Thanks! Indeed, I have considered that. When I implement the bias servo, I will sense current from a 1M resistor from plate-to-servo board and will eliminate the parallel resistor. I also plan to try tying the 1M to the source of the CCS mosfet, which will further increase gain in the driver and make bias harder to maintain (but still should be easy for the bias servo).
There are probably lots of ways of improving stability on this. The plate voltage was actually quite stable when I derived the bias from a tap on the lower feedback "resistor" (actually two trim pots). It was only when I added the p-channel fet that this became difficult. With the parts I had on hand, I just made a simple bias supply out of a small trim pot and a cheap +15V switching supply. The plate drift I am seeing could be from the drift on the output of that supply. In other words, this was quick and dirty and I guarantee it is not the best that can be done. I think the method you suggest would add considerable stability.
Earlier, I had considered using an op-amp follower instead of a p-channel fet follower on the cathode of the input stage. My idea was to have a trim pot inside the feedback loop of the op-amp so that the voltage drop across the adjustable resistor would provide the bias, but the op-amp feedback would keep impedance at the cathode very low.
The reason I'm going with the servo is that I designed a bias servo PCB with 4 channels, and that is perfect to bias this stage and the output stage in a SE stereo amp. I have already tested the design in other amps and I know it will work well, so it's the path of least resistance for me, but it's not a solution that everyone will like.
This all ended up working out way better than I expected.
Throughout these experiments, I could hear the amp improving in its ability to play hard rock/electronic music cranked to 11, but losing warmth on something like Norah Jones. It's just kind of the tradeoff.
Luckily, I think I'll be able to build switchable plate loads into drivers on the final amp so I can choose whether I want "SE tube sound" or low distortion each time before I power up.
This is what I had in mind. A quick sketch attached. As per previous post on the 300B DHT thread, something similar is what I tested with the 6e5p. I believe George is developing a board on these lines with his UNSET design?
Anyhow, JC Morrison is using DC feedback loop into the screen to help tracking the anode voltage to improve stability of his gm multiplier block. Check it out. Also played with Frank B “best pentode” stage which works well.
I think I could implement this on the 300B as a good driver to increase a bit the gain from the D3a in triode mode.
I will probably rig this up and do some tests. The performance at large voltage swing (i.e.200Vpp) should be promising.
Any further suggestions to refine this?
I think it makes sense you reuse the servo boards if you have already developed them.
Cheers
Ale
I like it. The only way to tell how well it works is to build it and see how stable the bias is. If it proves very stable, you could maybe attach the upper feedback resistor to the source of the CCS mosfet (the mu output). If the bias scheme is good enough to provide bias stability into that, it should offer even lower distortion since the pentode plate will only see the CCS as the load.
General trends I saw in my experiments with CCS-loaded pentodes were that higher plate idle voltages offered lower distortion and lower screen voltages offered lower distortion.
If I were doing this, I'd be tempted to run up a CCS-loaded 47, run as pentode with plate-grid feedback, and see how that worked. That'd be a very pretty solution.
In case you didn't notice, tube looks are important to me. Those D3As are pretty too. That'd be a lot of eye candy in one amp.
Good suggestions, thank you. I will rig something up and test. Best way to tweak and improve. The 47 would be a good option but don’t want to modify current 300B amp driver board as will leave me without music until I get it up and running.
I do also like the nice glowing valves, the thoriated tungsten are obviously on top of the list. I’ve been running 01a and 10Y/VT-25 and 801a for some time. Also as output stage I had the 814. I think this is still a good option as an output stage as per your design or even with the driver design I posted before.
Cheers
Ale
I do also like the nice glowing valves, the thoriated tungsten are obviously on top of the list. I’ve been running 01a and 10Y/VT-25 and 801a for some time. Also as output stage I had the 814. I think this is still a good option as an output stage as per your design or even with the driver design I posted before.
Cheers
Ale
814 looks like a great candidate for this approach.
I was looking at the 4P1L datasheet (never used them myself) as well and it looks like a better candidate than the 47. The 47 is pretty low gain(100 or so), even as a CCS-loaded pentode, so I'm not sure if distortion would be that low if you used it, but 4P1L seems to have a gain of over 300 and more linear curves. The 4P1L also brings the suppressor grid out on its own pin so you could try biasing the suppressor to square up the curves even more and lower distortion further.
