Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[danlavry]

Terry Demol wrote on Sat, 21 May 2005 03:28

danlavry wrote on Thu, 19 May 2005 17:06 "Hi Dan,   ....I'm sure you are aware of all the usual defences... ...and I'm still seated firmly in camp one...   Cheers, Terry

I said:

Instead of arguing the merits of Case 1 and the merits of Case 2, let me ask you to do so. Can you (a design engineer) convince me (a design engineer) with a TECHNICAL ARGUMENT that Case 1 is better than Case 2?

You replied:

I'm sure you
are aware of all the usual defenses but they can usually,
as you have stated with the latest gen of good sounding chips,
be very well refuted with measurements.

Terry,
Please do not assume that I am aware of any of the “usual defenses” and also do not assume that all the other readers here know any of the “usual defenses”. I asked you if you can come up with a TECHNICAL ARGUMENT that case 1 is better than 2, and you completely side stepped it!  

Yes, I heard many arguments about why case 1 (open loop) is better, and some of the arguments were unbelievably ridicules. I asked you to build a case why the open loop approach is better (with good solid reasons). You repeated the old story: you believe that specs are not everything, and that you believe that there is something else that makes good sound that we do not know.

When I made medical amplifiers, supplies, telecom gear, instrumentation equipment... there was always a rather clear understanding of what makes for good results, and it could always be reduced to solid specifications. It was not always easy to measure, but the goal and the parameters were always very clear.

But since I have been doing audio, a long time now, the statement “it sounds better but I do not know why” has been part of the landscape.

As long as “But I do not know why” is an integral part of the “equation”, the people that advocate an open loop approach, can not sell their method as a fundamentally better way to go. All they can say is: “so far it sounded better”, which is a very subjective opinion, and can be a self serving one.

Here is a challenging statement for you:  when it comes to transparency, measurements do count, and waveform in = waveform out is the goal.  

I, too, have done my share of “low feedback” as well as “much feedback” designs. I too have listened a lot, and have my sonic conclusions. So instead of you coming out with SONIC arguments (it sounds better) which I am completely aware of (and there was no need to repeat), why don’t you list what the benefits of the open loop approach are. Let us see if I can take apart those arguments. I promise that I will follow such an interaction with a list of benefits for the “lots of feedback” camp, and give you a chance to take apart my arguments (if you can). But lets talk about the TECHNICAL, not sonic issues.

Regards
Dan Lavry

[Terry Demol]

danlavry wrote on Sat, 21 May 2005 20:48

Terry Demol wrote on Sat, 21 May 2005 03:28 danlavry wrote on Thu, 19 May 2005 17:06 "Hi Dan,   ....I'm sure you are aware of all the usual defences... ...and I'm still seated firmly in camp one...   Cheers, Terry I said: Instead of arguing the merits of Case 1 and the merits of Case 2, let me ask you to do so. Can you (a design engineer) convince me (a design engineer) with a TECHNICAL ARGUMENT that Case 1 is better than Case 2? You replied: I'm sure you are aware of all the usual defenses but they can usually, as you have stated with the latest gen of good sounding chips, be very well refuted with measurements. Terry, Please do not assume that I am aware of any of the “usual defenses”and also do not assume that all the other readers here know any of the “usual defenses”. I asked you if you can come up with a TECHNICAL ARGUMENT that case 1 is better than 2, and you completely side stepped it!   Yes, I heard many arguments about why case 1 (open loop) is better, and some of the arguments were unbelievably ridicules. I asked you to build a case why the open loop approach is better (with good solid reasons). You repeated the old story: you believe that specs are not everything, and that you believe that there is something else that makes good sound that we do not know. When I made medical amplifiers, supplies, telecom gear, instrumentation equipment... there was always a rather clear understanding of what makes for good results, and it could always be reduced to solid specifications. It was not always easy to measure, but the goal and the parameters were always very clear. But since I have been doing audio, a long time now, the statement “it sounds better but I do not know why” has been part of the landscape. As long as “But I do not know why” is an integral part of the “equation”, the people that advocate an open loop approach, can not sell their method as a fundamentally better way to go. All they can say is: “so far it sounded better”, which is a very subjective opinion, and can be a self serving one. Here is a challenging statement for you:  when it comes to transparency, measurements do count, and waveform in = waveform out is the goal.   I, too, have done my share of “low feedback” as well as “much feedback” designs. I too have listened a lot, and have my sonic conclusions. So instead of you coming out with SONIC arguments (it sounds better) which I am completely aware of (and there was no need to repeat), why don’t you list what the benefits of the open loop approach are. Let us see if I can take apart those arguments. I promise that I will follow such an interaction with a list of benefits for the “lots of feedback” camp, and give you a chance to take apart my arguments (if you can). But lets talk about the TECHNICAL, not sonic issues.

