Why zero feedback sounds better?

Collapse
This topic is closed.
X
X
 
  • Time
  • Show
Clear All
new posts
  • Dennis H
    Ultra Senior Member
    • Aug 2002
    • 3791

    Why zero feedback sounds better?

    I haven't waded through it yet but this guy wrote his Masters thesis on the subject.



    ABSTRACT
    A NEW METHODOLOGY FOR AUDIO FREQUENCY
    POWER AMPLIFIER TESTING BASED ON PSYCHO-ACOUSTIC
    DATA THAT BETTER CORRELATES WITH SOUND QUALITY.
    By
    Daniel H. Cheever
    University of New Hampshire, December 2001
    There exists general agreement that the commonly accepted test and
    measurement protocols for audio frequency power amplifiers fail to correlate with the
    subjectively accessed devices sound quality. A review of the history of audio testing was
    undertaken to reveal if prior art has produced tests that better correlate with sound
    quality. A universal concept emerged, one that calls for stronger weight of the higher
    order, more aurally discordant harmonic distortion products, over the low order, more
    benign harmonics. Separately a study of the psychoacoustics of the ear resulted in a
    mathematical derivation of the ears intrinsic aural distortion. The two are combined and
    offer a methodology for weighing the harmonics based on a dimensionless figure of merit
    that quantifies the amplifier’s harmonic distortion envelopes departure from the ears aural
    masking, named Total Aural Disconsonance or T.A.D. It is shown both analytically and
    through actual device measurements that the application of negative feedback, regardless
    of level, results in poorer T.A.D. figures. Two amplifiers of opposing standard
    measurement results are fully tested and subjectively analyzed and results show that the
    T.A.D. method outperforms classic T.H.D and I.M. for characterizing amplifier quality.
  • Victor
    Senior Member
    • Apr 2002
    • 338

    #2
    Dennis,

    I read through the thesis. It is very interesting in many ways; particularly I liked the historic overview of the amplifier testing methodologies. There are however, as I see it, problems in this thesis. First it is the premise of the thesis itself. Mr. Cheever thinks that we need to account for the ear’s capabilities and sensitivities; I would submit that we absolutely do not.

    The purpose of the amplifier is not to ‘sound good’, - it is to faithfully increase the amplitude of the signal with the corresponding change in current and do that without any spectral anomalies. I also would take issue with Mr. Cheever listening test procedure. He attempted to ‘compare’ the 1.5 W triode amp of the ‘20s era to the quite modern solid state 200 W Hafler. The Blind Listening Test was set-up with 5 subjects. As it turned out all of them preferred the sound of the tube amplifier. From that test Mr. Cheever concludes that there is something wrong with the existing testing methodology and the new approach is needed.

    Well, this conclusion is plain wrong. What does this test show? It shows absolutely nothing of substance. It shows that 5 people preferred the sound of the tube amp, - so what? We all know that most people actual prefer the sound of the kind of a distortion that the tube amps produce. Furthermore, using 5 people in the test is not statistically significant to draw any conclusions. Well, in any case, if I wanted a system that ‘sounds’ pleasing to my ears, why use a power amplifier to accomplish that? It is a lot easier to make a low power circuit that will give you all the distortion that you ears may desire. Mr. Cheever writes that his thesis is

    “…investigation of the hypotheses that the current accepted measurements that qualify fidelity of an amplifier fail to correlated well with a subjective sound quality.”

    Mr. Cheever is then proceeds to suggest the alternative way. I am not at all surprised that there is no correlation between how the amplifiers supposedly sounds and the specs. Reason being that the amplifiers are not supposed to have any sound at all, so the point that Mr. Cheever is raising is rather moot. The essence of Fourier Theory will back up what I am saying and you cannot really argue with that, as Fourier does apply here.

    As you read the thesis you can’t help but wonder that there is something wrong here. My feeing is that Mr. Cheever embraces the idea that an amplifier somehow has magical ‘sound’ characteristics. It is almost like a finger print, unique to each amplifier. Nothing can be further from the truth. He quotes Bob Harley from Stereophile and Mr. Pass and others who have gone on record to support the same ideas. He does not quote anybody else who may disagree with his premises. There are many who will in fact disagree. I am one of them.

