My Vacuum Tube Frequency Synthesizer

[BobW]

I'm looking at using a tuning fork crystal in an oscillator, and I'm trying to figure out the maximum safe drive level. The 38 kHz crystals that I have on hand are these:
Datasheet: https://www.mouser.com/ds/2/137/epson_1705528-1165749.pdf

They're rated for a maximum drive of 1µW. I'm trying to relate that to maximum rms voltage across the crystal. Looking for some sort of effective resistance at resonance, I find the only resistance value on the data sheet to be the motional resistance. Is this the number that I need? For my crystal, the motional resistance is 35k. Using the P=v^2/r formula, I get a maximum rms voltage across the crystal of 0.187 volts. Does that sound right?

I'd been thinking of using inverse parallel diodes across the crystal to protect it, but even Schottkys would have too high a forward voltage for this. Plus, the specified motional resistance is a maximum value. If the motional resistance is smaller, then the maximum voltage would also have to decrease. This implies that I should be limiting current rather than voltage, but that could be a problem in the circuit that I have in mind.

FYI: Yes I'm aware of several circuits specifically designed for these crystals, but I need to use these in a vacuum tube oscillator, which makes life more interesting.

 

[BobW]

Update:
I found an old appnote on using these crystals with a FET. A triode isn't that much different. So I was able to adapt the circuit.

Works like a charm, but I had to add the back to back zeners to limit the voltage in the feedback loop. Without it, the voltage builds up to many tens of volts, and then the crystal pops. I wasn't fast enough disconnecting the HT line one time, and so I blew one crystal.

 

[BobW]

Further update:
An even better solution uses a pentode instead of a triode, with the feedback taken from the screen grid, and the output taken from the plate.

With the output and feedback isolated from each other this way, the plate load can vary widely with no effect on the feedback level, and the oscillator remains completely stable. Plus, the plate can deliver a much higher output voltage & current this way. These pentodes are an amazing invention. I wouldn't be surprised to see triodes become obsolete.

 

[dr pepper]

A microcontroller manufacturer, I think it was microchip show in their datasheet if you run 38kc watch type xtals on timer1 you put a 220k resistor in series at 5v ti limit the drive.
I killed a couple of 60kc xtals a while back with too much drive, only took a few uW.
Depletion mode fets work much like a thermionic triode.

 

[BobW]

Yes. The crystals that I have, are rated at 1 µW maximum drive, which is extremely low.

I must say that I'm surprised at how well it works with a vacuum tube, either triode or pentode. In the past, I've had a lot of problems with these crystals trying to get them to oscillate at all, using logic gate oscillator circuits. It appears that the values of the passive components (feedback resistor, damping resistor, and loading capacitors) are critical. Once you get them figured out, it works well.
BTW, this is the source of my information:
https://www.aldinc.com/pdf/UltraLowPowerCrystalOsc.pdf

Anyway, my goal is to build a 20 kHz oscillator as part of a PLL frequency synthesizer project, but I decided to get the oscillator working with the crystals that I have on hand, which happen to be 38kHz. Now that I know it works, I'll order some 20kHz crystals, and then move on with the next phase of the project.

 

[dr pepper]

Gain & phase have to be right for oscillation.
Tubes being high impedance might be helping.
What tube are you using?, here in the Uk it'd be something like an Ef86.

 

[schmitt trigger]

Indeed pentodes were an amazing improvement over triodes.

As a matter of fact and taking a cue from vacuum tubes, "tetrode" transistors were indeed investigated. There is an oral history on the website http://www.semiconductormuseum.com/Transistors/GE/OralHistories/Suran/Suran_Index.htm
They did not work, but the invention evolved into the unijunction transistor. I won't spoil it for you, read the website. Fascinating!!

That will answer the ages-old question: how come the unijunction's terminals are labeled Base1, Base2 and Emitter?

 

[BobW]

Thanks for the unijunction link. Interesting article. I hadn't realized that it was an accidental discovery.

What tube are you using?, here in the Uk it'd be something like an Ef86.

