Archive for category Amateur Radio
A Circuit Cellar reader asked for a better explanation of the parasitic capacitors inside a quartz crystal can than I provided in my April 2017 Circuit Cellar column. Here’s the schematic, with values for a 32 kHz tuning fork resonator and the original caption:
I wrote this about the caps:
The value of C0 in the model’s middle branch corresponds to the capacitance between the electrodes plated onto the quartz. The 3.57954 MHz crystal in the title photo, with two silver electrodes deposited on a flat insulating disk, closely resembles an ordinary capacitor. You can measure a crystal’s C0 using a capacitance meter.
However, each of those electrodes also has a capacitance to the resonator’s metal case. The top branch of the model shows two capacitors in series, with Cpar representing half the total parasitic capacitance measured between both leads and the case. Grounding the case, represented by the conductor between the capacitors, by soldering it to the ground plane of an RF circuit eliminates any signal transfer through those capacitors. They will appear as shunt capacitors between the pins and ground; in critical applications, you must add their capacitance to the external load capacitors.
And this about the measurement technique, using a fixture on my AADE LC meter:
With the resonator case captured under the clip on the far right and both its leads held by the clip to the upper left, the meter measures Cpar, the lead-to-case parasitic capacitance. The meter will display twice the value of each parasitic capacitor, at least to a good approximation, because they are in parallel. The five resonators averaged 0.45 pF, with each lead having about 0.25 pF of capacitance to the case. Obviously, measuring half a picofarad requires careful zeroing and a stable fixture: a not-quite-tight banana jack nut caused baffling errors during my first few measurements.
With the resonator repositioned as shown in Photo 2, with one lead under each clip, the meter measures C0, the lead-to-lead capacitance. After careful zeroing, the resonators averaged 0.85 pF.
Although the parasitic lead-to-case capacitors are in parallel with C0, their equivalent capacitance is only Cpar/4 = 0.1 pF. That’s close enough to the measurement error for C0, so I ignored it by rounding C0 upward.
He quite correctly pointed out:
With both leads connected together on one side, then that essentially constitutes a single electrical element inside the case. And with the case being a single element, this configuration in the test fixture seems like a single capacitor with one lead being the case and the other lead being the pins, with a vacuum dielectric.
I would think the meter would display the total capacitance rather than twice the value … It makes sense to me to later say Cpar/2 when the leads are not connected together.
Here’s my second pass at the problem:
… the two-capacitor model comes from the common-case condition, where each lead displays a (nominally equal) parasitic capacitance to the case, because the crystal mounting is reasonably symmetric inside the can. It’s easiest to measure the total capacitance with the leads shorted together, because it’s in the low pF range, then divide by two to get the value of each lead-to-case cap.
For “real” RF circuits with larger (HC-49 -ish) crystals in parallel-resonance mode, you ground the case and subtract the parasitic capacitance at each lead from the external load capacitors. That’s the usual situation for microprocessor clock oscillators: the crystal sits across the clock amplifier pins, with two more-or-less equal caps from the pins to ground. You should subtract the internal parasitic caps from the clock’s specified load caps, but in practice the values are so small and the cap tolerance so large that it mostly doesn’t matter.
Un-grounding the case puts those two parasitic caps in series, just as with two discrete caps, so the lead-to-lead capacitance is (or should be!) half of each: 1/4 of the both-leads-to-case value.
Re-reading yet again says I glossed over the effect of having C0 in parallel with the Cpar/2 caps, but methinks dragging those complications into the model benefits only the theoreticians among us (or those working very close to the edge of the possible).
To make it worse, I also botched the QEX reference, which should be Jan/Feb 2016, not 2017. Verily, having a column go read-only makes the errors jump right off the page. [sigh]
At least I can point to this and amend as needed.
With an LM75 atop the 125 MHz oscillator and the whole thing wrapped in foam:
Let it cool overnight in the Basement Laboratory, fire it up, record the temperature every 30 seconds, and get the slightly chunky blue curve:
Because we know this is one of those exponential-approach problems, the equation looks like:
Temp(t) = Tfinal + (Tinit - Tfinal) × e-t/τ
We can find the time constant by either going through the hassle of an RMS curve fit or just winging it by assuming:
- The initial temperature, which is 22.5 °C = close to 22.7 °C ambient
- The final temperature (call it 42 °C)
- Any good data point will suffice
The point at 480 s is a nice, round 40 °C, so plug ’em in:
40.0 = 42.0 + (22.7 - 42.0) × e-480/τ
Turning the crank produces τ = 212 s, which looks about right.
Trying it again with the 36.125 °C point at 240 s pops out 200.0 °C.
Time for a third opinion!
Because we live in the future, the ever-so-smooth red curve comes from unleashing LibreOffice Calc’s Goal Seek to find a time constant that minimizes the RMS Error. After a moment, it suggests 199.4 s, which I’ll accept as definitive.
