6LR8 push pull amp build

[gadget73]

I'll get the schematic scanned tomorrow.

12dB of feedback now without changing anything. Handy, since that was my initial target anyway.

24 watts both channels driven
2.55 volts for full power out. Not entirely thrilled with this, but its a relatively low gain front end. The triodes have a mu of 40 and its a single gain stage. Only so much to be done given that constraint.

Response at 1 watt
20hz -0.74dB
20khz -0.58dB

Full power bandwidth -1dB points
18.5 hz
27 khz

that tracks extremely close to the 1 watt figures, so its just a basic response thing, not a transformer limitation. I could probably squeak a little more on the top end but at the expense of stability.

Speaking of that, best I could do is unloaded to 0.047uF. If I was willing to significantly impact the HF balance it probably could have been a bit better but THD performance up top would have suffered. These transformers really like to ring, and most of the tricks I know had counter-intuitive results. A step network off the voltage amp made it worse, as did adding a grid stopper to the input. About the only thing that worked for the ringing was a phase advance cap on the feedback resistor, and for stability it ended up with a Zoebel across the output.

THD figures, 1 watt
20hz 0.35%
100hz 0.11%
1khz 0.2%
10khz 0.3%
20khz 0.55%

nothing to brag about, but not awful.

Whats it sound like? Surprisingly clean and a pinch on the bright side if anything. On paper its "ooh, down a half DB above where any of us can hear, what crap" but this thing has all the top end sparkle I'm capable of getting. Absolutely nothing embarrassing in the bass department either. Despite the lower feedback level it doesn't have that slightly mushy warm sound that is often associated with that.

If I were going to re-do it, maybe I'd do a conventional voltage amp into a split load inverter, but unknown how much better that might be.

Something else that I did with this, no tie from B- to chassis ground at all, and the chassis is grounded using an IEC socket. I'm still debating this, and may add a pair of diodes with a bypass cap between B- and chassis to keep DC isolation but allow for HF trash to drain off.

 

[gadget73]

other questionably relevant data point
Power supply regulation from idle to full scream is just shy of a 10% drop on the plates, and just shy of 15% on the screens.

Bias supply dips almost 4%.

Some of that is probably the thermistor across the primary. I'm sure the bias supply isn't actually drawing any extra current so its just reflecting drop caused by the increased B+ load. Either way I'm not going to worry about it.

 

[gadget73]

10khz squares. It actually looks slightly better without the Zoebel but stability is crap.



Before any tuning, just into the dummy loads. Its quite bouncy.

 

[wparks]

Speaking of that, best I could do is unloaded to 0.047uF. If I was willing to significantly impact the HF balance it probably could have been a bit better but THD performance up top would have suffered. These transformers really like to ring, and most of the tricks I know had counter-intuitive results. A step network off the voltage amp made it worse, as did adding a grid stopper to the input. About the only thing that worked for the ringing was a phase advance cap on the feedback resistor, and for stability it ended up with a Zobel across the output.

That's pretty bizarre.
For the ringing in the transformer, isn't the phase advance cap about the only thing that will improve that? and if that cannot quash it, plan B is feedback from the plate? That would normally be the first thing to set, because it's value also impacts the recipe needed for stability right? Looks like you got it under control pretty well- A little bounce in the suspension, maybe needs a new pair of shocks, but I'd ride it!

I would not normally expect either a voltage amp shelf/step or a grid stopper to have any affect on transformer ringing and would only be applied after the phase advance cap is chosen that gives the flattest FR, but you may be saying though that the shelf and stopper had no effect on the stability- and that is curious. Zobels are ugly, but sometimes you don't have a choice- and they can be a great ace in the hole for a problem child.

Stability to 0.047uF is nothing to sneeze at! Can't wait to see your values.

An excellent build with tremendous results!

 

[gadget73]

The phase cap across the feedback resistor sorted it very nicely, but sometimes adding a fair size grid stop to the input will form a low pass and knock down what goes in. Not quite fixing it, but preventing some of those high frequencies from getting in at all will make it have less overshoot and ringing. I tried a 10k and it got worse, which I don't understand. I got similar results from a 39pF through a resistor to ground off the voltage amp plate. I also tried adding grid stops between the coupling caps and pentode grids, also made it worse. So the whole thing is built in a textbook "do not do this" sort of manner, but thats what it wanted to behave.