The original triode version used a general purpose 6C4 (EC90). Then I switched to a 6AU6 (EF94?) pentode. I'm sure it would work with an EF86 too. I think it would work with just about any triode or pentode, because there's nothing critical about the circuit, once the correct passive components are picked. I've now switched to yet another tube, a 6U8 (ECF82) triode-pentode dual section tube, because this project is going to get more complex, and it'll save space to combine more functions into single envelopes. The crystal oscillator now uses the pentode section of the 6U8, and the triode section is used as a blocking oscillator that's synchronized to a chosen subharmonic of the crystal oscillator, so that I can generate various accurate reference frequencies from the master crystal oscillator. The blocking oscillator was last night's project. This evening's project is to build a JK flip-flop, probably with a 12AT7 (ECC81) dual triode. That will be an interesting challenge. If I can get that working satisfactorily (i.e., switching fast enough), then the rest of the project should be fairly straightforward.

 

[schmitt trigger]

An Eccles-Jordan flip flop?
That should be straightforward.

 

[BobW]

Yes, except that because it's direct coupled I have to figure out the biasing and make sure that the input levels are compatible with the output levels. At the moment I'm trying to figure out whether I can get away with a single supply, or whether I'll need to add one or more bias supplies.

 

[dr pepper]

Sounds like a plan, no doubt it will work and work well.
So do you plan to use thermionics in the final application, or is this a prototyping stage.
I've seen the Ecf82 used in Fm receivers, cant recall what function, maybe the Fm front end.

 

[Diver300]

I'm looking at using a tuning fork crystal in an oscillator, and I'm trying to figure out the maximum safe drive level. The 38 kHz crystals that I have on hand are these:
Datasheet: https://www.mouser.com/ds/2/137/epson_1705528-1165749.pdf

They're rated for a maximum drive of 1µW. I'm trying to relate that to maximum rms voltage across the crystal. Looking for some sort of effective resistance at resonance, I find the only resistance value on the data sheet to be the motional resistance. Is this the number that I need? For my crystal, the motional resistance is 35k. Using the P=v^2/r formula, I get a maximum rms voltage across the crystal of 0.187 volts. Does that sound right?

It depends what load capacitance you have. I can't find that Epson crystal at Mouser, but they typically have a load capacitance of 12.5 pF, and around a 30 kΩ ESR (equivalent series resistance).

The crystal's equivalent circuit is a tiny capacitor 2 fF or 0.002 pF in series with a huge inductor and the series resistance. At the correct operating frequency, the overall effect of the capacitor and the inductor is and inductor that resonates with the 12.5 pF of the external circuit at 38 kHz. Ignoring the resistance for a moment, the impedance of the crystal will be -j/ωC and the impedance of the external circuit will be j/ωC

At 38 kHz, the magnitude of the impedance will be around 335 kΩ.

Now looking at the series resistance, if the current is such that 1 μW is generated in the 30 kΩ, the current will be around 6 μA, and the RMS voltage across the crystal will be 1.93 V.

That sounds a lot more realistic than 0.187 V, in my experience of crystals. The 5.1 V zeners will probably keep the power near to what it should be.

I don't think that you need the 6.8 MΩ resistor.

 

[BobW]

At 38 kHz, the magnitude of the impedance will be around 335 kΩ.

Thanks. That makes more sense. The datasheet seems to be short on details. I would have thought that they would either provide resonant impedance or else all of the motional parameters. They don't give the motional inductance, though I guess it can be calculated from the other parameters.

I included the 6.8 Meg. resistor to stay as close to the app note recommendations as possible. Once I got the oscillator working well, I left it alone. Since the tube is inherently in its linear region it wouldn't seem to be necessary. I'll have another look and see if it starts up reliably without it.

On the flip-flop adventure, I did some digging through various ancient texts (IBM Model 604 Manual, and others). There appears to be very consistent tube logic design standards which are a big help. The logic levels from several manufacturers are identical at +50V for low level, and +150V for high level. Even the biasing components appear to be identical:
Plate (anode) load resistor: 20k
Plate-grid feedback: 200k
Grid bias: 200k
Plate supply: +150V
Grid bias supply: -100V
I might be able to use a cathode bias resistor to avoid the negative grid bias supply, but it's probably not a good idea to reinvent the wheel. Adding a low current negative supply won't be a big hardship.

 

[schmitt trigger]

Hopefully you can share some of those vintage flip flop schematics!