The spreadsheet looks like this:
|Time s||Temp °C||Exp App||Error²|
Exp App column is the exponential equation, assuming the three variables at the top, the
Error² column is the squared error between the measurement and the equation, and the
RMS Error cell contains the square root of the average of those squared errors.
The Goal Seeker couldn’t push
RMS Error to zero and gave up with
Tau = 199.4. That’s sensitive to the initial and final temperatures, but close enough to my back of the envelope to remind me not to screw around with extensive calculations when “two minutes” will suffice.
Basically, after five time constants = 1000 s = 15 minutes, the oscillator is stable enough to not worry about.
I let the DDS cool down overnight, turned it on, and recorded the frequency offset as a function of temperature as it heated up again:
The reduced spacing between the points as the temperature increases shows how fast the oscillator heats up. I zero-beat the 10 MHz output, scribbled the temperature, noted the offset, and iterated as fast as I could. The clump of data over on the right end comes from the previous session with essentially stable temperatures.
I only had to throw out two data points to get such a beautiful fit; the gaps should be obvious.
The fit seems fine from room (well, basement) ambient up to hotter than you’d really like to treat the DDS, so using the quadratic equation should allow on-the-fly temperature compensation. Assuming, of course, the equation matches some version of reality close to the one prevailing in the Basement Laboratory, which remains to be seen.
In truth, it probably doesn’t, because the temperature was changing so rapidly the observations all run a bit behind reality. You’d want a temperature-controlled environment around the PCB to let the oscillator stabilize after each increment, then take the measurements. I am so not going to go there.
The original data:
A day of jockeying the AD9850 DDS oscillator shows an interesting relation between the frequency offset and the oscillator temperature:
Now, as it turns out, the one lonely little dot off the line happened just after I lit the board up after a tweak, so the oscillator temperature hadn’t stabilized. Tossing it out produces a much nicer fit:
Looks like I made it up, doesn’t it?
The first-order coefficient shows the frequency varies by -36 Hz/°C. The actual oscillator frequency decreases with increasing temperature, which means the compensating offset must become more negative to make the oscillator frequency variable match reality. In previous iterations, I’ve gotten this wrong.
For example, at 42.5 °C the oscillator runs at:
125.000000 MHz - 412 Hz = 124.999588 MHz
Dividing that into 232 = 34.35985169 count/Hz, which is the coefficient converting a desired frequency into the DDS delta phase register value. Then, to get 10.000000 MHz at the DDS output, you multiply:
10×106 × 34.35985169 = 343.598517×106
Stuff that into the DDS and away it goes.
Warmed half a degree to 43.0 °C, the oscillator runs at:
125.000000 MHz - 430 Hz = 124.999570 MHz
That’s 18 Hz lower, so the coefficient becomes 34.35985667, and the corresponding delta phase for a 10 MHz output is 343.598567×106.
Obviously, you need Pretty Good Precision in your arithmetic to get those answers.
After insulating the DDS module to reduce the effect of passing breezes, I thought the oscillator temperature would track the ambient temperature fairly closely, because of the more-or-less constant power dissipation inside the foam blanket. Which turned out to be the case:
The little dingle-dangle shows startup conditions, where the oscillator warms up at a constant room temperature. The outlier dot sits 0.125 °C to the right of the lowest pair of points, being really conspicuous, which was another hint it didn’t belong with the rest of the contestants.
So, given the ambient temperature, the oscillator temperature will stabilize at 0.97 × ambient + 20.24, which is close enough to a nice, even 20 °C hotter.
The insulation blanket reduces short-term variations due to breezes, which, given the -36 Hz/°C = 0.29 ppm temperature coefficient, makes good sense; you can watch the DDS output frequency blow in the breeze. It does, however, increase the oscillator temperature enough to drop the frequency by 720 Hz, so you probably shouldn’t use the DDS oscillator without compensating for at least its zero-th order offset at whatever temperature you expect.
Of course, that’s over a teeny-tiny temperature range, where nearly anything would be linear.
The original data:
While that’s happening, you compare the DDS output to a reference frequency on an oscilloscope:
The top trace (and scope trigger) is the GPS-locked 10 MHz reference, the lower trace is the AD9850 DDS output (not through the MAX4165 buffer amp, because bandwidth). If the frequencies aren’t identical, the DDS trace will crawl left or right with respect to the reference: leftward if the DDS frequency is too high, rightward if it’s too low. If the DDS frequency is way off, then the waveform may scamper or run, with the distinct possibility of aliasing on digital scopes; you have been warned.
The joystick acts as a bidirectional switch, rather than an analog input, with the loop determining the step increment and timing. The ad-hoc axis orientation lets you (well, me) push the joystick against the waveform crawl, which gradually slows down and stops when the offset value makes the DDS output match the reference.
The OLED displays the current status:
The lurid red glow along the bottom is lens flare from the amber LED showing the relay is turned on. The slightly dimmer characters across the middle of the display show how the refresh interacts with the camera shutter at 1/30 s exposure.
