain's performance is usually no better than its weakest link.

In contrast, IC vendors look at what they can do within the confines of an IC layout. It's easier for them to control impedances within an IC, for example, than it is for you to do so on a pc board. IC vendors also look at how they can compensate for deficiencies in one internal block by changing the performance or design margins within another block in the same IC. Because the end user sees just the IC pins, these internal trade-offs are often inconsequential to you as long as the signals at the inputs and outputs meet target specifications. Sometimes, traditional discrete and building-block device specifications, such as distortion or SNR at key points in the signal-processing chain, are not even available in an IC design, which concentrates on black-box I/O performance.

IC and building-block device vendors have the luxury of potentially employing a relatively large number of active devices to accomplish in monolithic form what would be impractical in discrete designs. Rather than just trying to reproduce a discrete design as an IC, vendors can look at topologies that take advantage of the attributes of monolithic design, such as using devices that inherently have matching temperature coefficients or adding a few transistors to avoid needing passive components with impractical values.

You cannot solve all problems within an IC. In fact, the IC, with its single die, makes some problems potentially far worse. The local isolator signal can easily get into a sensitive front end, or the low-noise amplifier can get feedthrough from a mixer product. You must examine the performance not only under intended conditions, but also under undesirable but common ones, such as overload or strong adjacent-channel signals, to see if the component is sufficiently well-behaved.

Finally, consider reuse. If your IC has many functions, you may be unable to apply that device again in a different design unless much commonality exists between them. With lower level functional blocks, such as a small-scale IC or even a discrete-based design, you may be able to reuse or adapt parts that you successfully implement and put into production when it comes time for a second generation or new application. This approach can save you development and test time, but recognize that these virtues may be irrelevant in the fast-changing world of RF applications and components.

Alternatively, partition your design so that functions that you can most likely reuse are separate from application-specific functions. IF stages, for example, have a lot of commonality across disparate applications and situations, so they may be a good candidate for this partitioning.

GaAs vs silicon: continued

Vendors of silicon-based and those of GaAs-based ICs each post strong arguments in favor of their technologies for RF functions. In general, GaAs produces better inductors, lower noise front ends, and better switching circuitry compared with silicon-based ICs. However, silicon vendors can design around and compensate for some of these weaknesses, which may be sufficient for your needs. The vendors can also put extra functions onto one IC. A silicon device tends to be less expensive than a comparable GaAs device, but there's the rub: It's hard to find truly comparable devices. When you look at system costs, the difference between the two technologies is often small in the popular 800- to 2800-MHz region.

The trend toward 3V operation from 5V also affects the silicon-vs-GaAs equation. Although GaAs is generally better than silicon at 3V, silicon can usually deliver sufficient performance for many applications. And, although it may seem to defy intuition, IC designers, with the support of good application engineering, maintain good dynamic range and SNR, despite the decrease in operating-supply voltage from 12 to 5V and now to 3V.

If your system involves a transmitting stage and a power amplifier and you are battery-constrained, you need to carefully look at that stage. The power amplifier consumes the most power, so an efficient design and the GaAs-vs-silicon choice are more critical. Note that the power amplifier in an Advanced Mobile Phone Service (AMPS) cellular-phone system may require 45 to 50% of the circuit power. You may decide to use a slightly more efficient GaAs device or boost the power-amplifier supply using a $2 dc/dc converter to achieve greater operating efficiency.

Increasingly, RF vendors don't limit themselves to choosing silicon or GaAs but offer variations of both technologies. This approach either allows vendors to put together a portfolio of discrete devices and functional building blocks that lets you choose the best trade-offs or to provide chip sets that use both technologies as needed.

Before you choose a custom IC, chip set, building block, or discrete device, ask yourself some tough questions. On the application and marketing side, start with determining your anticipated volume, time to market, and product life cycle. Also, you must differentiate your product from competitors' products. Be honest about your own technical expertise, too: How much RF-design knowledge do you have, and how much time do you have to obtain more knowledge if needed? Is this an area in which you want to develop expertise, or would you rather devote your project team's hours to some other product-design factor?

Work on the target-product specifications, identifying the priority parameters and determining their desired and minimum acceptable values. For example, in most communications systems, a better RF specification makes your product stand out. If you build radar receivers, you want more front-end sensitivity and lower noise. But, although improvements are good, they reach a point of diminishing returns, or the user even develops a "don't-care" attitude, such as when one cellular phone has a better front end than another. Sometimes, meeting the specification is good enough, and there is little point in striving to exceed it. In the real world, your compromise may not matter if the price is right.

