Low cost <1 ppb 10 MHz OXCOs on ebay

[rhb]

There are a slew of references based on the CTI OSC5A2B02 10MHz 5V Square Wave OCXO Crystal Oscillator very cheap (~$20 USD).  After placing in an enclosures to protect against drafts, the two I am testing are holding <0.1 ppb day to day stability once trimmed to the thermal environment.  Bare modules are less than $3, though I don't yet know what I'm going to get.

The one shown  lacks fiber fill to damp airflow and is very stable lying flat, but shifts frequency quite a bit if placed long edge vertical.  Lay it flat again and it's back where it was.  Another filled with cotton balls doesn't care about orientation.

This one will get a mat of fiber insulation and a temperature controller.  The dramatic change in frequency from convection makes keeping a copy of this pointless except as an example of what not to do.

The yellow trace is one of Leo Bodnar's GPSDOs.  The other trace is the OXCO output.  I attached the two screen dumps so you can calculate the frequency difference independently of my 5386A.  I know from experience that when the error is <0.01 Hz I can't possibly get it better.

One of the great virtues of packaging them in a box is the ability to use it with any T&M kit that takes an external 10 MHz reference via a backpanel  port.

Have Fun!
Reg

[james_s]

I see the boards, but I don't see any bare oscillators. I wonder what these came from? Cell sites?

[I wanted a rude username]

They tend to be "gently reworked" off various PCBs as you can see in this photo from an AliExpress listing. The thought of saw-induced vibrations in such a precision part gives me the screaming abdabs.

[SCSKITS]

I have bare boards for these low cost OCXOs here: : https://www.ebay.com/itm/155419526549 if anyone is interested.

Also re-did the 53131A OCXO module (which was originally from the HP design/schematics) but changed out the expensive and sometimes hard to get AD7243ARZ DAC with a less expensive 5V DAC.

ed

[rhb]

 Limited to a 10 s gate, the 5386A is  really not  much more accurate than ~0.1 Hz.  So for serious study I need to develop an appropriate instrument by feeding a GPSDO clock to the timer inputs of a fast MCU dev board.

At the moment I'm leaning towards putting aluminum plates thermally bonded to Peltier devices with old CPU  heat sinks and fans attached for precise temperature control.  The optimal temperature is just high enough to keep the OXCO oven from turning on, avoiding generating  transient thermal events generated by the OXCO heater.

I rather fear I am among the lost as I'm about to see if the fabled <$150 rubidium standards exist.

Pray for my salvation and that I shall be delivered from the debauchery of becoming a Time-Nut. ;-)
Reg

BTW this is the board design I bought though possibly not this vendor.

https://www.ebay.com/itm/144978139139

[ch_scr]

Limited to a 10 s gate, the 5386A is  really not  much more accurate than ~0.1 Hz.  So for serious study I need to develop an appropriate instrument by feeding a GPSDO clock to the timer inputs of a fast MCU dev board.
[...]

If you don't want to have to invent the wheel again (or as a second source sanity check), there is this project on here, just a heads up in case you haven't seen it.

[807]

@rhb...were you able to tune the xtal to 10MHz without modifying the resistors?

As from this previous thread, the module bought by me & others wouldn't tune to exactly 10MHz without modification...

https://www.eevblog.com/forum/projects/cheap-chinese-ocxo-frequency-reference-modules/msg4728044/#msg4728044

[rhb]

@rhb...were you able to tune the xtal to 10MHz without modifying the resistors?

As from this previous thread, the module bought by me & others wouldn't tune to exactly 10MHz without modification...

https://www.eevblog.com/forum/projects/cheap-chinese-ocxo-frequency-reference-modules/msg4728044/#msg4728044

I had no trouble with either of mine.  We'll see when I get the other 3.

The bare modules need power and control voltage.  I can set the two I have to within 0 002 Hz but it still drifts despite batting inside and the unit in a beer box filled with batting.  Generally 0.03 Hz  or less over 24 hrs.  I need to attach an AR488 to log data.

I also need to make a test fixture for the bare  modules with the ability to select modules and automatically set them.  A single LM399 with TL082s for feeding each OXCO a stable Vref.

I'd like to close this thread and shift to the other one.

Reg

[MIS42N]

The bare modules need power and control voltage.  I can set the two I have to within 0 002 Hz but it still drifts despite batting inside and the unit in a beer box filled with batting.  Generally 0.03 Hz  or less over 24 hrs.  I need to attach an AR488 to log data.

The stability of these is more dependent on the stability of the control voltage than the case temperature. The sensitivity to control is around 0.1V/Hz (or about 1 part per million per volt, whichever way you like). So a change of 1 part per billion only requires 1mV change. If the control voltage is around 2V (most are) and the supply is 5V that equates to a supply change of 2.5mV. Most run of the mill regulators like the 7805 change output 1 or 2mV per degree C.

