Odd question but I'm hoping someone may have experience to guide my tuning.
I've built a set of EL34 monoblocks which I'm steadily refining, mainly tuning the feedback for best response and the input stages for optimum values per the loadlines. By the book I'm aiming for minimum distortion across the board. OK cool, makes sense... but it sucks some of the magic out of the soundstage depth and width.
Does it sound better, objectively? Yes. Does it sound better subjectively? Ehhh... I'd like to get a better balance between the frequency response and pleasant room-filling depth you get when it's close but not quite right.
Detail, tone, and extension in high and low is fairly consistent so I'm happy with that, now I'm just fine tuning the final character - which is mainly the imaging and holographic qualities.
So the question is: rather than chasing minimum distortion in all categories, which distortion should we "allow" or focus on? We all hear about the even harmonics being the "good" ones, so do I allow a few % of 2nd on my loadline calcs? 4th? Etc.
My first iteration was poorly optimized for the output load, had too much feedback, was borderline unstable, and was cathode biased- but had some impressive depth when listening. I'm trying to get it back to that depth but without the faults.
I've built a set of EL34 monoblocks which I'm steadily refining, mainly tuning the feedback for best response and the input stages for optimum values per the loadlines. By the book I'm aiming for minimum distortion across the board. OK cool, makes sense... but it sucks some of the magic out of the soundstage depth and width.
Does it sound better, objectively? Yes. Does it sound better subjectively? Ehhh... I'd like to get a better balance between the frequency response and pleasant room-filling depth you get when it's close but not quite right.
Detail, tone, and extension in high and low is fairly consistent so I'm happy with that, now I'm just fine tuning the final character - which is mainly the imaging and holographic qualities.
So the question is: rather than chasing minimum distortion in all categories, which distortion should we "allow" or focus on? We all hear about the even harmonics being the "good" ones, so do I allow a few % of 2nd on my loadline calcs? 4th? Etc.
My first iteration was poorly optimized for the output load, had too much feedback, was borderline unstable, and was cathode biased- but had some impressive depth when listening. I'm trying to get it back to that depth but without the faults.
A tube amplifier must be done as good as possible without negative feedback ... after that you can apply just the minimum necessary
And yes , minimum distortion is the goal , but not using negative feedback to cover design faults, then it will not suck the magic 🙂
And yes , minimum distortion is the goal , but not using negative feedback to cover design faults, then it will not suck the magic 🙂
Is this a push-pull or single-ended design? Can you share the schematic?
I am of the opinion that, yes, you can go too far in "perfecting" a design by the numbers and end up with an amp that isn't very gratifying to listen to. OTOH, you can also design a "flawed" amp that does one thing well but leaves a lot of other capabilities on the table.
Single-ended amps, in particular, are famous (or notorious, depending on your POV) for measuring poorly but sounding very pleasing in the right circumstances. I've heard a very well-designed 300B that I thought was dull-sounding, and very sloppy 300B amps that were musically very gratifying. I have always suspected that some well-known manufacturers, like Gordon Rankin and Dennis Had, design more by ear than by the numbers, choosing operating points and passive components that produce a musically engaging product.
I am of the opinion that, yes, you can go too far in "perfecting" a design by the numbers and end up with an amp that isn't very gratifying to listen to. OTOH, you can also design a "flawed" amp that does one thing well but leaves a lot of other capabilities on the table.
Single-ended amps, in particular, are famous (or notorious, depending on your POV) for measuring poorly but sounding very pleasing in the right circumstances. I've heard a very well-designed 300B that I thought was dull-sounding, and very sloppy 300B amps that were musically very gratifying. I have always suspected that some well-known manufacturers, like Gordon Rankin and Dennis Had, design more by ear than by the numbers, choosing operating points and passive components that produce a musically engaging product.
Last edited:
Agreed. Measurements are meaningless if it doesn't sound good subjectively. That said even if it sounded good Version 1.0 had a lot of flaws that needed addressing and could barely put out a square wave at any level.
Push pull quad. Here's my latest schematic.
