And now a set of curves at the currents we're more likely to use. Transconductance is well over 50mA/V, so a source follower using these devices will have better driving ability than any cathode follower using a single valve. At 10mA, the drain looks to be about 40k, so with 50mA/V (all WAGs), that makes mu = 2000, so there's plenty of feedback available in a source follower. My curve tracer won't go >100V, but there's a definite upward curving trend to the curves, so any source follower delivering a large voltage swing will probably want another FET wrapped around it to keep Vds constant, and a CCS load to keep gm constant.
Attachments
Hey-Hey!!!,
Thanks for the FQP curves. I have the 1N60's qued up for use in the cascode PI of my amp. Triode on the bottom, MOSFET on top. If they pop there's 2N90's in my box...
I think the low voltgage region in that circuit should just get labeled 'here be monsters'
cheers,
Douglas
OTOH, the 30V g-s voltage is well protected by an 18V Zener.
Thanks for the FQP curves. I have the 1N60's qued up for use in the cascode PI of my amp. Triode on the bottom, MOSFET on top. If they pop there's 2N90's in my box...
I think the low voltgage region in that circuit should just get labeled 'here be monsters' cheers,
Douglas
OTOH, the 30V g-s voltage is well protected by an 18V Zener.
I set this amp up on my breadboard to test all sorts of screen grid experiments, but there have been several diversions along the way. Last weekend one of my power supplies blew up and I spent most of my time fixing it.
Today I decided to test out something that I have been wanting to try for 2 years. What will the typical audio tube do in A2? I have been told (by several people) that it will distort badly. I had never tried it, so I decided to test the same old 6L6 that I have been torturing for almost 2 years. The amp was tested in triode mode for its baseline performance. I set the bias for some more realistic operating conditions, 400 volts (across the tube) and 60 mA. All measurements were made with a 5K transformer into an 8 ohm load. The drive was increased until 5% distortion was reached. Power was 4.5 watts. Clipping was visible at just over 6 watts. Distortion at 1 watt was 1.9% mostly 2nd, with some 3rd, and 4th visible.
Next I added a PowerDrive circuit and fired up the amp. It was plainly obvious that A2 is not a bad thing with a 6L6GC. I was attempting to measure the peak grid current by observing the voltage drop across the grid stopper (scope in differential mode). Of course there is no drop until you get close to A2. As you approach A2 the grid current shows up in pulses. At the edge of clipping (over 10 watts!) I was seeing a distorted sine wave of 10 volts P-P across my 100 ohm grid stopper! The grid stopper was a source of nonlinearity, so it had to go. With the grid stopper eliminated, the clipping point was just over 11 watts! This was from a 6L6GC in triode. The grid voltage was swinging from -75 volts to +15 volts.
At 5 watts I was seeing 2.8% distortion, mostly second. Cranking the power up until I hit 5% brought 9.5 watts. Clipping was reached at 11.1 watts. The distortion at 1 watt was 1.1%, almost entirely second. The third was -60db. There was some 60, 120, and 180 Hz from the ground loop created by the UPS running the computer that limits the measurement floor, so the distortion was likely under 1%.
Next, I will try UL mode. Then I will try some other tubes. I may even try this out with a black plate RCA (I'll be nice to it , I promise) and a Sovtek 6L6WXT (they can take more abuse) as well as some other tube types. It would also be nice to try different load impedances.
I will get back to the screen grid experiments, but I now have removed the grid current limitation that was mentioned heavilly in the earlier posts. This was a useful diversion.
In the process of processing the photo to fit the requirements, the other little experiment got lost in the reduced resolution. The LED in the cathode of the 12AT7 driver. Many people slam the 12AT7 as a driver in an SE amp, but they are pretty linear with a CCS load, especially with high impedance load (mosfet follower).
Today I decided to test out something that I have been wanting to try for 2 years. What will the typical audio tube do in A2? I have been told (by several people) that it will distort badly. I had never tried it, so I decided to test the same old 6L6 that I have been torturing for almost 2 years. The amp was tested in triode mode for its baseline performance. I set the bias for some more realistic operating conditions, 400 volts (across the tube) and 60 mA. All measurements were made with a 5K transformer into an 8 ohm load. The drive was increased until 5% distortion was reached. Power was 4.5 watts. Clipping was visible at just over 6 watts. Distortion at 1 watt was 1.9% mostly 2nd, with some 3rd, and 4th visible.
Next I added a PowerDrive circuit and fired up the amp. It was plainly obvious that A2 is not a bad thing with a 6L6GC. I was attempting to measure the peak grid current by observing the voltage drop across the grid stopper (scope in differential mode). Of course there is no drop until you get close to A2. As you approach A2 the grid current shows up in pulses. At the edge of clipping (over 10 watts!) I was seeing a distorted sine wave of 10 volts P-P across my 100 ohm grid stopper! The grid stopper was a source of nonlinearity, so it had to go. With the grid stopper eliminated, the clipping point was just over 11 watts! This was from a 6L6GC in triode. The grid voltage was swinging from -75 volts to +15 volts.