Also consider the flexibility and freedom you need. If one performance specification must really stand out or if you need to adjust it depending on circumstances, you may find that the discrete devices or lower level functional blocks are the best choice. In many applications, they may be the only choice for meeting your requirements, because most ICs with higher levels of integration, by design, do not allow for many choices beyond their basic ones (see box "Radio frequency‚megahertz").

Finally, you can't ignore testing. Although testing is part of every design effort, communications-related or not, a communications system usually must meet some regulatory standards in addition to whatever test you plan to do. If you buy an ASIC or chip set, much of this test burden falls on the vendor; if you use building blocks or discrete components, all of the testing needed to meet regulatory approval is part of your challenge, and you need to plan and budget accordingly.

Table 1 looks at the four design approaches and how each one relates to various design criteria; you can undoubtedly find exceptions to these descriptions. Note that the trend along a row is not necessarily monotonic, steadily increasing or decreasing as you traverse from discrete devices to single-IC ASICs. For example, system costs are generally higher with discretes than they are with the other choices, but in some applications, a clever design and a less-than-$1 discrete can outperform a sophisticated IC.

Overwhelming choices

Although diversity may seem like the hot buzz word these days, it's a real aspect of RF-component selection. The breadth and depth of available components can cause you to spend much of your project time deciding which component category and components within that group make the most sense for your application. Some vendors concentrate their offerings in one group and some of their offerings in a similar, second one; others have a portfolio that spans from discrete components to chip sets and even ASICs.

For a design engineer, working with a narrower range supplier has advantages and disadvantages when it comes to focus, experience, expertise, and commitment. Recognize that some vendors of highly integrated building blocks or chip sets prefer to deal primarily with large-volume OEMs and are ill-prepared to support lower volume applications or those designers who seek to adapt a targeted component to another, somewhat different application. Be sure that the vendor offers basic support--an evaluation kit for a building-block component and a reference design for chip sets. Otherwise, you waste time uncovering subtleties that the vendor should already have found. The evaluation kit for M/A-Com's AM50-0003 wide-dynamic range, low-noise amplifier for 800- to 1000-MHz cellular-phone applications represents such a demonstration board (Figure 1).

A sampling of vendor devices gives you an idea of the range of choices available for basic RF-design strategy. Anadigics, for example, offers the AWT1903 GaAs power amplifier, which meets the linearity requirements of 1900-MHz CDMA applications. With 26-dB gain and 28-dBm output power, the 5V amplifier uses three Class AB stages and an adjustable bias circuit that lets you optimize efficiency at different output-power levels.

If you need an IF subsystem, the 3V AD608 from Analog Devices comprises a mixer, a logarithmic/limiting amplifier, and an 80-dB-range received-signal-strength indicator (RSSI), providing a hard-limited, 400-mVp-p output for Personal Handiphone System (PHS), Global System for Mobile communications (GSM), TDMA, and similar applications (Figure 2). In a typical application, the AD608 receives the 240-MHz output of a surface-acoustic-wave (SAW) filter and downconverts this output to a 10.7-MHz IF signal with 24-dB gain.

For working at the basic level, NEC's UPA8XX series, which is available through California Eastern Laboratories, provides dual-transistor arrays in a 1.25×2-mm, six-pin package. The transistors' pinouts are independent on the package, so you can use them as dual transistors or in cascode arrangement (Figure 3a), and some versions are also available in a cascade variation (Figure 3b). The cascode UPA801T and UPA810T offer 7-GHz f_T and 1.2-dB noise figure at 1 GHz, which allow you to achieve amplifier performance of 16-dB gain, 0.2V V_CE, and 1-mA collector current.

Applications such as keyless-entry systems do not need much flexibility, but they do need compactness and low power. GEC Plessey Semiconductors' KESTX01 amplitude-shift-keying, transmitter for 400 to 460 MHz combines a VCO, a PLL, and a 1-mW power amplifier in a 14-pin package; the corresponding KESRX01 is a single-conversion superheterodyne receiver in a 24-pin package. Using the pair, your design can achieve range of about 50m and data rates as high as 5 kbps.

For more complex applications, such as wireless LANs and personal-communications-service (PCS) systems, the Harris HFA3724 IF/quadrature modulator/demodulator IC (Figure 4) operates in both transmit and receive directions between baseband signals and IFs spanning 10 to 400 MHz. The 5V, 80-lead IC integrates a variety of requisite filters and amplifier stages. The IC's I/Q amplitude and phase balance are 0.2 dB and 2º, respectively.

Featuring moderate functional integration, the 1.5- to 2.5-GHz HPMX-5001 upconverter/downconverter from Hewlett-Packard handles both directions of the RF-to-IF path for a variety of applications, such as Digital European Cordless Telephone/ Telecommunications (DECT); PCS; and industrial, scientific, and medical (ISM) band handsets and base stations. The device includes a VCO, a PLL prescaler, and mixers (Figure 5) and provides 2-dBm transmit output at 1900 MHz. You can also use the companion HPMX-5002 IF modulator/demodulator in your design.