Well you can challenge other peoples low/0 FB designs
all you like however, remember we have designed a 0
global/interstage FB amplifier stage that measures better at
25V RMS OP and a gain of 30 than what your Gold box does at
around 5V (ref stereophile test DA2002).

They're measuremts.

Our DA analog stages are much more linear again (less gain).

So it appears you are knocking our design approach
but we are getting *linearity* as good or better than
you are acheiving with closed loop designs anyway.

Sorry to get blunt on you Dan but I don't really take to
that agressive style you have adopted and I don't suffer
fools easily.

Best you find someone else to argue with.

I'm happy to have intelligent, curteous, even if technically
opposed *discussions* not heated arguments.

Cheers,

Terry

[danlavry]

"Well you can challenge other peoples low/0 FB designs
all you like however, remember we have designed a 0
global/interstage FB amplifier stage that measures better at
25V RMS OP and a gain of 30 than what your Gold box does at
around 5V (ref stereophile test DA2002).

They're measuremts.

Our DA analog stages are much more linear again (less gain).

So it appears you are knocking our design approach
but we are getting *linearity* as good or better than
you are acheiving with closed loop designs anyway.

Sorry to get blunt on you Dan but I don't really take to
that agressive style you have adopted and I don't suffer
fools easily.

Best you find someone else to argue with.

I'm happy to have intelligent, curteous, even if technically
opposed *discussions* not heated arguments.

Cheers,

Terry

Terry,

First, I did not challenge your design. You stated a good distortion number at 1KHz and 25V an I what did I say? I said:

“I agree with much of what you said, and indeed, a 0.0007% at a gain of 30 is great. I do not know what the performance is at other gains and frequencies, but I am sure the design is very fine.”

Given that I did not see your specs, my comments are very nice!

My experience with “no feedback” is such that you can optimize a design at one condition (such as 25V and 1KHz), and when you go to different amplitudes, temperature or frequencies, the specs get much lower. In fact you, yourself, mentioned that at higher frequencies, the specs go down a bit. You come here talking about 0.0007% and that is great. I have built a closed loop attenuator breadboard this weekend with 0.00019% at 1KHz, at unity gain with less than 2.5uV noise over 0-55dB attenuation range. It was all done with closed loop approach (much feedback), and measured by Audio Precision test system (with it’s closed loop internal amplifier circuits, like all the test gear I know of) but I do not consider it a foundation to an argument about the concept of feedback. Neither should you put all on a specific design when arguing a CONCEPT.

But I did not ask for details about your specific design or any specific design. Again, my statement was “I am sure the design is very fine” is not a challenge to your design. So why get so defensive?

I, too, have designed many circuits for sonic advantages. I, too, believe that an engineer needs to know what specs are more important, and what trade-offs to make. Therefore, I did not take issue with your looking into adding some harmonic distortions, and stating it to be for sonic reasons.  

I was suggesting what you said you will be happy to do:

have intelligent, courteous, even if technically opposed *discussions* not heated arguments.

This is exactly what I offered. An intelligent, courteous discussion, not heated arguments, and yes about opposed point of views. The discussion is not about specific gear, not is it appropriate to compare a power amp (all analog) against a DA (both digital and analog). The discussion is about a CONCEPT of feedback, the pluses and minuses (if any).

I stated very clearly that it is the end result that counts, and I would not and do not attack a piece of gear that accomplished the goal, be it open loop, close loop, made with bipolar, FET’s, or carrot juice in a sardine can. You can not say, I am attacking your gear, which I do not know anything about.

Again, there are many in audio that seem to consider the method of getting the results as important as the results themselves. One such issue has been the use of negative feedback. In my may years of audio, I have not met anyone that is willing to take on the challenge of TECHNICALLY defending the statements against too much feedback, or explaining why the use of little feedback is a preferred method.  The arguments were, so far, all SONIC - “it sounds better”.

There is nothing heated about what I am saying or said in this thread. But to use your style: “Sorry to get blunt on you” Terry, but I did not expect you to stand up and defend the open loop approach. I was 99% sure you would decline. For 25 years I did not hear a single technical reason why feedback is bad. I am extremely interested in hearing the TECHNICAL arguments supporting the notion that less feedback is better. I suspect there are none.

