Using Solid State devices in support of Vacuum Tubes

[bob p]

The use of solid-state devices in support of vacuum tubes is a topic that comes up now and then -- especially when the truly knowledgeable people are willing to cast pearls my way.

Along these lines, I've encountered a circuit that seems to use SS devices in this fashion, so I thought I'd bring the topic up for discussion. I could really use some help with the theory, so I'm hoping that people will chime in and offer help.

Here's a snipped from the schematic of an expensive HiFi amp that appears to be using a transistor to provide some sort of voltage regulation (maybe a constant current supply) to the cathodes of the phase inverter:

Can anyone comment on the theory behind this application? thanks!

[R.G.]

I always seem to be sticking my foot into it this way, so I'll take a swing.

It looks like it's a deceptively-drawin JFET constant current source. A MOSFET hooked up that way would not do anything until its breakdown voltage on the drain was exceeded. If you mentally sketch in a JFET, it makes perfect sense as a constant current source.

[R.G.]

I was somewhat thunderstruck at seeing this thread here, as I had just popped into put up some more MOSFET Heresies, extending the line at GEO.

Those of you who read GEO know about my theories and the practical proofs of them in using power MOSFETs as substitutes for cathode followers. There seem to have been a number of successful applications of that as tone stack drivers and as follower drivers for output tubes.

I toyed with the idea of a MOSFET phase inverter but put it aside because I thought the larg-ish Cgss would severely limit frequency response in a voltage-gain situation.

I finally got around to simulating this. I put a pair of IRF820 2A/800V MOSFETs in the sim in a standard diffamp-style inverter and checked their frequency response. It was as expected disappointing, down 6db at 340Hz. But the low frequency gain was over 60db, which amazed me.

I remembered something about linearizing diffamps by reducing the transconductance of the devices with unbypassed source resistors, so I put in 510R resistors in each source before the 6.8K joint resistor to ground. Response was now down to about 35db - useful in a PI. But what was really astonishing was that the high frequency -6db point was now over 140kHz.

It appears that trading unnecessarily high gain for input impedance lets you use a power MOSFET as a gain sub in a diffamp-style phase inverter.

There's a lot more work to do to get the thing right, but it looks like this is another place where MOSFETs can free up a tube to make good tube sounds otherwise. Promising anyway.

The circuit was a diffamp with 100K drain resistors to +300V, 510R source resistors joined at at 6.8K resistor to ground. The gates were tied to a variable DC bias source, which turned out to need about +22Vdc on it, and the second gate was tied to the bias voltage with 10nF across its 1M bias resistor. Input was through 50K (to fake a driving tube's plate resistance) and a 100nF cap.

[R.G.]

Then there's the uC to set and rebias output tubes. That worked pretty much first time. Reads the cathode current, outputs a positive voltage that's inverted and supplied to the grid resistors by a high voltage inverting amplifier.

[R.G.]

Then there's the power on sequencer/tube saver.

Using a MOSFET switch in the B+ line, you delay power on for long enough for the tubes to get hot. Then turn on the B+. Slowly.

Whenever cathode current goes over X you turn off B+ to save the tubes, OT and PT.

The uC also reads a thermistor for internal cabinet temp and turns on a fan. Very slowly at first, more if temp doesn't go down.

[bob p]

Originally Posted by R.G.

I always seem to be sticking my foot into it this way, so I'll take a swing.

It looks like it's a deceptively-drawin JFET constant current source. A MOSFET hooked up that way would not do anything until its breakdown voltage on the drain was exceeded. If you mentally sketch in a JFET, it makes perfect sense as a constant current source.

Do you mean an N-channel JFET, which would have a symbol like this?

I'm still confused by this, as I don't know where R13 in the diagram above (150R resistor) would connect to a JFET. As you said, its kind of deceptively drawn. Is R13 connected to a body connection? If so, can the device be a JFET? (I thought they had no body connections.)


The schematic lists the part as Q1 1038, but it doesn't say anything more. The parts list for the amp only lists the amp maker's internal part number of 30011016, and no generic part number. The parts list only has a description of "(SS) FET, Green/White/Blue." Interestingly, Q1 and Q2 have the same part number, though Q2 is listed as Q2 1036 in another location on the schematic. The part could be either "1036" or "1038" if one of them is a typo.