Of course, using a D3A as a CCS-loaded pentode with local feedback would probably give you as much closed-loop gain as you could possibly want with low distortion, similar to the results I'm getting with the 6BN11, but probably even better. Whether you apply feedback to the pentode stage itself or involve the output stage as well is a matter of preference, but both can work very well.
In my experience so far, the nice thing is that the harmonic profile of these pentode stages is a waterfall, even when I'm being heavy-handed with the feedback. So if you like the coloring of SET amps, you can target a particular amount of distortion by carefully choosing the plate load of the pentode to give the right amount of excess gain for feedback.
I plan on having a switch in the final version of this amp that has one driver plate load that minimizes distortion and one that gives the amp "SET sound," just by switching the driver plate load.
I was looking at the 4P1L datasheet (never used them myself) as well and it looks like a better candidate than the 47. The 47 is pretty low gain(100 or so), even as a CCS-loaded pentode, so I'm not sure if distortion would be that low if you used it, but 4P1L seems to have a gain of over 300 and more linear curves. The 4P1L also brings the suppressor grid out on its own pin so you could try biasing the suppressor to square up the curves even more and lower distortion further.
Of course, using a D3A as a CCS-loaded pentode with local feedback would probably give you as much closed-loop gain as you could possibly want with low distortion, similar to the results I'm getting with the 6BN11, but probably even better. Whether you apply feedback to the pentode stage itself or involve the output stage as well is a matter of preference, but both can work very well.
In my experience so far, the nice thing is that the harmonic profile of these pentode stages is a waterfall, even when I'm being heavy-handed with the feedback. So if you like the coloring of SET amps, you can target a particular amount of distortion by carefully choosing the plate load of the pentode to give the right amount of excess gain for feedback.
I plan on having a switch in the final version of this amp that has one driver plate load that minimizes distortion and one that gives the amp "SET sound," just by switching the driver plate load.
Good suggestions, thank you.
I used the 4P1L extensively over the years, great valve. As you say, it may be useful to try +12V on G3 as recommended by Anatoliy.
I built a single 4P1L pentode SE amp 6 years ago with just a LND150 driver and plate to grid feedback as per your initial tests. It sounded really well and could squeeze as much power as possible from one single bottle:
Robustiano (V0.7) – Bartola(R) Valves
The 47 has a lovely sound. Gary Pimm implemented a fantastic design with them in PP with similar feedback arrangement. I listened to a clone of it and was impressed with the sound. I always wanted to build one, however other projects got into the way.
As you say in your blog, using high resistance transmitting triodes is a good path with this topology. I have a bunch of 811 as well of T40s around. Need to find them but they would be great candidates for this!
Cheers
Ale
I used the 4P1L extensively over the years, great valve. As you say, it may be useful to try +12V on G3 as recommended by Anatoliy.
I built a single 4P1L pentode SE amp 6 years ago with just a LND150 driver and plate to grid feedback as per your initial tests. It sounded really well and could squeeze as much power as possible from one single bottle:
Robustiano (V0.7) – Bartola(R) Valves
The 47 has a lovely sound. Gary Pimm implemented a fantastic design with them in PP with similar feedback arrangement. I listened to a clone of it and was impressed with the sound. I always wanted to build one, however other projects got into the way.
As you say in your blog, using high resistance transmitting triodes is a good path with this topology. I have a bunch of 811 as well of T40s around. Need to find them but they would be great candidates for this!
Cheers
Ale
Hi SpreadSpectrum and company.
I am attempting to implement something similar, plate to cathode feedback in my 801A A2 amplifier. I am a novice when it comes to implementing NFB, so I would appreciate your feedback (no pun intended). NFB was taken from the 801A plate to the cathode of my pentode input, the EF37A. There is no cap blocking DC from the 801A plate, as such, DC current on the EF37A cathode is part of the bias. Unsure if this will have negative effects, but the sound is significantly improved with the cap omitted.
The anode current of the EF37A is set by a cascode CCS and the plate voltage / load is determined by a 470K resistor to ground. The cathode is degenerated.
801A / EF37A / source follower V+ are Maida regulated, negative supply is shunt regulated, 801A filament supply is Rod Coleman's.
