EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: uski on November 17, 2015, 05:37:52 pm
-
Hi,
Being curious, I've been searching for good information about the internals of commercially available TCXOs - but to no avail.
I know there's some type of temperature compensating circuit (yea... obvious) but that's about it.
Looks like there are some trade secrets here...
Does anyone know where to get more information ? How is the temperature compensation implemented ?
Thanks :)
-
Have read about it before, don't know what's actually used in practice.
One method is to compare the fundamental and 3rd harmonic modes. The crystal is cut for the 1st, and has the best tempco around that point. The 3rd has two important differences: it's not a perfect harmonic, but a few percent low (leading to a ~100s kHz difference in a 10MHz crystal), and it has a much stronger tempco. So that frequency can be sensed, and some compensation applied as a result, perhaps digitally.
The oscillator circuit will look quite strange. It needs a low distortion (AGC controlled) amplifier, tuned to the fundamental. This is mixed with one tuned to the 3rd harmonic. Diplexers allow both to operate independently, with low crosstalk. Or something like that.
The compensation circuit will consist of a counter which measures the harmonic in terms of the fundamental, and generate a correction voltage into a varactor or something like that. (Noteworthy that the harmonic's frequency will shift with reactance, just as the fundamental does. Must be something about the ratio that works particularly well here.)
Other methods could be used (like dumb old thermistors), but I don't know how good the relative accuracy of everything is.
You've also got the problem of compensating everything else (the varactor has a tempco, and you have to prepare a conversion table for the whole system!). The extra lab time, and manufacturing assurance (the crystal cuts need to be that much more precisely matched, otherwise the required compensation will be wrong), are well worth the added cost, when you need the extra accuracy.
Tim
-
Here's a book on the subject.
https://archive.org/details/CrystalOscillatorDesignTemperatureCompensation
Ed
-
Curious about it too. If you look at the "accurate" temperature compensated RTC's, it is simply done with a temperature sensor and a look-up table of the well known temperature curve of the crystal. Or they have a calibration facility in eeprom.
-
here is a document from someone at V-Tech, pretty interesting, circuit in later half.
[/url]http://scholar.lib.vt.edu/theses/available/etd-11262001-111453/unrestricted/etd.pdf (http://scholar.lib.vt.edu/theses/available/etd-11262001-111453/unrestricted/etd.pdf)[url]
-
here is a document from someone at V-Tech, pretty interesting, circuit in later half.
[/url]http://scholar.lib.vt.edu/theses/available/etd-11262001-111453/unrestricted/etd.pdf (http://scholar.lib.vt.edu/theses/available/etd-11262001-111453/unrestricted/etd.pdf)[url]
I'll warn you I don't have a reference or bibliography ready..so I'm going by memory :)
...but recently (last couple of years) I remember reading on a manufacturer's website touting "their" oscillators that they were better than the competition that was using "digital methods" to compensate for temperature based frequency variations.
It sounded like a temperature based lookup table where depending on temperature different compenstation component values are switched in or out of circuit.
The side effect is that you have "steps" in the frequency stability.
Considering the size of the modern SMD TCXO's , it's almost "magic". :)
-
Fromt he datasheet of the Crystek CVT32
2.5ppm for about $8
"Designed to meet today's
requirements for precision
operation and small layout
applications. Temperature
compensation is accomplished
through digital technology.
Standard packaging is 1k
tape and reel."
-
When I last investigated this a few years ago, I was led to believe that there were apparently two types of compensation, one digitally controlled and the other analogue.
A problem with digitally based TCXO correction is that they introduce step changes rather than a smooth transition which for digital coding applications can cause otherwise unexplained occasional data loss due to the sudden phase and/or frequency changes. In certain circumstances, it could result in loss of clock and bit synchronisation which neither block nor convolution FEC can correct for.
-
When I last investigated this a few years ago, I was led to believe that there were apparently two types of compensation, one digitally controlled and the other analogue.
A problem with digitally based TCXO correction is that they introduce step changes rather than a smooth transition which for digital coding applications can cause otherwise unexplained occasional data loss due to the sudden phase and/or frequency changes. In certain circumstances, it could result in loss of clock and bit synchronisation which neither block nor convolution FEC can correct for.