Regards
Dan Lavry
www.lavryengineering.com

[Terry Demol]

danlavry wrote on Mon, 23 May 2005 19:41

"Well you can challenge other peoples low/0 FB designs all you like however, remember we have designed a 0 global/interstage FB amplifier stage that measures better at 25V RMS OP and a gain of 30 than what your Gold box does at around 5V (ref stereophile test DA2002). They're measuremts. Our DA analog stages are much more linear again (less gain). So it appears you are knocking our design approach but we are getting *linearity* as good or better than you are acheiving with closed loop designs anyway. Sorry to get blunt on you Dan but I don't really take to that agressive style you have adopted and I don't suffer fools easily. Best you find someone else to argue with. I'm happy to have intelligent, curteous, even if technically opposed *discussions* not heated arguments. Cheers, Terry Terry, First, I did not challenge your design. You stated a good distortion number at 1KHz and 25V an I what did I say? I said: “I agree with much of what you said, and indeed, a 0.0007% at a gain of 30 is great. I do not know what the performance is at other gains and frequencies, but I am sure the design is very fine.” Given that I did not see your specs, my comments are very nice! My experience with “no feedback” is such that you can optimize a design at one condition (such as 25V and 1KHz), and when you go to different amplitudes, temperature or frequencies, the specs get much lower.

Dan,

We did not experience this. The fundamental effect of
incresed THD with increased gain and level was observed.
The circuit performed very much as I predicted, actually
quite a bit better... I was somewhat shocked when I measured
it.

Quote:

In fact you, yourself, mentioned that at higher frequencies, the specs go down a bit.

Only slightly, at 20kHz around 0.0016% I just dug up my notes
(this was some time ago) At 40V RMS it is about 0.0026% at 10kHz

In comparison most opamp based approaches will have
significant degradation of linearity at higher frequencies
due to the internal miller compensation C reducing OLG.
looking at data sheets up to 10x.  

Quote:

You come here talking about 0.0007% and that is great. I have built a closed loop attenuator breadboard this weekend with 0.00019% at 1KHz, at unity gain with less than 2.5uV noise over 0-55dB attenuation range.

Closed loop / unity gain / 1kHz is akin to "down hill with a
tail wind". With careful design and good EE practice adhered
to, using something like an AD797 and low enough R values to
minimise Johnson noise will bring these numbers.

Quote:

It was all done with closed loop approach (much feedback), and measured by Audio Precision test system (with it’s closed loop internal amplifier circuits, like all the test gear I know of) but I do not consider it a foundation to an argument about the concept of feedback. Neither should you put all on a specific design when arguing a CONCEPT.

I was quite clear in stating that:

a) The usual argument against 0 or low FB is poor linearity;
hence my design example of excellent linearity.

b) The usual argument against 0 or low FB is that the poor
linearity results in euphonics and that is why people like it;
again hence my example.

c) I was quite clear in stating that, quote; "it is difficult
to have a really bullet proof technical defence of case 1"
ie; low or 0 FB.

You appear to have not read any of this.

You said "I am polarizing the discussion into 2 extreme cases,
for the sake of raising a point (or presenting a question)."

*YOU* are polarising the technical discussion; not me. I have
already clearly stated it is difficult to technically
defend 0 x FB design approach.

*I* am merely stating that I choose to use 0 or low FB
designs because so far, they sound better. Period.

I have done plenty of stuff with AD797's and similar.
And I can easily design curcuits that will measure around
-120dB by using existing designs and adding FB. They already
have incredible open loop linearity. Even a bit of localised
FB will put existing designs very well into this territory.

But the bottom line for me (us) is sound AND numbers.
Audio Engineers rarely ask me about numbers. They just
plug it in and listen.

Quote:

Again, there are many in audio that seem to consider the method of getting the results as important as the results themselves. One such issue has been the use of negative feedback. In my may years of audio, I have not met anyone that is willing to take on the challenge of TECHNICALLY defending the statements against too much feedback, or explaining why the use of little feedback is a preferred method.  The arguments were, so far, all SONIC - “it sounds better”.

Here are some that make some sense:

a) Less susceptibility to RF
b) Less susceptibility to slew induced distortions when used
with RF heavy signals such as DAC I-V's etc.
c) Ability to drive any load without the need for OP decoupling.
d) Flatter distortion versus frequency curve as mentioned above
e) More benign distortion character ie; usually H2 + H3 and
the rest are much lower (depends on design). Often high global
FB designs have increased upper harmonics
f) Low TIM as discussed by Otala / Cordell etc
g) Very fast settling time (no miller C).

But all of these issues can be effectively got around with
closed loop designs with careful attention.

There are numerous others including the usual statements such
that the OP cannot correct the IP because the IP has already
happened... somewhat ridiculous.

And recently I saw some multi tone tests that were exploring
distortions due to dynamic phase shift of closed loop systems.

However the results were still pretty inconclusive.

