[jamesgarvin]

I read comments from some amplifier designers/manufacturers that negative feedback should be avoided at all costs, only resulting in compromised sound without a commensurate benefit. I presume that negative feedback serves some potentially useful purpose, given that, at some point in history, respected manufacturers employed the technique. I understand from reviews of the well regarded RM-200 (which I plan to purchase in the near future) that it employs a small amount of negative feedback.

Is negative feedback always bad sonically? Are they situations when it is appropriate? Why do some designers/manufacturers advocate against its use? Just food for some discussion.

[jon_010101]

This has become a surprisingly heated issue in recent days.  But, there are many misconceptions about negative feedback.  The claim made by some manufacturers is that feedback is a "crutch" for bad design, and a way to make a bad amp measure well.  In reality, the opposite is often true.  It takes a lot more engineering to do a proper feedback amp, and higher quality transformers.  Feedback is not a sign of a bad design, and can yield excellent results.  There is a reason why vintage Scott, McIntosh, Marantz, and HK Citation amps are still revered -- careful circuit tuning, brilliantly good transformers, and well-executed negative feedback.  When implemented properly, it leads to a less-colored sound, better damping, lower noise, and lower distortion.  There are times when it can be omitted -- for example, with triode amplifiers driving *very* efficient speakers -- but (in my experience) adding a bit of corrective feedback can significantly improve the listening experience . 

[WGH]

A tube amp designer's view on feedback can be found on Jim Hagerman's blog as he thinks about the pros and cons of feedback while designing a new amp:
http://www.hagtech.com/blog07.html

Search for the word 'feedback'.

Wayne

[Airborn]

jamesgarvin,

Here's some positive feedback.  Welcome to AudioCircle.  Nice first post.  

Ralph Karsten of Atma-Sphere gave his opinions on this topic in this article:  http://www.enjoythemusic.com/magazine/manufacture/0707/

Manley Labs apparently thinks Negative Feedback is OK: http://www.manleylabs.com/about99.html#negative%20feedback

[Daryl]

Anyone who says negative feedback is bad should be avoided at all costs.

This is the work of satan midirecting the attention of the common man from valid science toward witchcraft.

Then what you have are amplifiers costing more than your car that achieve only a few watts of output but have outrageous distortion, output impedance and noise levels.

The designers of these turkeys who have often only achieved the level of novice are hailed as 'experts' and everything they say is quoted all over the web as if it were handed down by god though they may as well have farted.

The more negative feedback the lower the output impedance and distortion.

Feeback loops will become problematic when the feedback becomes positive or an intermediate amplifer stage clips.

Both conditions cause the feedback loop to become unstable (loose it's grip) and distort.

Therefore the quality of an amplifier and the competency of the designer will indicated by how high the level of negative feedback that can be achieved and over how broad a range of operating conditions the feedback loop can maintain the utmost in stability.

Those are the design goals of an amplifer, more feedback and more stability.

The propagation delay through the amplifier circuit rotates the phase of the feedback until finally at some freqeuncy the feedback will become positive.

By this frequency the feeback level must be less than one or the amplifier will oscillate.

You design the amplifier with very high gain at low frequency and a first order rolloff which causes the gain fall as frequency increases so that the feedback will be less than one before the phase rotates to positive.

The faster the circuit the higher the feedback can be.

At 100hz you will have a lot of feedback and consequently amplifier control of the load but at 10,000hz you might have close to 100 times less (depending upon at what frequency the gain starts it's decline) so knowing output impedance at 10khz is something to look for.

The phase of the feedback is not only affected by delay but also the frequency response of the feedback loop since the feedback loops phase is implied by it's frequency response (this is known as a minimum phase relationship).

Feedback must remain negative while it's level is high so you must sculpt the response of the circuit to maintain phase at the most stable point throught the spectrum, this is known as 'phase compensation'.

The open loop transfer function (frequency/phase response) once phase compensated for maximum stability can still be influenced by the load impedance presented to the amplifier and as load impedance becomes more chaotic it can influence the feedback loop into instability.

To combat this phenomenon you want the output impedance of the final amplifier stage (without the feedback loop active or 'open loop') to be as low as possible.

The lower the final stages output impedance the less the load impedance can influence the phase of the feedback loop and the more stable the amplifier can remain with more severe load impedances.

As the voltage and current changes at the output of the amplifer it's final stages transfer function will change also.