With this setup, I am getting excellent sound. I am detecting no bias point drift on the EF37A using a 470K plate load.
THD @1W 0.25%
Output Z 2.15 (great for the 801A with a 6.5K:8 OPT)
Output power unclipped 5.2W at less than 1W THD
My schematic is below, wondering if there is anything I have overlooked. My goals are not quite so lofty as your SpreadSpectrum, I would be very happy with my amplifier if stable in these conditions
Thanks!
I am attempting to implement something similar, plate to cathode feedback in my 801A A2 amplifier. I am a novice when it comes to implementing NFB, so I would appreciate your feedback (no pun intended). NFB was taken from the 801A plate to the cathode of my pentode input, the EF37A. There is no cap blocking DC from the 801A plate, as such, DC current on the EF37A cathode is part of the bias. Unsure if this will have negative effects, but the sound is significantly improved with the cap omitted.
The anode current of the EF37A is set by a cascode CCS and the plate voltage / load is determined by a 470K resistor to ground. The cathode is degenerated.
801A / EF37A / source follower V+ are Maida regulated, negative supply is shunt regulated, 801A filament supply is Rod Coleman's.
With this setup, I am getting excellent sound. I am detecting no bias point drift on the EF37A using a 470K plate load.
THD @1W 0.25%
Output Z 2.15 (great for the 801A with a 6.5K:8 OPT)
Output power unclipped 5.2W at less than 1W THD
My schematic is below, wondering if there is anything I have overlooked. My goals are not quite so lofty as your SpreadSpectrum, I would be very happy with my amplifier if stable in these conditions
Thanks!
Last edited:
I like it! It seems like you are getting great results.
Cap coupling the feedback gives the advantage of reduced resistor heat, but it adds another time constant in the feedback loop. My philosophy is to use no more time constants in the feedback loop than are absolutely necessary, so I just lived with the heat, but I tried to minimize it with a 220k feedback resistor.
You have used a 50k resistor, so more heat. But this also makes the cathode degeneration in the input stage less significant in comparison to the higher feedback network currents. The only real drawback there is the heat and some lost output power since the feedback resistor is a parallel load.
33pF sets the cutoff frequency at ~96kHz, which is nice and high. How do square waves look? I had no ringing whatsoever but I set mine lower at ~65kHz. I still haven't gotten any more small-value caps to test and see what I can get away with. What kind of OT are you using?
One thing I'd look at is if the screen voltage sags at all at high output. I like how the screen is referenced to the cathode, but I'd test to make sure that the screen supply is stiff enough to keep voltage constant up to high output levels.
Is there a reason for the 1k grid resistor on the 801A? You will probably see a lot less distortion at higher levels without it.
ETA: As far as bias stability goes, I only had issues with mine when I added the p-channel fet to the cathode and made a bias supply out of a trimpot in the parts pile and a cheap switching supply. If the input tube plate voltage seems stable, you're probably good. I'm going to add a bias servo to mine since I already designed boards for that and I wanted that for the output tube anyway. It's a four channel board so I know what to use the other two channels for.
Cap coupling the feedback gives the advantage of reduced resistor heat, but it adds another time constant in the feedback loop. My philosophy is to use no more time constants in the feedback loop than are absolutely necessary, so I just lived with the heat, but I tried to minimize it with a 220k feedback resistor.
You have used a 50k resistor, so more heat. But this also makes the cathode degeneration in the input stage less significant in comparison to the higher feedback network currents. The only real drawback there is the heat and some lost output power since the feedback resistor is a parallel load.
33pF sets the cutoff frequency at ~96kHz, which is nice and high. How do square waves look? I had no ringing whatsoever but I set mine lower at ~65kHz. I still haven't gotten any more small-value caps to test and see what I can get away with. What kind of OT are you using?
One thing I'd look at is if the screen voltage sags at all at high output. I like how the screen is referenced to the cathode, but I'd test to make sure that the screen supply is stiff enough to keep voltage constant up to high output levels.
Is there a reason for the 1k grid resistor on the 801A? You will probably see a lot less distortion at higher levels without it.
