Tube Tester/Matcher Talk

[R.G.]

I'm aware that there are some of these. But I haven't gone and read about them in any detail.

I'm sure that they're fine instruments. The work to be done doesn't have much mystery - set the conditions, measure the results, store the results, move on. Each different design will have its own particular mix of characteristics. I'm mostly doing this for the fun of designing it.

This one grew out of the matching speculations in another thread, as opposed to tracing curves. Tracing curves is interesting, and produces nice pictures, but I spent too much time trying to figure out what this or that wiggle in the display meant and trying to deduce what the devil that shape meant. The point of this one is that it makes it possible to just take the individual data points as a mass of data, and then from that compute things about the tube - like current per grid voltage change over both macro and micro ranges. And then to figure the rate of change of the conductance over plate and grid voltage variations, etc. Basically, whatever you can think of to compute about a tube. It seems like a huge amount of data, but it's hard to buy a spinning hard disk under a terabyte these days, and even wimpy PCs run at a billion operations per second, so they can do a good job of hashing together answers from a mass of data.

And of course, if you ever get the data points for a tube into a computer, you don't have to erase it.

Hmm. I suppose that if one wanted to do that, one could make an adjustable heater supply, and measure the effect of varying heater voltage on plate current, as well as whether this affected the gm over the plate current ranges. It's a fertile field.

[martin manning]

Sure, designing your own is great, but you should do some benchmarking. The uTracer does all of the things you mention, but you might come up with something better. Re tracing vs. measuring characteristics at a point, I have found that nonsense readings can result from a test at a given operating point due to oscillation, which tracing will quickly reveal.

[R.G.]

I had actually (edit) thought (end edit) a bit about how to deal with noise and the possibility of oscillation. The simple approach that occurred to me is to add on a module or two to detect and measure that. Noise and oscillation are largely AC phenomena. The thought design so far is to just have the computer gather a lot of data easily on a succession of tubes just placed into the socket and replaced with the next one when the metaphorical green light comes on. That is, the matching background of this approach is to get a flat of about 100 12AX7s and plug them in one at a time (sticking a numbered label on them as you do this so they can't hide their identity later).

The time to test is dominated by the time to plug/unplug a tube and the heater warm up time per tube, with warm-up being the biggie. If you built a fixture instead of a single socket, you could perhaps put 10 sockets in place, and have the machine sense a tube in the socket by reading heater current presence. Conceptually, the machine would light an RGB LED by each socket with red for "no tube sensed", blue for "waiting for warmup or testing" and green for "data gathered, swap tubes". With half a dozen tube sockets in place, you could have the human feeding tubes to the tester occupied almost full time swapping tubes while the machine managed sensing, warmup, and data gathering. I just spotted an issue with how the machine would preserve per-tube identity, but ... oh, OK, put a dymo label printer on the PC running tests and have it spit out a numbered tube label as it ran a test. The human would have to stick the label onto the tube, but the machine could then make the bridge between socket #5 and tube ID Batch_7_12Dec20#18 for later digestion.

Obviously I'm making this up as I go along, but then that's what design is about.

But back to AC stuff. One could allow the machine to un-short a plate resistor and sense the AC coming off the plate through a cap leading to a signal conditioning clot of opamps. High gain opamp for sensing noise, lower gain for reading AC on the plate with grid and plate supply held steady, and AC magnitude out of the plate for a steady cycling of grid voltage up and down under program control. It's easy enough to make an AC peak detector that isn't perfect, but would tell you things like the AC gain from grid to plate with a nominal plate load, amount of noise, and amount of heater feed through. That's kind of far afield from a tube matcher, but not conceptually difficult.

[roberto]

Post Post Scriptum:
http://www.valvewizard.co.uk/analyser.html

R.G., I suggest you to deeply read at least this. First time I saw it months ago I considered as a very good starting point to implement more things.

[R.G.]

OK, scanned over it. It's one variant of what I have in mind, same overall approach, different details.

Differences:
- Different approach to pulsing the plate and screen voltages high, then low. That one uses a digital controller to pulse-charge storage caps up to a desired high voltage, then connects them to the tube being tested for the measurement. My approach is to use a DAC, then simply amplify that up to the desired voltage for as long as needed, then back down - much the same approach as that one uses for grid voltages. In fact, I had in mind a similar DAC-> amplifier with asymmetrical supplies for all of the voltages.
- Different current sensing. My idea was to just measure this directly with Hall effect modules, that one uses resistors and opamps to convert/scale currents.
- Different controller, probably. Although I haven't got that deeply into it, my controller would probably be a Raspberry Pi instead of two Arduinos. Not much difference excepting that like programming the Pi better than the Arduino.

Overall it's like I mentioned - the actual work to be done by the hardware is remarkably the same: set the measurement conditions, make the measurement, un-set the conditions, store the data, then do the next measurement. Rinse and repeat. It's fairly simple to compute and display the curves if you just like curves.

I probably ought to finish the block diagram for clarity.

[R.G.]

Forgot: Simulation says that the high voltage amp will run the plate/screen output from 0V up to 400+ in 200uS; discharge is faster, about 50uS. This needs to be tested on the initial prototype, of course, but it suggests that running one test point per couple of milliseconds is not prevented by the amplifiers.

[R.G.]

Started the PCB layout for the plate and screen driver amplifiers. It's a mechanical task, like martial arts practice, but it makes the mind consider things during the work. I'm not used to having to use high wattage resistors just because the voltage is high, but the process of layout snapped my mind back to worrying about each part's power dissipation. Turns out that the voltage divider resistors for the voltage splitter P-MOSFETs need to be at least 1W each! I was just worrying about the MOSFETs.

This path of thinking got me back into worrying about what's on the high voltage amp board, and it occurred to me that the DAC, an ADC converter, and possibly a floating current sense setup should be on the amp board, as these are much physically smaller than the power parts. And it made me realize that it would make sense to put an isolated I2C digital interface on the board so that the digital controller is galvanically isolated from the high voltage stuff, and does not participate in any power/voltage/ground loops the amplifier gets into.

[R.G.]

Putting the control/sense modules on the amp board opened up some ideas. The board as it sits now is about 2.5" x 5" at the first rough pass. The board can also carry on it:
- I2C to DAC module ($2.50)
- I2C to quad ADC module ($1.25)
- Isolation amplifier to read the output plate/screen current while it's floating up at ~500V ($5.00)
- single-chip I2C isolator, so the digital stuff can be isolated from your controller/computer ($1.50)
- DC-DC converter to provide isolated logic voltage for the above stuff. (~$1.50)

This is about $11 for the detailed sensing, on top of the actual cost of parts for the amplifier.

I'm leaning toward making a set of PCBs with single function per board. That is, one plate supply board as above which can make 0-500V plate or screen signal, and a 0 to -75V board for grids, which will be significantly cheaper to build.

I probably should think about a controllable heater supply too.