HP E1938A OCXO teardown / repair

Last update: Thu Jul 3 16:47:11 CEST 2025

Background & documentation references

There is a considerable amount of interesthing documentation available on the E1938A here. The Papers mentioned towards the end of this web page I found especially interesting as they also describe the actual practical implementation of the OCXO. Richard Karlquist, one of the designers of the E1938A has a lot of interesting information on his personal web page here. Tom Van Baak's LeapSecond.com website has more info on the E1938A.

Ongoing work

July 3, 2025

Been offline for a bit, visiting HAMRADIO 2025 in Friedrichshafen, Germany.
That said, I am back to a sealed oven, as detailed below. Thanks to Rob, I have added a nice (but albeit provisional) picture of "what LED indicates what" as well.


A local shop has supplied me with a new O-ring for the oven. Unfortunately Viton O-rings were not in stock, but the natural rubber O-ring is speced for 105 degrees centigrade. Short of using the sealing lacquer HP were using for the 3 Torx screws in the oven lid, I used German "Sicherunglack". Should do for now I think. If it lasts for another 20 years "it is not my problem anymore".


DS4, orange LED, is switched on continously after powerup, during oven warmup it blinks. Once the temperature setpoint has been reached DS4 will be switched off. With the oven insulation removed and when blowing cool air over the oven, DS4 will switch on momentarily during heating. This corresponds to the current drawn from the 5V power supply. At 'oven at setpoint' the current drawn is about ~ 1.5A from the 5V supply rail. This observation was at about 30 degrees centigrade ambient, in case anybody is interested.

DS7, green LED, blinks at a rate of 1 Hz (? I think, not having measured it). Appears to be a heartbeat indicator.

DS6, red LED, is on if something is wrong. Sofar I saw it light up while the oven heating problem was present. I expect that it is a generic "failure detected" indication. Have no way to prove that theory though.

DS5, yellow LED, "no idea" for now.

S1 is the reset switch for the PIC MCU. And J101 is the TTL-level serial port used to communicate with NGOOMM.

June 25, 2025

I got sidetracked a bit, spent my hobbytime on repairing two Advantest spectrum analysers. Now that they both work again, back to the E1938A.

Very interesting phenomenon today: upon powering up the oven heater started and regulation kicked in (as designed, I assume). That was a one-time effect though. In other words: intermittent problem. Great. Not. Based on some consideration, and while muttering the famous "Who Dares Wins" after maesuring things in the thermistor circuit, I decided that the thermistor flex wrapped around the crystal cavity must be flakey. A bit of "what else can it be?" Thanks to Paul PA1OM, for serving as a willing sparring partner on this control systems issue. Control systems were never my favourite subject in tech uni :-)


Flex disconnected from the oven feedthrough pins and replace by an adjustment pot. The flex did not improve during desoldering, but was knackered already, so...


That proved interesting. Bottom DMM shows the divided-by-two reference voltage generated within the oven. Top DMM shows the voltage from the thermistors, or rather: from the adjustment pot in this case. And the heater draws considerable power.


Long story short: I cut the flex wire connecting the thermistors to the feed-through pins and replaced them with some Tefzel insulated wire wrap wire. I guess Rick might take a dim view of this but lacking an original replacement flex you have to make do with what you have available...


Test setup running now the flex has been bodged..


During warmup the output voltage on the thermistor bridge slowly drops (top DMM). The reference voltage (2.5V divided by two, and produced inside the oven) shows no major change (bottom DMM). Note the 100.93 degrees centigrade on the oven label.


There are now a green and a yellow LED blinking, the red LED is no longer lit. Appears to me like a good sign.


Shown on the left the original parameters set on the main PCB's PIC. On the right the monitoring of the oven control loop.

Remaining questions/actions

June 16, 2025

Back to the project now that temperatures here have become more reasonable. I was getting very curious on what the NGOCOMM software tool can or cannot do to help, I decided make that run. Proved a bit more work than anticipated as it is a Win32 executable and Windows 11 will not run it. A WindowsXP VM in VirtualBox helped. An FTDI USB-serial TTL converter connected to J101 was used for communication.

Lacking any docs, it took a bit of tinkering to get NGOCOMM talking to the E1938A. Trick is to first do a Sync PIC and subsequently a Get RED. This fills the NGOCOMM display with the parameters of the PIC controller.



I guessed selecting the Ramping option should allow me to get heater control. Alas, nothing visible happened, no heating current was drawn.


Far more interesting was opening the control loop using the Loop Control Opn button, followed by a Send RED.
Immediately the heaters started drawing about 3.3 A at 5V. Closing the control loop with the oven hot immediately disabled the heaters.

The Loop Status window shows various measurements. With the Cnv checkbox checked the hexadecimal values are translated into something more human readable.

Observations sofar (unordered)

Open questions

June 11, 2025

First an answer to the question on the time-nuts list on why I was (maybe still am) suspicious about the flex circuits. Pictured below is the bottom side of the oven, showing the hermetically sealed interconnect pins into the oven cavity. I have to admit that sofar none of the measurements have to shown any issues with respect to the flex circuits. Apart from appearing brittle and making me very reluctant to really touch them.


