Hey everyone,
I've been working a while on understanding what goes into a push-pull amp. Attached is what I've come up with so far, based on my understanding and simulations. It should be a mostly Class-A, ultra-linear, around 30W per channel before clipping, and I intend to build it as mono blocks. I chose tubes which are readily available and tried to keep things simple, where possible. I used local and global feedback and the gain should be around 27.
I would like some feedback and suggestions for improvement on the design. I'm sure that somewhere I've probably made a major mistake or two, and I may have gone overboard trying to optimize the stability and distortion in the simulator. I'm no electrical engineer, so I appreciate your support, let me know what you think!
Thanks,
John
I've been working a while on understanding what goes into a push-pull amp. Attached is what I've come up with so far, based on my understanding and simulations. It should be a mostly Class-A, ultra-linear, around 30W per channel before clipping, and I intend to build it as mono blocks. I chose tubes which are readily available and tried to keep things simple, where possible. I used local and global feedback and the gain should be around 27.
I would like some feedback and suggestions for improvement on the design. I'm sure that somewhere I've probably made a major mistake or two, and I may have gone overboard trying to optimize the stability and distortion in the simulator. I'm no electrical engineer, so I appreciate your support, let me know what you think!
Thanks,
John
Last edited:
Hi John,
I’m working on a lookalike, based on the Music Machine diagram. Have you already built a prototype? I’ve never tried feeding the phase shifter from the ul-taps, I use the B+ instead. Perhaps it’s worth trying. I’m using another 10M45S in the anode of the driver stage to lower 2nd harmonics. A balancing pot in the phase shifter anode helps too.
Do you have LTSpice files for this to compare with mine?
Regards, Gerrit
I’m working on a lookalike, based on the Music Machine diagram. Have you already built a prototype? I’ve never tried feeding the phase shifter from the ul-taps, I use the B+ instead. Perhaps it’s worth trying. I’m using another 10M45S in the anode of the driver stage to lower 2nd harmonics. A balancing pot in the phase shifter anode helps too.
Do you have LTSpice files for this to compare with mine?
Regards, Gerrit
Actually, it'll probably blow a fuse. High voltage windings 3 and 5 are shorted.
The corrected power supply is attached.
Gerrit,
No, I haven't prototyped it, so far I have only simulated this in LTSpice and I attached that. Hopefully, everything necessary is there in the file. With the CCS, I don't think any additional balancing is needed for the phase splitter. I asked about that in another thread, and was pointed to the thread linked below. It is a pretty thorough explanation with validation.
Balance in CCS Long Tailed Pairs
Feeding from the UL taps adds some negative feedback from the output tubes to the drivers, similar to what a lot of people on this forum call Schade or plate-to-plate feedback.
Thank you, lcsaszar!
Gerrit,
No, I haven't prototyped it, so far I have only simulated this in LTSpice and I attached that. Hopefully, everything necessary is there in the file. With the CCS, I don't think any additional balancing is needed for the phase splitter. I asked about that in another thread, and was pointed to the thread linked below. It is a pretty thorough explanation with validation.
Balance in CCS Long Tailed Pairs
Feeding from the UL taps adds some negative feedback from the output tubes to the drivers, similar to what a lot of people on this forum call Schade or plate-to-plate feedback.
Thank you, lcsaszar!
Last edited:
The unintended connection between D1 and D2 is now fixed, but I agree with Koda that you will have 1KV rather than 500V. Why double up on the capacitors? I’d just use full-wave rectification (2 diodes) with the center tap grounded.
EDIT: Koda posted while I was still typing, but we agree.
The V1b cathode follower seems unnecessary and just adds another coupling capacitor - you have three in the feedback loop which will surely lead to low frequency instability. C12 and C13 in the power supply decoupling look to be too low in value, would expect to see a few uF here? The grid stoppers of 75k are very high, normally about 1k. 75k will cause excessive high frequency roll-off.
Ok, is this power supply looking a little better? I made a couple miscalculations between my modeling and reality, but I think I have it a little more correct now. I also eliminated a resistor stage to minimize the series resistance.
There were also a few other errors copying between my simulation and drawing I corrected, particularly in the bias supply. C12 and C13 are another mistake, I planned to use 470u and 220u respectively, which are probably overkill. Sorry about the errors.
The 75k grid resistors do roll off the HF, but I didn't think it was excessive in the simulation. I've attached plots with 1k and 75k for each stage. The measurement points are in the LTSpice file. Is there a reason to amplify the higher than audible frequencies?
V1B has been one of my deepest frustrations with this project. Because I'm building this as monoblocks, I hate to leave one half of a triode unused, but maybe that is what I need to do. If I use it as a gain stage, I have too much gain. I've tried different topologies and tubes. I'm still not certain what to do there.
There were also a few other errors copying between my simulation and drawing I corrected, particularly in the bias supply. C12 and C13 are another mistake, I planned to use 470u and 220u respectively, which are probably overkill. Sorry about the errors.
