HP 2804A : Quartz Thermometer ... teardown

[psykok]

Hi,

I just scored a nice  Hp 2804A Quartz Thermometer, don't ask why I bought that because I have no usage for this kind of instrument ... it's really just for the fun

Unfortunatly I don't get any probe but I found I workaround in the service manual to test it. Actually it's pretty easy, you only need a 28.5MHz source at 200mVRMS to emulate the probe.

And yes it works :



Sorry I tuned it on before taking it a part




















Regards,

Alex

[PA0PBZ]

Interesting frequency on the TCXO, I wonder why...

[VK5RC]

A story I heard many years ago was that the guys at HP were trying to get really frequency stable XOs and were trying different cuts of crystal and kept on getting blocked by the temperature coefficients, but they then realised they had found a great temperature probe instead!

[psykok]

Interesting frequency on the TCXO, I wonder why...

I'm not an expert but from what I understand (from the manual) the big TCXO is used as reference for the system because in the difference with a standard thermometer the probe used for this one are build from a quartz
more details:
http://en.wikipedia.org/wiki/Quartz_thermometer
http://www.metrictest.com/catalog/pdfs/product_pdfs/hp_2804a.pdf

0°C will be around 28.5MHz (this value comes from the calibration guide), each increase of 1000Hz will be a addition °C.
The resolution of this instrument is very high 0.001°C

[Vgkid]

This is pretty much a frequency counter, i wonder if yu could emulate one today. If i recall you can still get the crystals used in the probes.

[psykok]

your are right it's more a infrequence counter.

I don't know if the probes are still available from HP or from other provider but I think that the price will be for sure amazing high

Alex

[PA0PBZ]

I'm not an expert but from what I understand (from the manual) the big TCXO is used as reference for the system because in the difference with a standard thermometer the probe used for this one are build from a quartz

Yes I understand, but why not use 10.000000 like any other frequency counter?

[psykok]

You think that 10.000000 or 10.0000001 will make any differents?

Sent from my HTC One using Tapatalk

[PA0PBZ]

You think that 10.000000 or 10.0000001 will make any differents?

No, that is why I don't understand...

[SeanB]

Probably they aimed for an exact frequency, and the oscillator is marked with what it actually delivers. Better than a 10MHz crystal with 0.1% accuracy frequency wise in any case.

[echen1024]

See how it's typewritered or otherwise stamped on like the SN? That shows that that is the actual frequency of the device. It prolly was designed to be a 10MHz crystal, but that was the value they measured/.

[andy_silicon]

You think that 10.000000 or 10.0000001 will make any differents?

No, that is why I don't understand...

Being a slight timenut ...

The OCXO has a plastic plug for adjustment - and thus the chosen frequency is intentional - not random.

It is common when performing accurate characterization of high stability oscillators to compare against a known high stability oscillator. This is performed by mixing the output from the reference and unit under test.

This ( after filtering ) gives a difference frequency and assuming that the reference used is a 10Mhz reference - that would give  a difference frequency of around 0.1Hz.

Precise measurement of the variations of 0.1Hz is much easier ( interval measurement ) than 10Mhz.
You thus gain a potential improvement in measurement ability of 1x10^7

Hope that makes sense.

[PA0PBZ]

Hope that makes sense.

Yes, but wouldn't it make more sense to offset the reference so you end up with a tcxo with a 'normal' frequency?
I still think there must be some hidden reason to use this frequency in the 2804A.

[andy_silicon]

Hope that makes sense.

Yes, but wouldn't it make more sense to offset the reference so you end up with a tcxo with a 'normal' frequency?
I still think there must be some hidden reason to use this frequency in the 2804A.

We are all guessing unless the original design team stick their head into the thread :-)

If it were my super selected 'perfect' oscillator - I wouldn't adjust it.

