Why Let an Amplifier Sound Good when You can Force it to?

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Some people say feedback amplifiers sound best, some say they have some sort of characteristic sound. I am all confused and figure I'll try to see it in an enlightened manner. LOL

Looks to me like an amplifier that is forced to perform by feedback may at worst fail in a cascade of burned components if the feedback loop goes crazy and measures weren't taken to assure the amplifier can't overdrive stages outside the design safety zones.
 
Administrator
Joined 2007
Paid Member
Well this will end up a bun fight before the day is out :D

Assuming we are talking about power amplifiers then I think there is some truth in saying that various topologies and component choices do influence the subjective outcome.

As to feedback loops 'failing' in some way, well I would counter that by asking why would they (fail) ? They are passive parts, usually lightly stressed, they would be the last suspect's under a fault condition in my experience.

If you mean failing to maintain stable operation when faced with adverse drive conditions, well that's a different ball game. A competent design should be just fine when overdriven to any normal extent. If it flips and latches to a rail for example, in the face of modest overdrive then the designer has failed in their job.

You can spin all this a 1001 ways and still not be happy, or still try and argue for one implementation over another.
 
Looks to me like an amplifier that is forced to perform by feedback may at worst fail in a cascade of burned components if the feedback loop goes crazy and measures weren't taken to assure the amplifier can't overdrive stages outside the design safety zones.

That's why engineers have slide rules. They are charged with making sure this stuff doesn't happen.

A properly damped amplifier with carefully parsed gain structure will exhibit minimum latch up.
 
Member
Joined 2011
Paid Member
There's a famous audio circuit wizard called Nelson Pass, who designs and builds and sells amplifiers under the brands "Pass Labs" and "First Watt". He's also very active in the DIY audio movement, participates regularly on this website, and is the driving force behind the Burning Amp festivals.

I grabbed the image below from the Products page of his First Watt website. Notice that many of his amplifiers use feedback, and many use no feedback at all. Think about that for a few minutes.

You could search the internet for listening reviews of a few First Watt feedback amps, and a few of their non-feedback amps. What do reviewers say? You could ask your audio buddies who own one (or more!) of the First Watt amps, will they please let you listen to the amp, using music that YOU are familiar with?

Bob Cordell's power amp book gives example schematics for amplifiers with feedback, and other amplifiers with no feedback at all. You could experiment with those, either on the lab bench, or in circuit simulation software, or both.

_
 

Attachments

  • firstwatt.png
    firstwatt.png
    31.1 KB · Views: 1,429
Listening tests at the HiFi levels of nowdays mostly deal with placebo effect, if not totally biased not really done double blind.
Serious unbaised tests prove they are all the same. Simply because audio sources, loudspeakers and audotoriums are of a much lower quality.
Save your money and saliva and go for simple well proven designs, stay away from sensational assertions especially with stratospheric LTSpice results as well as hundred of pages books.
Here is all you need:
ESP Ron Elliot Elliott Sound Products - The Audio Pages (Main Index)
 
I haven't seen any amplifiers that have no feedback.

Even an emitter resistor, or cathode resistor, provides local feedback. That resistor also can make the device operate in a much more linear way.

Antique tube circuits never had global feedback. They had local degeneration. Early transistor circuits, which were not that different from tube circuits, used local degeneration just like their tube predecessors.

Even the super simple MOSFET follower has local degeneration because of its source resistor.

the Nelson Pass "Stasis" amplifier is a two stage amplifier. The output stage is a push pull Sziklai pair. It is not included in the global feedback. It definitely used local feedback.

An amplifier with no degeneration at all would not provide predictable performance. It would also be very nonlinear.
 
I don't think this can be answered. There are different kinds and amounts of feedback that can be applied to various gain stages which in turn give rise to different harmonic compositions at different loads and power levels. This then in turn favors or discriminates against different types of music or sounds. Example people pay big money for a 2A3 amp with 1% distortion because it makes simple vocal music sound really smooth and intimate but try the same amp on a big orchestral peace and you will say it sounds really congested.

Here are three companies, some of them use feedback and others don't, they are all commercially successful:

Ayre Acoustics - never uses any feedback loops, only local degeneration
Mr. Pass (Pass Labs, First Watt, ...) - sometimes yes sometimes no depending on the circuit
Mark Levinson - uses lots of global feedback

Then that takes you down the road of asking, are you in the precision instrumentation business or the entertainment business. Mr. Pass says he is in the entertainment business and I think what makes him successful is that he -> builds -> listens -> tweaks and improves -> listens again -> once he feels he has done the best he can reasonably do after some number of iterations he moves on to something new.
 
As far as runaway amps that can burn up a transistor I was thinking of a closed loop voltage feedback amp, used to flatten cheaper not so flat transistors. Sometimes cold solder joints, a bad resistor lead connection etc could break the loop open and allow the amp to drive to saturation. Mainly worried about using inexpensive driver transistors that could be driven to excessive wattage by a design no longer restricting the driver to a narrow low current, flat gain part of the hfe graph.

I figured out how to stretch the cheaper curvy transistors with voltage feedback and using some of them in the <100ma (Ic) range as a part of a Darlington. If the loop were to break open, some configurations could push the delicate transistors to unintended high drive levels and even start a cascade of cooked transistors.

I backed out of that and decided to use more robust and expensive components since I am not making a million of them. I seem to be encountering many analog power amps that could be called a high power opamp.
 
I haven't seen any amplifiers that have no feedback.

