1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.

    Dismiss Notice

opamps THD+N/IMD/SNR measurements don't mean jack IRL, so let it go humm'kay?

Discussion in 'Sound Science' started by leeperry, Dec 31, 2011.
1 2 3 4 5 6
8 9
  1. jcx
    KG's knock on bad PNP in op amps only applies to old, dirt cheap "jelly bean" parts - pay a couple of bucks each for parts designed in the last decade, built on vertically isolated, SOI, lift off processes and you get as good a match of NPN, PNP as possible given the different charge carrier mobilities - discretes don't come any closer
    and the modern parts benefit from much better internal designs, driven by manufacturers competing for $ in medical imaging, A/DSL markets wanting low distortion at much higher frequency than audio
    TI's "headphone amp" chip - TPA6120 (rebadged THS6012 DSL driver chip) is "faster" providing closed loop gain at higher frequency than the ft of the output Q KG used in the highly respected DynaHi discrete amp
    typical audiophile rants about feedback usually contain simple falsehoods about feedback properties - like the "good" type of local feedback - "degeneration" doesn't cause harmonic multiplication - it absolutely does cause "new" harmonics - and often the feedback factor for the single device is low so that you are in the "bad" part of the harmonic multiplication curves
  2. Willakan

    It would, were it not for the fact that these high-order harmonics are not a problem with even cheap audio opamps and that these high-order harmonics are readily measurable. Other than that...you don't really believe the stuff linked on this thread supports what you've said, do you? You have read it other than to pick one or two lines to quote out of context later? Right?
    It is easily verifiable that feedback is not implemented badly in the cheap audio opamps you despise (it is used in such vast quantities as to diminish all distortion to very low levels) and hence your argument is redundant. But we knew that already.
  3. leeperry

    No worries! As you prolly noticed, english isn't my native tongue so stuff gets lost in translation from time to time. My point was that each time I was sold gear based on killer specs, I felt lied to. And then an EE who only designs discrete gear explains me how easy it is to get killer specs out of opamps disregarding all subjective SQ aspects altogether...and then this guy on gearslutz says that no one has ever proven that IMD/THD/THD+N specs could translate into human perceptions. Reason why I created a thread here, in the hope that someone would come up w/ hard data to support this. I'm not an EE and most of what you mentioned goes way above my head, still I'm getting sick and tired of all those self-proclaimed audiophile companies that don't even bother making elaborated sales pitches anymore...they simply cut/paste the DAC chip specs(à la Calyx/Sabre) et voilà, naive customers are bagged [​IMG]

    Oh god, this chip sounds bad....these headamp IC's chips are horrid...shrill, narrow SS, über bright. Team Discrete to the rescue! and I don't care that THD is 0.000000000001% image.gif
  4. Currawong Contributor


    The problem with this "unlikely to be audible" comment I see regularly is, if you're listening to music, then you're hearing everything, including the distortion, regardless if it is so small a person couldn't be reliably tested to discern the difference. I didn't like what I heard using my friend's Benchmark DAC 1's headphone outputs, which was one of the older models that used 5532s, so I'm curious if there is a measurable explanation. If the performance changes under load, introducing distortion, then this could be a possible explanation for my displeasure, especially as many headphones don't have a flat impedance. This has nothing to do with what I can and cannot determine under testing, as I'm well aware other, unexpected displeasure in the arrangements of my audio rig were below, or even way below the threshold at which I'd be able to reliably determine the difference under testing. The point is that people roll OPAMPs in various components and like, or don't like the results of the often subtle changes. A person doesn't have to pass a test for their feelings to be valid, but it might be interesting and valuable to understand what is going on through measurements.
    If you, and others, are going to continually declare anything below DBT-threshold audible to be invalid, which is what you've been doing in effect, maybe I'll have to declare any technical discussion of science or electrical matters from anyone not suitably qualified and/or experienced to be invalid and all discussion of DBT from anyone not a formal expert on applying it equally invalid, which is going to take almost everyone out of this discussion, even me! [​IMG]
  5. maverickronin

    No one's proved we don't all live in the matrix either.  Proof is for math and alcohol.  What we have is evidence that such measurements predict human perceptions.
  6. Willakan


    This is assuming the changes are not produced by bias mechanisms. I'll admit there is a grey area where changes can be so difficult to perceive that confirming them is very difficult without incredibly exhaustive blind testing, which most of us won't have time for.
    The reason I can be confident here is because the statement "unlikely to be audible," in this case, is staggeringly conservative. Running with the example: Let's take the absolute worst-case distortion of the NE5532 into loads it is only just designed to deal with at audible frequency extremes: 0.003% into 20khz. If we look at the audibility of THD, the classic claimed threshold is 1%, which we'll say is a bit on the high side - to give the audiophile the benefit of the doubt, you can use the thresholds done when tests were conducted with artificial tones - music makes distortion far more difficult to hear due to the masking effect, whilst distortion can be heard with pure tones in (large) fractions of a percent. Let's also forget that 20khz is hardly easy to hear, let alone judge distortion at - you can probably get away with over five times as much distortion as the midrange.
    So if we take the worst case distortion into a very high frequency, treat that frequency as if it is much lower AND use the thresholds normally specified for test tones: 0.003% is still much too low to hear. Furthermore, "hearing" is used here in its most analytical sense - this is the sort of hearing that requires listening very carefully to sounds of very short duration to distinguish a minute difference, not listening to something and, say, criticising the sound of the treble. This is the sort of difference whereby if it was "inserted" into your system between listening sessions you are very unlikely to notice it.
  7. Paul Clark

