I know you're keen on the "e-linear" connection. I'm still working with conventional triode mode here. Biased at 9mA per side and a 47K series input resistor I get a respectably high load impedance on the 5687 and I can swing nearly double the voltage I need.
Is there a compelling reason to run significant bias current in the source followers? I calculate that only a fraction of a milliamp is needed to overcome the grid capacitance, even accounting for Mr. Miller. I was thinking of setting the bias at 2mA to reduce waste heat. Naturally, much more current is available in the positive-going direction.
I'm looking at some SiC MOSFETS for the followers. They have very low Crss.
Opinion?
Is there a compelling reason to run significant bias current in the source followers? I calculate that only a fraction of a milliamp is needed to overcome the grid capacitance, even accounting for Mr. Miller. I was thinking of setting the bias at 2mA to reduce waste heat. Naturally, much more current is available in the positive-going direction.
I'm looking at some SiC MOSFETS for the followers. They have very low Crss.
Opinion?
The more current, the higher the gm, and for that I would waste a fair bit o' heat.
I built a very simple g2 supply for those S258Q, 6CB5A amps; just hung a MOSFET on a voltage divider with a zener to protect the gate. When I put a resistor between g2/source and ground and bled off another 20 mA the regulation improved( as well as the output impedance decreasing due to more gm, and output Z being basically 1/gm). It sounded better too...🙂 ( now don't throw me in the same pail as moodwin for that remark ).
It is a basic realization arrived at from the original E-Linear experiments/development that drives high output Z elements( like pentodes, or cascode with triode on the bottom/mosfet on top). The fet on top builds a lot less 'here be Monsters' area on the plate curves of the cascode too.
cheers,
Douglas
I built a very simple g2 supply for those S258Q, 6CB5A amps; just hung a MOSFET on a voltage divider with a zener to protect the gate. When I put a resistor between g2/source and ground and bled off another 20 mA the regulation improved( as well as the output impedance decreasing due to more gm, and output Z being basically 1/gm). It sounded better too...🙂 ( now don't throw me in the same pail as moodwin for that remark ).
It is a basic realization arrived at from the original E-Linear experiments/development that drives high output Z elements( like pentodes, or cascode with triode on the bottom/mosfet on top). The fet on top builds a lot less 'here be Monsters' area on the plate curves of the cascode too.
cheers,
Douglas
True about the transconductance.
I have to think about cascoding the driver. It depends on whether that lets me swing less current in the 5687. At some point it's too much sand. Hell, I could solve some more problems by using a JFET cascode differential in the input. Or remove the tubes entirely...
I have to think about cascoding the driver. It depends on whether that lets me swing less current in the 5687. At some point it's too much sand. Hell, I could solve some more problems by using a JFET cascode differential in the input. Or remove the tubes entirely...
Gary Pimm built an amp with cascode MOSFET elements where ever a pentode was called for in his Tabor amp. Everything SS. Basically an N-channel pentode amp( including the OPT ).
If you think I have not considered doing away with the tube input stage, and substituting a cascode element myself, I'll tell you that you are mistaken. The finals I would leave as tubes; doing away with that much heat is a pain in the tucchus.
cheers,
Douglas
The 'thing' with a triode driving a shunt FB-ed output stage is the triode plate will suck up a fair bit o' that FB. I like the linearizing effect on the finals, and it is augmented by lowering their output Z( effective, in-circuit plate resistance ). There was a thread a while back on here comparing CFB to shunt/Schade FP, and somebody posted a set of MOSFET curves, with instead of having horizontal drain curves, they were vertical, like a triode with no curvature at low current and near zero plate resistance( the 'slant' ).
cheers,
Douglas
Schade and CFB exactly equivalent
I think that is it...🙂
cheers,
Douglas
Schade and CFB exactly equivalent
I think that is it...🙂
Now then...if we took the usually-grounded grid of the input LTP, and instead attached it to a common cathode R the finals are sharing, what would happen?
cheers,
Dougals
cheers,
Dougals
I'm trying to stay focused here. Turning an idea into a working piece of equipment is a lot of work. If you want to see a finished product one day, you'd best not keep distracting me, hah.
IIRC, under very limited conditions it's possible to use a partially-bypassed common cathode resistor to null out third-order distortion in the output stage. Didn't S. Bench publish a design based on that principle? I don't remember. Doing it willy-nilly sounds like a recipe for awfulness.
IIRC, under very limited conditions it's possible to use a partially-bypassed common cathode resistor to null out third-order distortion in the output stage. Didn't S. Bench publish a design based on that principle? I don't remember. Doing it willy-nilly sounds like a recipe for awfulness.
It'll at least give alternating positive and negative feedback with signal peaks. Too entertaining for me.
All good fortune,
Chris
All good fortune,
Chris
I was going to refer you, Chris, to a thread from earlier this year talking about the effect of a shared unbiased cathode resistor in the output stage. But I see you posted to that thread, so no need.
There was a time, like forty years ago, when I would have been able to analyze this problem if forced to by a professor. Now, not so much.
Too entertaining for you and me both.
There was a time, like forty years ago, when I would have been able to analyze this problem if forced to by a professor. Now, not so much.
Too entertaining for you and me both.
You are not the only one with a pile of parts, including a pair of S-271-S sitting on a bench Henry...🙂 Have you solid CCS boards? DN2540's to fill them?
cheers,
Douglas
I'll need current sources. Eventually I'll build them. I'm not sure what physical form the driver will take yet.
Is there a compelling reason to use a current source in the tail of the second stage of a two-stage differential amplifier? By going to a common tail resistor I can put current source loads on the plates, which raises the effective plate load resistance by a factor of three in this circuit. Since the second-stage input voltages are already balanced, the improvement due to a CCS in the tail, if at all, would be a second-order effect. I would think any harm from that would be outweighed by the benefit of substantially reduced AC current swing.
