 
|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
|
|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
In Reply to: Re: transient test - Part II posted by john curl on July 30, 2003 at 22:58:33:
Hi JC,>leads to adding compensation, and thereby slowing the gain bandwidth
>of the total circuit below that of a simple follower.This actually makes sense to me. Considered as part of a global
feedback loop, in that context, closing the loop, gets you the
linearity.I ran some frequency sweeps for various followers
Circuit MidBand Gain -3dB Frequency Simple .796764 .563312 2.1574Mhz Sziklai uncomp potentially unstable potentially unstable potentially unstable Sziklai comp (3900pF) .991296 .700846 569.105Khz Darlington .966362 .683218 1.2323Mhz Hawksford uncomp .951225 .672516 3.8531Mhz
Hawksform comp (190pF) .951225 .672516 3.0639Mhz It looks like the Hawksford would be very interesting to
investigate as it has a wider bandwidth then a simple follower
and may be more linear as well. Even its uncompensated form shows
only a slight hump rising from .951225v to 1.0077v at 1.875Mhz.
Follow Ups:
For another follower twist how about measuring the difference between follower output and input voltage and subtracting it from the input. This could be done quite easily by multiplying the measured difference voltage with a transconductance and simply add that gm to the gm of the preceeding voltage gain stage. I tried this in simulation and it worked nicely, for example at a given output power the third harmonic distortion at the follower output is -60dBc (0.1%). At the follower input the third harmonic is -80dBc (0.01%). Add the differencer/subtractor and the situation reverses - the 3HD at the follower output is now -80dBc and at the input it's -60dBc.This was discussed somewhat over at diyaudio.com (see link) sparked by patents held by Bruce Candy of Halcro. This technique probably predates Halcro but it's likely they use it. One intriguing possibility is that it should be possible to implement such a scheme so that it can be easily disabled and you could compare the sonic qualities of with vs. without.
John & Charles - you must have seen this, if you haven't tried it yourself. What do you think?
Regards
13DoW
- http://www.diyaudio.com/forums/showthread.php?s=&threadid=6362&highlight=halcro (Open in New Window)
I'm not a big fan of the "error correction" schemes. Basically it is just feedback with gain in the feedback loop itself. Since I don't like feedback in the first place, I'm not particularly interested in adding more feedback.The first "error correction" amplifier I heard of was by Quad in their (in)famous "Current-Dumping" amplifier from the early '70s. They sure used a lot of obfuscation to hide what they were really doing. Next up was the "Trans-Nova-Twin" designs by Jim Strickland of Acoustat, with a similar degree of obfuscation. I think both of these pre-dated Hawksford's article where this approach was clarified and put into formulae. Then came Cordell's AES paper, which was basically a construction project based on Hawksford's paper. I seem to recall a few garage-type companies selling amps with "error correction" output stages. But the first one to achieve any success at all was the Halcro. Of course, he lied about how it worked (denying that it used error correction), which he got away with until the patents became public.
When I was designing the Theta Dreadnaught, I simulated some of these types of circuits. However, I didn't see any gains worth pursuing. But then again, I'm not fixated on super-low values of steady-state harmonic distortion.
Best regards,
Charles Hansen
Charles,I don't know if what I described qualifies as NFB but that's probably a semantic dead-end. What I'm thinking of has no forward gain and there's no virtual earth but, anyway, I was curious if you'd tried it. On a related topic, I've been toying with a power amp design and set myself a goal to make it as simple and linear as possible without any overall NFB. I'm going for a diff pair voltage gain stage and balanced follower outputs - hence my interest in the follower related threads. From what I've read about your amps they are simple (not many stages) and have no overall NFB and are balanced.
From a circuit design point of view one limitation I see is the finite output resistance of the diff pair stage driving the output power FETs - the non-linear input capacitance sets the distortion limit. An obvious thing to do is to add a second set of followers to buffer the voltage gain stage from the output stage and that improves things as expected. However, I recall a review of one of your amps where bandwidth was discussed with comments from you that the limitation is in driving the input capacitance of the output power FET followers. You tried adding a buffer which improved the bandwidth as expected but the sonic performance was much degraded. As the bandwidth without a buffer was adequate, if not stellar, you left it alone for superior sonics. I'd be interested any comments or advice on driving followers simply.
Can you really 'design' it well or do you just have to build & tweak ..etc.TIA
13DoW
Hello 13DoW,If I followed you correctly, you were suggesting a variation on Hawksford's error correction. Assuming the output stage has unity gain, you use a diff amp to compare the input and output of the output stage. The difference signal is amplified and fed back to the input to provide error correction. On the semantic front, I think the error correction approach absolutely should be classified as "feedback". A signal is literally "fed back" to an earlier stage to try and correct non-linearities. I don't have the paper in front of me, but Hawksford said something to the effect of "this technique allows greater feedback levels while maintaining stability, simply because the loop is shorter".
What I've tried to do with my own circuits is what Einstein suggested, "Everything should be made as simple as possible, but not simpler."
I start with an input voltage-gain stage. To avoid a coupling cap, we have to "turn the signal around" and get the DC level back towards zero. I've done this with both folded cascodes and also current mirrors. Then we need a buffer to drive the low impedance load. For buffers, I've used vertical FETs, lateral FETs, lateral FETs driving bipolars, and triple emitter followers. At this point I would strongly suggest avoiding vertical FETs (such as the Americans make). Their non-linear input capacitance really screws things up. The lateral FETs aren't easy to use. You have to parallel at least 4 of them to get any reasonable transconductance. Then you have matching problems. That leaves more complex schemes as the only viable option. Driving emitter followers with source followers is nice, but you have to carefully watch the thermal issues. A triple emitter follower is probably the most practical. The double emitter follower used by most designers (e.g., Douglas Self) doesn't have enough current gain to avoid loading the voltage gain stage. I'm not sure why Self hasn't yet figured that out, as he is generally thorough (if somewhat biased).
Next, I balance the whole thing as there are many advantages to this approach. Finally, I make everything complementary, as there are also many advantages to be gained this way.
Good luck and have fun,
Charles Hansen
Charles,
thankyou for the answer. I have not read the AES papers so I don't know who claims what. As an aside I met Prof. Hawksford once, the name sounded like someone in tweed with, perhaps, half-frame glasses. He is nothing like that.I like the idea of as simple as possible. I plan for output source followers with class A current source pull-downs, there is no cross-over between devices so no cross-over distortion and balanced outputs will cancel any asymmetry. The diff pair is made linear by heavy degeneration to give a desired linearity for the maximum allowed input signal and the reduction in gm is offset by using large resistive loads (ie the saturated output swing is 90V pkdiff but only a maximum of 30V pkdiff). The common-mode requirements of the two stages are totally different but can be accommodated by having two independent power supplies offset so that the output common-mode of the voltage gain stage is the optimum input common-mode voltage for the follower. It is very simple, though it requires full class A and two power supplies but that's OK for a one-off DIY effort and I can easily save the equivalent of the few hundred watts of dissipation by following the family around the house and turning the lights off after them.
Because I've optimized for the output common mode (which doesn't have to be 0V as it's balanced) the input common-mode is nowhere near 0V so input ac coupling is required. Ah, but maybe a folded cascode might fix that .. I'll think on, but not for too long or this will never make it from the simulator to the workshop.Cheers,
13DoW
You guys should write a book first :)I notice in D. Self's book that he has a triple EF output stage
in Figs 5-13, 5-25 page 122,136. Then later, Fig. 6-16 page 176
he goes back to a conventional double emitter follower output
stage.No reasoning given as at first he talks up the triple and then
doesn't "follow" :) thru with it.
These EF triples can be difficult to compensate. There is some discussion of it in:http://www.tagmclarenaudio.com/dev/white/wp2.asp#refs
it only took me a couple of months to figure out, the first time around. Didn't even need an output inductor. Besides, what else are you going to do with your time -- watch TV?Cheers,
Charles Hansen
the "Hawksford circuit"?Thanks,
Charles Hansen
I'm happy to see some diffusion of this circuit techniqueas a follower the input impedance can be very high while canceling most of the output Q hfe variation ( ccs bias input Q )
where it really shines is in a VAS stage with moderate emitter degeneration, the darlington like input impedance is combined with cascode like output swing linearity
Reading the thread, it seems you know a lot about MOSFETs. Is that part of your job?Best regards,
Charles Hansen
as an analog engineer I've followed audio electronics design for 20+ yrs, so far nobody’s paying me to design the stuff thoughI've designed PWM DC motor controllers, and programmable current drivers for solenoid valves using mosfets, collecting info and models along the way with audio amps in mind, but most of my work has been with low level signals
now moving from signal conditioning/acquisition to motion control/DSP programming - there's a lot more to control theory, esp for nonlinear systems in the last 10-20 yrs than undergrad EEs ever see – which means I get to learn a lot more; multiple domain modeling, coordinated multi-axis force/motion control, nonlinear stiffness, geometric constraints, sensor noise all combine to make feedback amplifier design look simple in comparison
what I’ve seen so far makes it obvious that low distortion analog electronics design isn’t really where audio can be moved forward, the entire 50+ yr old component audio paradigm should be scrapped with loudspeaker/DSP/nonlinear control and room correction integrated from the ground up, baby steps have been made but DSP horsepower and nonlinear control theory have really moved forward in the last decade
(this doesn't mean I'm going to stop thinking about low distortion analog design anytime soon now, I just think it is a sideshow to real advances in audio reproduction)
Here it is. V4 is just there to eliminate any need for
a coupling capacitor which removes the need to move the
simulation window way out past the point where this cap
would DC stabilize.Note that the split DC supply lets you move the top of
R5 to pretty much 0 VDC anyway.
Do you have a reference to Hawksford's original article?Basically it's just an emitter follower driving an emitter follower. The only twist is that the supply voltage on the rail of the first emitter follower is modulated by the audio signal. This only affects the operation of the circuit via the Early effect, whereby the collector current is very slightly dependent upon the collector-emitter voltage.
In a conventional emitter follower the Early effect comes into play, since the collector voltage is constant while the emitter voltage has the audio signal impressed upon it. In this circuit, the second emitter follower is essentially keeping the C-E voltage constant across the first emitter follower via a bootstrap operation.
The only other thing to note is that the two transistors are complementary. This should tend to slightly reduce the even-order distortion compared to a pair of same-sex devices.
Not a bad circuit, but certainly not a fundamentally different circuit than a normal pair of emitter followers (and also not an "inside-out Sziklai")
Thanks,
Charles Hansen
Hi,Here is a partial bootstrap variation. I noticed in simulation
that Q1 had small negative base current on the tips of the negative
swing of the signal. Also wanted to eliminate the compensation. So
here is one approach to the problem. Other ways?Gain at 1Khz = .955714, -3db = .675690
-3db frequency = 2.7062Mhz which is still way out past the single
emitter follower.It is more linear than the single emitter follower case and doesn't
require any compensation. Not as linear as the Sziklai.Thoughts?
By the way, your reference to the extended bandwidth of this circuit vis-a-vis a single emitter follower is due to the combination of the source impedance (100 ohms in your example) and the interelectrode capacitances of the input device. The bootstrapped circuit forces all of the electrodes to "follow" the input, hence no charging current is required to charge the capacitances (that still exist, by the way). If you reduce the source impedance to zero, these bandwidth differences in your simulations will disappear. What does it all mean? Build it and listen and find out...Good luck,
Charles Hansen
The only way to know which circuit is best is to build it up and listen. The Sziklai is more "linear" because it has forward gain inside the loop that increases the amount of feedback. If you think this will make the amp sound better, I've got a Crown DC300 I'd be glad to sell you. I don't like bootstrapping on general principles, so I haven't ever tried a circuit like this (and don't plan to), but I could be wrong. Maybe it's the best sounding thing since Edison invented the gramophone. You'd have to build it up and listen to find out for sure.You've already found one problem with the first circuit, namely the fact that it only put 0.7 volts across the first transistor. Look at any of the transistor curves, and you'll see this is not a good way to run a transistor. 2 or 3 volts might be OK, but not 0.7 volts.
I use simulations about once a year, but usually spend nearly all my development time on real circuits and listening tests. It's the only way to know how something is going to sound.
Good luck,
Charles Hansen
"The only way to know which circuit is best is to build it up and listen."Ah, yes. In the immortal words of Pete Townshend: "We won't get fooled again!"
"The Sziklai is more "linear" because it has forward gain inside the loop that increases the amount of feedback. If you think this will make the amp sound better, I've got a Crown DC300 I'd be glad to sell you."
Charlie, the DC300 is a notorious TIM producer (jc will correct me if I'm wrong here). I'm not sure we've determined the Sziklai is capable of TIM.
I think if feedback is good, you should use as much as possible. Use a million dB if you can. If feedback is bad, you shouldn't use any. I admire people that stick to their guns and go all out, in whichever direction they desire. On the other hand, I really don't care for the approach that "a little bit of feedback is good, but not too much". Kind of sounds like a chef salting a soup. There are some phenomena when two opposing effects must be balanced, but this happens fairly rarely, in my experience.The proof, as always, is in the pudding. I've heard some great sounding amps that used feedback, although I think they were all tube amps. It seems that feedback has less of a negative sonic effect on tube circuits for some reason. But I've always wondered what those amps would sound like if they didn't use feedback (you'd probably have to do an extensive re-design of course). I don't think I've ever heard a well executed tube amp without feedback. Have you? Could be interesting...
Cheers,
Charles Hansen
"On the other hand, I really don't care for the approach that "a little bit of feedback is good, but not too much".Both Williamson and Otala advocated this approach, but they did it for different reasons. Williamson for stability reasons (circa 1947) and Otala for TIM reasons (circa 1970). I think once you get the TIM thing nailed, it's all-or-nothing. This is what Baxandall demonstrated circa 1978. Both stability and TIM will militate against this approach, but the design challenge is facinating.
"But I've always wondered what those amps would sound like if they didn't use feedback (you'd probably have to do an extensive re-design of course). I don't think I've ever heard a well executed tube amp without feedback. Have you? Could be interesting..."
I can't think of a single famous-name golden-era amp that was designed without loop feedback. Even the Brook 300B amp used it, although its designers were suspicious of it (AE, June 1947). Somebody correct me if I'm wrong.
With modern amps it's a different story. If you've ever heard an Atmasphere you've heard one. The latest contender is the Antique Sound Labs' Hurricane, as reviewed by HP. The Hurricane is a classic Williamson circuit (with octal tubes no less, just like the original) minus the loop feedback. There is no allowance made for excess gain so the sensitivity goes to the moon (in the British style).
Virtually all of the SET amps that have appeared in modern times are feedbackless designs. Most of these trace their lineage to the Western Electric 91A theater amp:
http://www.aloha-audio.com/library/schematics/we91a.jpg
Sometimes you will see feedback controls in these modern-day SET amps that allow up to 10dB to be applied. Unfortunately this violates the multiplication hump rule, but how many of these designers understand what that means? They hear a degradation and then exclaim: "Feedback is terrible!" A little knowledge can be dangerous.
Have you heard any of these? I heard an Atmasphere about 15 years ago. It was pretty good, but since the amp is so atypical in so many other ways, it's hard to get a handle on what was contributing what to the overall sound.I don't really consider the SET amps to be valid designs. They really only can work well under an extremely limited set of circumstances. Comparing them to normal amps is like comparing apples and oranges.
On the other hand, I didn't know the Hurricane didn't use feedback. Now I'm curious to hear one. Of course HP (and many others) are raving about how the thing sounds. So this could be an encouraging sign. Of course the difficulty here is the output impedance. I would guess it would be several ohms without feedback, which blurs the picture because measurable interactions will now occur with the load.
This isn't so bad with a solid state design. All of the Ayre amps have an output impedance of less than 0.5 ohms, and most are less than half that. Then the interaction with the load becomes relatively insignificant.
So Scott, why don't you see what happens when you pull the feedback out of one of your awesome amps! Let me know...
Best regards,
Charles Hansen
I've heard Atmaspheres and I've heard Hurricanes, but I will let the reviewers make the public pronouncements about their qualities. Trust me, doing so is a no-win situation for a mfr.Output impedance is a real issue with tube amps, even those with feedback, because the feedback is not always fully operative due to load reactance. The definitive solution is to use a great many output tubes.
Charlie, I pulled the feedback from the EL34 version of my V8 some three years ago. Its got a bigger soundstage and is more vivid in the midrange, but I can't say as I like it any better overall. I am having some luck, believe it or not, with 6550 beam tubes using *local* feedback. This converts the horizontal plate curves to triode-like diagonal curves. These 6550 amps don't use any global feedback. They were used to drive the Avalon Opus at the recent HE2003.
But I have not given up on feedback. I find it a facinating challenge. The reason I think it can work is because it is used throughout the human body. Even the ear uses it. It is what protects the inner ear from overload. The touch of a finger on a piano key is a servo operation through and through. There is even compensation by integration. So I believe feedback can work. It's just difficult.
Hi,> I am having some luck, believe it or not, with 6550 beam tubes using
> *local* feedbackVery interesting. The pentode with local feedback could actually
be a more linear triode than a triode. How much local feedback is
this?I also note the switch from el34s to 6550s.
The amount varies with tube type. We can use whatever is needed to bring the sensitivity of the tube down to the level of its triode connection. This way, they both have the same drive requirement.As for linearity vs the triode connection, that again depends on the tube type, as well as the amount of feedback appied. It can be better, yes. Esp if the pentode is made to operate halfway to a CF.
you might be right, but I'm not convinced. For example, in the ear feedback is used only for relatively long-term level changes. It is not trying to operate on each individual part of a waveform as happens in a feedback amplifier.The example of a piano player also doesn't sway me. It would seem to me that short time scale eye-to-hand feedback is used when *learning* to play the piano, but a skilled pianist only uses feedback in a longer time scale sense. In fact, it seems likely that the musical differences between a beginner and a skilled player are actually due to feedback. The beginner has to find each note tentatively, using a clumsy series of feedback controlled iterations to move the fingers, while the skilled player instead uses muscle entrainment to play each note confidently.
If this is true (I don't know if it is or not), then the analog as applied to amplifiers would be to use feedback for long time scale functions, such as DC servos or thermal feedback (in solid-state amps), but avoid using feedback to control the audio signal on a moment-by-moment basis.
Interesting stuff, and I don't claim to have definitive answers here (though I do have strong suspicions). However, I do like your approach of trying to make amplifiers mimic living beings!
Best regards,
Charles Hansen
The idea originates at MIT circa 1942:"Thus Wiener and Bigelow discovered the closed loop of information necessary to correct any action--the negative feedback loop--and they generalised this discovery in terms of the human organism....Their purpose was to approach the study of living organisms from the viewpoint of a servomechanisms engineer and, conversely, to consider servomechanisms with the experience of the physiologist." http://pespmc1.vub.ac.be/CYBSHIST.html
"The beginner has to find each note tentatively, using a clumsy series of feedback controlled iterations to move the fingers, while the skilled player instead uses muscle entrainment to play each note confidently."
Muscle entrainment is an enhancement of the overall feedback process. The skilled player uses *conscious feedback* to make micro-corrections in real-time while depending on his muscle memory to walk (less consciously) through the notes. When a tennis player returns a serve standing on one leg there is more to it than muscle memory. When a golfer corrects his forearm rotation in mid-release there is more to it than muscle memory. When a cat swats an escaping bird in mid-flight there is more to it than muscle memory.
Here is a quote from "Consciousness as a Feedback Interface":
http://www.innerworlds.50megs.com/consciousness.htm
"The speeds at which neuromagnetic signals are propagated, together with their capacity to recruit/alter multiple modalities suggests that the underlying mechanism has been selected to make instant choices on which specific portions to recruit in order to facilitate the behaviors acted out of the State, and to do so quickly....When it comes to response to threats, or sighting prey, the evolutionary advantages are obvious."
I still don't see any examples of feedback in living organisms extending to frequencies above a few tens of Hz. Think about a cat attacking a bird. In mid-spring he will have to adjust his attack as the bird takes flight. Human reaction times (at least in relation to driving) are usually quoted as around 0.25 to 0.5 seconds. Even if we give the cat a full order of magnitude advantage (just to be *really* safe), we are still only talking about a maximum of 20 to 40 Hz.So that seems interesting. Unless you can find some example of reaction times (feedback) much, much faster than 50 mS, it would seem that feedback as used in living organisms is used only *below* the audio range. Maybe that's the principle that should be adopted for audio electronics as well...
Best regards,
Charles Hansen
True, we know that neurochemical response times are limited by the transmission time across the synaptic gap to the order of .5 to 2mS. By comparison, the propagation of action potentials is much faster. Look again at:http://www.innerworlds.50megs.com/consciousness.htm
"...an action potential can travel a full centimeter (a couple of orders of magnitude larger than a synaptic gap) in about 1.3 msec. The brain's electrical responses, therefore, happen orders of magnitude more quickly than do it's chemical ones (10)."
But that's just the tip of the iceberg:
"In what could turn out to be one of the most important discoveries in cognitive studies of our decade, it has been found that there are five million magnetite crystals per gram in the human brain (1). Interestingly, The meninges, (the membrane that envelops the brain), has twenty times that number. These ‘biomagnetite' crystals demonstrate two interesting features. The first is that their shapes do not occur in nature, suggesting that they were formed in the tissue, rather than being absorbed from outside. The other is that these crystals appear to be oriented so as to maximize their magnetic moment, which tends to give groups of these crystals the capacity to act as a system....This system, we speculate, is what makes the selection of which neural areas to recruit, so that States (of consciousness) can elicit the appropriate phenomenological, behavioral, and affective responses."