The 75k grid resistors do roll off the HF, but I didn't think it was excessive in the simulation. I've attached plots with 1k and 75k for each stage. The measurement points are in the LTSpice file. Is there a reason to amplify the higher than audible frequencies?
V1B has been one of my deepest frustrations with this project. Because I'm building this as monoblocks, I hate to leave one half of a triode unused, but maybe that is what I need to do. If I use it as a gain stage, I have too much gain. I've tried different topologies and tubes. I'm still not certain what to do there.
The PS looks better now. 500V... Now that you are only using one cap (instead of two in series) you'll need 600V caps to be safe... You might want to go back to the first model, but with the new CT/diode arrangement. Go with the 1k version IMHO. My amps are flat to ~50kHz. You might not hear it, but with the right speakers, you'll feel it.
Last edited:
Another reason to go with 1k grid stoppers is that with 75k there will be more phase shift at high frequencies leading to stability issues when feedback is applied. With a straight pentode output stage there may be some advantages in going to a higher value (SY used 47k in his Red Light District amplifier), but values typically much lower with triode/ultralinear due to higher input capacitance.
Paralleling the 1st two triodes is the other thing that I've played around with. This eliminates the extra coupling capacitor and doesn't leave a wasted triode. I seem to get more distortion with this topology, but it's still pretty low. I guess this design is pretty similar to the one that Salas drew, only with the different supply for the phase splitter.
Have you thought about direct coupling the voltage amp stage to the inverter/driver stage? This would eliminate a coupling capacitor and thus eliminate an R/C time constant so that it would be a bit easier to stabilize the amp when the FB loop is closed.
There are several design issues to work through if you pursue this. See for example the Eico HF-60 for how this can be done.
There are several design issues to work through if you pursue this. See for example the Eico HF-60 for how this can be done.
The 6SN7 stage is operating pretty far from its most linear region. You might try about twice as much current, and see what you think. An excellent use for your dual triode input would be as a differential input, but even here type 12AT7 isn't the most linear choice. I do like your tight coupling of 6SN7 driver and output stage. Could it be expanded to include the whole output transformer?
All good fortune,
Chris
All good fortune,
Chris
Hi Chris,
You're right, I need to look at the 6SN7 a little more. Somewhere along the line, I ended up with a 6SN7GT datasheet that has much lower specifications than what I'm finding now for 6SN7 tubes. If I can get a little more drive out of the 6SN7, I might be able to go back to the 12AY7 and/or use the entire output transformer. I will recheck and update it this evening.
I've looked at using a differential input driving the 6SN7, but I was struggling to find a way to implement the GNFB with that topology. Do you have an example of how to implement that topology?
You're right, I need to look at the 6SN7 a little more. Somewhere along the line, I ended up with a 6SN7GT datasheet that has much lower specifications than what I'm finding now for 6SN7 tubes. If I can get a little more drive out of the 6SN7, I might be able to go back to the 12AY7 and/or use the entire output transformer. I will recheck and update it this evening.
I've looked at using a differential input driving the 6SN7, but I was struggling to find a way to implement the GNFB with that topology. Do you have an example of how to implement that topology?
It would look a lot like your second (6SN7) stage, with cathodes connected together and floating on a current source. Then you'd treat the two grids as if they were the inverting and non-inverting inputs of an "op-amp", which they are.
Long loop feedback to the inverting input would be by resistive voltage divider from the transformer secondary, as usual. One easy but important improvement possible is to lift signal ground from the divider network and the non-inverting input grid return "input" resistor from chassis ground by about 10 Ohms or so.
Reasons for this, and lots of other important but too often overlooked circuit elements, like output Zobels, etc. are covered in Bob Cordell's book. Very highly recommended for advancing DIYer's.
All good fortune,
Chris
You might also consider single triodes like 6AB4 (one half 12AT7) in 7-pin base, or triode coupled pentodes like 12BY7, D3a or 6CL6. Check the HK Citation schematic. I have similar goals in mind for a future amplifier and I’m following your design evolution with interest; hope to learn from the process.
Ok, that sent me back to the drawing board and simulator for a little while. I was finally able to figure out the differential input, which I had tried earlier on, but I was not able to figure out how to get the GNFB incorporated properly. I prefer to go this route for the lower distortion and I assume stability, as well.
The 10 Ohm resistance does seem important. With the proper setup, the CCS seems to do all of the work balancing the second tube of the phase splitter. It doesn't seem necessary to actually have it connected via divider to the input.
I also made an attempt at the zobel on the output. Paying more attention to the local feedback at the output stage, it appears to be better to still feed the 6SN7's from the HT rail, rather than solely from the anode or grid, so I reincorporated that.
Does this seem to be heading in the right direction?
The 10 Ohm resistance does seem important. With the proper setup, the CCS seems to do all of the work balancing the second tube of the phase splitter. It doesn't seem necessary to actually have it connected via divider to the input.
I also made an attempt at the zobel on the output. Paying more attention to the local feedback at the output stage, it appears to be better to still feed the 6SN7's from the HT rail, rather than solely from the anode or grid, so I reincorporated that.
Does this seem to be heading in the right direction?
Last edited: