What can cause DCXO frequency drift?

[sisomso]

I designed a 13bit DCXO (38.4MHz, 0.006ppm/step) with switch cap structure. I observed slow frequency drift for ~0.01 ppm during 1 hour period. Also, there is ~0.002ppm frequency variation in 2~3 second observation window. The part was put in the oven. So temp is fairly stable. I am wondering what can cause this kind of drift and variation? 1/f noise? If so, how to translate the 1/f noise to the output frequency?

Thanks in advance!

 

[chuckey]

When going for these really low drifts, it is important to remember that any mechanical strain that was introduced into any component during manufacturing or assembley will slowly work them selves out with an effect on the final frequency. I would run a time trial and actually measure the frequency every hour for a week, with some luck it should stabilize at the design frequency. Is your counter locked to a good source?
Frank

 

[RCinFLA]

First question is what is your reference. Do you have confidence in it?

Most variation is temp. A ovenized oscillator is usually set to about 55-65 deg C where the AT cut crystal has it high side deflection point yielding the flattest freq vs. temp over a small temp range.

A poorly designed crystal can have shifts in frequency. Too little drive level in the oscillator can increase freq shifts.

If this unit is brand new there may be drift for some period unless it had burn in by manufacturer.

 

[sisomso]

Hi Frank,

Thank you for your reply.

The counter uses an external high precision 10MHz reference. And I tried a commercial VCTCXO on the same counter, which shows very small variation (~0.0005ppm). So I think the counter should be accurrate enough for the DCXO.

As for the manufacturing effects, the DXCO has been used for couple of months for other PLL related testing. The slow drift doesn't affect the PLL performance. However, this is the first time I am trying to characterize the DCXO itself for AFC purpose in a system point of view. I assume two months time is long enough for all the manufacturing effects going away?

Thanks!

---------- Post added at 12:57 ---------- Previous post was at 12:44 ----------

RCinFLA,

Thanks. I will try to increase the drive level of the oscillator and measure again. Also I will try at 60C temp.

If it's temp, the spec of Xtal shows +/-10E-6~+/-30E-6 from -20C to 75C. I translate this into approximately 0.6ppm/C (worst case). For 0.006ppm variation the temp drift should be 0.01C. I assume no oven can control the temp this good. How can someone claim the DCXO is monotonic by testing, if drift caused by temp change is much greater than ppm/step?

Did I make any wrong assumption?

 

[FvM]

I assume no oven can control the temp this good. How can someone claim the DCXO is monotonic by testing, if drift caused by temp change is much greater than ppm/step?

You can at least for short time by adding sufficient heat capacity to the oscillator enclosure and put a thermal insulation arround it. I even assume, that you can build a thermostat with 0.01K stability.

 

[RCinFLA]

An AT cut crystal has a third order Bechmann curve. See attached. A given crystal has a curve based on the mechanical accuracy of the angle of the cut of the crystal blank from the large quartz bar.

Each crystal is unique (but fits a given 3rd order curve) and must be tested by at least a two temperature/frequency points to determine which curve in the AT Bechmann family it best fits.

If the crystal mount is done correctly so not to stress the quartz blank and the quartz blank is processed with good quality, it will only take two points to determine which Bechmann curve it matches. Many calibration tests take more then two temp points just to be sure and average out inaccuracies in absolute temp and freq readings.

There is likely some temp compensation via oscillator circuit temperature dependent capacitors to flatten out the curve. An ovenized oscillator relies on the temp of the oscillator/crystal environment being held constant, again usually in the 55-65 deg C range at the second 3rd order inflection point where the freq shift versus temp change is minimized.

A TCXO, like found in a cellphone or GPS receiver, has a reverse Bechmann IC that drives a varactor to warp the crystal oscillator and flatten out the frequency versus temperature.

 

[sisomso]

You can at least for short time by adding sufficient heat capacity to the oscillator enclosure and put a thermal insulation arround it. I even assume, that you can build a thermostat with 0.01K stability.

FvM, thank you for your reply. However, this is a bench test. I am not sure how complicated the thermal insulation setup is. All I got is an oven and a thermometer with 0.1C display. I guess I will study the AFC temperature compensation algorithm first.

Thanks!

---------- Post added at 14:13 ---------- Previous post was at 14:12 ----------

Very useful information. Now I am wondering if an on chip DCXO without any temperature compensation scheme can be used in any communication system with 0.1ppm spec? How does the AFC algorithm deal with the temperature dependence of the DCXO? I think I will read more about AFC. Thank you!

