A New Level of Integration in RF Components
By Mike Press and Mark Moffat, RFMD®

By the title of this paper, one might assume that it is about to describe yet another system-on-a-chip (SOC) design, or even something beyond that. Not so. The intention is to describe a level of integration below that of an SOC, and explain why this matters.

Bluetooth® was the first high-volume wireless standard which was implemented in an SOC, followed by wireless LAN. Cellular SOCs are a relatively recent phenomenon and only started volume shipments in the last year or so, though highly integrated transceivers have been shipping for years (see Figure 1). The common characteristics of all three are that they target very high volume markets (tens or hundreds of millions per year, see Figure 2) and adhere to well-defined industry standards.

Given these very large volumes, it is not hard to see why the cost to develop highly integrated ASICs can be justified. These SOCs are one extreme of the RF integration continuum.

At the other extreme, components rather than systems solutions, RF component integration has not advanced much in the last 10 years. Designers working on systems which are not served by dedicated, application and standards-specific chipsets or SOCs are still purchasing discrete mixers, amplifiers and synthesizers, much as they have always done. Vendors of these components often provide system-level block diagrams showing different part numbers for each functional block – usually with no one component having more than a single function.

In the superhet radio block diagram shown in Figure 3, the functions typically implemented as separate blocks are shown in green boxes. In this example, seven separate integrated circuits (ICs) are needed, though some systems could have more.

Extremes in Integration
Why is there a large difference in integration levels? Why are some systems completely integrated, and others not at all? Well, the cost to develop these chips is vastly different. For example, development of a complete cellular transceiver could cost on the order of $30M, and the resulting IC is optimized for just one application and standard — an ASIC. A typical component-level product would have a development cost of under $1M, but can be sufficiently general purpose in nature to be applied toward many different markets (i.e., it is not application specific). Further, there are many markets that are too small to justify the expense of developing an ASIC, and although many of which do not comply with any particular industry standard, they often require proprietary implementations of a desired function. Some such markets are listed in Figure 4.

Another reason why the general purpose market has not progressed much might be due to the RFIC industry’s direction in the 1990s. During this period when new standards started to emerge, many of which promised big volumes, RFIC design expertise was relatively scarce. Many companies that had this needed capability focused deployment of their engineers in high-volume markets, leaving the component market relatively underpopulated. Consequently, the component market was not under as much competitive pressure to innovate. In fairness to those vendors, some of those innovations are expensive to develop.

RFMD® has developed several cellular transceivers, shipped over 100 million of them and generated hundreds of millions of dollars in revenue in doing so.

The technology developed in those cellular programs can now be leveraged to implement innovative products for lower volume markets. The challenge is: How do we offer higher levels of integration than are currently available without making the target market too narrow to generate a return on the investment? We do not want to make another ASIC, but rather, a device that offers smaller markets some of the desired integration technologies developed for high volume applications.

Optimized vs. Flexible
An optimized design is required in hyper-competitive markets, where the pressure on the usual key performance parameters (e.g. power, size and price) is most intense. The level of integration is so high that it is improbable that such a product can be adapted for any other than the target application. Indeed, since software is often a part of these products, and suppliers avoid writing custom code for niche applications, the authors of this paper are unaware of a single instance of a true SOC being used in anything but its intended application. Which is to say, it is close to 100 percent optimized, and 0 percent flexible.

What Makes a Radio an ASIC?
The major system-level factors which make a radio specific to a particular application are:

• carrier frequency
• dynamic range
• channel bandwidth

Most RF circuits are inherently broadband, so they will work from around DC up to the maximum practical frequency provided by the process technology. Dynamic range is implemented by gain in the receive and transmit chains. Gain is relatively cheap to add since gain blocks are low cost and plentiful. The limitation of channel bandwidth is usually set by fixed filters, such as SAW filters, which are implemented off-chip.

Integration Direction
Typically, integration is done in a horizontal fashion – the receive chain is integrated and the transmit chain is integrated. Integration in the horizontal plane requires that the IC designer determine the gain, carrier and bandwidth in advance, and constrain them. Taking this approach results in an ASIC, the exact, inflexible approach we want to avoid.

RFMD instead decided to integrate vertical slices through a radio block diagram. This way, broadband components common to all radio systems could be integrated without any limitation on applicability in multiple markets. The components which are not integrated are only those that would limit the type of system that could be designed (i.e., the narrowband channel filters and gain blocks). Since these systems usually use filters that cannot be integrated anyway, they simply need to add gain blocks, which are among the lowest cost and most broadly available of RF components.