N.B.: Normally, you know the DDS clock oscillator frequency with some accuracy. Dividing that value into 232 (for the AD9850) gives you the delta-phase count / frequency ratio that converts a desired DDS output frequency into the delta-phase value telling the DDS to make it happen.
In this case, I want the output frequency to be exactly 10.000000 MHz, so I’m adjusting the oscillator frequency (nominal 125 MHz + offset), calculating the corresponding count-to-Hz ratio, multiplying the ratio by 10.000000 MHz, stuffing the ensuing count into the DDS, and eyeballing what happens. When the oscillator frequency variable matches the actual oscillator frequency, then the actual output will 10.000000 MHz and the ratio will be correct.
Got it? Took me a while.
Although the intent is to tune for best frequency match and move on, you (well, I) can use this to accumulate a table of frequency offset vs. temperature pairs, from which a (presumably simple) formula can be conjured to render this step unnecessary.
The Arduino source code as a GitHub Gist:
Disabling the display by activating its powersave option reveals 60 Hz pulses from the USB port on the Arduino Nano:
Unplugging the USB cable, leaving just the +5 VDC power supply and coax cable to the oscilloscope, solves most of the problem:
A closer look shows some (relatively) low frequency noise remains in full effect:
Disabling the display while measuring the crystal seems sensible, although, to avoid surprises, a pushbutton should start the process. Unplugging the USB port puts a real crimp in the data collection, although that’s probably survivable with a USB isolator, one of which is on the way around the planet.
The remaining low-level chop requires more thought. Somewhat to my surprise, holding the Arduino Reset button down doesn’t change much of anything, so it’s not a firmware thing.
Those 10 µF coupling caps gotta go.
With the OLED dark and the USB carrying data:
Compare that to the first pass:
Tamping down the noise seems to reduce the overall amplitude variation, but it also makes the capacitor-in and capacitor-out curves more consistent. There may be other things going on that I haven’t accounted for.
The peak frequencies differ by 0.2 Hz, which is probably due to a few degrees of temperature difference. Obviously, it’s badly in need of a temperature calibration & correction.
The Big Ideas: the DDS output, being more-or-less constant, needs a variable-gain amp to set the crystal drive level. The amp also fixes the impedance mismatch between the DDS output and the crystal, which may not be much of a problem for the (very) high ESR quartz tuning fork resonators in play.
The AD8950 DDS output feeds a 70 MHz (-ish) elliptical reconstruction filter chopping off image frequencies descending from the 125 MHz sampling clock, with a 100 Ω (-ish) output impedance that’s just about purely resistive at 60 kHz. An on-board 3.9 kΩ resistor (labeled with 392 on their schematic) sets the full-scale output current to 10 mA for a peak voltage of 1 V. The module uses only the + output of the differential pair, which means the sine wave runs from 0 V to 1 V: 1 Vpp = 500 mVpeak = 353 mVrms (ignoring the 500 mV offset).
Pin header J3 normally sports a jumper to connect the 3.9 kΩ RSET resistor, but you can insert an external resistor to increase the resistance and decrease the output current:
IOUT = 32 × 1.248 V / RSET
A little hot-melt glue action produced a suitable lashup from a 5 kΩ trimpot:
The pillars of green wire insulation forestall screwdriver shorts to the bare pin headers, although that’s less of risk with the upper insulating foam sheet in place:
A 5 kΩ trimpot can vary the output voltage downward by a factor of 2 = -6 dB, more or less.
All the quartz tuning fork resonator specs I’ve found, none of which may apply to the units on hand, seem to require no more than 1 µW drive. Given a resonator’s equivalent series resistance of around 20 kΩ (for real!), the drive voltage will be 150 mV (-ish):
1 µW = V² / 20 kΩ, so V = sqrt(20×10³) = 141 mV
The nominal version of the crystal tester had a 50 Ω input impedance, so I picked a MAX4165 op amp with mojo sufficient for anything over 25 Ω; in retrospect, a lighter load than 48 Ω would be fine.
In any event, the amp looks like this:
What looks like a DIP switch is really the 3×2 jumper header just to the right of the foam insulation, in front of the SOT23 space transformer PCB carrying the MAX4165. No jumper = 0 dB gain, then 6 dB steps upward from there. The -6 dB trimpot range gives more-or-less continuous output tweakage across 24 dB, -6 dB to +18 dB, which is certainly excessive. The 24 Ω terminating resistors provide 6 dB loss into the crystal, so the effective range is -12 to +12 dB, with 0 dB = 350 mVrms and -6 dB = 150 mVrms (-ish) at the crystal.
It’s a non-inverting amplifier, which (also in retrospect) probably isn’t a win:
- Yet Another Bypass Cap on the cold end of the gain-setting resistors
- Overly elaborate VCC/2 biasing to maintain sufficiently high input impedance
I’m reasonably sure all those big caps contribute to some low-level motorboating, but haven’t tracked it down.