Hydrogen maser / rubidium / cesium whatever they used. ( in fact good OCXO would be better for short term measurements )
 

[rf-loop]

This TCXO stability is quite poor. It is in class 0.1ppm/month. So this last 1Hz do not mean nearly anything. It is just stamped there in manufacturing phase and it is some kind of compromize for this eguipment adjust it to this value IF never need adjust. (mainly if there have been some stupid novice who have turned it to wrong or some kind of accidental other things. It is recommended value to adjust, as service manual tell and there is used equipment whats accuracy is 10 times more than device under adjustment. (manual service calibration procedure). So its real target (only for "one time adjustment" is 10.0000001Hz.

As manual tell, normally this TCXO do not need adjust during equipment life time! (and related to its stability specs, it  is "no matter" what is exact frequency because normally "compensation" in use is made by thumb switches on front panel. Just quess how much it have drifted after years of use.  This have also inputs for 10MHz/1MHz house standard if need (it leads better accuracy (less drift))

For accurate temperature measurements this unit is just electronic recyclable junk if do not have sensors and individual calibration modules. Also it may have some mechanical or electronic parts what can salvage for other HP equipments restorations.

This all can understand if read service manual and understand even low level basics about frequency references and principle how this equipment and these sensors works. No need think "this or that" reasons. There is not reason so that it really need be continuously exactly 10000001Hz.

But yes, tight final word just for this "why it is stamped 1Hz off from 10MHz" can give only person who have been there in HP or TCXO manufacturer.
But what ever deep true reason it do not matter much if it is drifetd some HZ up or down. But hourly/daily short term stability need be enough good.

[amyk]

There's an impressive amount of gold in there.

[andy_silicon]

Do I see date stamps on the ICs of 1977/1978 ?

[PA0PBZ]

But yes, tight final word just for this "why it is stamped 1Hz off from 10MHz" can give only person who have been there in HP or TCXO manufacturer.

Thank you for the long answer, I totally agree with you. My only question was Why the 1Hz, because whenever you see a tcxo it is generally always 10Mhz.
Anyway, I think the answer will be a mistery forever.

[psykok]

Do I see date stamps on the ICs of 1977/1978 ?

Yes  most of them are between 1976 and 1978

[alderwood]

Re the HP2804A quartz thermometer.

I have one and a probe (without the correct calibration module).

Someday I'd like to calibrate it, but for now I can make very high resolution relative temperature measurements.

It does have GPIB, so I could conceivably run a calibration against as standard and difference the readback temperature
versus actual values in the external PC.

It would be a lot more work to reverse engineer the storage of the cal constants in the plug in modules.

The 2804A was developed and produced at the (long gone) Frequency and Time division in Santa Clara, CA.

The HP2804A is descended from the Dymec 2800A quartz thermometer which you can read about at
http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1965-03.pdf

Dymec = an early manufacturing arm of HP.

On page 3 of this HP Journal document shows the analytically determined and then experimentally tweaked values of cutting the quartz crystal to zero the second and third order terms at one angle of 11.17 degrees and the other at 9.39 degrees.

So HP did not simply accidentally arrive at this crystal cut, it was calculated and then refined by experimentation. 

One can't simply substitute a different standard crystal and get the correct temperature performance as linear temperature coefficient crystals are uncommon.

I read a story that the product was proposed to upper management with NO market research.  When asked who would buy a high resolution, very accurate thermometer, the engineers said something like "I don't know, but someone will want it."

So the complete progression is dymec DY-2800A and dymec DY-2801A (two channel) and eventually the HP branded HP 2804A.

In 1965 the one channel 2800A, with one sensor went for $2,250 and the two channel, with two sensors, went for $3250.

In my experience, the probes are rare and even rarer with calibration modules, but there is always someone on Ebay selling a HP2804A without probes.

It seems people hold onto the instrument and discard the valuable part, the probe..

BTW, I pulled the cover on mine and the clock shows 10.000 MHZ, so the 10.0000001 MHZ marked on the other version appears to be something the vendor wanted to write on the tag.