Quote:

There is nothing heated about what I am saying or said in this thread. But to use your style: “Sorry to get blunt on you” Terry, but I did not expect you to stand up and defend the open loop approach. I was 99% sure you would decline. For 25 years I did not hear a single technical reason why feedback is bad. I am extremely interested in hearing the TECHNICAL arguments supporting the notion that less feedback is better. I suspect there are none.

Dan, please read the prior posts carefully, it will save me a
lot of time, of which I have very little currently...

I must apologise for my somewhat heated reply, I had spent
18hrs straight upgrading a mix console, finish 5am Mon morning
so I have been a little short on patience.

Regards,

Terry

[danlavry]

"I was quite clear in stating that:

a) The usual argument against 0 or low FB is poor linearity;
hence my design example of excellent linearity.

b) The usual argument against 0 or low FB is that the poor
linearity results in euphonics and that is why people like it;
again hence my example.

c) I was quite clear in stating that, quote; "it is difficult
to have a really bullet proof technical defence of case 1"
ie; low or 0 FB.

*YOU* are polarising the technical discussion; not me. I have
already clearly stated it is difficult to technically
defend 0 x FB design approach.

*I* am merely stating that I choose to use 0 or low FB
designs because so far, they sound better. Period.

Regards,
Terry"
[/quote]

Your reply is very fine, and of course I will have to take time to answer all that you put in the last massage. My initial reply is to restate that I am not talking about a specific design, certainly not about your design that I did not measure myself. Also, I have no issue with how one goes about achieving results.

You are correct to say that I am polarizing the discussion, and I clearly stated so. My suggestion to polarize the discussion is to provides what I consider a very good and clear format, which is a service to this technical forum. I am not insisting that you are the one to be the defender of one side, but you are here, you said that you prefer camp 1, and I tend to side with camp 2...

Here is a point that needs much clarifying in my view: (do not be offended, my writing is aimed to reach many of the less technical types in the forum)

Negative feedback is a concept, where one feeds some (or all) of the output signal out of phase to the input. The more output signal one feeds back, the closer the matching of the output waveform to the input waveform. Therefor, amplifying the signal by a huge amount enables better matching of the waves.

It is correct to observe that in the real world, the act of adding huge gain my introduce non linearity to the device or circuit, but the greatness of the negative concept is that when closing the feedback loop (applying a lot of feedback), the non linearity, including that of the open loop gain, is hug-ly reduced. The idea is, in conceptual terms, to just have enough gain over the bandwidth of the signal.

The real world practice of negative feedback may take a lot of forms. There are numerous circuit architectures and device types out there. A stand alone silicone bipolar transistor is a very non linear device. Put the emitter to ground and drive the base at .4V and there is zero current. At 1V the transistor is saturated. There is a very narrow range of operating voltage where a few milivolts will make a large current change. It is all very non linear, temperature dependent, and the device tolerances are often in the hundreds of percent...

We have learned a long time ago how to overcome those problems. Transistors (bipolar or Fets), can be made to behave when we use them in circuits with negative feedback. The first practical circuit called for insertion of an emitter resistor. Such a practice amount to a localized negative feedback, greatly overcoming the grounded emitter circuit limitations. The performance limitation of a highly non linear, device dependent, temperature dependent .2V input range can be made to be many volts range of almost linear, almost device and temperature independent behaviour...

Of course, there are limitation to how far one can correct that transistor. Some of the limitations are in circuit design and values, but one limitation is the transistor device itself. A transistor current gain is not infinite, thus one is limited in terms of how huge the feedback can be.

So designers often go through more than a single amplification stage to build up enough gain. Again, conceptually, the more gain the better. Of course there are practical limitations on how much gain one can achieve, though some OPamps are offering 120dB gain (one million).

At this point I am not arguing the merits of a few stage's with localized feedback in each stage against the "opamp approach" which is one overall feedback path. I am just trying to point out, for starters, that a lot of people in audio believe that negative feedback is an undesirable factor, but the circuits used ALL HAVE NEGATIVE FEEDBACK, and plenty of it!

It would be good of you to help me make it clear to the readers, that the statement "No negative feedback" is fundamentally incorrect. It should be substituted by "different circuit implementations of negative feedback". No negative feedback means no resistors in the emitters or sources. I did not see your circuit, but if you took out all the negative feedback mechanisms (including current feedback), what would happen to the performance?

I heard a well know power amplifier maker claims no negative feedeback. All he does is to "convert input voltage wave to a current wave, and steer the current into a high value resistor to build a lot of voltage...". Well, how do you make a low distortion voltage to current conversion? A circuit with negative feedback...