Is there a compelling reason to use a current source in the tail of the second stage of a two-stage differential amplifier? By going to a common tail resistor I can put current source loads on the plates, which raises the effective plate load resistance by a factor of three in this circuit. Since the second-stage input voltages are already balanced, the improvement due to a CCS in the tail, if at all, would be a second-order effect. I would think any harm from that would be outweighed by the benefit of substantially reduced AC current swing.
post deleted,
wanted to say "dont forget the internal plate resistance", but you allready did.
wanted to say "dont forget the internal plate resistance", but you allready did.
Last edited:
There is a further interesting bit you can try, recall that the 12BH7 stage in McIntosh amps is few positive FB from the OPT so it can deliver larger voltage swing. You can try that connection, but instead of resistors that will let the FB through, apply current reg loads. I've advocated doing this for the Mc amps, but for the average bear they are too valuable to experiment on...and he who gets a Mc amp already has the best so why would they modify...LOL
This does preclude a second CCS-ed tail load stage, and I'll have to agree with you that it isn't needed.
cheers,
Douglas
This does preclude a second CCS-ed tail load stage, and I'll have to agree with you that it isn't needed.
cheers,
Douglas
The scheme I have here in SPICE has 8.6mA CCS loads on the plates of the 5687. There is an additional 0.4mA coming from the 390K shunt feedback resistors, per tube.
At clipping, the 807 grids have 97V peak-to-peak signal versus 215V peak-to-peak on the 5687 plates. The difference is 6.9dB, which conforms to Schade's recommendation.
There are 47K series resistors between the 5687 plates and the source follower gates. The load current is 2.5mA peak-to-peak corresponding to an effective driver plate load of 86K. The signal current in the tail resistor is 12uA peak-to-peak and is noticeably distorted compared to a clean-looking sinewave with a CCS. But I don't think this is an issue.
So the driver tubes are swinging +-1.25mA against a bias current of 9.0mA per side. I think this is a pretty heathy operating point. If I feed the CCS loads from +525V I have enough headroom to drive the 807 grids to almost twice the clipping level. The waveform when I do that is ugly because of feedback from the clipped output tubes.
The closed-loop gain is 30dB and there is 8.5dB of global feedback.
I've specified +-150V supplies for the source followers and their current sources. I'm going to need heatsinks to dissipate up to a couple of watts per device.
Given that the 5687 plate resistance is probably around 4K Ohms, I feel comfortable with this design and don't see the need to go with cascodes or pentodes. I'm tempted to build it. Too bad I can't bloody click "Build" in SPICE and 3D print myself a prototype.
At clipping, the 807 grids have 97V peak-to-peak signal versus 215V peak-to-peak on the 5687 plates. The difference is 6.9dB, which conforms to Schade's recommendation.
There are 47K series resistors between the 5687 plates and the source follower gates. The load current is 2.5mA peak-to-peak corresponding to an effective driver plate load of 86K. The signal current in the tail resistor is 12uA peak-to-peak and is noticeably distorted compared to a clean-looking sinewave with a CCS. But I don't think this is an issue.
So the driver tubes are swinging +-1.25mA against a bias current of 9.0mA per side. I think this is a pretty heathy operating point. If I feed the CCS loads from +525V I have enough headroom to drive the 807 grids to almost twice the clipping level. The waveform when I do that is ugly because of feedback from the clipped output tubes.
The closed-loop gain is 30dB and there is 8.5dB of global feedback.
I've specified +-150V supplies for the source followers and their current sources. I'm going to need heatsinks to dissipate up to a couple of watts per device.
Given that the 5687 plate resistance is probably around 4K Ohms, I feel comfortable with this design and don't see the need to go with cascodes or pentodes. I'm tempted to build it. Too bad I can't bloody click "Build" in SPICE and 3D print myself a prototype.
Yeah, that is too bad...I wish it were possible to a point. Throw tubes, transformers and resistors into the 'build tank' then press build would put a lot of good parts on the 'used' market when folks tried to just engineer with Spice.
cheers,
Douglas
I have to think hard about how to actually build this thing. I don't have a workshop which makes everything much more challenging. I never expected to end up in such a modest house. But real estate prices just keep going up and time flies and whatever. Ever since I was a kid no matter what I was building I wanted to be bloody perfect. So it takes me a long time.
Like I said, I feel pretty good about the basic circuit. Now I need to solve a slew of other problems. It gets harder in three days when work starts again. Ugh, work. Anyway. Good progress so far.
Like I said, I feel pretty good about the basic circuit. Now I need to solve a slew of other problems. It gets harder in three days when work starts again. Ugh, work. Anyway. Good progress so far.
Turns out my 30 year-old HP digital scope is broken. I went to use it last year and the TV screen was on the fritz. It's an OEM Hitachi green CRT assembly. According to the interwebz the problem is dried up capacitors on the display board. So I just ordered a slew of new caps from Digi-Key.
While I was at it I ordered ten each of TO-92 and TO-220 DN2540s, and four expensive Cree SiC MOSFETS.
Hopefully the caps will fix my scope. I also have a Tek 475A that's in beautiful mechanical shape but Dog knows how long that's going to keep working. I don't want to think about trying to fix it when it breaks.
While I was at it I ordered ten each of TO-92 and TO-220 DN2540s, and four expensive Cree SiC MOSFETS.
Hopefully the caps will fix my scope. I also have a Tek 475A that's in beautiful mechanical shape but Dog knows how long that's going to keep working. I don't want to think about trying to fix it when it breaks.