Also from a cold start they drift significantly for an hour then a bit for two days. After two days they are pretty stable. May take a week if just installed (maybe heat from soldering or something?).

For a square wave into 50Ω I use a 74HC04 buffer - see the details around the OSC5A2B02 at https://www.eevblog.com/forum/projects/budget-gpsdo-a-work-in-progress/msg4329907/#msg4329907 - the output of the OCXO is HCMOS so not designed to drive into test leads.

[james_s]

It's pretty well known that crystal oscillators take a while to stabilize after not being used for a while, IIRC it has something to do with the crystal wafer itself outgassing or something. I agree about the control voltage, from looking at the picture my first thought was the cheap little pot used on those boards probably has a lot more drift than the oscillator.

[Datman]

The stability of these is more dependent on the stability of the control voltage than the case temperature. The sensitivity to control is around 0.1V/Hz (or about 1 part per million per volt, whichever way you like). So a change of 1 part per billion only requires 1mV change. If the control voltage is around 2V (most are) and the supply is 5V that equates to a supply change of 2.5mV. Most run of the mill regulators like the 7805 change output 1 or 2mV per degree C.

I made a TCXO using a OSC5A2B02 and a TL431.

[GigaJoe]

"Limited to a 10 s gate, the 5386A is  really not  much more accurate than ~0.1 Hz. "

that not true , this type of counters are second generation, able to measure phase shift between signals, with minimal jitter and 10M source usually are, it perfectly measure extended 11 digits.   in practice im able to compare with 100 and 1000 sec  period with other counters, with the same result.

CETC institute has a lot of documents in PDF  publicity available but i think around 2016 access to the web site was blocked.  as it basically close to military.  i may mistake but as far as remember  OSC5A - AT cut , OC5SC - SC cut, with better long therm.

12volt OC12SC36 - has better volt sensing circuit, in my practice approx 0.3mV to adjust greater accuracy. and that as well force to use very well stable source of adjusted voltage.  the best result i got  when voltage source thermally connected to OCXO case,   basically thermally stable,  with some additional thermal isolation i had drift in -2-10 \Y (really no. remember, but perfectly fine for 8d direct counter) as well it help in  1-decrease power consumption. 2-much better temperature control. 3-decrase jitter due to temperature deviation.

"cheap little pot" - use the same, not affecting at all,  in a correct circuit   adjustment has to be a small fraction of total regulated voltage,  so ideal rough adjustment inside the heated chamber, and superfine outside.

[rhb]

The stability of these is more dependent on the stability of the control voltage than the case temperature. The sensitivity to control is around 0.1V/Hz (or about 1 part per million per volt, whichever way you like). So a change of 1 part per billion only requires 1mV change. If the control voltage is around 2V (most are) and the supply is 5V that equates to a supply change of 2.5mV. Most run of the mill regulators like the 7805 change output 1 or 2mV per degree C.

I made a TCXO using a OSC5A2B02 and a TL431.

Pray tell, tell us more.

[Datman]

It's very simple:
The ATtiny85 makes the LED flashing at three different frequencies according to the temperature. When it is stable, the working temperature has been reached.

[807]

It's very simple:
The ATtiny85 makes the LED flashing at three different frequencies according to the temperature. When it is stable, the working temperature has been reached.

Why did you increase the TL431 reference voltage to 4v? Did you find that you couldn't adjust the xtal to 10MHz using 2.5v?

Most of the results on here for the xtal ref voltage seem to be less than 2v.

[Datman]

0 to 4V -> -2 to +2 ppm:

[MIS42N]

I made a TCXO using a OSC5A2B02 and a TL431.

I looked at the specs for the TL431, it did not seem to be any better than a run of the mill LDO regulator but used for a different purpose. Might as well take the divider straight off the +5V supply. Or did I miss something.

If the OSC5A2B02 has been used frequently, it takes about 15 minutes from cold start to reach some form of stability. The first 3 minutes it can be in error by more than 10Hz. After 15 minutes it is less than 1Hz out, usually around 0.1Hz. It will change frequency if its orientation is changed. Tip it upside down and then right way up, it can settle at a different frequency. So once it is set up it is not a good idea to move it.

[james_s]

The TL431 has been around for ages, and it's a shunt regulator which is designed as a voltage reference. If you use a power rail as a reference then your reference is affected by other loads on that rail.

[MIS42N]

The TL431 has been around for ages, and it's a shunt regulator which is designed as a voltage reference. If you use a power rail as a reference then your reference is affected by other loads on that rail.

Of course!! I made that comment late at night without too much thought. In fact I had the problem with my GPSDO design, and solved it by running the OCXO and its control voltage off a separate LDO. The OCXO provides a fairly constant load so using its +5V also for the control voltage source is OK with the advantage that the OCXO has a stable +5V supply. I also made provision for a precision voltage reference but it didn't yield much improvement so usually left off.

