[For a copy of this article in PDF format, which displays figures and equations, click

here

. Requires Adobe Acrobat Reader,

free download.

]

The rapid growth in wireless technology has placed a variety of new demands for components on system designers. Because the frequency source for a wireless system is a central element to the design, it is often one of the first issues confronted during product development.

Three principal factors play into the selection of the frequency source for a wireless application:

• Accuracy: For wireless applications, the source must provide an accurate frequency with extremely tight tolerances, typically ±25 ppm or less.
• Cost: The component cost should reflect not only the per-unit price, but also the cost of any ancillary components and testing required to deliver equivalent performance.
• Size: Since many wireless applications involve portable or handheld devices, the frequency source should occupy the smallest possible footprint.

Frequency sources can be provided either as standalone crystals, which require an oscillator circuit to be incorporated on the IC; or as integrated, pre-packaged oscillators.

This article illustrates the inherent advantages of using an oscillator instead of a crystal, which in most cases will more than make up for the relatively minor per-unit cost advantage, if any, of a crystal-only approach.

At first glance, it might seem straightforward to obtain a crystal with the required accuracy for a given wireless device. However, it is important to look behind the methods used to measure these crystal attributes, and at the behavior of crystals under load, to truly understand the potential behavior of a crystal in a real-world application.

The two most important contributing factors in considering a crystal's performance are variances due to measurement methods and variances in load capacitance. The frequency of a crystal shifts in response to a change in load, an attribute known as trim sensitivity, usually expressed in units of ppm/pF.

Testing of crystals using a crystal impedance meter is performed at the desired load capacitance. However, the accuracy of the impedance meter is typically ±2 percent of the specified load. For a specified load of 20 pF, the load experienced by the crystal may vary from 19.6 pF to 20.4 pF.

Take a look at an application requiring a frequency source with an accuracy of ±25 ppm. At first glance, we could meet this specification using a crystal with a tolerance of ±10 ppm, a stability of ±10 ppm, and an aging rate of ±5.0 ppm. We'll assume the crystal has a trim sensitivity of 13 ppm/pF, a typical value.

When one factors in the ±2 percent accuracy of the impedance meter, the actual frequency reading will only be accurate to ±5.2 ppm. Since this error is additive to the inherent tolerance of the crystal, the real tolerance is now ±15.2 ppm, compared to the crystal's rated tolerance of ±10 ppm.

We might try to tighten the accuracy range by specifying a crystal with a tolerance of ±10 ppm, a stability of ±6.0 ppm, and an aging rate of ±3.0 ppm, which, when compared with our ±25 ppm specification, would seem to allow ±6.0 ppm of cushion to account for the inherent testing variation. Unfortunately, even when the specs are tightened this way, crystal accuracy may still not be sufficient, because we have not yet factored in frequency inaccuracy due to variations in load capacitance.

The load capacitance experienced by a crystal installed in a circuit varies in response to several factors. First among these is the tolerance of the load capacitors themselves. In a circuit with two 33 pF capacitors, each with 1 percent tolerance, and a stray capacitance at a constant 3.5 pF, the desired load would be 20 pF, using the equation illustrated in Figure 1.

However, the actual load capacitance, based on capacitor tolerance of 1 percent, would be 20 pF, ±0.165 pF, or 19.835 to 20.165 pF. Assuming the 13 ppm/pF trim sensitivity used before, the variation in crystal frequency from capacitor tolerance alone would be ±2.15 ppm.

Note that we assumed a constant stray capacitance for this calculation. In the real world, however, stray capacitance varies between different boards and between different ICs. Typical values for these variances might be ±1.3 pF from board to board and ±0.7 pF from IC to IC, for a total stray capacitance variation of ±2.0 pF. When we combine the variation in stray capacitance with that for load capacitance, the overall load capacitance experienced by the crystal may vary from 17.835 pF to 22.165 pF (20.0 pF ±2.165 pF).

Again, using the 13 ppm/pF trim sensitivity value, this works out to a potential variance in crystal frequency of ±28.15 ppm, well in excess of our specified ±25 ppm.

These calculations imply that considerations of board and IC capacitance variances will push even a highly accurate crystal outside the desired accuracy range. Indeed, when we plug in values for a crystal with extremely tight tolerances — one rated at ±18 ppm including tolerance, stability, and aging, while allowing ±2.0 ppm testing error — the required tolerance for board variation would be ±0.4 pF to meet the ±25 ppm specification. This level of accuracy in board capacitance is difficult, if not impossible, to achieve in the real world of manufacturing.

As can be seen from the above calculations, much of the variation that pushes even a tight-tolerance crystal outside desired specification parameters results from the external circuit. In contrast, a pre-packaged oscillator provides the crystal already installed in its own circuit.

This allows the manufacturer to trim the crystal frequency in the circuit that will actually be used in the application, eliminating the additional variance that arises when the crystal is used in a different circuit. Therefore, the resulting tolerance reflects much more accurately the tolerance that can be expected in the desired application.

In addition to eliminating a key source of external variability, the use of pre-packaged oscillators sidesteps the additional (often extensive) design work required to develop a discrete oscillator for a packaged crystal with tight frequency accuracy.

When the total system cost is considered (including design cost), the decision swings even more strongly toward the pre-packaged oscillator. For example, it's instructive to compare the cost of a crystal designed for 802.11a, with a frequency accuracy of ±18 ppm inclusive of tolerance, stability, and aging, with that of a pre-packaged crystal oscillator with an accuracy of ±25 ppm inclusive of tolerance, stability, aging, and load and voltage change.

In quantities of 100,000, the price for the crystal is only about $0.01 to $0.02 less than that of the oscillator. When one factors in the cost of the capacitors required for the crystal-only approach (as well as additional design, manufacturing, and sourcing/acquisition costs), the small per-unit advantage quickly disappears.

This is especially true with 3^rd overtone crystals (usually 40 MHz and above), which require two additional external components and generally cost a minimum of $0.05 more than the equivalent oscillator in large volume. Of course, as discussed above, a ±25 ppm level of accuracy may still be impossible with the ±18 ppm crystal once all of the additional sources of variance are assessed.

Of the three factors (accuracy, cost, and size) that we identified as critical for the selection of frequency sources for wireless devices, only size remains to be considered. There certainly will be a size penalty when one selects an oscillator over a crystal; but is it significant?

Compare a small-footprint crystal illustrated in Figure 2 with its oscillator equivalent, illustrated in Figure 3.

The size penalty is a mere 0.05 mm in the height dimension alone, which most wireless devices can accommodate. Since size differences between most other crystal-oscillator pairs are similar to this example, one can see that the size penalty in most cases will be a small consideration relative to the cost and accuracy concerns.

Conventional crystals, when compared with pre-packaged oscillators, will have far greater difficulty meeting the tight accuracy tolerances required for wireless devices. The relatively small cost differential between the two options disappears (and may even favor the pre-packaged oscillator) once additional component, design, and manufacturing costs are factored into the crystal approach. Finally, pre-packaged oscillators are only slightly larger than their crystal counterparts, and usually only in a single dimension.

For nearly all wireless applications, pre-packaged oscillators represent the best and most cost effective design option.

Roger Burns is the engineering sales manager for Fox Electronics (www.foxonline.com). He has also held a number of other positions in the technical support and field applications areas at Fox. Before joining Fox, Roger was in the U.S. Navy's nuclear engineering program and attended Edison College in Fort Myers, Florida. He can be reached at rogerb@foxonline.com.