    The rest of the thesis is actually a very interesting study of the human hearing which is quite scientific, but I cannot comment on it as I luck the necessary expertise.

    We then arrive at the core of the thesis where Mr. Cheever proposes a novel method of measurement. He calls it TAD, or the Total Aural Dissonance. His math does make sense, and I must say that his TAD is in a way a figure of merit that will tell you how good the sound of an amp is, providing that we can believe that the study of hearing presented in the thesis is universally true. That is, - it is true for all humans. This is a big ‘if’ in my view and is a deal breaker for me. The TAD is a measure of the deviations or the differences of harmonics between the once that the ear ‘likes’ the once that actually generated by the amplifier under test. I must say it is interesting way of looking at hearing.

    In the end, I am convinced that the entire thesis misses the point of what the amplifiers are supposed to be doing and it introduces an unnecessary complexity into the well-established and proven methodology.

    Victor

    Comment

    • Dennis H
      Ultra Senior Member
      • Aug 2002
      • 3791

      #3
      I just skimmed through it too. I think there's some merit to his premise that higher order distortion is much more "disturbing" than lower order and a THD measurement misses the mark as very small amounts of high order distortion can sound subjectively bad. His proposed formula for "equally bad" levels of distortion is approximately proportional to 1/(N^11) where N is the harmonic number.

      I would think you could apply the same reasoning to metal cone drivers with a severe cone breakup as well. Many people describe them a "analytical" sounding. That could be because the cone breakup resonance, even if well above the passband, still manifests itself as small amounts (and it doesn't take much) of high order distortion in the passband.

      Comment

      • ThomasW
        Moderator Emeritus
        • Aug 2000
        • 10934

        #4
        I can't wait to read Jon's reply about that article :smackbutt:




        theAudioWorx
        Klone-Audio

        IB subwoofer FAQ page


        "Complicated equipment and light reflectors and various other items of hardware are enough, to my mind, to prevent the birdie from coming out." ...... Henri Cartier-Bresson

        Comment

        • Dennis H
          Ultra Senior Member
          • Aug 2002
          • 3791

          #5
          I'm looking forward to Jon's reply as well. He's been a big proponent of no/low feedback designs but I've never really understood why. Less higher order distortion, even if it's at the expense of more lower order distortion, seems to offer a possible explanation for why they sound better.

          Comment

          • JonMarsh
            Mad Max Moderator
            • Aug 2000
            • 15270

            #6
            Well, it's nine thirty Saturday night, and I just got back from my business trip- haven't read the thesis quoted above, but I did download it for future reference.

            Just off hand, comparing a 2 Watt triode amp and a Hafler solid state amp sounds kind of dufus even to me....

            He may have a relevant idea, but a very weird way of trying to demonstrate it. :roll: Besides, everyone knows that if you want to compare feedback to non feedback, don't mess around with some silly glow bottles, just go get yourself an Ayre, Theta Digital, or BAT. :LOL:

            Seriously, though, even good old Peter Baxandall (he of the famous tone control design from the fifties and many other things) did some basic analysis, and his conclusion was that even the application of moderate to modest amounts of negative feedback could be a significant problem, because what it does quite literally even in the best of cases is take low order distortion products and push them up to higher orders. Now, if you have very low distortion to begin with, and if you can safely use a LOT of feedback (especially, wide bandwidth), then you can reduce the measured distortion to very low levels in the audio bandwidth.

            For an example of this, see Robert Cordell's paper on a low power wideband MOSFET amplifier which he claimed to use feedforward error correction. To my eyes, it doesn't- but it does use a nested feedback loop around the bias and output stage, and then a fairly high gain and wideband "main" feedback loop. In effect, for the range out to about 35-40 kHz, the roll off rate is two pole, and the gain quite high. Halcro's "patented" circuits look suspiciously similar in some regards, so I have to wonder how their patents might hold up to a challenge.