[David Hess]

The Tektronix 7D15 universal timer/counter has the same issue with the reference oscillator being deliberately calibrated off frequency.  I think andy_silicon has it right with the reason being for easier calibration but the HP2804A service manual should say something.

From page 5-12 of the 7D15 service manual:

The displayed waveform will drift slowly to the left or right.  This represents a positive or negative clock error.  A drift to the left represents a positive error and a drift to the right represents a negative error.  With a Time Base sweep of 0.1 microseconds per division, a drift of one division per second equals a clock error of 0.1 Hz.

The 5 MHz oscillator in the 7D15 has a frequency error to which is should be set.  This frequency error is printed on the oscillator.  The relation of "clock error" to "frequency error" is equal to the formula: frequency error/5 = clock error.  For example: a frequency error of -1.5 Hz is marked on the 7D15 oscillator.  This represents a clock error of -1.5/5 or 0.3 Hz.  The oscillator frequency, therefore, should be adjusted for a drift of three divisions per second.  Since the error is negative, the drift should be to the right.

[alderwood]

I popped the cover on my second spare parts 2804a and the oscillator has 10.000 MHZ stamped on it at well.

None of the 0.1Hz resolution seen on the other oscillator.

I believe it was an overzealous technician at the oscillator vendor that printed on the 0.1hz resolution on the oscillator in question.

Plus as others have pointed out, the 2804A will work from a 1MHZ external reference, and the manual refers to it as such.

The manual on page 1-2 "for high accuracy measurements an external time base may be used.  The 2804A Quartz Thermometer requires an input voltage of 1.0V rms into 1K ohm via the rear panel BNC from a source of 1 MHZ."
 
This is spec'ed as 1MHZ, so scaling up by 10 gets to the 10MHZ desired, not 10MHZ + 0.1HZ.

[alexwhittemore]

Being a slight timenut ...

The OCXO has a plastic plug for adjustment - and thus the chosen frequency is intentional - not random.

It is common when performing accurate characterization of high stability oscillators to compare against a known high stability oscillator. This is performed by mixing the output from the reference and unit under test.

This ( after filtering ) gives a difference frequency and assuming that the reference used is a 10Mhz reference - that would give  a difference frequency of around 0.1Hz.

Precise measurement of the variations of 0.1Hz is much easier ( interval measurement ) than 10Mhz.
You thus gain a potential improvement in measurement ability of 1x10^7

Hope that makes sense.

That is fucking awesome! Like a null detection setup for measuring frequency vs a reference instead of voltage.

I learned something today, thankyou!

[alderwood]

Another way to get high accuracy on an TCXO adjustment is to use the oscilloscope procedure shown on page 5-7 of the HP5335A counter service manual (05335-90005.pdf), (may be on the Agilent/Keysight website).

The reference frequency is put in the external sync input of the scope and the time base to be adjusted is put to the input of the scope and the scope is set to trigger on the external sync. 

This allows one to use different reference frequencies as they need be sub harmonically related to the oscillator being adjusted.

The picture in the manual shows 100KHz, 1MHz, 5MHz or 10MHZ can be used to adjust the 10MHZ time base out of the HP5335A.

If the two frequencies(reference and timebase) are exactly harmonically related, the displayed signal will not shift as the trigger will happen at exactly the same time in the waveform for each sweep. 

The manual has the user time the sidewise movement of the trace to determine the actual error.

[alderwood]

If one is interested in temperature sensing quartz crystals, go to http://www.statek.com/prod_sensors.php

They make some linear temperature sensing crystals that oscillate at frequencies such as 172kHz and 262kHz.

The crystal used in the HP2804A is in the 28MHz range, so these statek crystals are in no direct way usable with the HP2804A.

The slope of frequency change per C degree is 46.4ppm/C so at 172KHZ, a 1 degree C change moves the frequency about 8Hz, so to resolve 1 millidegree requires detecting a 0.008Hz deviation.

This requires a time interval measurement over a number of cycles of the 172KHZ waveform to achieve the 1millidegree resolution of the HP2804A.