Design Features November 21, 1996

 PCS RF-stage design: tackling the trade-offs

Bill Schweber, Technical Editor

  If the true challenge in any competitive engineering design is juggling the conflicting trade-offs among different approaches and components, then RF-stage design requires the skills of a master juggler. You have to consider issues of efficiency, biasing, impedance matching, linearity, passive component location, noise, and even type of modulation as you try to achieve that "best" compromise.

  Adding to your design challenge, RF circuitry does not forgive little things. The small values of stray capacitance and inductance that you can safely ignore at lower frequencies (for example, below 100 MHz) are significant factors in the gigahertz area; changing the placement of a component by even a millimeter can cause headaches. Inconsistencies in material characteristics within a single circuit board, or among different batches, can cause variations that upset a carefully tuned design.

  In the past, you could overcome many RF-design difficulties by tweaking a few components, which was practical because production volumes were lower. Today's RF circuitry uses mostly tiny, nonadjustable components, crowded into surface-mount designs that leave little room or opportunity for adjustment. The high production volume of Personal Communications Service (PCS) devices, such as cellular phones, spread-spectrum systems, and even simple keyless electronic locks, means that designs have to work "as built," not "after adjustment."

  The good news is that the newest RF-power amplifiers in the 1W range and their complementary front-end components provide far better performance than do previous generations. They are easier to use, require fewer pesky support components, and come with more thorough specifications. You can evaluate discrete, modular, and IC approaches to find the best performance and cost fit for your situation. In addition, improved component models and design tools let you obtain a much better preliminary analysis of your basic design and illuminate areas of potential difficulty.

Class and dynamic range

  Whenever you have multiple design trade-offs, begin to resolve them by identifying those areas in which you have fixed requirements, either by choice or adherence to a standard. For example, the type of modulation you use affects potential transmitter stage efficiency, because it constrains the biasing of the amplifier and its operating class. In general, unfiltered Class A is the most linear but least efficient, and Class D is the least linear and most efficient; common modes, such as Class AB, Class B, Class BC, and Class C, define intermediate positions on the linearity/efficiency trade-off scale.

  The US cellular Advanced Mobile Phone System (AMPS) standard uses FM and therefore can tolerate the distortion of Class B or BC, with typical resulting efficiency of 50 to 60%. Similarly, the widely used Global System for Mobile (GSM) standard uses an FM variant. In contrast, the newer time-division multiple access (TDMA) system (IS-54/IS-136) uses a version of differential quadrature phase- shift keying (DQSPK) modulation, which demands better linearity and, thus, Class A or AB design with efficiency of 40 to 50%. Most stringent is the spread-spectrum code division multiple access (CDMA) system (IS-95) now being field tested in the United States, which uses a version of offset QSPK modulation. To maintain phase information and lower the noise floor (which affects the number of supportable simultaneous users in the band), this technique needs the high linearity of a 30 to 40% efficient Class A design.

  Vendors are responding to these needs with power amplifiers characterized for these specific applications. For example, the AM52-000x family from M/A-COM has eight-pin SOIC devices for AMPS, GSM, Digital Communications System (DCS)-1800, and Personal Handyphone System (PHS) systems. Figure 1 shows the schematic and performance of the AM52-0001 member when configured for AMPS applications.

  Modulation type and amplifier class are just part of the picture. You also have to consider what other demands the system puts on the amplifier. Cellular phones vary their power output, primarily to restrict operating range and, thus, define cell boundaries. The required dynamic range varies: it is 20 dB for AMPS, GSM, and IS-54, but 60 dB for IS-95. Your design must have sufficiently low distortion at these levels. You can choose to implement the simple but less power-efficient technique of varying the drive to the power amplifier or the more complex but more efficient approach of varying the amplifier gain itself to achieve the dynamic range requirements. The cellular standards also define allowable distortion and out-of-band spectral components, for which you have to carefully plan.

  There are still more complications to the efficiency story. Unlike AMPS, the TDMA transmit duty cycle is not 100%, and TDMA and CDMA power levels are constantly changing during a call. Therefore, your overall power usage is less than a worst-case, maximum-output calculation indicates. You also must control the power ramp-up and ramp-down slew rates to meet the standard's specifications, intended to minimize harmonic distortion while quickly bringing the phone to its active mode.