ICs are also making some well-known architectures practical that for years were too expensive or quirky to implement with other approaches. Maxim's MAX2102 directly tunes the range from 950 MHz to the 2150-MHz L band and brings the received signal down to baseband using a broadband I/Q downconverter. The 5V, 28-pin SOIC includes a low-noise amplifier with companion 50-dB-range AGC, two downconverter mixers, an oscillator buffer with direct and 90° outputs, and a PLL prescaler. The device targets carrier powers of ­19 to ­69 dBm and has a 13.2-dB noise figure.

Expanding applications, such as Global Positioning System (GPS) receivers, give vendors the incentive to provide devices that, although targeting one application, you can use in others as well. Motorola's MRFIC1501R2 GaAs low-noise amplifier, for example, operates as high as 1.6 GHz for GPS, but you can also use it for ISM and mobile- radio bands (Figure 6). With a 5V supply, the SO-8 device provides 18-dB typical gain, has a 1.1-dB noise figure, and consumes 30 mW.

Even vendors whose reputations are primarily for their passive devices are expanding into basic active components and building blocks. Murata offers a family of GaAs low-noise amplifiers that you can use at 2.4 GHz for front-end amplifiers or drivers for the power stage. The company's XMIA2-M5 consumes 6 mA at 3V and provides 16-dB gain with a 1.7-dB noise figure.

CDMA cellular phones require highly linear transmit stages to avoid frequency spillover into adjacent channels, and some vendors' devices target these requirements. RF Micro Devices offers the high-power RF2146 linear amplifier for use in CDMA and spread-spectrum systems at frequencies as high as 2 GHz. This device produces 28-dBm output power with 18.5-dB gain with efficiency of 37 to 40%.

National's LMX3161 DECT radio-transceiver IC in a 48-pin package shows how vendors are moving up the integration ladder. The IC operates from 3 to 5.5V, with RF sensitivity as low as ­93 dBm and a system noise figure of 6.5 dB, and you can optimize its performance for either sensitivity or intercept point. The IC's transmitter includes a 1.1-GHz PLL and a 2-GHz frequency doubler, and the receiver within the device has a 2-GHz low-noise mixer, an IF amplifier, a limiting amplifier, an RSSI, and an internal voltage regulator that eases your supply's requirements.

Siemens provides a major building block with its TDA 6060XS multistandard modulator and PLL. This IC combines a video modulator and digital PLL, which sets the modulator frequency at 30 to 950 MHz in 250-kHz increments. It also includes a sound frequency and amplitude modulator, which is programmable for standard carriers, such as 4.5, 5.5, 6, and 6.5 MHz. The 28-pin TSSOP device targets VCRs, set-top boxes, satellite receivers, and similar applications.

For the power-amplifier stage in GSM phones, Temic offers a series of power amplifiers built as GaAs monolithic microwave ICs (MMICs). The TST0900, for example, provides 35-dBm output at efficiency greater than 45% from 880 to 915 MHz. Its internal dc/dc converter generates the required negative bias voltage and operates from a 5V supply. Other family members feature different output vs efficiency trade-offs and 3V operation.

Also using GaAs, Triquint Semiconductor has 3V receiver ICs for PCS-band CDMA applications. The TQ9228 for 1930- to 1990-MHz operation includes a low-noise amplifier, a mixer, and an IF-amplifier stage (Figure 7). It takes a low-side local oscillator and produces an 85-MHz differential IF signal. You can set the gain using an analog signal from 7 to 27 dB; noise figure is 2.6 dB at high gain and 16 dB at low gain.

Carefully investigate devices that vendors define as "complete," because this term is flexible. Some "complete" devices require a few discrete components, such as low-noise amplifiers or power amplifiers; most require many passive components, for example. Where you stand on the validity of the completeness claim depends on where you sit: If you are a vendor, the component may seem complete compared with alternatives; if you're an OEM, the number of passive and small active devices you still need may make you wonder about that claim.

Reference

1. Robertson, Dave, "Selecting Mixed-Signal Components for Digital Communication Systems--An Introduction," Analog Dialogue, Analog Devices Inc, Volumes 30-3 (1996) , 30-4 (1996), 31-1 (1997), 31-2 (1997).

Acknowledgments

Thanks to Doug Grant of Analog Devices Inc, Hans Dropmann of Maxim Integrated Products, and Mark Burkett of California Eastern Laboratories for their insight and ideas.

| EDN Access | Feedback | Table of Contents |