Using this approach, RFMD has changed the equation in the “integration vs. flexibility” tradeoff.

The first example of this approach is in a family of products which integrate common RF front end functions. The IC offers a higher level of integration than is currently available (a minimum of one package is saved compared to any other implementation) without any compromise in flexibility. The radio designer is given access to a broadband mixer (or two in the duplex version), and a fractional-n synthesizer with VCO.

Multiple VCOs and LO path divider options allow the device to operate with a LO signal from 300MHz up to 2.4GHz, RF and IF signals from 100MHz to 2.5GHz – the most broadband product of its type on the market. However, the device is easily applied to narrowband applications. Channel bandwidth is set with SAW filters. VCOs are internal, but an external VCO can be used, if desired, for better phase noise than can be obtained from the on-chip implementation. The synthesizer is a fractional-n type, capable of tuning to a precision of 1.5Hz across the entire band. Phase noise is typically -135dBm/Hz at 1MHz offset for a 1GHz LO. In-band noise depends on the loop filter bandwidth and reference frequency, but an RMS phase error of 1.3 degrees can typically be achieved at 2.4GHz, with improved performance at lower LO frequencies. The reference frequency input can use a crystal of 10MHz to 52MHz, or can be overdriven with an external reference signal of 10MHz to 104MHz. Internal reference dividers can be programmed to reduce the internal reference frequency for the device. The maximum phase detector frequency is 52MHz. The loop filter is off-chip, allowing it to be optimized for the application. A high performance op-amp is provided on-chip, allowing active loop filters to be implemented (this may be disabled for applications in which a passive loop filter is required).

Mixer linearity is programmable up to +20dBm. At maximum linearity, the mixer, PLL and VCO together consume 72mA from a 3 volt supply, which can be further reduced for lower linearity requirements. The device is supplied in a 5x5mm package.

Three versions of the part will be available:

• Dual mixer version for Half/Full-Duplex or Diversity (RF2051)
• Single mixer version (RF2052)
• Single mixer version with external VCO capability (RF2053)

Full duplex operation is achievable using the RF2051, provided that the required LO frequency is the same for both Rx and Tx mixers. Half-duplex mode is also achievable on this part. Two sets of PLL registers are provided, so that switching between two LO frequencies is achieved by simply toggling the state of a hardware pin. For applications only requiring a single mixer or requiring higher mixer isolation than can be obtained from a monolithic implementation, the RF2052 is available.

Because the PLL is a fractional-n design, simple digital modulation can be performed using the device, as long as the modulation bandwidth is less than the PLL loop bandwidth. FSK can be implemented by programming the two required frequencies into the two PLL registers and toggling an external pin at the required rate. Even GMSK can be implemented by continually streaming the n-divide ratio into the part via the serial bus.

The RF2051 is shown in an example application in Figure 6. The device can be used in front of a transceiver chipset to relocate it to another frequency band. The simple elegance of the implementation is evident. Tools are provided with the device to assist in programming the synthesizer and determining the best match to the mixer ports.

The device could also find application in software-defined radios and diversity systems because either mixer can be used in up-conversion or down-conversion configurations, and can be operated over a wide range of frequencies.

There are many advantages to this approach of integrating commonly used radio functions in a flexible manner:

• Fewer packages mean a lower solution area. We expect designs using this device can be implemented in half the size of discrete, component-based solutions.

• Faster radio development time. By integrating the RF blocks, we also take care of the interfacing between them. This ensures that the mixers are always driven with the optimal LO power. Shorter interface paths also mean less LO-RF isolation and other layout-dependent RF issues. Because the part is broadband in nature, a radio design can be re-used for a different frequency application by simply changing external filtering and matching components, thus reducing the cycle time.

• Higher yielding final products. The overall performance of the part is guaranteed. This is an improvement over a discrete design in which the designer needs to verify and prove that acceptable performance is obtainable from the design for all combinations of performance variation for each of the individual components used.

Overall, this partitioning is expected to save designers’ time, reduce risk and board area without any compromise in flexibility or performance.

The products mentioned above are sampling now, with production volumes available around the middle of 2008. Other products with similar benefits will be announced later this year.

Authors
Mike Press is a strategic marketing manager at RFMD. Mark Moffat is the director of the Emerging Products Line and has been with RFMD for nine years, working on products for the broadband and cellular markets.

RFMD
www.rfmd.com
TXTLINX.COM124
Email this article to a friend!