            You know, the only reason I go on about this biz about NFB amps is just because they sound very clean and transparent- and I've yet to hear a conventional design using much feedback that sounds as natural. That doesn't mean you can't have an enjoyable musical experience, and let's face it, with many program sources (average CD players, movie sound tracks) it's almost a case of GIGO.

            But ya gotta know, it's interesting how much more listenable almost all sources are with a system configured that way- you couldn't get me to go back. I mean, my Aragon's make pretty good subwoofer amps, but I'd never listen to them full range anymore unless it was because my V5 Ayre was being sent back to the factory for an update- which should happen in a week or so! :B

            Even some things that might be considered "torture test" CDs', like the Hooters, or Avril Lavigne sound clearer and less fatiguing.

            So, to do a real test, I'd suggest comparing something like a Bryston, Krell, or Aragon against an Ayre, Theta Digital, or BAT amp. Let's face it, it would help if the guys making the more conentional amps had to go to the lengths that the NFB folks do in order to get the overall linearity without using a feedback loop. It's not easy, but I'm pretty sure that if you did it that way, then used a teeny bit of feedback, you'd be a lot better off than with the more conventional circuit approaches otherwise used.

            Nelson Pass is kind of in that area in between, but even he acknowledges that you may well like the versions of his circuits which use less feedback better. And while John Curl does design "conventional" high grade feedback circuits for customers like Parasound, he's told me all his own stuff is NFB these days- and he acknowledges the influence Charles Hansen and his efforts and successes at Ayre have had over the years.

            Just trust your ears, Luke.


            ~Jon




            Earth First!
            _______________________________
            We'll screw up the other planets later....
            the AudioWorx
            Natalie P
            M8ta
            Modula Neo DCC
            Modula MT XE
            Modula Xtreme
            Isiris
            Wavecor Ardent

            SMJ
            Minerva Monitor
            Calliope
            Ardent D

            In Development...
            Isiris Mk II updates- in final test stage!
            Obi-Wan
            Saint-Saëns Symphonique/AKA SMJ-40
            Modula PWB
            Calliope CC Supreme
            Natalie P Ultra
            Natalie P Supreme
            Janus BP1 Sub


            Resistance is not futile, it is Volts divided by Amperes...
            Just ask Mr. Ohm....

            Comment

            • Victor
              Senior Member
              • Apr 2002
              • 338

              #7
              I do not think that feedback is all that bad. In fact I would go as far as to say that a linear system in an audio sence cannot exist without feedback. I have yet to see a truly ‘feedbackless’ design that works. Well, I think I must define what ‘works’ for me. In a nutshell, - the THD+N down to at least 16 bits is all I ask for.

              I know that there are amplifiers (power or line level kind) out there that claim not to have any feedback or very little of it. Little feedback is all that is necessary to produce a successful design. However, ‘no feedback’ design approach will be very hard-pressed to produce THD numbers better then 12-14 bits. The notable exception is the ‘buffer’ type circuits that have no gain. They are useful as line drivers but nothing else.

              Another possible approach is a ’current conveyer’. Strictly speaking it has no feedback, however it is very difficult to get it to work and consequently there is only one such IC on the market that I am aware off. Building this topology with discrete devices is possible but will not produce desired results as the device matching will have to have ‘stratospheric’ precision.

              Now, let’s come back to design that I have seen that claim to have ‘no global feedback’. Global is the operative word here. This statement gives a perception that there is no feedback at all. This is not so. There is feedback, just it is not global. Here is what is commonly done; - the amplifier circuit is naturally broken into 2 parts. The first part contains gain and it indeed employs a feedback. The 2-nd part is an output driver with no gain and it has no feedback. Hence the design can be seen as lucking Global Feedback. However, as we all see this is really not the case, as the feedback is used where it is needed, - around the gain stage.
              The examples of such approach are many, the famous Pass A75 comes to mind as well as many BAT designs and there are others. The penalty paid in this approach is the inferior THD performance by at least an order of magnitude or worst comparing to the same amplifier with global feedback.

