uTracer (Micro Curve Tracer)

[martin manning]

This issue with the Quick Test turned out to be quite a puzzle. I ran a dozens of trials to find out where and when the problem would occur, and found a way around it (adding some delay to the measurement sequence), but I'm really only playing the role of “Watson” here.

The real problem was (ok now, hold on to something…) that the transistors that open the discharge paths for the anode and screen supply reservoir capacitors (the "flush valves" for when the voltages need to be reduced) weren't switching off fast enough. They have to be switched off when a measurement is made so that the discharge current doesn’t show up as part of the tube current, and evidently the variation from part-to-part can be enough that some transistors are just a bit too slow. Recall that measurements are snatched in the space of 1ms, so they have to be quick! Residual discharge current was causing the sensed voltage and current to be a little high.

Faster transistors might be found and substituted, but Ronald (Dekker) came up with a simple mod to the circuit, placing a 10nF cap across the base resistors of the two discharge transistors (the resistors are effectively in the same position as a grid stopper) to make them switch a bit more smartly. The result of this can be tested rigorously by connecting 1% resistors from the anode and screen terminals to the cathode terminal, and letting the uTracer measure them. With no added delay the result is now very good- the uTracer-measured resistance is now within 1% of what I measure with my DMM!

[martin manning]

This uTracer is now sporting a new switch, which allows connecting the anode and screen jacks directly to the reservoir capacitors instead of to the pulsed voltage sources. This allows a "continuous" mode of operation, where the anode and screen voltages on the tube are run up and held constant for each step in the same way as the grid bias. The boost converters are only able to supply a few mA of continuous current, but this is enough for small signal tubes, and for operating magic eyes (if I should ever run across one of those...). What I'm thinking is that I might be able to learn something about the noise level in a given tube this way, either by listening to it or by measuring the noise voltage. As a bonus, the continuous mode switch allows easy access to the boost converter voltages, for calibration purposes, without having to open up the box. In the gut shot you can see the added switch and the blue (anode) and orange (screen) wires soldered to the reservoir cap leads, which is the recommended place to tap into the HV supplies.

Experiments with this will have to wait, though, since I managed to blow out the grid bias circuit. I believe the output transistors were inadvertently damaged a few days ago when I let a patch cord that was plugged into the grid terminal contact the case (ground). At least it's a cheap fix- only about $0.50.

[martin manning]

Back in business after replacing the grid bias circuit’s output transistors, and the two Op-Amps for good measure.

Now that the anode and screen voltages have been squared away, I’m whining about the grid voltage. It appears to have a small offset from the input value when a tube is connected and under test. Measuring the grid voltage by connecting a DMM from the grid to the cathode is easy enough, and from this a correction to the input value could be found (much the same as the anode and screen voltages need to be increased a bit to hit the set point values exactly). On my uTracer a target of -14V for a 6L6GC requires an input value of -14.6V. Unfortunately, the DVM is measuring a very noisy voltage and converting it to DC, where the uTracer actually switches off its boost converters just before making a measurement (so the noise should be gone).

For typical power tubes, the offset I’m measuring would make the reported anode current 4-5 percentage points high. For small-signal tubes like a 12AX7 where the Vg and anode current are low, the difference in anode current could be tens of percent’s. It’s unlikely that gm and mu are much affected by this, just the “emission” number.

Since this noise is on the grid voltage except during the measurement cycle, examining tube noise in continuous mode would not be simple.

[martin manning]

Turns out that trying to measure the grid bias voltage with a DVM is misleading, for the reasons mentioned above. Looking for another way to assess the accuracy of the bias voltage at the Quick Test center point, I ran an anode voltage sweep at constant grid voltage on a 12AX7 (for a typical small signal tube) and a 6L6GC (for a typical power pentode) with the grid bias set at a nominal value in the GUI and compared the resulting anode current to running the same test with the grid bias voltage supplied by a battery. The battery is a couple of 9V's and a 10k pot so I can accurately dial in the desired -14V and -2V bias voltages. In the plots below, the 6L6 results show virtually the same anode current, and the 12AX7 results (both sections) are within 6%. This is much better than what was suggested by the measurements above.

So, it looks like just setting the nominal reference point bias voltage produces results that are quite accurate, and the same goes for the curve traces. Good thing, because the grid bias is set open-loop. Some improvement might even be made to the calibration by using two points (instead of the single point that is used now), and this is "on the list" of improvements to be included in the next release of the GUI. That might bring down the error seen in the 12AX7 test, but these are small numbers- the uTracer grid bias voltage would only need to change by ~40mV to match the battery-bias anode current.

[martin manning]

Here's another interesting trace: a 6V6 with and without a screen resistor.

One set of curves was traced in the normal way setting screen voltage at 250 and sweeping plate and grid 1 voltages.

Another trace was then done with a 1k resistor between the screen supply and the screen pin. The result is that the curves are pulled down by the reduced screen voltage (Vg2 = 250 - Ig2*1000) which becomes very non-linear as the anode voltage reaches the knee in each grid curve and the screen current rises. Maximum screen current is reduced considerably.