The other stages as well but the others should be designed for more stability, in the final you must also consider power efficiency so it's variation will be much higher than the other stages.

The amplifier must be designed to maintain stability for all output voltages and currents.

Finally high levels of high frequency information can cause intermediate stages of the amplifier to clip before the feedback loop can get it under control due to the extreme gain of the system.

An audio signal doesn't necessarily have a lot of high frequency information but the culprit might also be unexpected high frequency noise from preceding componets (especially digital one's).

The amplifier must be designed to handle ample levels of high frequency information without causing interstage clipping or what is referred to as 'transient intermodulation'.

With negative feeback the output noise will essentially be the 'input noise' of the first stage multiplied by the overall closed loop gain of the amplifier so with negative feedback you need only design a low noise input stage that fixes the noise level for the amplifer.

Negative feedback also reduces the amplifiers distortion and output impedance by the ratio of open loop gain to closed loop gain.

That is why negative feedback is used those are the considerations necessary to properly design an amplifier.

Lots of folks talking about amplifers but few know what constitutes a good amplifier or what to look for.

As a result the best amplifer and designer is a matter of whom is the best salesman/con-man and rather than the market driving performance levels it is driving foolish features and obsurd claims because the customers demand it.

[GMuffley]

To say feedback is categorically bad is categorically wrong.  It is entirely dependent on the circuit, the application and skill of the designer.  There certainly are amps that sound fine with no feedback, but there are always design trade-offs, such as greater susceptibility to impedance variations of the loudspeaker it is connected to and greater output impedance.  Look at the Citation II for example; it has over 30db of feedback, but there are multiple loops and the open loop bandwith of the circuit is very wide.  It works very well.  Ultralinear output stages, including the RM-9 series, tend to sound better with some level of feedback.  Roger's nifty variable feedback option demonstrates very well how specific levels of feedback alters not only the gain of the amp, but the manner in which it works with a particular speaker load.

[jon_010101]

The designers of these turkeys who have only achieved the level of novice are hailed as 'experts' and everything they say is quoted all over the web as if it were handed down by god though they may as well have farted.

You should check out the recent "Negative Feedback Circuits" thread on the tubes forum at AudioKarma    Quick Summary: Anti-NFB amp-builder picks fight with wrong crowd, hilarity ensues.

[Roger A. Modjeski]

This has become a surprisingly heated issue in recent days.  But, there are many misconceptions about negative feedback.  The claim made by some manufacturers is that feedback is a "crutch" for bad design, and a way to make a bad amp measure well.  In reality, the opposite is often true.  It takes a lot more engineering to do a proper feedback amp, and higher quality transformers.  Feedback is not a sign of a bad design, and can yield excellent results.  There is a reason why vintage Scott, McIntosh, Marantz, and HK Citation amps are still revered -- careful circuit tuning, brilliantly good transformers, and well-executed negative feedback.  When implemented properly, it leads to a less-colored sound, better damping, lower noise, and lower distortion.  There are times when it can be omitted -- for example, with triode amplifiers driving *very* efficient speakers -- but (in my experience) adding a bit of corrective feedback can significantly improve the listening experience . 

I agree with the above and will add the following:

It is my belief that certain amplifiers sound really bad when they meet a load that, due to feedback, makes them misbehave. These are called conditionally stable amplifiers. Under many, but not all conditions they behave properly but meeting up with difficult loads they do strange and unpredictable things. This may remind you of some person you know who is conditionally stable. 

By my experience only 10% of the amplifier designers in the high-end industry have any formal training in electrical engineering. Being an engineer of another stripe doesn't count, they don't teach topics like feedback or low noise design to non-electrical students. Stable feedback is difficult to implement even for those who know how. jon_10101 makes several good points. To quote him in reference to tube designs one needs "brilliantly good transformers" for as little as 12 dB of feedback.  50% of the time I spend on a complete design is for the transformers and the winding details. I wind all the prototypes myself and during the process I decide the best way to put down the wire to minimize the leakage inductance and parallel capacitance. For absolute stability the output transformer has to go out beyond 60 KHz and even then there are things I do to increase the stability. Marantz's Sid Smith used a tertiary winding (fancy name for another winding) for the high frequency feedback. I used this technique in the RM-9. Other techniques, like standing the output tubes on top of the output winding work well. This type of feedback is quite successful in the RM-200 because I have balanced feedback to the balanced inputs. BTW the RM-200 performs exactly the same in balanced and un-balanced mode. I wouldn't have it any other way. The RM-9 and 10 have somewhat traditional feedback to the input tube cathode in the RM-10 or grid of the diff amp in the RM-9.