My latest tweaks have been focused on the input tubes. Originally feedback was -32db which I lowered to -24db (that fixed the square wave problems and oscillation), and in the process I eliminated the cathode bypass on the voltage amp to reduce the gain (that alone made a surprising difference in detail retrieval, who knew). Otherwise I just followed the loadlines for the 6SJ7 and 6SL7 for most headroom and lowest distortion in each stage.
Push pull quad. Here's my latest schematic.
My latest tweaks have been focused on the input tubes. Originally feedback was -32db which I lowered to -24db (that fixed the square wave problems and oscillation), and in the process I eliminated the cathode bypass on the voltage amp to reduce the gain (that alone made a surprising difference in detail retrieval, who knew). Otherwise I just followed the loadlines for the 6SJ7 and 6SL7 for most headroom and lowest distortion in each stage.
Attachments
I hate to be a stinker, but every design I did with low distortion as a goal sounded best. I can only advise others to design for low distortion and make careful, accurate measurements…
Of course this goal was always tempered with using the least amount of negative feedback as possible.
Of course this goal was always tempered with using the least amount of negative feedback as possible.
24dB feedback is still a lot. The Eico HF-60, a similar circuit, only used 21dB. You might try a lower-gain driver like a 6SN7 and reduce the feedback further. Just a thought. Take a look at the Eico HF-60. EF86 input stage might give you a bit more of what you're looking for, sonically, and would work well with a 6SN7 driver.
Nothing "stinky" about going for lowest distortion. ;-) In this case, though, low distortion could be achieved with different tube choices that might produce the sonic signature the OP is looking for. The HF-60 is a beautiful amp and has a very "holographic" sound, in my memory.
What does "lowest distortion" mean? Under 1% under 0.01%? And at what power output & at what frequency?
I've designed and built SE ended amps, spent hours tweaking the design, reducing hum & N, applied NFB for optimum frequency response but found when listening to the amp on my speakers in my room that I preferred it without NFB and with less than optimum frequency response.
I try follow good practice as regards layout, grounding, using a hefty power supply more than capable of supplying the power needed for the amplifier as well as center biasing stages in their linear region & with more than enough voltage swing and current/ low impedance to drive the OP stage but don't worry too much about ataining the best THD figures.
Andy.
I've designed and built SE ended amps, spent hours tweaking the design, reducing hum & N, applied NFB for optimum frequency response but found when listening to the amp on my speakers in my room that I preferred it without NFB and with less than optimum frequency response.
I try follow good practice as regards layout, grounding, using a hefty power supply more than capable of supplying the power needed for the amplifier as well as center biasing stages in their linear region & with more than enough voltage swing and current/ low impedance to drive the OP stage but don't worry too much about ataining the best THD figures.
Andy.
Consensus seems to be circling back to less feedback. Which I would love to do but the open loop gain on this design is nuts - with the cathode bypass in place it was 58db.
If I were to rework to fit 6SN7 in the driver stage what changes would I need to perform? Do I just need to adjust the plate and cathode resistors and leave the rest alone?
If I were to rework to fit 6SN7 in the driver stage what changes would I need to perform? Do I just need to adjust the plate and cathode resistors and leave the rest alone?
Well, you need to design the gain to match the amount of feedback and sensitivity you want. ;-) Feedback is not intended as a cure for too much gain, obviously. Here's a link to the Eico HF-60 schematic:
https://www.thetubestore.com/lib/thetubestore/schematics/Eico/Eico-HF-50-HF-60-Schematic.pdf
A tried-and-true circuit. I don't mean to rob you of the fun of designing your own amp, but why reinvent the wheel? ;-) The multi-tap feedback network is specific to the Acro OPT, but otherwise it should work. Or just leave the 6SJ7 and use the the 6SN7 long-tailed pair. I don't know what your plate voltage is on the 6SJ7 so you'll probably have to adjust the LTP cathode resistor to get the bias right.
https://www.thetubestore.com/lib/thetubestore/schematics/Eico/Eico-HF-50-HF-60-Schematic.pdf
A tried-and-true circuit. I don't mean to rob you of the fun of designing your own amp, but why reinvent the wheel? ;-) The multi-tap feedback network is specific to the Acro OPT, but otherwise it should work. Or just leave the 6SJ7 and use the the 6SN7 long-tailed pair. I don't know what your plate voltage is on the 6SJ7 so you'll probably have to adjust the LTP cathode resistor to get the bias right.