From this information, can you come up with the real/generic name for the part? I haven't had much luck, but then I don't know diddly about JFETs.

[bob p]

Originally Posted by R.G.

I was somewhat thunderstruck at seeing this thread here...

Well, if the truth be told I was trying to get your attention without having to resort to private messages.

Those of you who read GEO know about my theories and the practical proofs of them in using power MOSFETs as substitutes for cathode followers. There seem to have been a number of successful applications of that as tone stack drivers and as follower drivers for output tubes.

As it turns out, I have been planning on using a 12BH7 CF driver to drive the 3 P-P pairs of 6550 that will be used in the output section of each channel of the 100W Class A amp. I've been planning on using a CF driver primarily out of my desire to match impedance, and partly out of my desire to eliminate the problems of Class B bias shift and grid current clipping. (I'd rather have the output tubes driven into plate saturation than have to listen to grid current limiting.)

Being open minded about this, I suppose it might be worth looking into trying out a MOSFET for this sort of application if I had a little help. I know that the MOSFET will take care of the Class B bias shift and the grid current clipping. Am I correct in assuming that it would also be capable of driving the low impedance that's presented by the 3x6550 grids in parallel?

[bob p]

I toyed with the idea of a MOSFET phase inverter but put it aside because I thought the larg-ish Cgss would severely limit frequency response in a voltage-gain situation.

I finally got around to simulating this. I put a pair of IRF820 2A/800V MOSFETs in the sim in a standard diffamp-style inverter and checked their frequency response. It was as expected disappointing, down 6db at 340Hz. But the low frequency gain was over 60db, which amazed me.

I remembered something about linearizing diffamps by reducing the transconductance of the devices with unbypassed source resistors, so I put in 510R resistors in each source before the 6.8K joint resistor to ground. Response was now down to about 35db - useful in a PI. But what was really astonishing was that the high frequency -6db point was now over 140kHz.

By any chance, did the diagram that I posted above (that uses an unbypassed source resistor) help you along the road to this realization? Or was that just a coincidence?

Another thing I thought I'd ask is whether MOSFETs are better than JFETs in this sort of application, because JFETs have higher transconductance.

[bob p]

It appears that trading unnecessarily high gain for input impedance lets you use a power MOSFET as a gain sub in a diffamp-style phase inverter.

There's a lot more work to do to get the thing right, but it looks like this is another place where MOSFETs can free up a tube to make good tube sounds otherwise. Promising anyway.

The circuit was a diffamp with 100K drain resistors to +300V, 510R source resistors joined at at 6.8K resistor to ground. The gates were tied to a variable DC bias source, which turned out to need about +22Vdc on it, and the second gate was tied to the bias voltage with 10nF across its 1M bias resistor. Input was through 50K (to fake a driving tube's plate resistance) and a 100nF cap.

you know what they say -- a picture is worth 1000 words. <schematic hint>

My interest along these lines isn't based so much on the desire to eliminate tubes with SS devices as a cost-saving measure. To some extent, that's what led us all down the path to SS amps in the first place. What I'm more interested in personally is using SS devices to enhance the performance of tubes in performing their intended duties. To that extent, it would be very helpful if I could wrap my mind better around what's really going on with Q1 in that schematic snippet that I posted up at the top of this page. Admittedly, my knowledge of SS components is the weakest link in this discussion, and any help that anyone could offer would be greatly appreciated.

[bob p]

Originally Posted by R.G.

Then there's the uC to set and rebias output tubes. That worked pretty much first time. Reads the cathode current, outputs a positive voltage that's inverted and supplied to the grid resistors by a high voltage inverting amplifier.

Pardon my ignorance -- uC?

[bob p]

Originally Posted by R.G.

Then there's the power on sequencer/tube saver.

Using a MOSFET switch in the B+ line, you delay power on for long enough for the tubes to get hot. Then turn on the B+. Slowly.

Whenever cathode current goes over X you turn off B+ to save the tubes, OT and PT.