Being a curious fellow I have measured many of the amps mentioned by jon_10101 and recorded the results
in three (going on 4) composition notebooks with results of tests I have made on amplifiers that have crossed my path.  With 2-4 pages average that's 150 amps. Some I have owned, some my customers have owned and some people bring for me to evaluate. I can make over 75% of these amps oscillate into rather common loads. Sometimes the oscillations are "birdies" that only show at low frequencies due to the back EMF of cone woofers. Some will go into destructive oscillations with as little as 0.1 UF which can be the capacitance of many cables. I saw a respected solid-state amp go up in smoke with such a cable. Another amp went from drawing 50 watts from the wall to 500 watts when I put a very small 0.1 cap across the loutput. In cases like this (and there are many) the cap doesn't get hot at all but there are now 500 watts going in and nothing going out. About 30 seconds of this would likely cause a melt-down of the output transistors. These birdies and full power oscillations usually occur in the 20-50 KHz area so they cannot be heard but they might toast your tweeter. Birdies are little bursts of oscillation that appear on the main sine or music wave that I observe on my oscilloscope as I connect various loads. They are often rather small and look like a bird on a wire, hence the name. When they occur at low frequencies they show up repeatedly as the output crosses a specific voltage.

Of the brands mentioned by jon I found only the Marantz were up to my standards of stability, distortion and overload recovery. I could make all the others misbehave under various standards I have developed for my designs.

Here are my findings:

The HK Citation II had poor low frequency stability due to the feedback in the driver and the use of high capacitance coupling caps. Here is a perfect example where bigger caps are not better. Those who modify heed this warning: do not raisie or lower cap values unless you have the equipment to verify that your change is a good one. You could make the amp unstable in the low frequecy range.

HK went for ultra wide bandwidth but created other problems in the process. The amp takes several seconds to recover from a good clip. Given the age in which they were designed I imagine the designers didn't concern themselves with the clipping behavior believing they would never clip on the efficient speakers of the day. This poses a problem for current users who may indeed clip now and then. I designed for immediate recovery from overload as I can see 10-20 % clipping on a scope before it is audible to most listeners.

That's enough for now. Let me know if you want more on this topic.

[avahifi]

Good information posted by Daryl and Roger.

Frank Van Alstine

[Russell Dawkins]

I'd really enjoy hearing more, Roger.
It's good of you to be so free with your information and observations/musings.
Thank you for taking the time.

Have you ever built an amplifier of your own design, Daryl?

[Roger A. Modjeski]

Thanks to Frank Van Alstine for your comments. We have both been at this a long time and I know you do good work.

Russell, I often say too much. Do you really want more? If so, what in particular?

[Flyquail56]

Roger,

I would like to hear more about the testing procedure you mentioned for stability, overload recovery, etc.

I responded to claims about the HK Citation II in another forum and referred to your findings posted above. Well, I found out very quickly that the Cit II is a sacred cow to some people; a lively discussion ensued 

Having used various Dynaco tube amps for most of the last 30+ years and listened to a host of others, I have yet to find the distortion-free amplifier. Therefore I am not inclined to question your findings. Your writing here and on your website is very interesting and informative, so I would like to know more about the subject of these tests if you are willing to take the time.
Thank you.


Best regards,
Mike Hazel

[Roger A. Modjeski]

Mike,

The tests are very simple and revealing. Take the sacred cow Citation, drive it a few db  into clipping, reduce the input and watch how it recovers. The base line on the scope will travel up and down clipping the now reduced signal. This experiment can also be done with music and the woofer cone will walk around after clipping. As the cone walks around at about 0.5 cycles/second you can hear the top and bottom of the wave alternately clip. The cause is easy to surmise: too much feedback and too many low frequency roll-off points stacked on one another. Engineers call them poles and zeros. Non-engineers just hope no one will notice. This behavior was the one of the first things I noticed about the Citation amps. I also noticed that Marantz amps of the same period exhibited none of this behavior. Hats off to Sid Smith, an engineer by trade, who designed those.