At 5 watts I was seeing 2.8% distortion, mostly second. Cranking the power up until I hit 5% brought 9.5 watts. Clipping was reached at 11.1 watts. The distortion at 1 watt was 1.1%, almost entirely second. The third was -60db. There was some 60, 120, and 180 Hz from the ground loop created by the UPS running the computer that limits the measurement floor, so the distortion was likely under 1%.
Next, I will try UL mode. Then I will try some other tubes. I may even try this out with a black plate RCA (I'll be nice to it , I promise) and a Sovtek 6L6WXT (they can take more abuse) as well as some other tube types. It would also be nice to try different load impedances.
I will get back to the screen grid experiments, but I now have removed the grid current limitation that was mentioned heavilly in the earlier posts. This was a useful diversion.
In the process of processing the photo to fit the requirements, the other little experiment got lost in the reduced resolution. The LED in the cathode of the 12AT7 driver. Many people slam the 12AT7 as a driver in an SE amp, but they are pretty linear with a CCS load, especially with high impedance load (mosfet follower).
Attachments
Most of my testing has been with GE6201's (I have a few hundred of them). I have experimented with mostly vintage USA tubes that are from my surplus tube collection. I have also tried a few Mullard's and Telefunken's. They all have a sweet spot (combination of current and voltage) but it can be different from tube to tube.
Everyone keeps saying JJ, so I will get a pair on my next order to AES.
Everyone keeps saying JJ, so I will get a pair on my next order to AES.
I have been a little sidetracked by tax forms and a wedding in the family instead of testing lately. But I have worked out a better way to see what is going on in variable %UL without doing 100's of spectrograms which are hard to interpret.
Using the same test setup, I feed a very small 1 KHz signal to the input and measure output AC to compute incremental gain. Then I vary the DC input level across the linear range to make a graph of incremental gain versus signal level. This graph is then normalized to unity gain, so distortion shows up as deviation from unity. Plotting screen current on the same graph shows the correlation of gain with screen current. This graph then gets done for several values of %UL and also with the screen current (MOSFET drain current) returned to B++, 50% tap or 100% tap.
I should be able to get some graphs by the end of the month for the UL tests.
Later this summer, I hope to have an even better test setup available that can automate this sort of thing. My power supplies are computer controllable in Basic on the PC, and I am working on a test configuration that will allow me to read currents and differential voltages too. You might call it a SPICE simulator that uses real parts. And I will eventually have computer controlled load resistors too:
http://www.edn.com/contents/images/505072.pdf
Don
Using the same test setup, I feed a very small 1 KHz signal to the input and measure output AC to compute incremental gain. Then I vary the DC input level across the linear range to make a graph of incremental gain versus signal level. This graph is then normalized to unity gain, so distortion shows up as deviation from unity. Plotting screen current on the same graph shows the correlation of gain with screen current. This graph then gets done for several values of %UL and also with the screen current (MOSFET drain current) returned to B++, 50% tap or 100% tap.
I should be able to get some graphs by the end of the month for the UL tests.
Later this summer, I hope to have an even better test setup available that can automate this sort of thing. My power supplies are computer controllable in Basic on the PC, and I am working on a test configuration that will allow me to read currents and differential voltages too. You might call it a SPICE simulator that uses real parts. And I will eventually have computer controlled load resistors too:
http://www.edn.com/contents/images/505072.pdf
Don
Don, I read this interesting article about computer-controlled MOSFET load resistor. I think this could work for basic tests for audio amps, and probably for the specific testing that you mentioned, eliminating a lot of tedious work. But I would question the use of an active load for amplifier distortion, bandwidth and square-wave testing. The effective resistance is only as good as the behavior of the FET, opamps and DAC in the feedback loop. Your amplifier might be "better" than your load, and you might be measuring your load's performance instead.
Now THAT could work.
However the linked article (see the last item in the PDF) shows a clever way to use a MOSFET with feedback to emulate a fixed resistor. At low frequencies, this ought to work well. But the MOSFET with feedback loop would be working as hard as the amplifier when audio test signals are impressed across it. For Don's series of curves, the automation makes sense. For square-wave tests or distortion tests across the band, it would add another active part to the circuit.
However the linked article (see the last item in the PDF) shows a clever way to use a MOSFET with feedback to emulate a fixed resistor. At low frequencies, this ought to work well. But the MOSFET with feedback loop would be working as hard as the amplifier when audio test signals are impressed across it. For Don's series of curves, the automation makes sense. For square-wave tests or distortion tests across the band, it would add another active part to the circuit.