The uC also reads a thermistor for internal cabinet temp and turns on a fan. Very slowly at first, more if temp doesn't go down.

Regarding protection against bias failure -- this reminds me of an old method of bias protection: you use an extra triode and hold it at cutoff by connecting its grid to the bias supply voltage, and then wire the triode in series with a relay, so that the relay could shut down the circuit in the event of a bias failure. All that's missing from this design and your MOSFET suggestion are the explosive bolts.

[Enzo]

Pardon my ignorance -- uC?

micro-controller

[R.G.]

Originally Posted by bob p

Do you mean an N-channel JFET, which would have a symbol like this?

I'm still confused by this, as I don't know where R13 in the diagram above (150R resistor) would connect to a JFET. As you said, its kind of deceptively drawn. Is R13 connected to a body connection? If so, can the device be a JFET? (I thought they had no body connections.)

Yes, I meant n-channel JFET.

My speculation on it being a JFET is based on the fact that a MOSFET as drawn simply cannot work. It struck me as I looked at it that if you put a JFET in there where the MOSFET is and connected drain to drain, gate to gate, and source to source, that a JFET would in fact work. I therefore made the otherwise unsubstantiated leap that someone put an erroneous symbol in the schematic and it was really a JFET. I have seen a JFET used as a CCS in the source of MOSFETs before for high voltage operations, and this seems to fit well.

Originally Posted by bob p

The schematic lists the part as Q1 1038, but it doesn't say anything more. The parts list for the amp only lists the amp maker's internal part number of 30011016, and no generic part number. The parts list only has a description of "(SS) FET, Green/White/Blue."

From this information, can you come up with the real/generic name for the part? I haven't had much luck, but then I don't know diddly about JFETs.

It's probably not going to be any help. Those are in-house part numbers, and they won't help you at all unless you can get the manufacturer's cheat sheet for in-house to industry part types. Makers do this for two reasons:
1. They get to control and select what parts are used, which is a real and valid reason to do this.
2. They get to charge extra for their preselected parts in repairs.

You can think of a JFET as a low voltage but higher current version of a pentode. In common with tubes, JFETs with no reverse bias from gate to source conduct heavily. In common with tubes, you can bias them with a resistor from source to ground to cause a voltage drop from channel conduction. NOT in common with tubes is the fact that JFET current will pinch off at a value called Idss (current from drain to source with gate-source shorted). JFETs make near-perfect constant current sources with their gate and source shorted. You can make them variable CCSs by putting a resistor in series with the source and tying the gate to the far end. Varying the resistor varies the current down from Idss.

The CCS apps largely don't work with triodes and poorly with pentodes. But the biasing is similar.

JFETs usually have drain-source breakdowns from 20V-50V.

[R.G.]

As it turns out, I have been planning on using a 12BH7 CF driver to drive the 3 P-P pairs of 6550 that will be used in the output section of each channel of the 100W Class A amp. I've been planning on using a CF driver primarily out of my desire to match impedance, and partly out of my desire to eliminate the problems of Class B bias shift and grid current clipping. (I'd rather have the output tubes driven into plate saturation than have to listen to grid current limiting.)

Being open minded about this, I suppose it might be worth looking into trying out a MOSFET for this sort of application if I had a little help. I know that the MOSFET will take care of the Class B bias shift and the grid current clipping. Am I correct in assuming that it would also be capable of driving the low impedance that's presented by the 3x6550 grids in parallel?

Yes. A typical high voltage power MOSFET has a transconductance of about 0.5 to 2 amperes per volt. I'm not sure what the forward biased impedance of a 6550 grid is, but if it's as low as 1K, a typical IRF820 will only need enhancement of a fraction of a volt to pump that much current. The 820 is a 2A continuous device; the pulsed channel current is much larger. And both higher current and gainy-er devices are available. Put it this way - if a tube can drive it, the MOSFET won't even notice it, at least in terms of current drive.

By any chance, did the diagram that I posted above (that uses an unbypassed source resistor) help you along the road to this realization? Or was that just a coincidence?

It was coincidence, but that's what was so striking.

Another thing I thought I'd ask is whether MOSFETs are better than JFETs in this sort of application, because JFETs have higher transconductance.