If I lose the Zoebel the remaining bounce goes away and the squares look excellent but it oscillates unloaded. Wasn't keen on that.

 

[wparks]

One "should have been obvious" thing that bit me in the butt and I mentioned in the follow-up in that conversation we had with Dave- (don't know if you/he read the follow-up or not, he did not respond) is when I insert a restrictive grid stopper it will actually blunt the corner of the input square wave, so it will tend to make the 10kHz square wave response look better than it actually is- but the resonant "bell" that rings in the transformer simply is not being struck sharply enough with the blunted/rounded square wave to show you the actual response. If you install a restrictive grid stopper first, then tune the phase advance cap for a nice square wave result, then remove the grid stopper and you will be in for quite a surprise. I was. The feedback input into the cathode is not frequency restricted by the grid stopper. I try to follow Dave's advice, and even if the amp is unstable unloaded the phase advance cap is set first, then I follow with grid stopper, shelf, and output zobel in that order/preference to obtain stability. I'm not quite to the point where I follow Dave's advice and characterize distortion at the different frequencies- I'm happy enough if I manage to get a good square wave, a flat passband to 20kHz, AND it's stable to 0.02 to 0.03uF. But then my old man ears would never be able to hear the difference between several percentage points of THD.

 

[wdecho]

Excellent and interesting thread.

 

[gadget73]

Schematics in a hopefully legible format. I'll see if I can figure out how to get rid of the grid lines, that would improve the legibility somewhat.

Power supply schematic should be considered not 100% complete. I'm not changing the design, but I forgot to draw in the pair of 100 ohm from heater supply to ground, and there is no data on the chokes. Have to flip the amp around so I can read the numbers and find out what they are again. I know I had this info at one point but can't find it.

I will say the power transformer being very under-stressed makes it run very cool. I had it going for a good 4 hours last night and it was barely warm.

 

[gadget73]

less grid lines, finally found a use for this Copilot thing that keeps popping up on my work computer. Not sure if this is more readable or not.

 

[Tom Bavis]

Problem was that the feedback delivered about 15 mA (about 15V output, through 1K) at full output. This passed through the LED as well as the feedback resistor. But the input stage and phase inverter only draw about 6 mA, so both were actually going into saturation on negative peaks as cathode went to zero or negative - LED stopped conducting. Bypassing the LED would bypass this AC current, it averages to zero DC - bypassed resistor works as well.

 

[wparks]

I'll see if I can figure out how to get rid of the grid lines, that would improve the legibility somewhat.

I draw all of my schematics on blue grid paper as well, in light pencil. The best thing you can do is to retrace all of your pencil lines in dark ink. Black gel pen works good for this, but not every one likes to draw over pencil, so it's a bit trial and error. I will then erase the pencil lines, using a Pentel "Clic Eraser" that you can get in stock at Walmart. Those erasers are amazing at completely removing the pencil without damaging the ink. Then, when you scan it, either the grid lines are innocuous, or, in my free paint.net program, I do an adjustment to crank the contrast up and brightness down until the grid lines just completely disappear because they are so much lighter than the ink. (I add all the labels there too.) I know, it's a PITA, but the results are beautiful.

 

[Lavane]

Transmitter tubes are another intresting tube to use.
Knockbill built a PP amp using transmitter tubes for output that have two sections inside each tube. A PPP would be intresring to build using this type of tube.
Here is a picture of John's amp below.

 

[gadget73]

Somewhere I have a rough sketch of an amp design to use 815 tubes. Thats a pair of 2E26 in a single bottle with a pair of top caps. It was basically a Williamson with a cathode follower stage to drive the tubes into grid current. Amp was drawn up with no values on anything, and no power supply. Thats about as far as I got though.

I also scribbled up a front end thats a long tail pair inverter at the input with a voltage amp stage after it, feedback coming into the second grid. I like the idea but didn't spend any time to work out the details of how the grid can be at 0 volts with an LTP since I'd prefer to not have to deal with a coupling cap in the feedback path. I think the answer is a CCS tied to a negative supply rail, but thats another thing I have yet to experiment with.

then there was the other idea I had using a pentode up front and bringing feedback in to the screen grid.

too much time to think, not enough motivation to build

Problem was that the feedback delivered about 15 mA (about 15V output, through 1K) at full output. This passed through the LED as well as the feedback resistor. But the input stage and phase inverter only draw about 6 mA, so both were actually going into saturation on negative peaks as cathode went to zero or negative - LED stopped conducting. Bypassing the LED would bypass this AC current, it averages to zero DC - bypassed resistor works as well.