Me neither. Those who say that 'feedback is bad in audio' normally have to make a distinction between 'global negative feedback' and 'local negative feedback' and then exclude the latter from their strictures. Local negative feedback can't be avoided when using common-drain and common-gate configurations so ISTM a no-feedback approach would mean only pure common-source use of amplifying devices which would mean lots of gain and lashings of non-linearity.
 
- Feedback is an ideal universal mechanism, only imperfections in the elements comprising the feedback loops contribute to imperfections in the output of the system.
- Feedback is everywhere, even inside electronic gain devices such as transistors, at electron and molecular level.
- Feedback requires some element outputting a magnitude which is the sum or difference (and/or other math operations) between other 2 or more magnitudes: The error amplifier. Feedback requires the output of the system to be fed back to input of error amplifier.
- Feedback happens inherently in every amplification circuit, even in follower configurations (the simplest case). The conditions for a follower to be unstable do exist, and can be created in practice.
- In linear electronic circuits: there are lower and upper practical limits for feedback. Lower NFB loops have to work with higher error magnitudes, requiring wider linear range of operation of error amplifier (and earlier amplification stages). Higher NFB loops encounter the opposite problem, error magnitude becomes low enough to be disturbed by other magnitudes, as by: parasitic inductive and capacitive coupling between circuit elements, noise floor of components, drifts in error amplifier characteristics due to temperature.
- In switched-mode amplifier circuits: feedback is employed too. "It does not need to be stable in order to be linear". The interest for switched-mode amplification arises when the practical limits of feedback are explored (the raw conditions where linearity improvement takes place).

The following simulation pictures (based in a real circuit) show a class-D amplifier (modulator type: self-oscillating post-filter single-integrator) doing <0.007% 3rd harmonic at 5khz @ 50W @ 8ohm. The circuit is sized for up to 800W/2ohm output, using two N-ch TO-220 power MOSFET. This matches the performance of finest class AB projects shown in best solid state threads, while not being a stable NFB loop, but it is an unstable NFB loop constructed to be highly linear (this proves the 1st statement in this post).
 

Attachments

  • class_D_50W_THD_1.png
    class_D_50W_THD_1.png
    33 KB · Views: 611
  • class_D_50W_THD_2.png
    class_D_50W_THD_2.png
    34.9 KB · Views: 608
I was just going to say using negative feedback for linearization requires a precision comparator stage for the magic to work. Differential amp pairs on one die are a good idea for that. Sort of funny how the errors wiggle their way through the system until they stabilize at the comparator. There has to be some sort of high speed hunting and settling time, well above audio range.

I could not find but a few transistors even somewhat able to produce linear gain over the functional range without feedback. And they probably deteriorate with time and heat.
 
Last edited:
I think the whole low-overall-feedback and no-feedback movements can be traced back to 1966, when Daugherty and Greiner published "Some design objectives for audio power amplifiers", IEEE Transactions on Audio and Electroacoustics, vol. AU-14, nr. 1, March 1966, pages 43...48. Their conclusions on error signal overshoot were later repeated in a series of articles of the research group of Otala, with some refinements to the amplifier model.

In the appendix of their article, Daugherty and Greiner prove for a single-pole feedback amplifier driven by a first-order low-pass filtered step function that the error signal exhibits no overshoot when the open-loop bandwidth is greater than or equal to the bandwidth of the low-pass filter that preceeds the amplifier. Error signal means the difference between the input and feedback signals of the amplifier, which is the signal that drives the input stage.

That is, they have proven that a sufficient condition to prevent slewing is to make the open-loop bandwidth greater than the bandwidth of the filter in front of the amplifier. Given the fact that the achievable gain-bandwidth product in a practical amplifier is limited by non-dominant poles, a large open-loop bandwidth implies a relatively small low-frequency loop gain.

Unfortunately, in the article they present their sufficient condition as if it were a necessary and sufficient condition. It is by no means necessary. The fact that the error signal overshoots in an amplifier with a small open-loop bandwidth is no problem as long as the amplifier's input stage is designed to handle the overshoot.

When you increase the open-loop bandwidth and reduce the low-frequency loop gain of an amplifier by connecting a resistor in parallel with the capacitor that sets the dominant pole, all that happens is that the final value of the error signal gets larger. The overshoot, which is by definition the ratio of the initial peak to the final value, gets smaller but the required linear range of the input stage is not reduced at all. When you also take square wave input signals into account, the required linear range is reduced, but only by a factor of two.

Peter Garde pointed out these facts in his articles "Transient distortion in feedback amplifiers", Journal of the Audio Engineering Society, vol. 26, nr. 5, May 1978, pages 314...321 (reprinted from the Proceedings of the IREE Australia, October 1977) and "Slope distortion and amplifier design", Journal of the Audio Engineering Society, vol. 26, nr. 9, September 1978, pages 602...608 (reprinted from the Proceedings of the IREE Australia, December 1977). He also explained how the linear range of an input stage can be increased by local feedback.

Apparently the damage was already done by then, because 40 years later the idea that a feedback amplifier with a small open-loop bandwidth necessarily suffers from transient problems still exists in high-end audio circles. In fact, it has only become more extreme over the years, changing from 'use as much feedback as you can without making the open-loop bandwidth smaller than 20 kHz' to 'feedback is evil, avoid it at all costs'.
 
There are two approaches to feedback:
1. talk about it
2. learn about it
Sadly, many people choose option 1 and carefully avoid option 2. Doing it the other way round (2, then 1) means you have something useful to say.
So true.
There is such a gap between the "1 then 2" people versus the "2 then 1" people that dicussions are of no use.
All we can do is keep aware of their arguments just in case something new would make sense, otherwise, just ignore them.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.