What does this mean in terms of feedback?
"Changes in state make changes in sensory and cognitive modalities, and they in turn, trigger changes in state. We can reasonably conclude that there is a feedback mechanism whereby each modality is connected to the others."
So we have all these different modalities connected simultaneously as a system linked by feedback loops. What is the upshot?
"Magnetic signals are propagated with much greater speeds than those of action potentials moving through neurons. Contemporary physics requires that magnetic signals be propagated at a significant fraction of the velocity of light, so that the entire brain could be exposed to a neuromagnetic signal in vanishingly small amounts of time....We might also conclude that neuromagnetic signaling is the context in which consciousness occurs."
Not known with certainty yet, so I can't claim an ironclad argument, BUT, there is no particular reason why feedback cannot operate predictably at high speeds. The GPS, for example, is a system of high-speed feedback controlling smart bombs.
The magnetic crystals would tend to explain a lot of things that don't otherwise add up about the nervous system. The comment that consciousness may be tied up here is very intriguing.However, I never said that "feedback cannot operate predictably at high speeds". Obviously it works very well, even in audio amplifiers (at least from a measurement standpoint). It's just that my experience is that non-feedback amps sound more "real" and "organic" than do feedback amps. Your hypothesis that using feedback is the "correct" approach because it is used in the human body is very interesting, but I'm not convinced yet. Of course, I could be wrong. ;-)
Thanks for the fascinating dialog!
Charles Hansen
Here's another crazy thought: the notion that quantum entanglement is a feedback system wherein two entangled photons communicate their location more or less instantly at whatever distance.http://www.mtnmath.com/whatth/node54.html
"Your hypothesis that using feedback is the "correct" approach because it is used in the human body is very interesting, but I'm not convinced yet."
No hypothesis, really, just a close-to-home example. All I know is that feedback around a linear stage has no penalty in terms of multiplication. The question is, how linear must those stages be to instantiate your "organic" amplifier.
"Of course, I could be wrong. ;-)"
Me too. ;-
Hello Scott,You talked about applying feedback around a linear stage. This raises two questions:
1) How linear does the stage have to be in order to "safely" apply feedback?
2) If the stage is linear enough to safely apply feedback, is feedback even then necessary (or beneficial)?
For instance our V-5x power amp has a bandwidth of 200 kHz, an output impedance of around 0.15 ohms, with distortion around 0.1% at 100 watts, 0.01% at 10 watts, and buried in the noise at 1 watt. This is all achieved without feedback. Sure, those numbers would be even "better" if we were to apply feedback, but I'm not sure there would be any point. (I would even hazard a guess that these numbers are comparable to a well designed tube amp such as yours after the application of feedback.)
Best regards,
Charles Hansen
"1) How linear does the stage have to be in order to "safely" apply feedback?"As a start we need to ensure that the feedback reduces all harmonics below the level of the open-loop amplifier. If this does not occur, then the device is too nonlinear, or there is not enough feedback.
"2) If the stage is linear enough to safely apply feedback, is feedback even then necessary (or beneficial)?"
Let me answer that with another question: Which devices and circuits are linear enough to not need feedback of some kind? The beauty of this question is that the answers apply to feedback as well as nonfeedback amplifiers. So both approaches are linked in this way, and both can be pursued in parallel.
"Sure, those numbers would be even "better" if we were to apply feedback, but I'm not sure there would be any point."
You said it yourself: you're not sure. Neither am I. There's only one way to find out, and that is to become an expert in feedback. But this is a long road, which may or may not be a dead end. According to jc, that's what it is, and he certainly qualifies as an expert. BUT, I remain curious, and I don't like watching TV.
TV sucks!Keep us posted on your progress and conclusions!
Cheers,
Charles Hansen
For instance our V-5x power amp has a bandwidth of 200 kHz, an output impedance of around 0.15 ohms, with distortion around 0.1% at 100 watts, 0.01% at 10 watts, and buried in the noise at 1 watt. This is all achieved without feedback.But, as has been pointed out previously, if you're using followers, you're using feedback. It's the feedback of the followers which is giving you an acceptably low output impedance by significantly reducing the output impedance of the previous stage. It's the feedback of the followers which is making the devices used for the follower behave far more linear than they would if they were configured as common emitter amplifiers rather than common collector amplifiers and thereby not drastically increasing the amplifier's overall distortion.