So that is my first objection to the "no feedback" crowed. That "no feedback" and "feedback is bad" should be changed to "feedback is what saves us from the inherit non linearity of cheap pieces of silicone or tubes".

I am not saying the you personally are running around saying feedback is bad. Others do, and you can help make it clear that feedback is good, it is used by all, and the issue of when and where is a circuit design issue.

Regards
Dan Lavry
www.lavryengineering.com    

 

[Johnny B]

Dan,

Surely you are not meaning to deride and disparage tubes or valves, are you?

[danlavry]

Johnny B wrote on Wed, 25 May 2005 17:53

Dan, Surely you are not meaning to deride and disparage tubes or valves, are you?

I am not sure how you connect tubes to the discussion here, but my position is pretty consistent regarding all the parts and components. Tubes, like anything else, need to be looked at in context to the rest of the design and also with respect to the goals one wishes to achieve.

What I mean by "the rest of the design" is: One may linearize a tube to the point that it will sound much closer to a semiconductor, or one can choose to bring out the non linearity. One can have a design where the sonics is overwhelmingly influenced by the coupling transformers (typically required for tube designs), and so on...

A strong example for my statement about "wishes and goals": a tube amplifier for electric guitar is often made to utilize some tube specific saturation characteristics. That is as much a part of the sound as the guitar itself. I can not possibly have any objections to tubes in such context.

But when it comes to transparency (keeping the wave shape least altered through the gear), tubes are harder to work with than semiconductors. I am not anti tubes, I know they tend to introduce more distortions when used in typical circuits, and some people argue that they like it sonically, which is fine with me. I prefer a more transparent behaviour but I am not into dictating tastes.  

Regards    
Dan Lavry
www.lavryengineering.com

[Johnny B]

danlavry wrote on Wed, 25 May 2005 20:34

I am not anti tubes, I know they tend to introduce more distortions when used in typical circuits, and some people argue that they like it sonically, which is fine with me.

Yes, one might say, "All technology has its place...even the older technology!"

When certain technology is placed in the hands of a truly artistic engineer---great sounding results can be obtained.

We could also say, "Everyone keeps pushing the technology forward...esp. in respect to digital technology."

My only hope is that sound quality does not suffer...that in the procress of all this techno-advancement...sound quality improves.

How I would precisely define "sound quality" or "great" sound quality is something I find difficult, even the act of providing "real life" examples of what I consider great sound quality can be attacked with a logical argument that "what I like" can be reduced to a "subjective" level, hence, of little or no value.

With all things being as equal as possible, I think I still prefer the sound of older recordings over newer recordings, exactly why that is true for me, remains a puzzlement to my small brain.  
 
If I were smarter and far more talented than I am (I'm completely lacking in any form of talent, BTW) I might be able to understand why I prefer older recordings and be able to explain it, but alas, God saw fit to handicap me with a small brain.  








 

[Terry Demol]

danlavry wrote on Wed, 25 May 2005 17:01

"I was quite clear in stating that: At this point I am not arguing the merits of a few stage's with localized feedback in each stage against the "opamp approach" which is one overall feedback path. I am just trying to point out, for starters, that a lot of people in audio believe that negative feedback is an undesirable factor, but the circuits used ALL HAVE NEGATIVE FEEDBACK, and plenty of it! It would be good of you to help me make it clear to the readers, that the statement "No negative feedback" is fundamentally incorrect. It should be substituted by "different circuit implementations of negative feedback". No negative feedback means no resistors in the emitters or sources. I did not see your circuit, but if you took out all the negative feedback mechanisms (including current feedback), what would happen to the performance?

HI Dan,

Yes, technically you are correct.

My understanding is that if really correct terminology is
used, then it is impossible to make any amplification  circuit
without negative feedback (unless it is atransformer).

I will clarify; even a single triode that has the cathode
grounded (through a capacitor) and voltage input will have a
finite OP impedance due to the tubes internal negative
feedback. Dave Collins sent me a nice technical paper on this
subject (thanks DC).

So I believe there are only various levels of feedback.

The most common is global feedback. A proportion of the OP is
sent back to the IP but out of phase. This is probably the most
lamented form of feedback. Possibly the claim has grounds due
to it being the most invasive circuit wise. Global feedback
needs to have some kind of control on the open loop gain to
prevent oscillation due to phase shift,  at some frequency
the signal sent back to the IP will be + phase instead of -
because it has had sufficient phase shift through the device.

I will not get into phase margins, dominant pole, miller
compensation  etc... probably too techo for the purpose of this
discussion.

Next down the line is interstage feedback. This works on the
same principle but the OP is only sent back out of phase
between two adjecent stages as opposed to a whole opamp
(usually 3 stages or more).

