If "lifeless" means "output closely resembling the input," then high feedback in a tube amp will make it "lifeless."
How about this definition of lifeless:
Vocals fade into a bland background to the extent that lyrics become noticably harder to understand. (There's even a website: Kiss This Guy dedicated to misunderstood lyrics, and it would seem people are reporting the same things here.)
Whole passages go unheard even if you know they're there.
Highs seem to disappear.
No soundstage whatsoever.
Others have compared it to trying to listen with a heavy blanket thrown over the speakers. It's a good way of describing it. It's what I heard with 20db of gNFB applied to a hollow state design.
I have found no explanation for this, especially since it doesn't apply to BJT based designs. There, the more NFB, the better. For hollow state, not so.
Stability doesn't seem to be the problem, since the 807 based design was stable with 20db of gNFB, and it's stable with 7.0db gNFB. It just sounds much better with the latter level of gNFB than it did with the former.
I have found no explanation for this, and I'm sure it's not just me since I've seen the subject come up here before. It's one of the reasons I decided to include variable gNFB into my designs: let the listener decide how much to add (or not).
Hi Miles;
It can be explained as dynamic distortions that are created when say driver hits increased output tube's grid current through coupling cap, and trying to compensate it by feedback it even increases that nasty peaks. Also, a mediocre sagging power supply can be a reason of such "lifeless" sound, the same dynamic processes, especially when screen grids voltages are not stable. I once got "lifeless" sound with smeared soundstage trying to save on film capacitors that shunted 800V b+. As I understood, it was modulation of reactance of electrolytics by currents consumed by class AB amps. If the amp is designed properly, deep feedback never causes such "lifeless" effects, especially when it is nested, i.e. multiple loops in addition to the global one.
It can be explained as dynamic distortions that are created when say driver hits increased output tube's grid current through coupling cap, and trying to compensate it by feedback it even increases that nasty peaks. Also, a mediocre sagging power supply can be a reason of such "lifeless" sound, the same dynamic processes, especially when screen grids voltages are not stable. I once got "lifeless" sound with smeared soundstage trying to save on film capacitors that shunted 800V b+. As I understood, it was modulation of reactance of electrolytics by currents consumed by class AB amps. If the amp is designed properly, deep feedback never causes such "lifeless" effects, especially when it is nested, i.e. multiple loops in addition to the global one.
I don't hear that in any of my amps, sorry. But (to Wavebourn's point), I use low impedance drivers and solid, well-regulated power supplies.
I have a tube linestage on the bench with nearly 50dB of feedback- output is indistinguishable from the input, at least on my (pretty good) system and to my (reasonably good) ears..
I have a tube linestage on the bench with nearly 50dB of feedback- output is indistinguishable from the input, at least on my (pretty good) system and to my (reasonably good) ears..
Low amounts of feedback (6-10dB) are not a good way to reduce distortion, as they merely mix it up and create more high-order stuff. They are a good way of reducing gain and output impedance in an amp which has sufficiently low distortion that you could run it open-loop. Note that local feedback has exactly the same order-multiplying effect as global feedback, so you need either none or enough.
Thanks for the great stuff everybody.
Considering NFB sounded like some kind of Black Magic to me, I want to see and measure the effects. I use a scope and a HP distortion analyzer, both garage sale quality. I also use the spectrum analyzer function of the sound card based program TureRTA, and it's sweep function, a PC program (RoMac) that just returns the values of the harmonics.
I question the accuracy of all these tools. I definitely question my ability to correctly interpret the data. But, I do feel that I can at least do A-B comparisons. The cool part is you can hear what you see. But I also question the experience of my ears.
At the end of the day after using these tools my amps sound better. At least to me.
Regarding the cap(s) values used in the NFB network. A square wave will tell if you have reduced or eliminated ringing. The RTA sweep function will let me see the frequency response impact of the cap values.
It will also show the FR impact of decoupling caps but that's a topic I want to explore in a dedicated post.
Please keep in mind I'm a newbie and weight this info accordingly.