  Although cellular applications are driving many RF-component ad-vances, not all applications are cellular. RF designs span low-power, relatively simple applications, such as keyless locks using simple modulation (even plain and inefficient AM), to point-to-point wireless links that are designed to replace cabled paths. In these situations, you may have more choice over the modulation scheme and signal characteristics, which complicates your analysis but gives you the opportunity to reach a better overall solution.

  Biasing defines the operating point and class of the RF stage. Not only must it be at the proper design point for desired efficiency and distortion, it must also work with inevitable changes in supply voltage and temperature. You can even develop an amplifier configuration designed to let you easily change the bias point, to shift operation between classes for different applications and efficiency trade-offs.

  Increasingly, RF power-amplifier vendors are incorporating bias circuitry into their devices. This circuitry simplifies your design, especially for GaAs devices that otherwise require a negative bias voltage derived from a dc/dc converter just for this purpose.

  However, there are situations in which either you want more direct control over the bias point or your chosen amplifier does not integrate the bias circuitry. Analog IC vendors have addressed this issue with devices that provide improved bias performance. It's especially critical for Class AB amplifiers, which provide fairly low distortion (at much lower quiescent power levels than Class A) but only when biased around a fairly narrow operating point (Figure 2); not surprisingly, this point moves with temperature.

  IC vendors are addressing the bias problem to simplify the bias circuitry design and improve performance. For example, the MAX840 series from Maxim Integrated Products consists of low-noise, low-ripple devices for biasing GaAs FETs. These ICs provide a fixed –2V or adjustable –0.5 to –9.4V output, at as high as 4 mA, and operate from supplies between 2.5 and 10V. Linear Technology Corp addresses biasing for high-power devices with LT1166 automatic bias IC, which sets the Class AB operating point and tracks temperature changes. It has two independent loops that act in concert. A voltage-control loop maintains output voltage at the input value, and a current-control loop keeps the bias current at the desired value. Motorola's MDC5000T1, in an SO-143 package, acts as a dc feedback element and maintains a stable bias current for bipolar and field-effect transistors, operating with supply voltages as low as 1.8V.

  Some vendors have integrated biasing circuitry with other power-amplifier functions. The TPS9103 from Texas Instruments combines the negative bias supply with a high-side p-channel MOSFET switch that controls power delivery to the amplifier. Internal logic prevents the switch from turning on unless gate bias is present, to avoid destroying the power device.

It's nice to receive, too

  The receiver front end, often called the low-noise-amplifier (LNA) stage, faces some of the same trade-offs as the power amplifier but has several key differences. The transmit function is deterministic, taking a known, well-behaved internal signal and amplifying it before passing it to the antenna. In contrast, the LNA must capture a relatively weak RF signal corrupted by noise, fading, interference, distortion, and other poorly known characteristics.

  Power efficiency is not a major factor at the minuscule LNA signal levels, but other factors, such as internal device noise figure, gain, input return loss, and third-order intercept and intermodulation products are important. In full-duplex applications, which are increasingly common, you usually have to trade lower noise figure for desirable increased return loss. You can do this trade via the matching circuitry between the antenna and the amplifier input. For example, when matched for best noise figure, the NEC NE34018 GaAs heterojunction FET for 1-to-3 GHz applications yields 15.8-dB gain, –1.7-dB input return loss, and a noise figure of 0.62 dB; when matched for best return loss, the corresponding specifications are 20.6 dB, –17.9 dB, and 2.70 dB.

  When the potential product volume is high and the signal characteristics are carefully bounded, such as in the low-power license-free 2.4-GHz ISM band, vendors are providing devices that combine the power amplifier and LNA into one device. The 24-lead SSOP package TQ9205K from TriQuint Semiconductor, for example, provides an LNA and a power amplifier, along with a pair of transmit/receive (T/R) switches, for half-duplex operation (Figure 3). Its 21-dBm output power is less than the 1W maximum allowed by FCC specifications but adequate for a majority of ISM applications.

  Good RF design in the gigahertz region requires extra attention to detail. Passive components used in impedance matching, bypassing, and channel spectral filtering must have negligible and consistent internal parasitics. Fortunately, most of these components are now available in smaller sizes and surface-mount packages, which has the potential for greatly reducing stray parameters.