An AT cut crystal has a third order Bechmann curve. See attached. A given crystal has a curve based on the mechanical accuracy of the angle of the cut of the crystal blank from the large quartz bar.

Each crystal is unique (but fits a given 3rd order curve) and must be tested by at least a two temperature/frequency points to determine which curve in the AT Bechmann family it best fits.

If the crystal mount is done correctly so not to stress the quartz blank and the quartz blank is processed with good quality, it will only take two points to determine which Bechmann curve it matches. Many calibration tests take more then two temp points just to be sure and average out inaccuracies in absolute temp and freq readings.

There is likely some temp compensation via oscillator circuit temperature dependent capacitors to flatten out the curve. An ovenized oscillator relies on the temp of the oscillator/crystal environment being held constant, again usually in the 55-65 deg C range at the second 3rd order inflection point where the freq shift versus temp change is minimized.

A TCXO, like found in a cellphone or GPS receiver, has a reverse Bechmann IC that drives a varactor to warp the crystal oscillator and flatten out the frequency versus temperature.

 

[RCinFLA]

For cellphones, it depends on MA type. Typically for CDMA the TCXO is +/- 1.0 ppm, by itself, over 0 deg C to +50 deg C. The tighter it is the faster the initial lock on the CDMA signal. GSM is looser, like 10 ppm over temp.

Cellular uses AFC on the channel to correct for TCXO frequency error. Older phone used analog warp line to TCXO. New phones use Frac-N synthesizer to digitially tune out the frequency offset so there is no analog tuning line to TCXO. Between battery saver strobing, the phone saves the AFC offset so freq is accurate for next wakeup cycle. AFC tracking accuracy is usually better the +/- 0.1 ppm assuming the base stations are accurate.

GPS TCXO's are typically +/- 0.5 ppm. GPS requires a very accurate frequency source to speed satellite lock time. The receiver must search for the PRN code for expected satellites. Typical bandwidth for this PRN search is less then 500 Hz so the receiver must do successive frequency bin searches based on what frequency offset may be, based on the TCXO tolerance. The receiver also saves the AFC value for future initial offset calibration.

All crystals have an aging drift, due to material being ejected from blank. This can be deposited metal for the coupling spot or quartz 'dust' that is lodged in the quartz crystal lattice structure surface. Good quality crystals are ultrasonically cleaned to remove as much of the 'dust' as possible. Because material is falling off the blank, crystals usually rise in frequency as they age.

All TCXO also have a center frequency make tolerance. This is calibrated out in the cellphone or GPS receiver at the factory. This calibration is then modified by AFC averaging algorythms to account for aging drift.

---------- Post added at 15:50 ---------- Previous post was at 15:36 ----------

The lowest cost way to make a high stability frequency source is to use a GPS receiver with a 1 pulse per second digital output to 'dissipline' a OCXO. The absolute time accuracy of each pulse output is typically +/- 1 usec but over 15 to 30 minutes their average accuracy is better then 1 ppb. You can easily achieve a frequency accuracy of better then 10 ppb with this method.

 

[sisomso]

Here is what I think the AFC algorithm would be. Please correct me if I am wrong.

Assume I have a TCXO 10ppm over temp in an old GSM phone with an AFCDAC (0.01ppm/step) to tune the TCXO. The targeting frequency is F0. Assume the initial frequency is F1(init)=F0+1.0ppm. Base station senses the 1.0ppm difference, feedback a signal to set AFCDAC(new)=AFCDAC(old)-100. Ideally, F1(new)=F0. However, my question is: assuming temp changes (~2 deg C) during the feedback procedure, which results in a +0.2ppm drift on the TCXO, the F1(new)=F0+0.2ppm. The result after AFC is still out of spec (+/-0.1ppm).

Will this happen in real life? Or the update rate is high enough (1ms?) to prevent a 0.2ppm drift caused by temp change? If the carrier frequency becomes out of spec during a communication (phone call), will there be a drop call?

I am new to the system, really appreciate your help.

 

[RCinFLA]

That is not the way it works. It is solely the cellphone's Rx that makes measurement of the received signal from basestation and centers it up.

The Rx AFC is continously operating on the received signal to compensate for temperature change (usually caused by transmitter PA heating up).

The long term update on the calibration to account for TCXO aging is 'special sauce' by each cellphone manufacturer. If you screw it up you can make the phone inoperable or degraded in performance.