So while you can say you're achieving those numbers without global feedback (which isn't terribly difficult to do), you can't rightly say that you're achieving those numbers without any feedback because it's simply not true.
se
> I don't think I've ever heard a well executed tube amp without
> feedback. Have you? Could be interesting...Well I can't resist that...
Disclaimer: I hold nothing against feedback or its intended use...
this amp was merely an experiment to see if it was possible to build
an amp with no feedback.My current amp is a 2-stage homebrew. Driver is a 6c45 with
battery bias in the cathode. It is CCS'd ala a Gary Pimm
style CCS and DC coupled to the output stage. The Output stage
is a KR super tube (300B like) in Parafeed. It has its own entirely
seperate split power supply and is also CCS'd (using a KT90!).The driver has 0 ohms unbypassed and the output stage has 0 ohms
unbypassed. Both stages derive their bias from a power supply.
In the case of the driver, it is from the battery. In the case
of the output tube, it is from the delta between the split supplies.No feedback!
The soundstage goes out to the moon.
Sounds like an interesting, if somewhat impractical design. I don't keep up with the latest developments in decades old tubes, so you'll have to forgive my ignorance regarding a "Gary Pimm style" CCS and "Parafeed". How does the thing sound besides the soundstaging being very deep? What is the power output? What is the output impedance?I'm kind of curious about the lack of degeneration. Have you measured the distortion? Have you compared the circuit with and without degeneration, both with listening tests and measurements?
Let's assume for a moment that the circuit sounds much better without the degeneration, even though the measurements are basically the same. Then you would still have the question of what is causing the sonic difference. Is it the degeneration, or the presence of the resistors?
I've heard some speaker designers swear that getting rid of resistors in the crossover networks (by carefully matching the sensitivities of the raw drivers) made huge improvements in sound quality. The obvious conclusion there is that resistors sound bad. I don't have any details on that, nor any direct experience. It's possible that they were just using bad-sounding resistors, and that good-sounding resistors wouldn't have the same problem.
Another way to do a test like this would be to take a circuit that uses degeneration and then bypass the resistors with high-quality caps. Again another variable is introduced, but at least it would allow one to look at the same problem from a different angle.
Best regards,
Charles Hansen
Hi,I really like the sound.
> Let's assume for a moment that the circuit sounds much better
> without the degeneration, even though the measurements are basically
> the same. Then you would still have the question of what is causing
> the sonic difference. Is it the degeneration, or the presence of the
> resistors?It sounded sterile at first. Thru experimenting and listening
tests, I found that placing a 10 ohm carbon comp resistor in the
cathode of the output tube warmed it up. This caused about 1.4v
worth of bias and some tiny amount of feedback (or degeneration).
Increasing the value of the same type resistor caused too much of
the warmth. I didn't try other resistor types.I have an HP-334 distortion analyzer. It isn't in calibration,
but it can measure differences pretty well. There were no measureable
differences between with the carbon comp and without (not that
they weren't there - I couldn't measure them :)The amp is *really* sensitive to the components used in it, more
sensitive then other amps I've built. My belief is that it is the
sound of the resistor itself that is responsible for the sound
change - How's that for subjective effects!
So if I'm understanding you correctly, it sounded bad without any degeneration. But then you only added a little bit of degeneration to one of the two tubes to get the sound you wanted. Very interesting indeed, but difficult to draw any conclusions.You also noted that the amp is sensitive to the components used. I have found this to be true in our zero-feedback designs as well. My hypothesis is that just as the presence of feedback dominates the measured results, it also dominates the sonic results.
In other words, you can take a mediocre circuit, add a bunch of feedback and it will measure pretty well. The feedback is the dominant factor in the measurements. In the same way, adding feedback will give a strong "feedback sonic signature that dominates the overall sonics. Remove the feedback, and the signature goes away. The amp is now much more neutral and transparent, allowing component changes to be heard much more clearly.
Best regards,
Charles Hansen
"In other words, you can take a mediocre circuit, add a bunch of feedback and it will measure pretty well. The feedback is the dominant factor in the measurements."It's not just the feedback we're hearing, it's the nonlinearity in the mediocre circuit multiplied by the loop iterations. The idea is to apply feedback around a linear circuit.