              So, my position is that there is no way to design a low-noise, high precision amplifier without the feedback. In many ways the feedback is an ‘icing on the cake’. The 'cake' or the amplifier in question must be linear to begin with, before any feedback is applied. So it is entirely doable to design an open loop system with a THD at just below 1%. Applying ‘very little’ feedback would make this system to show a THD of 0.01% without any problems.

              Victor

              Comment

              • Victor
                Senior Member
                • Apr 2002
                • 338

                #8
                ..double post, - sorry.
                Vic

                Comment

                • Dennis H
                  Ultra Senior Member
                  • Aug 2002
                  • 3791

                  #9
                  what it does quite literally even in the best of cases is take low order distortion products and push them up to higher orders
                  I'm thinking maybe metal cones do a mechanical version of the same thing. Seas shows 2nd and 3rd order harmonic distortion in their published sweeps but nothing above. It would be interesting to see sweeps up to 10th order for something with a nasty breakup, like a W22, driven to loud volumes. Those high-order distortion products would land smack in the middle of the midrange - 4500/N.

                  Comment

                  • JonMarsh
                    Mad Max Moderator
                    • Aug 2000
                    • 15270

                    #10
                    Hello Victor,

                    First, your point re some none global feedback designs is correct, if you're considering some older Denon's which have used this technique, or the Borbley audio designs.

                    In the case of the Ayre and Theta amps, familiar as I am with their topologies, that's not the case at all. Each individual stage uses local degenerative feedback, such as emitter or cathode or source degeneration, and also use linearizing techniques such as cascode or folded cascode operation. There is no feedback loop around the input stage or the output stage in Ayre designs. In some of the Theta designs, there is a DC servo because of the temperature/bias issues with the MOSFETs used as drivers. This is the same approach I use in my own DAC, which uses a DC coupled circuit, lots of local degeneration and linearization, NO feedback loops, and rather high voltage power supplies for a 4 VRMS balanced output - plus and minus 30V. Believe me, it's loafing at normal output levels. And because it runs high class A current in a collector output configuration, all cable loading does is reduce the bandwidth in the region above 200 kHz; there's no possilibity of the local oscillation or back rectification of RFI products at the output.

                    Screw theory, though; just go listen. That's all it takes. Every one I've demo'd my Ayre amp for has bought one... so if you're not in the mood for considering a new purchase, don't listen to one. :W



                    Dennis, re the motor's in loudspeakers, the distortion products inherently are pretty low order. The stuff you see when you finally get up to the cone resonance is just that- resonance. It's non non-linearity in the BL transfer function of the driver or the driver suspension. If your crossover properly filters the HF area, the output at the HF peak can be over 70 dB down- this is my experience using the HiVi metal cone woofers also.

                    This is why one of the tests I do with finished crossovers is to listen to the woofer with low pass filter by itself, as well as measure it- this removes the masking effect of the tweeter, and any grunge or grit from the mid or woofer will be readily identifiable.

                    Regards,

                    Jon




                    Earth First!
                    _______________________________
                    We'll screw up the other planets later....
                    the AudioWorx
                    Natalie P
                    M8ta
                    Modula Neo DCC
                    Modula MT XE
                    Modula Xtreme
                    Isiris
                    Wavecor Ardent

                    SMJ
                    Minerva Monitor
                    Calliope
                    Ardent D

                    In Development...
                    Isiris Mk II updates- in final test stage!
                    Obi-Wan
                    Saint-Saëns Symphonique/AKA SMJ-40
                    Modula PWB
                    Calliope CC Supreme
                    Natalie P Ultra
                    Natalie P Supreme
                    Janus BP1 Sub


                    Resistance is not futile, it is Volts divided by Amperes...
                    Just ask Mr. Ohm....

