A fun tidbit about crystal oscillators is that they allowed “un-tethered” sound recording on motion picture film cameras. If both your sound recorder and your film camera are using a crystal oscillator as a reference for their motors - you can sync them up in post without needing them to be physically connected during filming.
An entire 1000 foot reel of 35 mm film only has about 15,000 frames on it, so one part in 30,000 would be good enough. When I worked in TV none of the sound equipment had ovens for their crystals.
We did have a sync generator with a crystal oven. I forget who made it. The sync generator has multiple outputs, the most important one being the color subcarrier, which is 3.579545 MHz for US NTSC (I still remember that number). It also puts out vertical and horizontal sweep signals. The stable timebase allowed us to free run for a day in case we lost the network signal for some reason. The network (NBC in our case) had a cesium clock in New York that they calibrated against WWV for time of day. We locked our clock to their signal, and all our equipment to our clock.
As a teenager, I was fascinated by live video effects and mixing multiple video streams. And I assumed that the TV station had equipment with huge amount of storage (this was in the eighties) and sync streams together.
During a visit at the Belgian national broadcast corporation, they showed the central clock generator to which all video sources were synced. It suddenly all made sense. :-)
for homelab application where extra space & power consumption is not a real concern, "temperature resistance" (tempco) is no longer relevant. you can get a constant temperature controller with +/- 0.01 degree range kind of spec for $65. verified using a reputable digital temperature sensor (outside the control loop) and the performance is pretty solid.
out of interest, what would the physical setup look like? Hard to imagine you could achieve isotropic temperature approaching +/- 0.01 degree over the size of a typical PCB.
Does it have to be isotopic though? The temperature must be constant over time, but a spatial gradient shouldn’t influence the stability of the crystal.
BTW, checkout my other comment in this thread about a GPSDO PCB with a resistor grid on the backside to evenly heat it.
But is the error constant or does it vary over time? If it's the former, it can be calibrated away. If it's the latter, what is the mechanism behind it?
It'll vary over time with the ambient temperature. When you set up a temperature control loop, you have a heater which creates a temperature gradient between it and the ambient temperature. This temperature gradient depends on the power that the heater is putting out, the thermal resistances of the box and any insulation around it. You then have a temperature sensor, which you will presumably put somewhere in the box, hopefully near the crystal, but it won't exactly be the crystal. Then, as the ambient temperature sensor varies, the power output of the heater will vary to try to keep the temperature that the sensor is seeing constant. But because that power also changes the thermal gradient, the thermal gradient in the system will also change, and so the temperature of the crystal is never completely insensitive to the ambient temperature, even if the temperature sensor reading doesn't change at all.
How big and/or meaningful this effect is depends on what the thermal resistances in the system are, and where the temperature sensor is relative to what you want to temperature control. Generally you want a very conductive box that's then insulated very well, since this means the temperature won't vary much across the box (all of the temperature gradient between the heater and ambient is 'taken up' by the insulation). But if you're talking sufficiently high precision this can be quite difficult to achieve.
Thanks for the explanation. Having thermal element (say, resistors) spread as a regular grid all over the backside of the PCB would help with reducing the gradient.
It makes me wonder if it would make sense to have a slow rotating fan inside the box. Not to get rid of excess heat but to compress the gradient against the well-insulated wall of the box. It's probably overkill for most cases, since there are other factor that influence frequency stability...
A fun tidbit about crystal oscillators is that they allowed “un-tethered” sound recording on motion picture film cameras. If both your sound recorder and your film camera are using a crystal oscillator as a reference for their motors - you can sync them up in post without needing them to be physically connected during filming.
I imagine that the accuracy requirements for those crystals are not quite as stringent as the one that I’m talking about here!
An entire 1000 foot reel of 35 mm film only has about 15,000 frames on it, so one part in 30,000 would be good enough. When I worked in TV none of the sound equipment had ovens for their crystals.
We did have a sync generator with a crystal oven. I forget who made it. The sync generator has multiple outputs, the most important one being the color subcarrier, which is 3.579545 MHz for US NTSC (I still remember that number). It also puts out vertical and horizontal sweep signals. The stable timebase allowed us to free run for a day in case we lost the network signal for some reason. The network (NBC in our case) had a cesium clock in New York that they calibrated against WWV for time of day. We locked our clock to their signal, and all our equipment to our clock.
As a teenager, I was fascinated by live video effects and mixing multiple video streams. And I assumed that the TV station had equipment with huge amount of storage (this was in the eighties) and sync streams together.
During a visit at the Belgian national broadcast corporation, they showed the central clock generator to which all video sources were synced. It suddenly all made sense. :-)
interesting teardown, thanks.
for homelab application where extra space & power consumption is not a real concern, "temperature resistance" (tempco) is no longer relevant. you can get a constant temperature controller with +/- 0.01 degree range kind of spec for $65. verified using a reputable digital temperature sensor (outside the control loop) and the performance is pretty solid.
I believe it.
This DIY GPSDO has a self-heating PCB to keep the temperature constant: https://www.paulvdiyblogs.net/2023/06/gpsdo-version-4.html?m....
It’s a long blog post, but in the August 24, 2023 update, he mentions that the PCB temperature stays rock solid at 52.9C.
out of interest, what would the physical setup look like? Hard to imagine you could achieve isotropic temperature approaching +/- 0.01 degree over the size of a typical PCB.
Does it have to be isotopic though? The temperature must be constant over time, but a spatial gradient shouldn’t influence the stability of the crystal.
BTW, checkout my other comment in this thread about a GPSDO PCB with a resistor grid on the backside to evenly heat it.
A spatial gradient between the crystal and the temperature sensor, if it varies, can cause an error.
But is the error constant or does it vary over time? If it's the former, it can be calibrated away. If it's the latter, what is the mechanism behind it?
It'll vary over time with the ambient temperature. When you set up a temperature control loop, you have a heater which creates a temperature gradient between it and the ambient temperature. This temperature gradient depends on the power that the heater is putting out, the thermal resistances of the box and any insulation around it. You then have a temperature sensor, which you will presumably put somewhere in the box, hopefully near the crystal, but it won't exactly be the crystal. Then, as the ambient temperature sensor varies, the power output of the heater will vary to try to keep the temperature that the sensor is seeing constant. But because that power also changes the thermal gradient, the thermal gradient in the system will also change, and so the temperature of the crystal is never completely insensitive to the ambient temperature, even if the temperature sensor reading doesn't change at all.
How big and/or meaningful this effect is depends on what the thermal resistances in the system are, and where the temperature sensor is relative to what you want to temperature control. Generally you want a very conductive box that's then insulated very well, since this means the temperature won't vary much across the box (all of the temperature gradient between the heater and ambient is 'taken up' by the insulation). But if you're talking sufficiently high precision this can be quite difficult to achieve.
Thanks for the explanation. Having thermal element (say, resistors) spread as a regular grid all over the backside of the PCB would help with reducing the gradient.
It makes me wonder if it would make sense to have a slow rotating fan inside the box. Not to get rid of excess heat but to compress the gradient against the well-insulated wall of the box. It's probably overkill for most cases, since there are other factor that influence frequency stability...
Was just talking about frequency references last night -- the ARRL Frequency Measurement Test is this Thursday evening
https://fmt.arrl.org/