Tube curve tracer project

[andrei]

For a while, I have been thinking about building a modern vacuum tube curve tracer - an affordable one, that is. I came up with a conceptual design that looks like it would fit within couple of hundred bucks parts budget. The idea is to simplify hardware and offload most tasks to computer software. Here is the plan:

Hardware:
Dispense with regulated plate supplies - use rectified AC as in tube testers of yore. Control tube current, not grid voltage, with a current source in a grounded grid configuration. This is both safer and easier to interface with DAC, as it could be done on the low-voltage side of the rail. Measure Vg vs Vp curve for a set Ip as plate voltage swings at 120Hz. Interface to host computer via cheap 16bit USB codec chip - PCM2902. Use a high-efficiency step-down switcher for heater supply. Do basic operating parameter settings (heater voltage, idle bias current, max plate voltage, power limiter resistance) with hardware controls, but let software to do a Ip current sweep (within set limits). That leaves one DAC channel unallocated, which could be used to control a screen supply. I have not decided on socket wiring strategy yet - could be simple point-to-point with patch wire.

Software:
Acquired data could be easily converted into the usual plate characteristics curves (Ip vs Vp at fixed Vg), or transfer characteristics (Ip vs Vg at fixed Vp). Include provisions to compare/average device characteristics. Display single dynamical transconductance figure for easy pass/fail testing. Interface to Curve Captor for Spice modeling.

Comments?

[andrei]

No one interested? Or does everyone own VacuTrace or Sophia?

[jackinnj]

http://www.tech-diy.com/MCTracer_no_...er_feb6_04.htm

[andrei]

I know about that one... Interfacing to computer is going to be the cheapest part of the tester, whatever glue chips you use. The main expense (and size) sits in the high-voltage supplies. This is where one can do better, I think. Schematic above will trace triode curves without the usual baggage in high-voltage supplies.

All the computer-interface tracers I've seen so far are brute-force designs. The plate supply is a regulated HT rectifier controlled by some sort of voltage sweep generator (DAC or analog). This is a lot of circuitry (and heat dissipation) for something that is available straight out of the wall - a sine wave voltage sweep. Separate grid bias supply is also not necessary if you run the tube in self-biasing configuration.

[jackinnj]

just remember that when you use the sweep "from the wall" all of your measurements must be close to synchronous -- not that difficult if you use a decent ADC and USB -- but remember the processing and communications overhead as you cache data, and the inacuracies of the line voltage over time.

you will find that when measuring tubes you have to integrate the measurements -- the resistors you use to drop the voltage to your ADC knock off several bits of useable information from the measurement in a random fashion (remember SQRT (4KTBR) ) -- these bits can be recaptured with multiple measurements and averaging and/or a low pass filter.

if you want to do it "off the wall" I suggest that you get a copy of Petrowsky's article from Glass Audio Vol 2, #2 in which he described a method using a scope -- if I recall correctly he used some discrete CMOS logic and transistor switches to change the resistance on a voltage regulator to derive the grid bias. he also got the retrace problem solved.

[andrei]

Originally posted by jackinnj
just remember that when you use the sweep "from the wall" all of your measurements must be close to synchronous -- not that difficult if you use a decent ADC and USB -- but remember the processing and communications overhead as you cache data, and the inacuracies of the line voltage over time.

Not an issue if I use a stereo audio codec - two channels are synchronous.

you will find that when measuring tubes you have to integrate the measurements -- the resistors you use to drop the voltage to your ADC knock off several bits of useable information from the measurement in a random fashion (remember SQRT (4KTBR) ) -- these bits can be recaptured with multiple measurements and averaging and/or a low pass filter.

1 MOhm resistance into 40kHz audio band gives 26uV thermal noise. Not an issue even for a 16-bit ADC, and I suspect that PCM2902 is effectively 14-bit or less anyway. Averaging is always a good idea, but statistics is easier to implement in software.

if you want to do it "off the wall" I suggest that you get a copy of Petrowsky's article from Glass Audio Vol 2, #2 in which he described a method using a scope -- if I recall correctly he used some discrete CMOS logic and transistor switches to change the resistance on a voltage regulator to derive the grid bias. he also got the retrace problem solved.

Is there a copy online somewhere?

[jackinnj]

Originally posted by andrei


Not an issue if I use a stereo audio codec - two channels are synchronous.