[martin manning]

People have been saying JJ's 6V6S is more like a 6L6, or a 5881, so I thought I'd trace one and see what its plate curves look like in comparison to a "real" 6V6-GT (GE). The colored curves are at the same Vg1 voltages as on the GE data sheet. As can be seen, it's different, with a softer knee (half-way to a pentode shape), and a very different screen current and screen current curve shape in the kink region.

At the 250V Va, 250V Vg2, and -12.5V Vg1 reference point it's close to 6V6-GT spec:

6V6-GT - 6V6S
Ia 45 - 44.5
Is 4.5 - 4.97
gm 4.1 - 3.83

The u-Tracer GUI has been getting regular updates, most recently the ability to customize the "pins" data in the saved set-up file (so you don't have to look them up), and a better grid bias calibration with three points instead of just one as in the original version.

[Phil_S]

I can't fathom how you do this, Martin, but it's, like, shazam! Since people compare the JJ to a 6L6, can you superimpose that on the 6L6 spec sheet?

[VacuumVoodoo]

People have been saying JJ's 6V6S is more like a 6L6, or a 5881, so I thought I'd trace one and see what its plate curves look like in comparison to a "real" 6V6-GT (GE). The colored curves are at the same Vg1 voltages as on the GE data sheet. As can be seen, it's different, with a softer knee (half-way to a pentode shape), and a very different screen current and screen current curve shape in the kink region.

At the 250V Va, 250V Vg2, and -12.5V Vg1 reference point it's close to 6V6-GT spec:

6V6-GT - 6V6S
Ia 45 - 44.5
Is 4.5 - 4.97
gm 4.1 - 3.83

The u-Tracer GUI has been getting regular updates, most recently the ability to customize the "pins" data in the saved set-up file (so you don't have to look them up), and a better grid bias calibration with three points instead of just one as in the original version.

This explains why JJ shows plate characteristic curves only for triode connection in their data sheet.

[martin manning]

I can't fathom how you do this, Martin, but it's, like, shazam! Since people compare the JJ to a 6L6, can you superimpose that on the 6L6 spec sheet?

I could, but I'd have to trace another set of curves at the grid voltages shown on the 6L6 data sheet. How about this:

At the 6V6 spec point a 6L6GC (reading the curves) has anode current at ~82mA, or nearly 2x the 6V6. So the JJ is very much closer to a 6V6 than a 6L6. Its rated anode dissipation is 14W, as compared to a 6V6's 12W and a 6L6's 30W, or 5881's 23W. Again much closer to a 6V6 than a 6L6.

Aleks, that was my thought too!

[matt h]

(deleted)

[Leo_Gnardo]

So the JJ is very much closer to a 6V6 than a 6L6. Its rated anode dissipation is 14W, as compared to a 6V6's 12W and a 6L6's 30W,

When I heard the claim about "being closer to a 6L6" I never thought a 6L6GC was intended, but rather the old school 6L6 and/or 6L6G with a, er, off the top of my head 19?20? W plate dissipation figure. You know, the original that the 5881 was the ruggedized version of.

Yup, exactly that, matt h. I find JJ 6V6's in self-bias amps settle themselves @ 19W. "Regular" 6V6's go @ 15W, whether old RCA's GE Sylvania or newer EH "firecracker" 6V6's.

JJ may print 14W in their rating, but the tube itself leads us to its 19W plate dissipation figure. I'm sure JJ 6V6 in fixed bias would last a good long time with 14-15W plate or even a tad lower.

[martin manning]

The JJ is a tough tube for sure, but it closely matches the 6V6 in plate current and transconductance at the 6V6 spec point, whereas a 6L6 has 80% more Ia, and 50% higher gm. I'm still more inclined to call it a ruggedized 6V6 than a 6L6 of any sort.

[matt h]

(deleted)

[VacuumVoodoo]

This also explains why they could run two of these under pretty harsh conditions in the Evil Robot amps not long ago locking the design to a single source of key component. Smart? Not really....

[martin manning]

Modern Russian reproduction vs. vintage spec for a pair of EH KT88.

For evaluation purposes the data sheet has characteristics listed at 250V plate and screen (as most do) with Ia expected to be 140mA. The Quick Test can easily handle that, but the u-Tracer is limited to 300V and ~200mA so it is not possible to do a full set of curves at 250V Vg2. Vg1 had to be -10V or less to keep the plate current within limits. The u-Tracer plot below shows the partial anode curves and the QT points.

KT88 heater current is 1.6A nominal, and the u-Tracer's internal supply is fused at 1.5A. I have successfully traced EL34's (1.5A), but I'm using an external supply here so no damage will be done and no fuses will be wasted. I can also observe heater current vs. spec, which isn't possible when using the internal supply.

This pair was sold as matched, and they are indeed very close in Ia (at 92% of spec) and in gm, but check out the level and and the variation in screen current vs. the MOV data sheet value! The data sheet's 3mA is listed with a qualifier, but I don't know if "approximate" was supposed to include 300-400%.