The trick is to have sufficient gain in the input, phase splitter and driver stage. About 24dB more than the desired closed loop gain. Then apply local feedback in each gain stage to get about 12dB more gain than the desired closed loop gain. Then close the loop, and use some phase compensation to get optimum step response. At least I did so.
Member
Joined 2009
Paid Member
Could be worth some careful listening before modifying for lower open loop gain so you can compare with confidence - nothing like direct experience (as long as it's safe).
I carefully optimize with less or no NFB, tweaking to lowest hum and dist. Then applies NFB to get low dist < 0.1% or better if possible. I find no contradiction between musicality and low dist, in fact less is better. But below a fairly
low figure my ears don't bother anymore. Also importent to use a design that is as insensitive as possible
for tubes spread in parameters to keep the good sound after future tubes swaps.
low figure my ears don't bother anymore. Also importent to use a design that is as insensitive as possible
for tubes spread in parameters to keep the good sound after future tubes swaps.
Like everything else, it's all a compromise.
Perhaps the best compromise for designing a simple tube audio power amp (or integrated amp) is:
1. Decide what you want the sensitivity of the finished amp to be. I usually aim for 1Vrms in = max power out to speaker
2, Design for lowest THD and noise open loop (before applying any NFB).
3. Try to aim for open loop gain that will give you your target sensitivity closed loop (after applying NFB) in the end.
4. Try to aim for either only a little bit of NFB (5 to 8dB), or a huge amount of NFB (20dB or more if you can do it). Note that careful compensation will need to be applied if you go the high NFB route, to suppress oscillations, etc. Also, using a lot of NFB that includes the OPT inside the FB loop means that OPT has to be of very high quality. That was the idea behind the Williamson amplifier. It used up to 30dB of global NFB and required a special design that would behave well with that much NFB around it. UTC, Partridge, Acrosound, Peerless made OPTs for Williamson designs that were never cheap (and are now collector's items).
Let's say you want to build a PP EL34-UL amp with 10dB NFB (max power out = ca. 20W)
1. Design the amp to reach full power out (either clipping or 1% THD) with 0.33Vrms input, with the lowest possible THD into the load.
2.. That way, after you've applied 10dB of NFB (gain/3) the closed loop gain will give you max power at 1Vrms input with 1/3 the THD.
I've found that I prefer the sound of a DHT output stage with no negative feedback loop -- BUT that only ever works with extremely good (EXPENSIVE) output transformers, like original Tango, vintage UTC LS series, or equivalent. Perhaps Hashimoto is in that league nowadays. I don't know. I don't know if the top-of-the-line Edcor OPTs play in that league. I'm pretty sure most Hammond OPTs do not, although there are one or two models that are reputed to be quite good.
Bottom line is, just like they used to say in the 1950s, the final sound quality will be heavily dependent on your choice of OPT -- especially if you're going for an amp with no global NFB around it, and especially if you're going for an amp with >24dB of NFB around it.
From my experience and from listening to what others have done, one of the most foolproof designs is a PP UL EL84 10W/channel amp with about 10dB of NFB (or only 6dB if your speakers present a very friendly load). PP EL84-Triode will yield about 5W/channel and can sound very nice with 6dB gNFB, or if your OPTs are extra-special-good, perhaps no gNFB (perhaps).
Perhaps the best compromise for designing a simple tube audio power amp (or integrated amp) is:
1. Decide what you want the sensitivity of the finished amp to be. I usually aim for 1Vrms in = max power out to speaker
2, Design for lowest THD and noise open loop (before applying any NFB).