A classic example of interstage FB is the CFP transistor pair
(complimnetary feedback pair) where a simple 2 transistor
arrangement is such that the OP of one device is fed back to
linearise the first device. These interstage FB arrangements
work very well and can linearise a simple circuit by a factor of
10x to 50x even more.  

Next down the line again is a single stage that has
degeneration. This is called local feedback. For a simple
transistor, a resistance is added to the emmitter which has
the effect of making the base to emitter voltage more linear
WRT current and temperature.

I suppose next down the line is a gain stage that has no
degeneration. Generally speaking, this is not workable
with bipolar transistors because they are too non linear,
somewhat workable with fets and quite workable with tubes,
especially triodes.

In fact some triodes show quite incredible linearity with no
degeneration. The 6SN7 comes to mind... from memory it can do
leass than 0.1% THD with 10V RMS and no degeneration.

Getting back to terminology, generally when people speak of
zero feedback it will mean no global or interstage feedback.
But usually some amount of degeneration is used which is
technically local feedback.

Also people will refer to OP emmitter followers as zero
feedback but technically they have 100% local feedback.

Most opemps or gain stages can use combinations of all the
above. Power amplifiers are a good example.

There are many other ways of linearisation, probably too much
to go into here; cascoding, bootstrapping, feedforward
correction (such as Hawksfords OP error correction) and more.

Quote:

I heard a well know power amplifier maker claims no negative feedeback. All he does is to "convert input voltage wave to a current wave, and steer the current into a high value resistor to build a lot of voltage...". Well, how do you make a low distortion voltage to current conversion? A circuit with negative feedback...

Exacty... and it becomes an argument of terminology.
The voltage needs to be buffered numerous times before being
able to drive a speaker which is all emmitter follower type
circuits. See above.

Quote:

So that is my first objection to the "no feedback" crowed. That "no feedback" and "feedback is bad" should be changed to "feedback is what saves us from the inherit non linearity of cheap pieces of silicone or tubes".

The real argument is within various types of FB. When I used
our amp gain stage as an example I was careful to say no global
or intestage FB. This is what is comonly referred to as "zero
feedback" but as can be seen here it is technically a misnomer
and complete threads have been wasted purely trying to establish
a correct terminology for "zero feedback" We don't need that here.

Cheers,

Terry

[danlavry]

"HI Dan,

Yes, technically you are correct.

My understanding is that if really correct terminology is
used, then it is impossible to make any amplification circuit
without negative feedback (unless it is atransformer)".

I do not see it as a terminology issue. I view it as a circuit design issue.

"I will clarify; even a single triode that has the cathode
grounded (through a capacitor) and voltage input will have a
finite OP impedance due to the tubes internal negative
feedback. Dave Collins sent me a nice technical paper on this
subject (thanks DC)."

The resistor at the triode is for DC biasing, and yes it is negative feedback action. Actually, the capacitor shunting the cathode biasing resistor is put there to eliminate the negative feedback for AC signals (audio). Why? Because a triode does not have much gain (pentode do not have much either). Removing the shunting capacitor, the tube would offer little gain, but much less distortions! The cap is there because needing gain was more important part of a compromise (between gain and distortions).
I did see some past designs that provided better specs, relied on the feedback of the tube’s cathode resistor, and of course required a lot more tubes to build up the required gain.

"The most common is global feedback…
Next down the line is interstage feedback…
Next down the line again is a single stage that has
degeneration.... "

Thank you for taking the time to “break down” the feedback to “global”, “interstage”, “single stage”…  Personally, I do not find such a breakdown to be very important, unless one includes many other specific hardware factors into the picture. Nor do I think that the comments about phase or stability can be addressed out of context to issues such as desired specs and performance, specific devices being used…

Frequency stability is an issue, until it is being resolved. The same statement holds for all other performance parameters, be it noise, distortions, flatness and the rest… Yes, Miller capacity is a major problem for some applications, and a no problem for other cases.

I really do not see how high quality OPamps and non OPamps circuits based on “global feedback” can be labeled inferior. Yes I recall the days when 1MHz bandwidth was a big deal. Today we have devices go up many thousands of times further…  We need not carry the parameters of 1950 technology on our backs…

The transistors and FETs used inside OPamps are made of very high frequency (ft), the time delay through the devices is very short, some Opamps rival the best discrete parts, (certainly so for low power applications)… It is all about circuit designs and packaging, with some advantages to OPamps and others to discrete.

But let’s get back to feedback:

There is a conceptual advantage of what you call “global feedback”. By conceptual I mean an advantage that is independent of contemporary devices. That advantage is fundamental. Say I need an analog amplifier with 20dB gain.