Yes, yes and yes. I understand all the concerns about distortion testing with an active load. The programming and readback D/As, and A/Ds, for the power supplies, and I and V measurements, are only 16 bit here as well (and most commercial programmable power supplies only have 12 bits besides). My intended usage of the system is for analyzing simple configurations without global feedback, where the distortion is generally high enough to swamp the test system errors.
The R-2R design and also servo-mechanical pots are also under consideration for higher freq. testing. I'm just starting out with something simple for limited testing initially.
The MOSFET resistor simulator design might be capable of surprising performance if implemented with fast OP Amps. The D/A used in it is a current reference in - to current output part, which is generally plenty fast. The digital code is stationary during applied signal testing. But it would obviously be a horse race against a high performance, low THD, GNFB amplifier.
Limiting the testing to the differential gain approach (with a slow sweep of DC signal level) with a low frequency, low amplitude AC test signal, should minimize the stress on accuracy and speed. I will just have to build it and see how far it can be pushed.
(Just as an interesting note, by replacing the current sensing resistor in the R simulator design with a cap., one gets a variable capacitor.)
One can also do a hybrid approach, using a CCS with a servo variable resistor (servo-mechanical or R-2R) from the output back to the voltage reference. That way, current out is proportional to voltage applied. This makes a fast power R device. Accuracy is more difficult, but could be moderately improved by using a matched diode in the V reference divider side, like used in current mirrors. ( simple stationary R value errors could be encoded in a table in the PC Basic program, but AC signal stability of the R value still depends on the diode matching quality)
Other interesting components for a PC test system include a DDS (direct digital synthesis) sinewave oscillator for generating test signals. (of course one can just use the sound card too, but they don't generally provide easy programming in Basic drivers)
A digitally controlled CCS is straight-forward of course.
Digital DVMs with optical RS-232 are abundant for voltage and current measurements, although generally a little slow, and only 16 bit (50,000 count) at best.
Don
The R-2R design and also servo-mechanical pots are also under consideration for higher freq. testing. I'm just starting out with something simple for limited testing initially.
The MOSFET resistor simulator design might be capable of surprising performance if implemented with fast OP Amps. The D/A used in it is a current reference in - to current output part, which is generally plenty fast. The digital code is stationary during applied signal testing. But it would obviously be a horse race against a high performance, low THD, GNFB amplifier.
Limiting the testing to the differential gain approach (with a slow sweep of DC signal level) with a low frequency, low amplitude AC test signal, should minimize the stress on accuracy and speed. I will just have to build it and see how far it can be pushed.
(Just as an interesting note, by replacing the current sensing resistor in the R simulator design with a cap., one gets a variable capacitor.)
One can also do a hybrid approach, using a CCS with a servo variable resistor (servo-mechanical or R-2R) from the output back to the voltage reference. That way, current out is proportional to voltage applied. This makes a fast power R device. Accuracy is more difficult, but could be moderately improved by using a matched diode in the V reference divider side, like used in current mirrors. ( simple stationary R value errors could be encoded in a table in the PC Basic program, but AC signal stability of the R value still depends on the diode matching quality)
Other interesting components for a PC test system include a DDS (direct digital synthesis) sinewave oscillator for generating test signals. (of course one can just use the sound card too, but they don't generally provide easy programming in Basic drivers)
A digitally controlled CCS is straight-forward of course.
Digital DVMs with optical RS-232 are abundant for voltage and current measurements, although generally a little slow, and only 16 bit (50,000 count) at best.
Don
Hi Jkeny,
I'm afraid I haven't gotten any more accomplished on this, at least not directly. The problem is that using sprectrum analyzer results are rather hard to interpret what is actually going on without doing a massive sweep thru all variables, and it is hard to do this in a controlled way. Things like operating point or bias have to change as one changes %tapping, and then you don't know which made the improvement or degradation on the spectrum graph. The other problem is not many will be convinced by the results if the conclusions aren't obvious.
I have been working on getting some automated power supplies, controlled by Visual Basic going. As I mentioned a while back, a much more illuminating test would be to measure incremental gain versus signal voltage. Then one could plot gain versus signal voltage by DC sweeping the input, with a small AC signal summed with it. One measures the AC signal output amplitude versus the DC sweep level to produce the gain graph then. Multiple gain sweeps with different %UL or load resistance could then be plotted on one graph.
So when I get this capability up and running, testing the UL circuit should be a piece of cake.
In the meantime however, I am fairly convinced what the results will finally show, its just that no one believes me!