If I read another one of your posts correctly, your limited experimentation with removing the feedback from your V-8 amplifier was probably a step forward sonically. Why not pursue that path to its logical conclusion?Best regards,
Charles Hansen
I am pursuing it, to the point of public demonstration. The EL34 triode V8 was exhibited at the 2001 VSAC show in Seattle. The 6550 pentode V8 was exhibited at HE2003 in San Franciso. Neither amp uses global feedback. So there is pursuit. :-At the same time, I am facinated by the concept of the "blameless" amplifier. The one we can insert in the signal path and not hear. I see no reason why feedback shouldn't be included if it can be made to work.
Hi,> But then you only added a little bit of degeneration to one of the
> two tubes to get the sound you wanted.Yes, added a tiny resistor to the output tube cathode and it made
a large difference in the sound...in the bass area. Since
the output tube operates at around -90v bias, the 1.4v additional
bias due to the resistor had to be insignificant.Forgot to mention output impedance which might have a lot to do
with it. The tube Rp is about 420. Adding the small R to that,
so 420 + (10*mu) =~ 460. Divide by turns ratio squared is about
1.5 ohms
Good work. Resistors suffer from the same sort of distortion that I seem to be measuring with wire. The harmonics are unusually higher order, probably due to a sharp break in the the transfer function. ' Keep on Truckin! '
Hi,Are any brands better for distortion. How about the super
resistors like Vishay or Caddock? How about types like WW or
metal film, carbon comp?
I didn't make the measurements myself, an associate of mine did, years ago. The measurement method is beyond discussion on this website, but it was similar to what I do to measure wires.
Hi,I have built up stuff with tubes. Now my aim is to find some
guiding principles that perhaps the "blameless amp" crowd has
missed.I'd like to go beyond the flavor of the month capacitor approach
to building amps and see if there aren't some solid explanations
for why things sound good and why they don't.You seem to advocate the no feedback approach, yet are willing to
use simple emitter followers which have a lot of feedback. In your
opinion, when is feedback good and when is it bad? Is it only good
over one stage (purely local) and you stop when it goes beyond that?
In other words never across 2 or more Vbe junctions?
local degeneration (whether in the form of an unbypassed emitter resistor or an emitter follower) is quite a bit different than a feedback loop around two or more stages. Calling both things "feedback" only serves to confuse. Unfortunately, we don't have enough widely accepted terms to use. "Degeneration" is probably the most precise term, but not very descriptive -- it was coined as the opposite of "regeneration", which was positive feedback used to increase the gain of RF receiving circuits before the heterodyne was developed. Besides, "degneration" is vaguely reminiscent of "degenerate", which has a clearly negative connotation (sort of like "pervert"). Who would want their circuit to be a "degenerate"?Cheers,
Charles Hansen
Not a bad circuit, but certainly not a fundamentally different circuit than a normal pair of emitter followers (and also not an "inside-out Sziklai")Sure it is. Same topology as Sziklai but output device becomes input device and input device becomes output device in terms of polarity. Emitter becomes collector and collector becomes emitter. Inside out.
se
nt
Charles, a push pull version of that design was in the National Semi NH2002 follower first made in the '60's. What else are we giving Hawksford credit for? ;-)
LH0002 you're thinking of? "LH" stood for "Linear Hybrid", which is why they are no longer made. They were actually separate dies that were bonded inside the same case. In those days there wasn't any way to make fast complementary transistors on the same die.This part did feature the complementary devices in series, but did not have Hawksford's bootstrap feature that keeps the voltage across the first device constant. Like I said, it's a clever idea, but probably too clever to actually sound good.
Best regards,
Charles Hansen
If I remember correctly, the '0002 had a pair of 2219's and 2907's, and maybe two or four resistors. My boss in a hybrid company was trying to second source the part.I recall it was a pretty fast bugger for it's time. One problem was I/O offset, spec'd in the low millivolt I think...Required matched transistors, usually off the same wafer in consecutive locations.
Cheers, John
If I remember correctly, the '0002 had a pair of 2219's and 2907's, and maybe two or four resistors.
se
Haven't seen that in 20 years..thanks..Course, the round package and leads were kovar..made them real easy to move through the production line, cause the trays all had magnetic posts.
Cheers, John
nt
Looks like an inside-out Sziklai. :)se
This post is made possible by the generous support of people like you and our sponsors:
Top of Page ]
[ Contact Us ]
[ Support/Wish List ]
[ Copyright © 2025, Audio Asylum, all rights reserved ]