I can do it by having a few stages of amplification with one feedback path utilizing the combined gain of all the stages within the feedback loop (you called it global).

I can do it in 4 stages, each one having its own individual feedback loop. Each stage (with its individual feedback) feeds the next stage (with its own individual feedback).

Which way is better?

The concept of negative feedback (assuming you can have a circuit with all the desired characteristics, such as stability) works best and does the most good when you “pile up” a lot of gain between the input and the output. Say you have an open loop gain of a million (120dB), which obviously takes some stages in series. Given that the design calls for 20dB gain, you have 100dB of negative feedback at your disposal. That 100dB, when used for negative feedback, is going to reduce the distortions due to “anything inside the loop” by a factor of 100000. And there are a lot of factors inside the loop that you want to correct for. The cross-over distortions of the output stage, the temperature dependent issues relating to all the components… So say you have a 10% distortion before you closed the loop, it becomes a 1 part per million problems (.00001%).  That is why that OPamp you mentioned (AD797) works so well. It takes advantage of the fundamental concept of using as much gain as possible between input and output, what you call “global feedback”.

Lets try a different aproach with 4 identical stages of 30dB each. The combined gain is 4*30dB=120dB, but each stage has only 30dB local gain, and we need to set aside 5dB per stage for signal gain, leaving only 25dB feedback gain. 25dB is a gain of about ONLY 18.
With such little gain (18) for negative feedback, each and every stage needs to be extremely linear. There is not much feedback to help circuit non linearities. To achieve .0001%, each stage needs to be linear to .0018% before feedback (the feedback of gain=18 will reduce the outcome .0018%/18=.0001%). Any single stage that does not  meet the desired goal will “ruin everything”, and an open loop .0018% stage is a difficult and costly proposition, relative to even a very costly OPamp, or a high loop gain implementation.

Of course my example is a bit extreme, and I stated it so to demonstrate a point.
On one hand, real life cases often have less than 120dB feedback gain, and the main reason is stability (at times, it is very difficult to stabilize more than 120dB of negative feedback gain). On the other hand, most circuit with localized feedback has more than 18 for open loop gain.

One needs to realize that negative feedback is one of the “mothers of all electronics engineering tools”, and that negative feedback action is strongest when there is a lot of gain within the feedback loop. Yes, there are limitations as to how far we can go, but the access gain (gain left for feedback) divides the distortions. An access gain is a good thing, and the more the merrier, as long as other practical considerations do not enter the picture. An argument against a lot of gain (the theoretical perfection achived at gain of infinity) should be acompanied with a reason to limit the gain. So far, I do not see a reason to try and go for low gain (that I can agree with).  

Negative feedback is a powerful concept of creating access gain to be utilized for correcting a slew of problem, and fundamentally, the more gain the merrier. Yes, there are practical limits, but that does not explain why we have to cut the gain way below what we CAN achieve.

Yes, the results is what counts, and I do not take issue with any gear that accomplishes good results. True, I am the one that is polarizing the issue of negative feedback vs. less feedback, though I did not start it – it was Bob Katz comment about low open loop gain. Only one day after receiving your first response, I saw a write-up about some new gear associated with a respected name tougting “low negative feedback” as an asset.

In fact, “if I had a $ for each time” that someone asks me if my gear is or is not using feedback… I feel there is some marketing driven “story” floating around about why less negative feedback is an advantage. And BTW, I believe most customers do not know the difference between global or local feedback.

And even I, with analog design being my first and foremost area of expertise, after over 35 years of circuit design, do not find it easy to draw clear lines between single stage, interstage, global… Add to it current feedback, voltage feedback... I see it as less about “vocabulary” and more as a buzz words floating around in the audio world.

Thank you for your factual response, and I am sure it has some educational value to many on the forum. I am glad to have you agree that at minimum, there are some serious misconceptions floating around, and that indeed feedback IS being used. That certainly narrowed the gap between what we say – we seem to agree on most facts.