They each have their own speciality. JFETs have interelectrode capacitances which are small enough to use directly with tubes without dramatic high frequency losses. Power MOSFETs have a very high Cgss (grid-source capacitance) which is crippling in some applications. It's like having a high-transconductance tube with a 1000pF cap from grid to cathode that you can't get rid of.

JFETs have high gain, low capacitance. On the down side, they are low voltage devices and have only a few milliamperes of current.

MOSFETs have monster current capabilities, high voltage, but the ugliness of a high Cgss. Actually, Zetex makes some TO-92 devices that are 450V and 0.7A/900mW and only have Cgss of 50pF. I expect that these could make a really high frequency capable PI.

Tubes have high voltage, low-ish capacitances; but low current limits, and that funny grid-cathode conduction thing.

you know what they say -- a picture is worth 1000 words. <schematic hint>

Yeah, yeah, I'll go draw it up. 8-)

My interest along these lines isn't based so much on the desire to eliminate tubes with SS devices as a cost-saving measure. To some extent, that's what led us all down the path to SS amps in the first place. What I'm more interested in personally is using SS devices to enhance the performance of tubes in performing their intended duties. To that extent, it would be very helpful if I could wrap my mind better around what's really going on with Q1 in that schematic snippet that I posted up at the top of this page. Admittedly, my knowledge of SS components is the weakest link in this discussion, and any help that anyone could offer would be greatly appreciated.

I'll take a look and see if I can write something up.

Regarding protection against bias failure -- this reminds me of an old method of bias protection: you use an extra triode and hold it at cutoff by connecting its grid to the bias supply voltage, and then wire the triode in series with a relay, so that the relay could shut down the circuit in the event of a bias failure. All that's missing from this design and your MOSFET suggestion are the explosive bolts.

You see, I always did want to design something with explosive bolts in it... 8-)

[bob p]

Thanks Enzo.

Originally Posted by R.G.

JFETs make near-perfect constant current sources with their gate and source shorted. You can make them variable CCSs by putting a resistor in series with the source and tying the gate to the far end. Varying the resistor varies the current down from Idss.

Well, I'm still trying to figure out how to interpret that non-standard drawing and match up the leads on a JFET to those in the original schematic snippet.

Looking at my original schematic snippet, does R12 connect to the Gate, R13 to the Source, and R10 to the Drain?

Originally Posted by R.G.

The CCS apps largely don't work with triodes and poorly with pentodes. But the biasing is similar.

JFETs usually have drain-source breakdowns from 20V-50V.

Is that to say that the variable CCS apps *do* work with triodes and pentodes?

[R.G.]

Thank you Bob. You prodded me to go do some research.

It is *possible* to make depletion mode MOSFETs, it's just that no one does.

Upon looking, I'll have to amend that to no one DID. IXYS released a line of depletion mode MOSFETs of suitable characteristics back in 2002 (just yesterday to me) that will fill the bill and I bet that's what your device is.

More later as I figure it out. It is possible that the schemo is correct. Think what happens if you mentally put a pentode in, plate to drain, grid to gate, cathode to source and ignore how the screen works for the moment.

[R.G.]

Rats. Looks like no one stocks them. They may be available only from IXYS. Too bad. That device is a direct replacement for a lot of tube positions.

[R.G.]

OK, I found available devices. Supertex makes the LND150 which is rated at 500V, up to 30ma, Idss=1-3ma and 740mW in a TO-92 package. They're available from Mouser for $0.55 each, and in stock. These would make good 12AX7 section replacements because they have Ciss of only 10pF.

I bet your current regulator device is a DN3535N3. This is a TO-92 device rated at 450V, 740mW, but with Idss of 200ma, so it would do the current regulation at anything up to that value if the voltage was low enough not to exceed the power rating. Their Ciss is 350pF, which makes them slow for amplification use, but they'd do the regulation just fine.

And I told you wrong. The schematic symbol is correct. That's the symbol for a depletion mode MOSFET. The little isolated bar in the middle of the channel indicates that. I missed that along the way. Now I know.