[GigaJoe]

TL431 can change output voltage up to 50 millivolts by internal ageing process ... it basic, extremely cheap reference. you not get maximum potential from OXCO using this type of reference.  I had some LM336 -5  thermally connected , as better version of TL ,  and even after 3-4month that has not enough stability , the frequency shifting as LM336  drifting ....   for a best result overall drift should be as low as possible.

I suggest something LT1021 . LT1027 ,  or similar  thermally connected to OCXO,   plasic  ok, used one ok  ... 
https://www.ebay.com/itm/324030185416

[rhb]

[snip]

I suggest something LT1021 . LT1027 ,  or similar  thermally connected to OCXO,   plasic  ok, used one ok  ... 
https://www.ebay.com/itm/324030185416

The voltage reference should be in complete thermal isolation  from the OXCOs as should be the voltage dividers for each OXCO.  These should be TL08x or similar voltage followers so there is no discernible load on the reference.

Each OXCO should be in its own oven set so the internal heater never turns on.  That will minimize thermal gradients.


Have Fun!
Reg

[Yrrah]

For those interested: this weekend my ready made board with a CTI 10.0 MHz OCXO by BG7TBL arrived. It looks all brand new but I guess the OCXO is salvaged from an obsolete system. I made some phase noise measurements with my home brew PN system (DG4RBF, UKW Berichte and Audiometer by DG8SAQ). In all cases the reference oscillator is a HP10811A as it is EFC controlled. All components are battery fed. As a sanity check I measured a Wenzel 501-08772 first. Then I measured an Isotemp 131 OCXO and the CTI (OSC5A2B04). Warm-up time > 30 min. The results are attached. The Wenzel is my "gold" standard: PN as per datasheet: @ 100Hz = -155 dBc, @1kHz = -165 dBc and at 10 kHz also -165 dBc. The HP10811A measures: @100 Hz -154 dBc , 1 kHz -163 dBc and 10 kHz -164.5 dBc. From brand reputation: both are probably even better than the specs. There is not much difference between the two. Pretty much in agreement with HP spec. I am not claiming accuracy within 1 dB. And if the difference is small no distinction is possible. The Isotemp measures: @ 100 Hz -149 dBc, @ 1kHz -155 dBc and @ 10 kHz -156 dBc. The CTI: @ 100 Hz -143 dBc, @ 1 kHz -154 dBc and @ 10 kHz -158 dBc. Quite close to one another and the distance to the 10811 is large enough for the numbers to be ok. Any one to confirm or to dispute these figures?

[Yrrah]

A couple of weeks ago I got 4 more CTI OCXO's (5A2B02), nicely removed form whatever system. Time for some additional noise spectra measurements. I made a test-fixture to accommodate the DUT's. No good looks but effective. Power (+5V) is provided by a 7806 followed by an RCRC filter (R=2.5 ohm, C=1000uF, 105C). In UKW Berichte (4/2022) I have shown that a good filter behind a mediocre (noise point of view) voltage regulator can give quite good results. The EFC voltage is generated by a LT3045 set at 5V. You need a range of about 1.5 - 2.2 V so I took a 500 ohm multiturn and about 1 kohm to ground and 2.2 kohm to +. The DUT is embedded in some sort of foam for thermal stability. Not quite good visible at the picture. Between cover and OCXO is also foam. This way the current draw drops to about 170 mA, once warmed up. Some additional caps here and there and that's it. No output filter, just the square wave. Before the noise measurement some basic tests were performed and the spectrum was observed. Just some soldering to exchange the DUT.
The photo shows the measurement set-up. Everything is battery fed. Exactly the same as described in previous post. The DUT is followed by a 5 dB attenuator.
The results are shown in the third picture (S1P files from Audiometer fed to Octave script to generate graph). No 5 is the 5A2B04 (on ready made pcb, refer to previous post). You clearly see some statistical variance between DUT's but they are all pretty good. No1 is best but the differences are small. The reference (HP10811-60111) can be seen in the previous post still better than any of the CTI's. Out of curiosity I measured no3 with a Minicircuits LPF: no difference. Now I'd like to see Adev measurements, well that's possibly for the future.

[rhb]

Thanks for taking the time to post these.  I'm hoping to buy a tinyPFA and run Adevs as soon as R&L stocks them.

For $3 each I think they are fantastic deals.  Every test that I've seen shows that they are quite good.  Not the best,  but likely much more stable long term because of aging.

Reg

[Yrrah]

Well, I think this kind of info/data is easier to retrieve here. Over the last two days I measure about 20 mHz drift (max) of OCXO no3 in test fixture. Measured with my HP5345A counter (always on, internal HP10811 clock). Nice stuff indeed. 
Harke