Regards
Dan Lavry
www.lavryengineering.com    

[mikepecchio]

hello,

I have a bit of experience with the TL0x chips.  Ive tried these in MANY circuits. Like others have said, the implementation is very important, and they can perform and measure well under the right conditions.

but there are 2 general problems with these particular opamps that are quantifiable.  poor output drive and rising distortion vs frequency.  the first can be made negligible if you design it into the right kind of circuit.  IMO you should not load a TL0x with less than about 10k. this makes it unusable in alot of audio applications. for example, in a virtual earth inverting configuration the feedback loop is in parallel with the load.  a 5k feedback resistor alone will make the opamp's distortion unacceptable even before you connect any load at all.

the other IMO more serious problem, easily measurable, is that no matter the configuration, the THD increases rather dramatically with frequency.  This spec is something you should consider when choosing an opamp.  I personally believe this has a big effect on the subjective sound of an opamp and IMO it isn't talked about enough.

coincidentally, the other opamp mentioned, the 5534, has one of the flattest THD vs. Freq. plots out there.  not intended as an endorsement for that chip, but I do find that part alot more useful.

regards,
mike p

[maxdimario]

Quote:

When I made medical amplifiers, supplies, telecom gear, instrumentation equipment... there was always a rather clear understanding of what makes for good results, and it could always be reduced to solid specifications. It was not always easy to measure, but the goal and the parameters were always very clear. But since I have been doing audio, a long time now, the statement “it sounds better but I do not know why” has been part of the landscape.

This is because an amplifier made for music reproduction is not a specific-purpose circuit such as the ones used in instrumentation equipment and the other types of circuits you mentioned.

The perfect music amplifier leaves intact all of the elements that the HUMAN HEARING APPARATUS -- brain included -- needs in order to give 'meaning' to the sound.

If you don't use your ears as the first and final test in designing, you will never achieve a good audio circuit.

The best audio circuits designed by experts in the field of sound reproduction tend to be very simple high-performance class a discrete, with minimal or no feedback.

this is true in high end hi-fi and in the best sounding recording gear.

the problem is that building an amplifier without reverting to feedback can be very expensive, especially when the load is problematic and varies in impedance depending on the output signal.

[Brian Roth]

Max, two points for me to make.

#1.  *IF* there subtle, yet audible differences between one piece of gear and another, but the measured specs don't reveal it, then our measuring processes are incomplete.  That was the gist of my intial questions in this thread.  If opamps are so hideous, why don't conventional testing methods reveal repeatable numbers that will verify what listeners perceive?  If these audible differences exist, then they can me measured if the right tools are used.  What are those tools????

#2.  ALL amplifiers I've ever seen use some sort of negative feedback, including such things as unbypassed cathode resistors in tube stages.

Bri

[maxdimario]

Brian,

specs need to be interpreted, and a circuit which is engineered on specific tolerances will work towards achieving better specs.

what if I were to say that .5% THD in a musically 'correct' circuit is more effective musically and realism-wise than a lesser circuit of .00002% THD?

Anyone with focus on specs would immediately 'correct' the circuit for fear of too high distortion, and lose the realism.

there are issues in sound reproduction that have not been addressed for the last 40 years, in part because of the implementation of neg feedback as a solution to most costing and engineering problems, in part because of the shift towards solid-state which has a tipically higher distortion per active component, and in part because of the emphasis on specs instead of ability to reproduce a sound convincingly.

As Dan said above, personal opinion can make things difficult for the engineer (who approaches from a purely technical standpoint).

you can get around these things once you start listening, but not by relying exclusively on specs.

There were early amplifiers that used special quality tubes and no feedback.



I believe that a moderate amount of local dc feedback does not harm the music too much and makes complex sounds sometimes more defined. In addition solid state amps need feedback because of the higher distortion per active component.

But the less feedback you need the better, as a rule of thumb.

to make a stable class a amp with no feedback you have to spend a lot of money on power supply not to mention individual selected components.


in a way multitracking has been fuelled by lesser circuits that make sounds 'smaller', because the small sounds lend themselves to layering in arrangements etc. as opposed to fewer, bigger more 'complete' sounds.

As far as there being 'subtle' differences between gear I disagree. There is a huge difference given that the microphone, preamp, summing amps, monitoring are all to very high standards.

on a lesser system everything resembles everything else.


There are significant differences between the authority a good discrete circuit has and the kind of audio path that has been used liberally in studios for the last 25 years.

[Crispin HT]

Dan, Terry, hi.

I picked up on this thread by chance, and thought I'd just add a bit:

Feedback is necessary to linearise an active design.  However if the basic approach (Global feedback) were perfect, everything should sound the same if you could match the circuit topologies.

Therefore perhaps we should look at what makes a Feedback system imperfect.  I suggest the following:
1) Lack of "enough" open loop gain at the frequency of use.
2) Inherent non-linearity of the device before feedback is applied.
3)Linearity of the passives used in the feedback process.
4)Time related factors, settling of outputs, impulse behaviour.
5)External factors outside the Feedback loop: grounding, DC, return currents, instability into loads, input overload, thermal and noise factors, and of course circuit topology.

In my experience once you've got a basically sound amp, it's point 5) that affects the performance, not the Feedback itself.

Cheers
Crispin HT

www.Crookwood.com