Considering NFB sounded like some kind of Black Magic to me, I want to see and measure the effects. I use a scope and a HP distortion analyzer, both garage sale quality. I also use the spectrum analyzer function of the sound card based program TureRTA, and it's sweep function, a PC program (RoMac) that just returns the values of the harmonics.
I question the accuracy of all these tools. I definitely question my ability to correctly interpret the data. But, I do feel that I can at least do A-B comparisons. The cool part is you can hear what you see. But I also question the experience of my ears.
At the end of the day after using these tools my amps sound better. At least to me.
Regarding the cap(s) values used in the NFB network. A square wave will tell if you have reduced or eliminated ringing. The RTA sweep function will let me see the frequency response impact of the cap values.
It will also show the FR impact of decoupling caps but that's a topic I want to explore in a dedicated post.
Please keep in mind I'm a newbie and weight this info accordingly.
As always, thanks for the responses.
These amps are not really Dynacos. I started with the Dynaco mark II circuit. Using a pair of vintage Motorola 300ma power transformers. A GZ34 rectifier instead of the 5U4GB which feeds into a JJ 40-20-20-20 cap to a 300ma choke. Power is then routed through an NTC thermistor to another 40-20-20-20 cap. I copied Joe's 6922 cascode 6922 cathodyne phase splitter front end. Tubes are EH6550's and JJ GZ34 and 6922's.
The output transformers are off the rack Hammond 1650n's. They have a strange way of wiring the secondary to make you chose output impedance. If you use a single speaker ground in the center you can wire it so you have 8 and 4 ohm outputs, but not a 16. I put the negative in the center and spaced them so you could use banana plugs. Joe mentions that you do not have to use feedback with his design, although you may if you want to.
Out of the cathodyne at 1k I have a perfect square wave. Although I have twice the AC voltage out of the positive side, it seems to work.
I put the amps between my SP3A and Snell type a's with my Rega planet as a source. With this setup they sound wonderful! I've had a few friends listen to them and they were really impressed. These babys are done.
Next up, A 7591 amp based around a pair of Acrosound TO300's and I will finish my 8b copy using a pair of Dynaco A470's.
I want to thank everyone that helped with advice on these amps. Especailly SY and Eli!
Kevin
These amps are not really Dynacos. I started with the Dynaco mark II circuit. Using a pair of vintage Motorola 300ma power transformers. A GZ34 rectifier instead of the 5U4GB which feeds into a JJ 40-20-20-20 cap to a 300ma choke. Power is then routed through an NTC thermistor to another 40-20-20-20 cap. I copied Joe's 6922 cascode 6922 cathodyne phase splitter front end. Tubes are EH6550's and JJ GZ34 and 6922's.
The output transformers are off the rack Hammond 1650n's. They have a strange way of wiring the secondary to make you chose output impedance. If you use a single speaker ground in the center you can wire it so you have 8 and 4 ohm outputs, but not a 16. I put the negative in the center and spaced them so you could use banana plugs. Joe mentions that you do not have to use feedback with his design, although you may if you want to.
Out of the cathodyne at 1k I have a perfect square wave. Although I have twice the AC voltage out of the positive side, it seems to work.
I put the amps between my SP3A and Snell type a's with my Rega planet as a source. With this setup they sound wonderful! I've had a few friends listen to them and they were really impressed. These babys are done.
Next up, A 7591 amp based around a pair of Acrosound TO300's and I will finish my 8b copy using a pair of Dynaco A470's.
I want to thank everyone that helped with advice on these amps. Especailly SY and Eli!
Kevin
A little help clearing up conflicting data!
Several month ago I started a post about understanding square waves. I was working with a 6BQ5. The subject, of coarse, turned to NFB. In this conversation df96 said for pentodes unless you get to over -20dB you are better off with none. The reason "At -6db you get lots of high order products arising from the feed back, but little suppression of them. Worst of all worlds"
This is supported in the Mullard "Tube Circuits for Audio Amplifiers". In which they state for pentodes you need to apply -21dB NFB.