Capacitive load stability tests are easy but dangerous to the amplifier, so do them at your own risk! Run a 2 KHz square wave through an amplifier at various power levels with various loads of 8 ohms in parallel with .1 to 10 uf. Be careful, some amplifiers will oscillate themselves into smoke, especially the transistor ones. The plates on a tube amp will start to glow cherry red. Don't leave the load on for more than a few seconds, just enough to see what happens. Do this only with a scope as a speaker's tweeter will likely go up in flames.

People who worship certain things often have little knowledge of that which they worship yet they will defend it to the end.

[Flyquail56]

Thanks Roger, very interesting. I don't generally read the audio mags, but since posting this request, I happened to find similar information in your comments in the 2002 review of the RM200 on Stereophile's website. Easy to understand even for a non-EE like myself (and a good read for anyone else interested in this subject).

Thanks again for taking time to elaborate.

Best regards,
Mike

[Roger A. Modjeski]

Good on ya' (that's a compliment in Australian slang) for doing some research. I have been expounding these stability tests for the last 25 years. Most amplifiers that hit my bench go into some form of bad behavior under the capacitive load test. I imagine that those designers who care will do the standard 2 uF in parallel with 8 ohms, 1-4 KHz square wave and hopefully over the full power range of the amp. Good amps will ring a bit at 50-100 KHz but the ring damps out in a few cycles. The not so good amps will keep ringing or go into full power oscillations that result in smoke. Where many designers go wrong is that the y assume if 2 uF is stable that anything below that will be also. I often find amplifiers who exhibit this behavior wiht 0.1 uf. This amount of capacitance is often found in high end cables and is the cause for some amps sounding very bad with certain cables or blowing up all together.

I appreciate that at least one magazine addressing behavior out of the 20-20Khz band. There are several ways to excite these oscillations. John Atkinson at Stereophile goes about it a different way that is equally good. He tests amps out to a few hundred KHz and reports any rise after the initial fall off that typically starts at 20-40 KHz.  His test done at low levels is less likely to blow up the amp. He has stated more than once that this is his last test because some amps die, making it their last test.

Some time ago a reader wrote to the Editor of Stereophile saying the tests were a waste of space. I take the opposite tack and read these informative tests with great care. From my point of view, John Atkinson is doing us a great service ferreting out the anomalies of equipment. For those readers who find the test report overwhelming just read his last paragraph where he states in simple terms what shortcomings he finds in the product and how the user can avoid them or move on to another product that will suit his needs better. For example, say an amplifier performs well in all the tests but noise. You would not want to use that amplifier with a high sensitivity speaker though it would be fine to use it with a less sensitive speaker.

Speaking of Stereophile, I tried the link available on my website and got to their home page but not to the review directly. Can you or anyone give us the link so I can update?

Roger

[pbrstreetgang]

Here ya go- http://www.stereophile.com/tubepoweramps/560/index.html

[lonewolfny42]

Here ya go- http://www.stereophile.com/tubepoweramps/560/index.html

I had a chance to hear this exact amp at our last Rave...and its still owned by Mr. Fremer. Its a really sweet sounding amp...nice work Roger.

[JoshK]

Roger,

I read your post with interest.  I don't have much experience one way or the other.  But, would you happen to have any texts that you think are worthwhile for learning about feedback and stability? In the context of audio or not is ok.  Particularly, ones that might also go into a little bit of filter theory.  Math is not a problem and might even be helpful for me.  (for example I have done some Fourier analysis in my education, but in a different context and I know what FFTs are)

I am building amps now for myself and trying to learn how to properly measure their performance (got two scopes).  Your description of your tests I find educational, but I don't think I'll try them all out until I understand them fully.  I wouldn't want to blow up some expensive tubes.

P.S. what is your experience/opinion on local feedback technics? How about multiple local feedback loops instead of a global FB loop, say cathode feedback plus plate to plate FB?

Josh

[pubul57]

So would converting the RM9 MKII to triode operation and NFB be a bad idea? Or would it still be better to have some NFB in this circuit? I am agnostic on the NFB issue.

[Roger A. Modjeski]

I enjoy reading old Lafayette and Allied radio catalogs, some of which I have had since age 12. They were big on specs and announcing how much feedback they employed. Lafayette had one claiming over 40 db feedback. That was the total in 3 isolated loops. One could never wrap that much feedback around an output transformer and obtain stability.  Well designed driver stages need no feedback and in some cases one can get the driver distortion to reduce the total distortion via cancellation.

Read the Triode post for info on RM-200 triode.