                    Comment

                    • Hank
                      Super Senior Member
                      • Jul 2002
                      • 1345

                      #11
                      Thomas, you Poobah, I was hoping you'd jump on that!
                      Just off hand, comparing a 2 Watt triode amp and a Hafler solid state amp sounds kind of dufus even to me....
                      Yes, for starters, apples/oranges doesn't work. I have two Hafler DH220's that I bought years ago. The first I "Poog'd" (who recognizes that?). The second is in a box awaiting me to install the Musical Concepts driver boards I bought a couple of years ago (okay, I got sidetracked into speaker building ops: ). I have my H.H. Scott LK-48-B integrated tube amp kit that I built as a mere lad of 17. Apples and Oranges.
                      This concept of NFB vs a little bit or a modicum of feedback is interesting, but guys, engineers as you are, music is even more complex. I play clarinet - you wouldn't believe the complexity of its overtones (beats the oboe 8O ). Sheesh - gimmee boxes that reproduce the complex structrue of a real instrument and I'm yours...or should I say, my dollars are yours.
                      Whew!

                      Comment

                      • Dennis H
                        Ultra Senior Member
                        • Aug 2002
                        • 3791

                        #12
                        Well, if the guy is anything like most of the grad students I knew, he couldn't afford the power cord of an Ayre, let alone a whole amp. He probably ate Ramen noodles for a month to pay for his little home made tube thingy so we should probably cut him a little slack.

                        To me, the most interesting part was the graphs of the ear's natural harmonics (harmonics not present in the music.) I take this to mean how loud the different distortion harmonics need to be to sound equally loud. In other words a 3rd order distortion, about 20dB quieter than a 2nd, sounds just as loud; a 4th, 30dB quieter than a 2nd, sounds just as loud; etc.

                        Comment

                        • Victor
                          Senior Member
                          • Apr 2002
                          • 338

                          #13
                          Jon,

                          Each individual stage uses local degenerative feedback, such as emitter or cathode or source degeneration, and also use linearizing techniques such as cascode or folded cascode operation. There is no feedback loop around the input stage or the output stage in Ayre designs.
                          This very interesting. I can see how the cascoding techniques will linearize the overall operation. However, what puzzles me is the lack of the overall feedback. I would imagine that the gain would have to be kept rather low for the system to behave properly. Also, that would mean that more then 2 signal stages maybe necessary before the power transistor buffer. Hmm… now as I think about it perhaps you can get away with 2 stages. I still think that you will be very hard pressed to get a THD better than 0.5 % with that approach. I would love to see the schematics!

                          In some of the Theta designs, there is a DC servo because of the temperature/bias issues with the MOSFETs used as drivers. This is the same approach I use in my own DAC, which uses a DC coupled circuit, lots of local degeneration and linearization, NO feedback loops, and rather high voltage power supplies for a 4 VRMS balanced output - plus and minus 30V.
                          Well, the DC servo does nor really qualify as a feedback in this discussion, so naturally it does not count. Jon, I hope I am not asking for too much, but can you describe in more detail the circuit you use. Naturally the schematics will be even better; - I hope I am not asking for too much.

                          I see… so you are not using the source follower type outputs. In your case you opted for the high impedance connection. Yes, I can see that the cable loading will become an issue in this case and cable length must be kept to a minimum. The feedback on the other hand would have helped you there as it would have reduced the output impedance. On the other hand you do not have any issues with the distortion generating source-follower connection.

                          Believe me, it's loafing at normal output levels. And because it runs high class A current in a collector output configuration, all cable loading does is reduce the bandwidth in the region above 200 kHz; there's no possilibity of the local oscillation or back rectification of RFI products at the output.
                          In fact I agree that this connection is preferred as long as you do not have to drive anything that is excessively capacitative, but still I think that without the feedback the output impedance will create the low-pass filter effect right inside the 20 kHz envelope. I am actually surprised that in your circuit this effect is only around the 200 kHz. This is great result. What output devices are you using? I would love to see your circuit!