This pretty much ices it - using the Supertex devices, it IS possible to do a very much stock-looking phase inverter out of depletion mode MOSFETs that will happily be linear outside the range of triode sections they replace. They will even bias with the same methods.

It is possible, using these devices, to replace every single tube in a tube power amp with a MOSFET.

That does not say you'll like the tone, only that the MOSFET will fit in the place and work. They'll probably be too linear, too clean if all the tubes are replaced. But selective replacement might point out the positions where tubes really contribute to the sound, and let us answer the question of what part of the amp sound is the output transformer.

ACK!!! I have to go model some circuits.

[bob p]

Originally Posted by R.G.

OK, I found available devices. Supertex makes the LND150 which is rated at 500V, up to 30ma, Idss=1-3ma and 740mW in a TO-92 package. They're available from Mouser for $0.55 each, and in stock. These would make good 12AX7 section replacements because they have Ciss of only 10pF.

I bet your current regulator device is a DN3535N3. This is a TO-92 device rated at 450V, 740mW, but with Idss of 200ma, so it would do the current regulation at anything up to that value if the voltage was low enough not to exceed the power rating. Their Ciss is 350pF, which makes them slow for amplification use, but they'd do the regulation just fine.

I've looked around for the DN3535N3 -- couldn't find it. I went to the Supertex site and found their Depletion Mode MOSFETs. The closest device I found listed was the DN3535, which comes in a TO-234AA package. This device is only rated at 300V, Idss 200mA. The part number for ordering is DN3535N8 and DN3535N8-G. Notably, the specs are different from those you've listed, so I'm thinking I might not have the right part, or maybe there's a typo somewhere. If you could provide a link to your data I'd appreciate it.

Now that we're homing in on the identity of the part, I wanted to ask about its role as a current regulator for the cathodes of the phase inverter. I'm still trying to determine just why the audio section of the amp was designed this way, as the audio section of this amp is otherwise a plain-jane circuit; it uses a 6N1P triode at the input, going to a pair of 6N1P that act as the phase inverter, and the PI drives a pair of fixed-bias 6550EH running in pentode mode with a plate voltage of about 420 VDC.

What escapes me is why an otherwise conventional amp is using a MOSFET there in the first place. I seem to be missing the problem that they're trying to solve.

[bob p]

FWIW my schematic diagram for the amp has a drawing date of 11-07-01, which suggests that somebody was making depletion mode MOSFETs earlier than we had thought. Does that help to narrow down the manufacturer and/or the identity of the part?

[R.G.]

Originally Posted by bob p

I've looked around for the DN3535N3 -- couldn't find it. I went to the Supertex site and found their Depletion Mode MOSFETs. ... Notably, the specs are different from those you've listed, so I'm thinking I might not have the right part, or maybe there's a typo somewhere. If you could provide a link to your data I'd appreciate it.

Sorry - I had cross-eye-itis when I typed that in, flipping from one window to another. The device is the DN3545N3-G. See:the Supertex datasheet.

Now that we're homing in on the identity of the part, I wanted to ask about its role as a current regulator for the cathodes of the phase inverter. I'm still trying to determine just why the audio section of the amp was designed this way, as the audio section of this amp is otherwise a plain-jane circuit; it uses a 6N1P triode at the input, going to a pair of 6N1P that act as the phase inverter, and the PI drives a pair of fixed-bias 6550EH running in pentode mode with a plate voltage of about 420 VDC.

What escapes me is why an otherwise conventional amp is using a MOSFET there in the first place. I seem to be missing the problem that they're trying to solve.

The big reason to use a CCS in the tail of a diffamp (which is what this is) is to force a high impedance between the cathodes, which forces the balance of the diffamp to be much, much better. The diffamp-derived PIs depend on the transfer of signal from the input cathode to the inactive-side cathode. The second side is effectively a common-grid stage. Anything less than infinite impedance to ground at the joined cathodes leaks signal from the sources and the second tube gets a smaller input signal. The diffamp puts out a bigger signal on the second side than the first. That's why a lot of guitar amps use unbalanced plate resistors in the PI - to try to rebalance the unbalanced output signal from not having a good, high tail CCS.

A CCS in the diffamp tail is the high accuracy way to do it. Using a MOSFET for that CCS is a good application of the parts.