 

[BobW]

This was my primary source for voltage levels and biasing:
http://bitsavers.org/pdf/ibm/604/227-7609-0_604_CE_man_1958.pdf
bitsavers.org appears to be a huge resource of ancient computing equipment info, but I haven't given it more than a cursory look.
The JK flip-flop I intend to use is a modification of a transistor circuit from one of my old electronics text books, because the IBM 604 manual doesn't give any examples that have J and K inputs. If/when I get it working, I'll post the circuit.

 

[schmitt trigger]

BobW;

What an incredible manual!
Back then 1958, when IBM was the world's largest computer maker....by a wide margin.

 

[BobW]

Since this thread has diverged from the original question, I've changed the title. Hope this doesn't make things confusing.

Here's an update of the circuit after tonight's festivities. The diagram now includes the blocking oscillator that's used as a frequency divider, and the JK flip-flop.

I wish everything worked as well as the blocking oscillator. Its pulse width is set by the transformer characteristics, and the repetition rate by the RC network. By adjusting its RC timing network values, it will divide the crystal oscillator by all kinds of different values, including non-integer values. I have it adjusted to divide by 3.5 at the moment, so that it produces 10.857142 kHz from the 38 kHz reference, which is close to the 10 kHz frequency that I'll eventually be working with. The transformer is one that I removed from a dead switching power supply. I don't know much about it except that primary and secondary inductance are both 20 mH, and turns ratio is 1:1.

The flip-flop is now working, but it was a bit tricky:
- The -100V grid bias is fairly critical. I found that if it varies by more than ±5V, it stops working.
- The blocking oscillator puts out a very good negative going pulse with very fast leading edge which is ideal, but I found that the pulse duration was too short to trigger the flip-flop. So I added the pulse stretcher. This still gives the fast leading edge, but lengthens the pulse enough that the flip-flop switches reliably.

The J & K inputs appear to work, but more testing will be necessary. I'm concerned that their setup time may be too slow. The flip-flop will be used to sample and hold the output of the VCO (yet to be designed), and so the J & K inputs will have to react quickly.

 

[atferrari]

Further update:
An even better solution uses a pentode instead of a triode, with the feedback taken from the screen grid, and the output taken from the plate.
View attachment 111325

With the output and feedback isolated from each other this way, the plate load can vary widely with no effect on the feedback level, and the oscillator remains completely stable. Plus, the plate can deliver a much higher output voltage & current this way. These pentodes are an amazing invention. I wouldn't be surprised to see triodes become obsolete.

Could you show a picture of your implementation?

 

[schmitt trigger]

What are you planning to stabilize your power supplies? Perhaps an OA2 tube?

 

[BobW]

So do you plan to use thermionics in the final application, or is this a prototyping stage.

If I can get it to work reliably, and with a reasonably small parts count, then the plan is to use tubes in the final build.

Could you show a picture of your implementation?

Pictures are attached. A couple of scope traces are included. In the first one, the scope shows the 38 kHz crystal oscillator waveform in the lower trace, and the 10.857 (divide by 3.5) blocking oscillator output waveform. Note that dividing by a non-integer number is not recommended, because it requires very fine and critical adjustment of the RC network. Integer division is fairly non-critical.

In the next photo, the scope shows the blocking oscillator trace again (upper), and this time the lower trace is the output of the JK flip-flop, which is triggered from the blocking oscillator output, and is running at half the blocking oscillator frequency. If the J and K inputs are tied to logic 1, or left unconnected as they are in this case, then the output alternates state on each trigger pulse.

Below are closeups of the breadboarded circuit. The upper breadboard is the crystal oscillator (left side) and blocking oscillator (right side). The 6U8 tube is lying on its side above the breadboard. The lower breadboard is the JK flip-flop, and the rectifier/filter for the -100V bias supply. The bias supply transformer is powered from a variac (grey box on the right) so that I can adjust the bias voltage accurately. The main power supply is the grey box in the upper part of the photo. It supplies both +150V and 6.3VAC for the filaments.

What are you planning to stabilize your power supplies? Perhaps an OA2 tube?

The mains power here is quite stable. So, I'm hoping that I won't have to add any special circuitry to stabilize the power supplies. But if I do, I'd probably look at a tube type regulator. It would be cheating to use a solid state regulator when everything else is built with tubes.