                          Victor

                          Comment

                          • JonMarsh
                            Mad Max Moderator
                            • Aug 2000
                            • 15270

                            #14
                            Hi Victor,

                            Don't have access to my personal stuff from at work, but I can provide a brief description-

                            First, like the Ayre modules, my NFB DAC/line amplifer is fully balanced input to output. The input stage uses paralleled 2SK389 and 2SJ109 in a sort of complementary differential configuration; the trick is in the arrangements for source degeneration and balancing. One version I've experimented with uses a more complex arrangement with multiple transistors and source biasing so that the input differential stage is actually A/B; unlike a conventional differential stage which only has a fixed bias current available, and a basically S shaped transfer function with it's attendent non-linearity at higher input levels, this balanced differential stage will increase the output drive with increasing differential inputs linearily. So, for example, the nominal "resting bias" could be in the range of 2 mA, with drive increasing to 4mA. Those are arbitrary values, and in practice the currents depend on the drain/collector load impedance you want to see driving the next stage.

                            Paralleling is done for noise rejection as well as having the desired IDSS capability.

                            The following complementary stage uses ZETEX FZT796A and FZT696B; these are very high linearity, low output capacitance 200V devices; again, paralled devices are used. The output collector load for the common stage sets the ouptut impedance of the line state/DAC buffer; this is a relatively low value, so the effective output impedance broad band is quite similar to most conventional designs using Op Amps and small series resistors to protect the OpAmp and limit capacitive loading. Remember, most so called wide band low THD op amps will most definitlely NOT tolerate connection to several nanofarads of capble capacitance; a few do, by means of having a capacitor from the output pin back to the voltage gain stage, so that the open loop gain is rolled off for stability. Mind you, these op amps don't run anything like the kind of class A current that this line stage uses- typically 1/10th to 1/40th.

                            The ouptut is balanced push pull, as is the input.

                            Not having a loop, there are no loop stability issues. With a moderate output capacitance (a couple of hundred PF, typical for good cables in many applications, inherent bandwidth of the circuit is way beyond the audio spectrum.

                            The power amp circuits are basically similar, but more complex in the use of cascodes and biasing arrangements. The output stage I've been working on uses the extended beta On Semi transistors which are copies of the equivalent Toshiba; a high frequency compensation method I borrowed from my old RF days for a project for a consulting customer 5 years ago stabilizes the output stage for use with capacitive loads without a series output inductor. The Ayre amps use the same technique, as Chas found it worked very well also. I'm still trying to figure out how to cram all the circuitry in the PCB space available in my Aragon amps, so I can retrofit them. It's just that the circuitry is a bit more complex, and also uses much more local regulation. And I've got a couple of speaker projects to finish up first!

                            For understandable reasons, I don't really want to publish detailed schematics about these circuits. I may write an article on them for AudioXpress- I'm still debating that.

                            Again, whether this might be your cup of tea would best be answered by demoing an Ayre, Theta, or BAT amp. Just my opinion. If you use a digital source, I'd recommend one with similar approaches and low jitter electronics. The cheapest really good CD player I know of is still the Ayre CX7, though a Music Fidelity A324 is very good for a more conventional design.

                            Best regards,
                            Jon




                            Earth First!
                            _______________________________
                            We'll screw up the other planets later....
                            the AudioWorx
                            Natalie P
                            M8ta
                            Modula Neo DCC
                            Modula MT XE
                            Modula Xtreme
                            Isiris
                            Wavecor Ardent

                            SMJ
                            Minerva Monitor
                            Calliope
                            Ardent D

                            In Development...
                            Isiris Mk II updates- in final test stage!
                            Obi-Wan
                            Saint-Saëns Symphonique/AKA SMJ-40
                            Modula PWB
                            Calliope CC Supreme
                            Natalie P Ultra
                            Natalie P Supreme
                            Janus BP1 Sub


                            Resistance is not futile, it is Volts divided by Amperes...
                            Just ask Mr. Ohm....

                            Comment

                            Working...
                            Searching...Please wait.
                            An unexpected error was returned: 'Your submission could not be processed because you have logged in since the previous page was loaded.

                            Please push the back button and reload the previous window.'
                            An unexpected error was returned: 'Your submission could not be processed because the token has expired.

                            Please push the back button and reload the previous window.'
                            An internal error has occurred and the module cannot be displayed.
                            There are no results that meet this criteria.
                            Search Result for "|||"