As to why they did it? No way to tell. It's a forward-looking way to do things. Tubes can't really get to a low enough saturation voltage to do that job. A depletion mode MOSFET is perfect.

[bob p]

Originally Posted by R.G.

The big reason to use a CCS in the tail of a diffamp (which is what this is) is to force a high impedance between the cathodes, which forces the balance of the diffamp to be much, much better. The diffamp-derived PIs depend on the transfer of signal from the input cathode to the inactive-side cathode. The second side is effectively a common-grid stage. Anything less than infinite impedance to ground at the joined cathodes leaks signal from the sources and the second tube gets a smaller input signal. The diffamp puts out a bigger signal on the second side than the first. That's why a lot of guitar amps use unbalanced plate resistors in the PI - to try to rebalance the unbalanced output signal from not having a good, high tail CCS.

Thanks. I appreciate that with a diffamp, the balance is improved by constant-current supply to the cathode. Traditionally (in the tube era), a high value Rk would be connected from a highly negative supply, or a constant-current pentode could be used.

On a related note, I have an old HP 739AR Frequency Response Test Set that uses a diffamp that's configured as I described above. Essentially, the 739AR is a variable-frequency constant-voltage signal source that's used for calibrating O-scopes, VTVMs, etc. Interestingly, it uses a diffamp that has a high value cathode resistor (220k) that's tied to -250V that's provided by stacked OA2 and OB2 voltage regulator tubes. (Alternatively, I guess that a constant current pentode could have been used.)

I guess that using the MOSFET in the audio amplifier circuit we've been talking about essentially amounts to the modern equivalent of the 1950s approach that used voltage regulator tubes to balance the diffamp.

Thanks for helping to put this into perspective.

A CCS in the diffamp tail is the high accuracy way to do it. Using a MOSFET for that CCS is a good application of the parts.

As to why they did it? No way to tell. It's a forward-looking way to do things. Tubes can't really get to a low enough saturation voltage to do that job. A depletion mode MOSFET is perfect.

Well, it looks like we've answered the question that prompted me to start this thread! Thanks!

The MOSFET example in the audio amp circuit looks like it definitely fits the bill in terms of using SS devices to help tubes to be better audio devices. Now I just can't resist asking: Is this application going to make it into the MOSFET Follies at GEO?

[R.G.]

Yeah, I guess I'm going to do the next installment there. New devices to play with!!!

[billings]

Originally Posted by bob p

My interest along these lines isn't based so much on the desire to eliminate tubes with SS devices as a cost-saving measure. To some extent, that's what led us all down the path to SS amps in the first place. What I'm more interested in personally is using SS devices to enhance the performance of tubes in performing their intended duties. To that extent, it would be very helpful if I could wrap my mind better around what's really going on with Q1 in that schematic snippet that I posted up at the top of this page. Admittedly, my knowledge of SS components is the weakest link in this discussion, and any help that anyone could offer would be greatly appreciated.

I'm no expert at any of this, but maybe I can be of some small assistance...

As R.G. said, that's got to be a current source in that schematic. Effectively, it's increasing the resistance in the tail of a long tailed pair so that no extra current over the bias current can be sourced or sunk from ground by either tube, it must be stolen from or lent to the other tube's bias current. Thus any signal swing on one tube is more precisely the opposite on the other tube than it would be if there were a current signal to ground.

As far as how the current source works, well, you can actually put a triode or pentode in the exact same position as Q1 in that schematic, just with a few different resistor values. I'm sure there's a better explanation than this, but it might be understood as an amplifier that uses that resistor between cathode/source and ground as a current sensor - if the load tries to draw more current, Vgs will go more negative and plate voltage will rise very quickly before current increases appreciably.

Just looking at the plate curves you can tell that a pentode makes a much better current source than a triode. R.G. has said in articles that in a certain sense a power MOSFET is like a high powered pentode, and the same applies in this case - it's even more apt for a depletion-mode MOSFET, like in the schematic posted.

There is a whole slew of different current sources you can make with transistors. If a hi-fier were resorting to them, I'm surprised they would do something as basic as the one in that schematic.