Why does it have to be a sudden change? Just incorporate a long time constant filter with time constant comparable to the thermal lag of the oscillator package.
-
When I last investigated this a few years ago, I was led to believe that there were apparently two types of compensation, one digitally controlled and the other analogue.
A problem with digitally based TCXO correction is that they introduce step changes rather than a smooth transition which for digital coding applications can cause otherwise unexplained occasional data loss due to the sudden phase and/or frequency changes. In certain circumstances, it could result in loss of clock and bit synchronisation which neither block nor convolution FEC can correct for.
Why does it have to be a sudden change? Just incorporate a long time constant filter with time constant comparable to the thermal lag of the oscillator package.
I don't know, but that was the perceived wisdom of some RF/comms gurus far better equipped in the grey matter department than me.
-
FWIW what I'm banging on about is alluded to in this article http://www.crystek.com/documents/appnotes/Understanding_TCXOs.pdf (http://www.crystek.com/documents/appnotes/Understanding_TCXOs.pdf)
-
When I last investigated this a few years ago, I was led to believe that there were apparently two types of compensation, one digitally controlled and the other analogue.
A problem with digitally based TCXO correction is that they introduce step changes rather than a smooth transition which for digital coding applications can cause otherwise unexplained occasional data loss due to the sudden phase and/or frequency changes. In certain circumstances, it could result in loss of clock and bit synchronisation which neither block nor convolution FEC can correct for.
Why does it have to be a sudden change? Just incorporate a long time constant filter with time constant comparable to the thermal lag of the oscillator package.
I don't know, but that was the perceived wisdom of some RF/comms gurus far better equipped in the grey matter department than me.
Once they read this thread I'm sure they'll get their act together and stop those $8 , 3x5 mm monstrocities from hopping about :)
-
FWIW what I'm banging on about is alluded to in this article http://www.crystek.com/documents/appnotes/Understanding_TCXOs.pdf (http://www.crystek.com/documents/appnotes/Understanding_TCXOs.pdf)
Yes, I see where it says that. I've heard stories of people pulling their hair out over "phase hits".
This article implies that these days the whole oscillator plus programmable analog temp comp is incorporated into a single ASIC:
http://defenseelectronicsmag.com/site-files/defenseelectronicsmag.com/files/archive/rfdesign.com/mag/609RFDF4.pdf (http://defenseelectronicsmag.com/site-files/defenseelectronicsmag.com/files/archive/rfdesign.com/mag/609RFDF4.pdf)
-
I'm going to read all the documents you linked, thanks guys :-+
I may attempt to open a TCXO and take a picture with a microscope to see the internals.
PS: $8 for 2.5ppm ? You can get TCXOs from well-known brands for much much much much less in high volume quantities, such as
+-0.5ppm: http://www.digikey.com/product-detail/en/KT2520K26000AAW18TAS/1253-1425-1-ND/5253696 (http://www.digikey.com/product-detail/en/KT2520K26000AAW18TAS/1253-1425-1-ND/5253696)
-
I'm going to read all the documents you linked, thanks guys :-+
I may attempt to open a TCXO and take a picture with a microscope to see the internals.
PS: $8 for 2.5ppm ? You can get TCXOs from well-known brands for much much much much less in high volume quantities, such as
+-0.5ppm: http://www.digikey.com/product-detail/en/KT2520K26000AAW18TAS/1253-1425-1-ND/5253696 (http://www.digikey.com/product-detail/en/KT2520K26000AAW18TAS/1253-1425-1-ND/5253696)
Just a quick comment: TCXOs have both standard tolerances and environmental characteristics such as temperature, ageing, supply voltage etc.
That 0.5ppm applies to the temperature variation. For tolerance out of the reel, they're +/-2ppm. "Trick for young players" I guess!
As an example, some 26MHz TCXOs 2ppm/0.5ppm I use, I pay $1.40ea for 1ku.
Edit: also check that the output spec is compatible with your application, some are CMOS, TTL etc square wave, but you'll also often find sine or clipped sine.