If you aren't regulating, you actually have to measure the grid, plate voltages and currents -- any error in the line voltage is multiplied in the secondary -- and if you look at an FFT of your line voltage you can see the harmonics and birdies which fuzz up your measurements. Put a data-logger on your mains, then have the wife crank up the dishwasher or her hair-blower and you'll see what I mean (fwiw, I learned this the hard way when doing black-and-white photography -- if you don't regulate the lamp output any perturbation -- like an air conditioner kicking in -- really screws up your results.)

and you will probably want to isolate the codec from the high voltage with an analog optocoupler (Siemens Vishay IL300, Avago (the newest iteration of HP's, the Agilent, now Avago semiconductor products, HCNR200, HCPL7840), Fairchild, Vactrol). you don't want to blow the CODEC and your laptop because some HV decided to stray across your PCB.

"Circuit Cellar' archives their articles and Dr. Steber from Illinois University wrote a great piece on using the sound card as an XY grapher for this type of application -- it was semiconductors, but you can do the programming -- you don't have to get an evaluation board from TI for the CODEC.

[Yvesm]

Hi Andrei

I do like that idea ... mainly because I've a somewhat similar "minimal harware rig" in my head

You may use a singe phase rectifier to allow a lower duty cycle for the tube itself and for the software too.

In my idea, I planned to generate a stairstep voltage for the grid by loading a large cap thru a large resistor (or better a CCS) each "other phase" of the main (slowly to get as much measuring points as possible ).
This can be done with just a diode, a res. and a cap.
Some form of "discharge/restart" must be included (the simplest is a "push button"' ! )

The drawback is that I should, as often as possible, measure three values simultaneously (Vg, Vp and Ip) but the voltage stability of the main is of "no concern". I planned to use a µcontroler with integrated ADC's. (No need for even a DAC !)

After sufficient data has been accumulated, drawing plate and/or grid curves at any scale is just a software affair.

This project is sleeping for now, I would be glad to chat about and to share ideas.

Yves.

[Tubes4e4]

Hi andrei,

Originally posted by andrei
No one interested? Or does everyone own VacuTrace or Sophia?

To both questions "no", but your idea of a very reduced design will very soon run into very down-to-earth technical problems regarding versatility and PSU problems.

Please have a look (and Mr. Walton, please, too) at
some plate curves I traced "by hand".

The range of tubes shown there is not wide (small signal pentodes to medium power pentodes), but the requirements/demands on PSU capabilities to cover them are immense, and I only succeeded by "intelligently" deciding what PSUs I have at hand to use for which feeds, often changing them several times during a tracing run to be used as optimal as possible (or simply to cover the voltage and current range of interest).

The only way to circumvent those PSU related problems in an automated "budget" design is to use intelligent methods of data sampling, which don´t depend on stable & capable PSUs. So you are on the right track

For automated "budget" designs (without any negative meaning regarding that "budget" word) I suggest to listen very closely to what Mr. Monmagnon might have to say

Tom

[andrei]

Originally posted by jackinnj
If you aren't regulating, you actually have to measure the grid, plate voltages and currents -- any error in the line voltage is multiplied in the secondary -- and if you look at an FFT of your line voltage you can see the harmonics and birdies which fuzz up your measurements.

I am aware of what the mains voltage actually looks like... But since Vp and Vg are sampled synchronously, the particular sweep waveform does not matter. Ip is going to be set with a high-compliance current source, so there is no need to sample it. I am willing to give up Ig measurement for the sake of economy.

and you will probably want to isolate the codec from the high voltage with an analog optocoupler (Siemens Vishay IL300, Avago (the newest iteration of HP's, the Agilent, now Avago semiconductor products, HCNR200, HCPL7840), Fairchild, Vactrol). you don't want to blow the CODEC and your laptop because some HV decided to stray across your PCB.

I'd rather isolate the digital bus... Does anyone know if there is off-the-shelf galvanic isolator for USB?

Originally posted by Tubes4e4
To both questions "no", but your idea of a very reduced design will very soon run into very down-to-earth technical problems regarding versatility and PSU problems.

Such as?

The baseline requirement for me is to be able to trace 300B as well as small-signal triodes like 6SN7, 6SL7. They do indeed have very different grid drive requirements, but self-bias arrangement takes care of that. As there is no HT regulators, most of the power dissipation is in the device under test, so power output is mostly limited by the tranformer. Having 500V max Vp will do, I think. You could switch to 250V tap for low voltage tubes like 6AS7. Heater supply is going to be a 30W/5A switcher using LT1074 - enough even for 6528.

Please have a look (and Mr. Walton, please, too) at
some plate curves I traced "by hand".

Good work .