ETA: As far as bias stability goes, I only had issues with mine when I added the p-channel fet to the cathode and made a bias supply out of a trimpot in the parts pile and a cheap switching supply. If the input tube plate voltage seems stable, you're probably good. I'm going to add a bias servo to mine since I already designed boards for that and I wanted that for the output tube anyway. It's a four channel board so I know what to use the other two channels for.
Last edited:
Lovely circuit. I'm a big fan of both valves myself!
I'd add as well a couple of points on the screen supply and the source follower at the output:
1. I'd change the 1Meg supply into the EF37 with a MOSFET deriving the DC voltage from the anode with a local feedback arrangement as I post earlier. You will guarantee better stability of this high gain stage. I think otherwise you will struggle to keep a stable screen with the current setup.
2. The Source Follower at the output has to be well decoupled to avoid oscillation. A 100uF + 100nF stacked film capacitors right at the +B and -B of the source follower with minimal leads into the FET and bottom CCS. Keep Colpitts away!. Also, I know is easier to implement a symmetrical 150V but you are asking more heat in the pass FET unnecessarily as you don't need more than +25V. Perhaps a dropping resistor on the +150V will help keeping the FET cooler.
Otherwise, looks great!
Ale
I'd add as well a couple of points on the screen supply and the source follower at the output:
1. I'd change the 1Meg supply into the EF37 with a MOSFET deriving the DC voltage from the anode with a local feedback arrangement as I post earlier. You will guarantee better stability of this high gain stage. I think otherwise you will struggle to keep a stable screen with the current setup.
2. The Source Follower at the output has to be well decoupled to avoid oscillation. A 100uF + 100nF stacked film capacitors right at the +B and -B of the source follower with minimal leads into the FET and bottom CCS. Keep Colpitts away!. Also, I know is easier to implement a symmetrical 150V but you are asking more heat in the pass FET unnecessarily as you don't need more than +25V. Perhaps a dropping resistor on the +150V will help keeping the FET cooler.
Otherwise, looks great!
Ale
Thank you SpreadSpectrum and Ale for the wonderful suggestions!
SpreadSpectrum - yes, the 50K feedback resistor does get quite hot, which I anticipated, it is a 20W wirewound. It is possible the value could be altered, I will runs some simulations and see if I can achieve a similar feedback fraction and bias point with a higher resistance part. If not, then perhaps I will use two 100K 20W in parallel to spread the dissipation around. I am glad to hear my DC blocking feedback capacitor is not necessary, the difference in clarity of sound is significant with the cap removed.
On the 1K resistor on the grid of the 801A, I had placed a low wattage part there to act as a fuse of sorts in case of a failure of the source follower FET. Maybe this is a paranoia, I think I will remove it with your recommendation.
As for the circuit changes suggested by you Ale, Vg2 stability, and square waves, let me take a look, run some simulations, and return with some measurements! I am working the weekend, so it will be some time, thanks for your input.
SpreadSpectrum - yes, the 50K feedback resistor does get quite hot, which I anticipated, it is a 20W wirewound. It is possible the value could be altered, I will runs some simulations and see if I can achieve a similar feedback fraction and bias point with a higher resistance part. If not, then perhaps I will use two 100K 20W in parallel to spread the dissipation around. I am glad to hear my DC blocking feedback capacitor is not necessary, the difference in clarity of sound is significant with the cap removed.
On the 1K resistor on the grid of the 801A, I had placed a low wattage part there to act as a fuse of sorts in case of a failure of the source follower FET. Maybe this is a paranoia, I think I will remove it with your recommendation.
As for the circuit changes suggested by you Ale, Vg2 stability, and square waves, let me take a look, run some simulations, and return with some measurements! I am working the weekend, so it will be some time, thanks for your input.
If you have a good bias point worked out and you don't mind designing for the heat, using the 50k resistor is fine.
One thing that concerns me a bit is the +150V supply to the mosfet will allow some significant stress on the 801A grid if you drive hard into clipping. Less headroom in that supply will enhance the safety of those beautiful 801As.
My amp requires up to ~+60V on the grid to hit saturation and I feed the mosfet with a +96V supply. That way, the feedback system will hit a hard stop there. The more supply you give it, the harder it will drive that grid to try and create more output, and with a 150V supply, it can probably really stress that grid.