This is what I think is going on in UL:
The %UL tapping is used to vary the effective Mu of the "triode" formed by screen feedback. Mu = gm * rp for a triode, and the cathode/grid1 gm is constant here. So effective rp changes with tapping. The tapping is selected to get the "triode" Rp to roughly match the best loading Z for the pentode for Max power out. Ie, the effective triode is then loaded down with Load Z = Rp for max power transfer from a triode.
Now this kind of matching produces very bad 3/2 power distortion, generally designers try to load a triode at several times Rp to keep distortion down. But since the usual UL tapping comes right off the xfmr at B+ DC levels, the screen distortion also becomes excessive due to the plate decending below the screen voltage half the time.
It fortuituusly turns out that the two distortions to the gain curve are of opposite nature. The 3/2 power law distortion increases current gain with tube turn on and the screen current decreases current gain with tube turn on. With proper tweaking of the %UL tap one can get these two distortions to roughly null each other out.
Only problem is that the 3/2 power distortion is related to load current and the screen current distortion is related to load (or plate actually) voltage. All is fine with a nominal load resistor. But when you put a reactive load like a real speaker on it, the two go out of phase. This gives nasty distortion with speakers, but passes the THD meter test with flying colors.
Now one CAN reduce the screen DC voltage with a tertiary xfmr winding or Mosfet driver scheme to reduce the screen current distortion. But then one is left with nasty 3/2 power distortion unless one increases the xfmr input Z to an appropriate level for the "triode". But if you do that, then the power output drops down to just triode level. So why bother with UL at All!
I plan to avoid it. (So you can see why I'm not too urgently spending time on UL testing.)
Don
I'm afraid I haven't gotten any more accomplished on this, at least not directly. The problem is that using sprectrum analyzer results are rather hard to interpret what is actually going on without doing a massive sweep thru all variables, and it is hard to do this in a controlled way. Things like operating point or bias have to change as one changes %tapping, and then you don't know which made the improvement or degradation on the spectrum graph. The other problem is not many will be convinced by the results if the conclusions aren't obvious.
I have been working on getting some automated power supplies, controlled by Visual Basic going. As I mentioned a while back, a much more illuminating test would be to measure incremental gain versus signal voltage. Then one could plot gain versus signal voltage by DC sweeping the input, with a small AC signal summed with it. One measures the AC signal output amplitude versus the DC sweep level to produce the gain graph then. Multiple gain sweeps with different %UL or load resistance could then be plotted on one graph.
So when I get this capability up and running, testing the UL circuit should be a piece of cake.
In the meantime however, I am fairly convinced what the results will finally show, its just that no one believes me!
This is what I think is going on in UL:
The %UL tapping is used to vary the effective Mu of the "triode" formed by screen feedback. Mu = gm * rp for a triode, and the cathode/grid1 gm is constant here. So effective rp changes with tapping. The tapping is selected to get the "triode" Rp to roughly match the best loading Z for the pentode for Max power out. Ie, the effective triode is then loaded down with Load Z = Rp for max power transfer from a triode.
Now this kind of matching produces very bad 3/2 power distortion, generally designers try to load a triode at several times Rp to keep distortion down. But since the usual UL tapping comes right off the xfmr at B+ DC levels, the screen distortion also becomes excessive due to the plate decending below the screen voltage half the time.
It fortuituusly turns out that the two distortions to the gain curve are of opposite nature. The 3/2 power law distortion increases current gain with tube turn on and the screen current decreases current gain with tube turn on. With proper tweaking of the %UL tap one can get these two distortions to roughly null each other out.
Only problem is that the 3/2 power distortion is related to load current and the screen current distortion is related to load (or plate actually) voltage. All is fine with a nominal load resistor. But when you put a reactive load like a real speaker on it, the two go out of phase. This gives nasty distortion with speakers, but passes the THD meter test with flying colors.
Now one CAN reduce the screen DC voltage with a tertiary xfmr winding or Mosfet driver scheme to reduce the screen current distortion. But then one is left with nasty 3/2 power distortion unless one increases the xfmr input Z to an appropriate level for the "triode". But if you do that, then the power output drops down to just triode level. So why bother with UL at All!
I plan to avoid it. (So you can see why I'm not too urgently spending time on UL testing.)
Don
This subject is still high on my list of things to do, although I don't plan on being nearly that scientific. I have a big box (hundreds) of sweep tubes that have 200 to 300 volt screen voltage ratings, but 550 to 1000 volt plate ratings. I started experimenting with mosfet circuits to apply the AC and DC voltages to the screen grid independently of each other back when this thread started.
Unfortunately, personal circumstances (friends and relatives keep getting cancer, 2 have died) have stopped most of my experiments. This situation is not likely to change for the rest of this year.
Unfortunately, personal circumstances (friends and relatives keep getting cancer, 2 have died) have stopped most of my experiments. This situation is not likely to change for the rest of this year.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.