Both of these statements ,df96 and Mullard, sound absolute. In this post the NFB value is discussed as being arbitrary.
HELP!!!!!!!
Several month ago I started a post about understanding square waves. I was working with a 6BQ5. The subject, of coarse, turned to NFB. In this conversation df96 said for pentodes unless you get to over -20dB you are better off with none. The reason "At -6db you get lots of high order products arising from the feed back, but little suppression of them. Worst of all worlds"
This is supported in the Mullard "Tube Circuits for Audio Amplifiers". In which they state for pentodes you need to apply -21dB NFB.
Both of these statements ,df96 and Mullard, sound absolute. In this post the NFB value is discussed as being arbitrary.
HELP!!!!!!!
It is arbitrary until you specify what the performance has to be. For example, if you use a straight pentode, no feedback around it, the source impedance will be high (not good for most speakers). The bass distortion, ditto. On the other hand, if you apply low levels of feedback, you MAY have less overall distortion, but higher order (perhaps more audible). But all of this depends on the specifics of the circuit; the numbers thrown around tend to refer to simplified models.
>20dB is a guideline number and certainly a safe bet for the majority of pentode designs. The creativity comes in figuring out the best overall way to achieve that.
>20dB is a guideline number and certainly a safe bet for the majority of pentode designs. The creativity comes in figuring out the best overall way to achieve that.
Thanks Stuart. As always your input is helpful. By the way, I didn't see that df96 already contributed to this post.
There is another aspect of NFB that I'm dying to explore. Perhaps you can help with this one.
When I put -16dB to -21dB of NFB on, a 25uf decoupling cap on the dirvers and 50uf on the outputs opens up the warm tones (100-700hz or so) but overly pronouces the bass. If I leave them off the warm tones are subdued but the bass is in balance. Playing with cap values I found 3.3uf on the drivers and 16uf on the outputs offers the the best balance. This is supported by a nice flat FR sweep.
The questions are,
Is it a mistake to use cap values this low?
Do I stick to the science (typical decoupling cap value equations) and fix something else?
Do I go with what sounds, and sweeps the best?
Thanks again for all your help.
Tom
There is another aspect of NFB that I'm dying to explore. Perhaps you can help with this one.
When I put -16dB to -21dB of NFB on, a 25uf decoupling cap on the dirvers and 50uf on the outputs opens up the warm tones (100-700hz or so) but overly pronouces the bass. If I leave them off the warm tones are subdued but the bass is in balance. Playing with cap values I found 3.3uf on the drivers and 16uf on the outputs offers the the best balance. This is supported by a nice flat FR sweep.
The questions are,
Is it a mistake to use cap values this low?
Do I stick to the science (typical decoupling cap value equations) and fix something else?
Do I go with what sounds, and sweeps the best?
Thanks again for all your help.
Tom
Interesting comments by Wavebourn in post #22. The reasons for conflicting opinions on high NFdbk perhaps revealed. Good advice in any case. Seeing as the screen voltage needs to be rock stable, I can't help but wonder what happens with an UL amp where the screen is rather unregulated and susceptible to phase shift.
20dB of feedback is not a cliff (e.g. 19dB horrible, 21dB excellent) but just a guideline. It all depends on what you are trying to do and how much open-loop distortion you have.
Once you have significant amounts of feedback you have to think about stability at both HF and LF ends of the audio spectrum. Changing decoupling caps will affect LF stability, and may cause LF or subsonic peaks without necessarily creating actual oscillation (motorboating). An LF peak can be seen in the Mullard frequency response plot for their 5-20. You can model this in Spice. Remember that the "science" includes both open-loop LF roll-off and closed-loop stability criteria - people sometimes forget the latter.
Circuit nodes which can act as both outputs and inputs (e.g. pentode screen grids) need good decoupling, otherwise you can get 3rd-order intermods created from 2nd-order nonlinearity. They won't show up on a 19+20kHz test; you would need 19.995+20.00kHz to expose this (or real music) so the difference frequency is low enough to see the lack of decoupling..