When I was testing my amp, I stopped at 19W and I only tested there for a short moment (you will notice that all my other plots have averaging, but not that one). I made that measurement as a quick on, capture, then off.
I really wanted to get a distortion figure at 20W but I chose to stop at 19W when I saw the grid waveform at the output tube. Tall spikes started forming with even tiny increases in amplitude and I didn't think it was wise to do that with a high-powered sine wave for a duration of seconds. In real music, durations in those conditions will be short if they even ever occur, and I'm limited to 36V above saturation. Maybe a little less would be better but I'm stacking 48V switching supplies for these experiments (and probably the final amp) so I am limited to multiples of 48.
It's probably a good idea to monitor the output tube grid waveform during testing as you will be able to see if you are doing something that could potentially cause damage to the tube. Watch for when the big spikes start forming.
I see your point with the +150V supply. Up to this point I have been adding just enough NFB to keep the 801A out of hard clipping, so it reaches maximum unclipped output at full volume with a 2Vrms input to the EF37A. The +150V supply is Maida regulated with an adjustable trimmer, so I can trade some voltage drop across the source follower for some voltage drop across the the Maida power MOSFET, I think that will be a good trade, the Maida will be well heat sinked. The 801A grid swings from around +55V to -90V, so I could reduce the supply to perhaps +100V or so. The +/- supplies are two halves of a dual secondary 150VAC toroid from Antek, I could probably afford to lose some voltage on both supplies, so maybe a smaller toroid is in order, thanks for pointing it out.
Will be very interested to see Corona in its final iteration, your results are very impressive! The 826 was not known to me before seeing this thread, they are very nice looking tubes.
Will be very interested to see Corona in its final iteration, your results are very impressive! The 826 was not known to me before seeing this thread, they are very nice looking tubes.
I have been using this method:
Leave the 150 volt drain supply as is. Choose an appropriately sized resistor to go between the mosfet drain and the supply with a bypass capacitor on the drain side.
The voltage at the drain will drop as the grid starts to draw current, but the PSRR of a follower is very high, so no distortion of the drive waveform occurs. At a predetermined point approaching excessive grid current the drain voltage will drop far enough to cause distortion.
The bypass cap can be sized such that a music peak will be fed by the cap but continuous operation in the "too much grid current" zone will cause distortion, but save the grid.
Leave the 150 volt drain supply as is. Choose an appropriately sized resistor to go between the mosfet drain and the supply with a bypass capacitor on the drain side.
The voltage at the drain will drop as the grid starts to draw current, but the PSRR of a follower is very high, so no distortion of the drive waveform occurs. At a predetermined point approaching excessive grid current the drain voltage will drop far enough to cause distortion.
The bypass cap can be sized such that a music peak will be fed by the cap but continuous operation in the "too much grid current" zone will cause distortion, but save the grid.
It's not paranoia!
The low-watt resistor acting as a fuse is worth keeping, as protection for hard-to-replace tubes - but it can go down to 100Ω. Power FETs almost always fail short-circuit! One cause of failure can be oscillation. Ale mentioned the supply decoupling, & that is well worthwhile. Another precaution is to use a 100Ω 1/16W at the FET's source, to prevent wiring/stray capacitance/grid loads from making the (Real part) of the FET gate's impedance from looking negative. This sounds arcane, I know! But it's easy to end up doing this, so 100Ω gate and source resistors are worth the trouble.
The source resistor protects in two ways, if it is dimensioned to blow out before a damaging amount of current flows.
Thanks, Rod, in that case I will simply lower the resistor's value to 100ohm, I will investigate the source resistor as well.
In regard to the stability of Vg2 in the CCS fed pentode stage, can the screen voltage simply be regulated? Since it was suggested to lower the source follower B+ to spare the 801A grid, this Maida regulated supply could easily be lowered to 100V, decoupled, and fed to the EF37A screen as well, fixing it to 100V, resolving two issues in one go. Too simple perhaps? I still have much to learn.
In regard to the stability of Vg2 in the CCS fed pentode stage, can the screen voltage simply be regulated? Since it was suggested to lower the source follower B+ to spare the 801A grid, this Maida regulated supply could easily be lowered to 100V, decoupled, and fed to the EF37A screen as well, fixing it to 100V, resolving two issues in one go. Too simple perhaps? I still have much to learn.