makes sense, driving the LED off at an audio frequency on the peaks. If I'd thought to try it at a lower frequency I suppose I'd have seen it blinking. Either way I appreciate it, it wasn't something I'd have thought of.

when I insert a restrictive grid stopper it will actually blunt the corner of the input square wave, so it will tend to make the 10kHz square wave response look better than it actually is-

yeah, thats the low pass action of the series resistance + Miller effect of the tube. I was trying to do that on purpose, and have had success using it on other things to knock down the amount of overshoot by filtering out some of the upper harmonics before they even get into the amp. Its not really a fix, but if it ends up at the same place I'm sort of OK with it. In this case it did the opposite of what I'd expect though and I don't understand that.

 

[gadget73]

going to retract my thinking that this is bright, that seems to be the shop DAC. Running it in my main rig and it doesn't have that.

 

[Lavane]

Maybe more of a SS sound? My SE amp sounds more like a SS amp according to a couple people that have heard it. I'm not sure I agree with that. It sounds good to me either way. It's not bright or harsh. The bass is decent and not loose or flabby.

 

[gadget73]

Not quite sure I'd say it sounds solid state, but it lost that slightly harsh edge that it had out in the shop. Different speakers and everything else too though so thats part of it I'm sure.

 

[gadget73]

after a side trek investigating solid state plate loads and the frustrations of buying the wrong insulator bushings:

Thread 'experience with MOSFET plate loads?'

Oct 21, 2026

Just curious if anyone has subbed a MOSFET CCS for a plate resistor on a small signal triode, and how it performed.

Also if this were say a floating paraphase inverter, would the inverter also get a CCS, or can / should that remain a plate resistor?

• gadget73
• Replies: 42
• Forum: Tube Audio

I have results

I'll post a schematic update at some point, but HF response is insignificantly better, 20khz is -0.35 dB now vs -0.58dB previously. That may be slight difference in tuning though.

THD below 3khz is slightly better, above its a bit worse. 1khz previously was 0.2%, now its 0.09%. 20khz previously was 0.55%, now its 0.86%. Again possibly slight tuning difference.

Stability and power bandwidth are unaffected. Square waves also look the same, or near enough.

Circuit mods as follows:
replace four 47K plate resistors with IXYS 10M45S using a 1.2K resistor to set current. It comes out about 2.6 mA.
feedback resistor goes from 1K to 1.5K in series with 390 ohms (1.9K total) with an 820pF phase advance cap.

Electronically its a pretty small change, mounting the CCS was the hard part.

The AC balance control now sits about 10:00 where it was about 1:00 previously. I could change one of the resistors to center the pot if I felt the need, but I really don't.

what I really wanted to achieve though, input sensitivity is now 1.35 volts RMS for full power output. Thats a perfectly acceptable value.

Subjectively it doesn't sound much different, which is also fine.

 

[Lavane]

What are the distortion specs at 1 watt and at full output?

 

[gadget73]

Notes out in the shop with exact numbers but well under 1% anywhere. Its worst at 20khz and I want to say thats like 0.5%? I didn't check at full power.

I noticed that the idle bias spikes at power-off. Probably a result of those 5K pots I'm using which makes the bias supply drain down too fast. I'd prefer shorter stems anyway, so I think I'll find some higher value short stem pots to go in there and re-work the supply to get similar adjustment range with less current flow. If I can keep the grid supply from collapsing before the screen supply does it will be good for the tubes long-term.

 

[gadget73]

Ordered some 15K pots to replace the 5K currently in there, when I get those in I'll work up the THD figures at various power levels. 15K pot should cut the bias supply current flow to 1/3 if I keep the same adjustment range, hopefully that will be enough to let the bias supply not discharge before the tubes stop conducting. If thats not quite enough, then I suppose strapping an extra filter cap in there is an option too.


in the meantime enjoy a modified schematic that the google AI box was kind enough to remove the grid lines from after I messed with the contrast some. One of these days I'll practice with that Doric lettering set I have and start making schematics with nice neat writing, in the meantime you get to learn why my kindergarten teacher told my mom to teach me to type 40 years ago.

I will admit to cheating a bit here, I don't have 1.9K resistors, but I do have 1.5K and 390 ohm. Close enough.