3. Try to aim for open loop gain that will give you your target sensitivity closed loop (after applying NFB) in the end.
4. Try to aim for either only a little bit of NFB (5 to 8dB), or a huge amount of NFB (20dB or more if you can do it). Note that careful compensation will need to be applied if you go the high NFB route, to suppress oscillations, etc. Also, using a lot of NFB that includes the OPT inside the FB loop means that OPT has to be of very high quality. That was the idea behind the Williamson amplifier. It used up to 30dB of global NFB and required a special design that would behave well with that much NFB around it. UTC, Partridge, Acrosound, Peerless made OPTs for Williamson designs that were never cheap (and are now collector's items).
Let's say you want to build a PP EL34-UL amp with 10dB NFB (max power out = ca. 20W)
1. Design the amp to reach full power out (either clipping or 1% THD) with 0.33Vrms input, with the lowest possible THD into the load.
2.. That way, after you've applied 10dB of NFB (gain/3) the closed loop gain will give you max power at 1Vrms input with 1/3 the THD.
I've found that I prefer the sound of a DHT output stage with no negative feedback loop -- BUT that only ever works with extremely good (EXPENSIVE) output transformers, like original Tango, vintage UTC LS series, or equivalent. Perhaps Hashimoto is in that league nowadays. I don't know. I don't know if the top-of-the-line Edcor OPTs play in that league. I'm pretty sure most Hammond OPTs do not, although there are one or two models that are reputed to be quite good.
Bottom line is, just like they used to say in the 1950s, the final sound quality will be heavily dependent on your choice of OPT -- especially if you're going for an amp with no global NFB around it, and especially if you're going for an amp with >24dB of NFB around it.
From my experience and from listening to what others have done, one of the most foolproof designs is a PP UL EL84 10W/channel amp with about 10dB of NFB (or only 6dB if your speakers present a very friendly load). PP EL84-Triode will yield about 5W/channel and can sound very nice with 6dB gNFB, or if your OPTs are extra-special-good, perhaps no gNFB (perhaps).
I figured out the solution. It was aided by y'all, indirectly.
As the original design used 30-odd db of feedback off the 8 ohm tap and I modified it to be -24 on the 4 ohm tap, I forgot a key thing: the value of the cap.
The original design used 1000pf and that was the correct value for -30db... but it was too much for -24, causing high frequency roll off above 12k.
The correct value for the new loop was 750pf, which reduced the roll off significantly to make it almost flat to 20k. Every pico counts in that part of the circuit, even +-50pf is enough to mess up response. 680pf was too little, 820 was too much.
The roll off was high enough to not be perceptible as a reduced treble, but it was enough to damp those high frequency cues that help soundstaging.
Now it's got that depth back I was chasing, woo!
As the original design used 30-odd db of feedback off the 8 ohm tap and I modified it to be -24 on the 4 ohm tap, I forgot a key thing: the value of the cap.
The original design used 1000pf and that was the correct value for -30db... but it was too much for -24, causing high frequency roll off above 12k.
The correct value for the new loop was 750pf, which reduced the roll off significantly to make it almost flat to 20k. Every pico counts in that part of the circuit, even +-50pf is enough to mess up response. 680pf was too little, 820 was too much.
The roll off was high enough to not be perceptible as a reduced treble, but it was enough to damp those high frequency cues that help soundstaging.
Now it's got that depth back I was chasing, woo!
Yes, excessivew rolloff in the highs would flatten the soundstage a bit. ;-) Glad you found the solution!
For what it is worth, when I did frequency sweeps with feedback disconnected I saw a slight boost in ultra high frequencies above the levels in the audible range... I strongly suspect that's why no feedback has that extra magic.
H
HAYK
Sydney Smith, the designer of Marantz tube amps, knew the secret. Look at 8b model spectrum. A perfect push pull must have only odd harmonics, to add even ones, the input tube must be exclusively distorting even ones to get a degrading harmonics. This why the input tube should be a "dirty" one. Model 8b uses IF detector 6BH6, model 9 uses 6DJ6 running at 80v P-K at low bias.
Or they just wanted to be cheap . EF86 was more expensive so they used common 6BH6 that is a good pentode anyway , not "dirty" .