Once you have significant amounts of feedback you have to think about stability at both HF and LF ends of the audio spectrum. Changing decoupling caps will affect LF stability, and may cause LF or subsonic peaks without necessarily creating actual oscillation (motorboating). An LF peak can be seen in the Mullard frequency response plot for their 5-20. You can model this in Spice. Remember that the "science" includes both open-loop LF roll-off and closed-loop stability criteria - people sometimes forget the latter.
Circuit nodes which can act as both outputs and inputs (e.g. pentode screen grids) need good decoupling, otherwise you can get 3rd-order intermods created from 2nd-order nonlinearity. They won't show up on a 19+20kHz test; you would need 19.995+20.00kHz to expose this (or real music) so the difference frequency is low enough to see the lack of decoupling..
Are you building this for yourself or for someone else? If the former, whatever sounds best to you is the best (a tired cliche, but nonetheless true). Beyond that, in my amps I want the measurements to be good (as well as the sonics), so I will believe the results of things like frequency sweeps before I'll believe theory.
I suspect the output transformer has a great deal to do with it, I use lots of feedback between my driver and output tube of my SE and it is full of life with an engaging sound. Also look at SY's post above.
Transistor amps don't have transformers either
I just think that including a band-limited and swing-limited device in a heavy feedback loop is not going to be a good thing for the sound.
As stated there is no "best" amount of NFB. I have recently decided I want a switch on the amp. For listening at low power levels I can use small amounts of NFB becase the amp does not distort at leves under say 1W. At higher power I find I want nore feedback.
So I think the answer depends on the volume setting.
It is also reasonable to have the loop around only the power section of the amp.
So may plan is to add a switch to short out part of the resistance in the NFB loop.
So I think the answer depends on the volume setting.
It is also reasonable to have the loop around only the power section of the amp.
So may plan is to add a switch to short out part of the resistance in the NFB loop.
First post, couldn't resist..
My hypothesis: In a conventional class AB tube amp, applying GNFB around it is going to make cutoff more abrupt because the feedback loop attempts to compensate the falling gain by increasing the magnitude of the AC driving voltage for both tubes (thanks to the phase inverter) which of course drives the tube about to turn off further into cutoff (while driving the other tube further away from cutoff, of course). This creates more higher order distortion components that did not exist before feedback was applied. Frequency compensation within the audio band makes this worse as there is less open loop gain at higher frequencies to remove these higher order products.
This does not happen in a complementary (Darlington) BJT emitter follower output stage because as the feedback loop attempts to compensate falling gain by increasing drive, it actually "fights" cutoff since the larger drive signal tries to increase the current through the device that is attempting to cut off. This makes the emitter waveform smoother.
Just my 2 cents..
My hypothesis: In a conventional class AB tube amp, applying GNFB around it is going to make cutoff more abrupt because the feedback loop attempts to compensate the falling gain by increasing the magnitude of the AC driving voltage for both tubes (thanks to the phase inverter) which of course drives the tube about to turn off further into cutoff (while driving the other tube further away from cutoff, of course). This creates more higher order distortion components that did not exist before feedback was applied. Frequency compensation within the audio band makes this worse as there is less open loop gain at higher frequencies to remove these higher order products.
This does not happen in a complementary (Darlington) BJT emitter follower output stage because as the feedback loop attempts to compensate falling gain by increasing drive, it actually "fights" cutoff since the larger drive signal tries to increase the current through the device that is attempting to cut off. This makes the emitter waveform smoother.
Just my 2 cents..
But it's negative feedback, so it can only ever reduce the gain, unless the phase lag turns it back into positive feedback
I visualise it like a rope that you are waggling at one end to try and get a nice motion at the other end. Adding an output transformer is like adding a weight in the middle of the rope..
Exactly the same happens in both cases. Driver as the result makes more "abrupt" signals in order to make less "abrupt" outputs from output tubes or transistors.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.