  You have to use any of these devices with care, because placement severely affects actual performance. Consider using a reference design (Reference 1), or adapting one if at all possible, because this design will probably save you time and frustration. Recognize that even a small change you make to a working design, whether conscious or not, may significantly change its operating characteristics. Even inconsistencies in base PC-board material and cladding can cause problems that are difficult to explain.

  Be sure to evaluate impedance matching, usually to 50ohms, early in your design. Most LNA outputs and power amplifier inputs are designed with this nominal value, but you probably need to match the LNA input and power amplifier output to your antenna impedance for maximum power transfer. RF filtering between the active stage and the antenna is also a factor. You may find yourself dealing with Smith charts and unusual passive-component topologies. Don't hesitate to seek help from experienced vendor application engineers who may have already solved similar interface challenges.

Everything is going to ICs, right?

There's no question that the high volume of cellular applications, coupled with process and device improvements, is driving most designs toward ICs for power-amplifier stages. These ICs offer superior and guaranteed performance, small size, ease of use, and increasingly high levels of integration. Even a moderate level of functional integration, such as an amplifier with a transmit/receive (T/R) switch, can save you space and money and still provide known performance.   In other cases, the increased integration doesn't provide different functions in the same package, but it does provide closely coupled multiple stages. The AWT-1900 power amplifier IC from Anadigics, for example, incorporates three stages of gain. To the designer, this 28-lead SSOP IC provides output power levels settable from –42 to +32 dBm with a 7-dBm input (figure). For outputs above 20 dBm, the efficiency of this device at 1800 MHz is better than 45%.   But don't rule out modules and discrete approaches. Depending on your volume, application specifics, and available time, you may find that you can design a lower-cost amplifier using these techniques. If your application has some unusual requirements that standard ICs do not satisfy, or if you need to allow for some adjustments for system uncertainties, discretes allow you to quickly get a tailored preliminary design in place with maximum flexibility. (Of course, making that design production-ready takes more time.)   Discrete designs also let you find alternate sources for most components. With general-purpose active devices, you may achieve a lower total-system cost than you get when attempting to adapt ICs optimized for other applications. Discrete designs also let you mix different active and passive technologies.   Multichip and hybrid modules offer you a middle approach, combining some of the features of discretes with the benefits of more integrated solutions. You can combine mixed technologies and design flexibility into a single installable component. This approach is especially practical for lower volume needs.   For the receiver front end, which has fewer design trade-offs than the power amplifier, discrete designs have a long and successful history. A well-designed low-noise amplifier using standard discrete devices may be the lowest cost approach that provides satisfactory performance for many applications.

PCS RF-stage design: tackling the trade-offs

The standard plastic DIP or SOIC package is often unsuitable for moderate-power RF amps, because the packages cannot handle the dissipation requirements and have too much stray inductance and capacitance. Therefore, the industry has traditionally used ceramic packages with stripline leads and heat-dissipating mounting surfaces and tabs (Figure A).   Driven by the need for low cost, and the well-defined, lower-power applications in cellular phones and spread-spectrum systems, vendors are increasingly shifting to plastic packages that resemble modified standard SO-8 and SO-16 devices. These surface-mount devices almost always combine some of their adjacent leads to provide a better heat sink and mounting tab than individual leads offer (Figure B).   Many packages also have a copper slug at the bottom, which is soldered to the circuit board for efficient heat transfer. OEMs have differing opinions on the slug technique: It offers very good thermal and RF impedance characteristics, but it hides the solder joint between the IC and the board. Some production and quality departments think such a hidden joint is poor practice; others think that good solder technique is more important than the ability to inspect each joint. Check what works, or is acceptable, in your organization.

Reference

1. Schweber, Bill, "Reference designs reshape design engineering," EDN, May 9, 1996, pg 73.

Acknowledgments

Thanks to Sam Hammond and Chris O'Connor of TriQuint Semiconductor Inc, Robert Bayruns of Anadigics Corp, and Joe Grimm, Mark Navarre, Walter Lau, and Steve Morris of California Eastern Labs for their time and insight.

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

| EDN Access | feedback | subscribe to EDN! |