Based on Bob Camp's questions on the current drawn by the heaters I have again pulled the 3 power FETs and measured things in a somewhat structured fashion (I hope..).

For the 'hot' measurements I allowed the oven enough time to reach 10MHz nominal, under the assumption that that is a reasonable indicator that the oven temperature has reached the correct value.
For the WARM_UP_HTR that took approximately 25 minutes, for the RIM_HTR and MASS_LID_HTR this took approximately 10 minutes.
In-between measurements I cooled the oven back to room temperature (~ 24 degrees C) using a fan.

The table below shows the outcome of it all. Measurements with respect to ACOM.

HeaterFETCurrent coldCurrent hotGate voltage coldGate voltage hot
WARM_UP_HTR Q80.88A0.75A--
MASS_LID_HTR Q92.24A1.90A-5.09V-4.88V
RIM_HTR Q101.24A1.05A-5.07V-4.84V

So, yes, the current drawn by the various heaters indeed drops when the oven heats up. Tempco of the heaters themselves I guess.

Based on the change in gate voltage it appears to me that the temperature sensing is working, the digital PI^2D is doing its thing etc. At least to a certain extent.

Again I noticed that ever so often on powerup the E1938A starts up drawing ~3A for a couple of seconds and then falls back to 60mA. For total lack of better ideas I get the feeling the PIC chip senses something on powerup that it does not like and shuts down all heating. With the FETs out, the DAC drives the gates of Q9 and Q10 based on the temperature it measures???? (lots of ?).

With the oven hot, I checked the VREF_2.5 on the DB25 connector. Initially what I measured with respect to VREFCOM did not make sense. Measured with respect to AGND it was 2.538V Not good, as it is not "the HP way" to be off like that. Some close inspection later: on my unit R200, a 0 Ohm SMD resistor was not populated. R200 should be at the empty pads just above the big DALE resistors, to the left of the DB15 connector. Once a fresh R200 had been installed, I had a nice 2.5018V for VREF_2.5 with respecto to the now proper VREFCOM.

All in all: lots of entertainment, but still no clue what is going on. My current gut feeling tells me "something fail safe kicking in in the PIC firmware". But why, what is causing it? Must be someting in the oven itself, based on what Morris wrote earlier.

June 10, 2025

I spent a bit more time on the E1938A. The "Hmmm" part in the June 6 write-up is (of course, in retrospect) going nowhere. I assumed I had isolated the thermistors from the rest of the circuit, and of course I had not. Duh..

Today I checked all supply voltages, they appear to be what they should be, so +5VD, +5VA etc. The VREF2.5 originating from within the oven is also reasonably OK at 2.534V. As the oven is not currently heating up, I assume the voltage reference will also be working properly when at the nominal ~100 degrC setpoint.

During testing I noticed that sometimes the E1938A 'boots' up drawing ~3.5A from the 5V rail. That is only for a few seconds, dropping back to say 50-70mA. More often than not it does not start with the high-current surge, instead just drawing its usual 50-70mA directly from power-up. I have yet to find any pattern in this behaviour, as I suspect that solving this will be the fix for the entire 'oven does not heat-up' issue.



Surgery in progress.

For good measure, I pulled the 3 power FETs that drive the 3 different heaters for inspection. The PCB material shows signs of heat stress under Q9 and Q10 (the ones with heatsink). Below the heatsinks there was some environment pollutant, now cleaned off. The FETs were identified as FETs by my little Chinese gizmo/component tester so I assume they are ok.


With the FETs out of circuit I could test the actual heaters by simply connecting the drain-source on the PCB (read: the tweezers on the photo is the D-S short). All three heaters, so WARM_UP_HTR, MASS_LID_HTR and RIM_HTR work. During warmup the current gradually reduces from ~3.5A to ~0.8A. All of this is good, proves all flex wires/heaters work. No shorts to ground, no unsound drawing of power. Flexes are unrepairable (by me) in case there had been a failure.

I noticed a CURR_LIM section on the schematic. Wondering if that was maybe kicking in inadvertently, I pulled R124 to see if that made a difference. It did not.

Allowing the oven to get really hot (I suspect around the nominal 100 degr C, do not currently have a contact thermometer available) makes the oscillator go to 10.000000 MHz. Likely good news I think. Get the oven fixed and the frequency will also be OK again.
What I also did: connect an FTDI TTL level serial-USB dongle to J101, the serial port. I think it is 9600 baud (not certain). Whatever it emits is, based on what I saw up til now, not intended for human consumption. Reading the Development Paper I understand the serial can be used to set oven control parameters etc. Maybe (I hope...) also read some status on what is going on. In any case I need more information on what it is the serial port can do, and maybe dedicated software to actually use it?

Also useful would be the meaning of the 4 LEDs DS4, DS5, DS6 & DS7. DS7 is blinking, looks like that syncs with what is emitted on the serial port. A heartbeat indicator maybe?

Any information is more than welcome! Suggestions too of course.

June 6, 2025

Bob Camp @time-nuts, far more eagle eyed than I, noticed that the temperature printed on the oven label does not match the main PCB label. Morris Odell @time-nuts, the original donor of the dead E1938A, has since confirmed that this mismatch is due to fitting a replacement oven assembly from a donor unit. The oven I have here has seen an ~20 year 24x7 operation. The mainboard I have is from the donor unit (as that(my) mainboard did not want to reach GPS lock when fitted in Morris' Z3815A. Morris confirmed the ~3.5A peak current on powerup, dropping to ~50mA steady state. The oven I have no longer warms up. Morris' replacement oven mounted on the original mainboard works just fine. It also worked fine on the mainboard I now have here. Provisional conclusion is that the 'not warming up' problem is somewhere inside the oven, so not on the mainboard.

Based on that I used my stereo microscope to take a closer look at the inside of the oven, especially at the 3 thermistors and the flex that interconnects them to the PCB inside the oven. Unfortunately my microscope does not have a camera mount, so please bear with me for the pictures made using the microscope...


Context for the next set of pictures. And the header pins/flex connection for the thermistors. The flexwire definitely has degraded over time.


The flex, once carefully bent, shows more of the cracking :-( Adjacent a close up of thermistor #1. The thermistors are mounted using a flexible polymer, my guess is that this is a thermally conductive silicone compound.


Thermistors #2 and #3 in close up.

Now the interesting part: thermistor #1 measures at 20.09 kOhm, #2 at 20.37 kOhm, #3 at 19.93 kOhm.

The corresponding section of the schematic is shown below.

Measuring the resistance on the header pins of the thermistor flex (PIN 1 and PIN2 at the right hand side of the schematic) shows 14.46 kOhm. Measuring on the DB15 connector Pins 2 and Pin8 (DB15 interconnects the oven to the mainboard) also shows ~14.46 kOhm.

Hmmmmmmm.....

First test


After wiring up the module to 5V and 12V and powering up, the initial current drawn from the 5V supply was ~3.5A. After a couple of minutes the current dropped to about 50mA. The outside of the stainless steel clamshell did not feel notably warmer than before powerup. The frequency produced was 9.9998935 MHz, so way off, and likely due to the oven not arriving at the right temperature. Which is also how it was 'advertised' "does not warm up".

Interestingly enough, after the first powerup and after cool down the initial 3.5A current did not re-occur. Which is unexpected.


Reset button and some status LEDs, some blinking. Unfortunately no information found yet on what each of them signifies.

Getting inside

Time to get inside and take a look at what is going on..

The stainless steel clamshell containing the actual oscillator oven is easily removed by removing 4 Torx screws and subsequently prying the bottom half of the shell loose with a small screw driver.
The light brown 'cookies' are insulation foam.


The flex PCB contains the heating wiring. Note the 3 Torx screws, sealed in some lacquer, to ensure the hermetic sealing that is referenced in the design Papers. Again a label with the temperature setting. Side note for the SI folks: 100 degr C aka 373.15 K is the temperature of boiling water. So when operational this oven is truly hot...


The actual OCXO stainless stell 'hockey puck' removed from the clam shell. Again a sticker with the temperature this particular crystal requires for optimal stability. To the left side 6 feed-through pins that connected the flexwire/DB15 connector to the electronics PCB inside the oven. Two more feed-hrough pins on the right side.


The sealing lacquer proved to the easily removable. It is very hard and brittle, it chipped off after some gently prying. Damaging the heater flex would be bad news indeed... Once the lacquer was removed, unscrewing the Torx screws was uneventful. Hurrah!


Lacquer chips and the inside of the lid of the stainless stell hockey puck. The flex to the left connects to the heater. The yellow discolouring on the stainless steel lid puzzles me, maybe out-gassing of plastics inside the hot oven set onto the lid?


"We're in like Flynn", to paraphrase 'Dave from Australia' (yes, EEVblog). All in all far easier than I had feared after reading the "hermetically sealed" story in the research & development Papers. In fact, I had dark suspicions of a "not repairable" unit. Quite a relief therefore. A truly elegant design! The green blobs on the centre ring surrounding the crystal are the 3 thermistors. They are connected to the electronics module via a small, 2 pin flex/header pin connector.


Flex PCB connecting to the crystal, with what looks like a trimmer capacitor to the left. The green blobs is glue fixing the thermistors to the crystal enclosure.


Another flex PCB is glued to the outside of the crystal enclosure, serving as the interconnect of the three thermistors. Each thermistor measured at about 18.8 kOhm. The resistance on the 2 header pins on the flex is about 14.5 kOhm.


All thermistors have their own portrait. Added numbers to identify one thermistor from the other.


The O-ring mentioned in the development Paper. It is totally shot, it breaks if you tap it on the table. Most likely degeneration of the O-ring's polymer due to heat.

As received

To establish reference on what exact model/revision of E1938A I received from Morris some pictures of the module in its 'as received' state. It fortunately survived its long trip from Australia to the Netherlands without incidents.


Note the temperature, 103.70 degrC on the label.


Detail pictures of the components used. Not sure what to think of the "100 C -560 pF" marking.


Detail pictures of the components used.


Detail pictures of the components used. Quite a bit of flux residu on the hand soldered components.


Detail pictures of the components used.