    KG's 2004 take on the TPA6120
    How does it compare with "The Wire" - LME49600?
    The LME49600 is the ideal solution for high output, high performance
    high fidelity head phone amplifiers. When placed in
    the feedback loop of the LME49710, LME49720 or
    LME49740 High Performance, High Fidelity audio operational
    amplifier, the LME49600 is able to drive 32 Ω headphones to
    a dissipation of greater than 500mW at 0.00003% THD+N
    while operating on ±15V power supply voltages.

  8. Chris J


    No,  discrete circuits need feedback to reduce distortion too.
    It's nowhere near that simple.
    There is one school of thought that goes like this:
    build a very linear open loop amp (i.e. an amp without feedback) then reduce the distortion even more with feedback.
    This is from some old Bryston technical paper.  Yes, it's a little off topic...........[​IMG]
    Strictly speaking you can build an Op Amp from discrete components or even vacuum tubes.
    I have a very old schematic for a vacuum tube Op Amp in .pdf form.


    No soup for me!
    yes, they use lots of feedback.

    This is the partial definition of an ideal Op Amp
     - infinite open loop voltage gain (i.e. gain before feedback)
     - infinite bandwidth
     - infinite input impedance
     - zero output impedance
    reference:  IC Op Amp Cookbook, Walter Jung
    in the real world Op Amps have an Open Loop gain of approx. 110 dB at 10-100 Hz, open loop gain typically drops off at higher frequencies.
    The difference between Open Loop gain and Closed Loop Gain (Closed Loop gain is the gain after feedback, i.e. the desired gain) is the amount of feedback applied to the Op Amp.
    So if you have a Closed Loop Gain of 20 dB then you would have 90 dBs of feedback at lower frequencies.
    At 10 kHz you would still have approx. 30 dB of feedback.
    reference: data sheet for LF353,  an Op Amp I basically picked more or less at random.
    I invite you to look at the Open Loop gain responses for your favourite Op Amp!
    Edit:   have you ever seen the distortion for an Op Amp when loaded down with a 32 ohm Headphone?  Just wondering...........
  9. maverickronin
    You should check the how the NJM4556s in the Objective2 do...
  10. jcx
    I don't think the TPA6120 is the end all, be all of headphone amps but it can be pretty good where it Vswing, Current output ability satisfies the headphone load sensitivity, Z  drive requirements
    pursuing the last vestiges of distortion I think a little thermal modulation can be seen in the datasheet plots, not surprising with a monolithic amp
    the "answer" is to multiloop - as suggested by Scott Wurcer, written up many times by Walt Jung
    you can choose a "good" fet input op amp and arrange the TPA6120 local gain to totally "unload" the input op amp - it only has to swing mV, uA to control the TPA which does the  headphone driving
    with local sub regulation of the input op amp supplies the input op amp doesn't "see" any load effects, its feedback corrects the few thermal, psrr errors of the output op amp
  11. Chris J


    Nice,  he actually measured and published THD and IMD at specific loads!
  12. Shike
    The problem with opamp rolling is you can not verify the integrity of the circuit the second you do it without measurements.  Furthermore, depending on the manufacture there's some questionable ones out there that I seriously doubt measure even their default builds.  THD+N isn't enough, but when you get loaded and unloaded THD+N, IMD, FR, SNR, crosstalk, etc. you get a pretty clear picture in comparison.
    If you compare to equally well built amps that offer the same performance in the above listed attributes and level-match them I believe you will have a hard time telling them apart though in a DBT.  To attain the same performance though each circuit will have to be optimized considering the opamp, it's silly to assume one circuit will work well with numerous topologies because they can fit in the same socket.
  13. Paul Clark
  14. JRG1990
    Used properly op-amps won't colour the sound and they will all sound exactly the same like there not even there, there are blind tests between entire dacs where no1 can tell which is which http://www.matrixhifi.com/molingordo5_pc_dac1_beh.htm , entirely different circuits and op-amps and no1 knew which was which, entirely different everything apart from speakers http://www.matrixhifi.com/ENG_marco.htm and no1 knew which set-up was which and your trying to debate theres audiable differences between op-amps.

    Exactly the problem maybe this where the idea that op-amps colour the sound come from,  op-amps have many requirements in a circuit  input bias values, configurations, gains, feedback circuits, load impedances, quiescent currents, speeds/bandwidths, compensation values, operating voltages , unity gain , common mode rejection ratio , power supply rejection ratio , trades offs in noise the op-amp rollers ignore all of these and could push the disortion of the op-amp well past 1%thd making it an audiable difference then by ear this is mistaken for a warmer sound or something silly, or they will cause oscillation which audiable artifacts could be mistaken for more detail and this won't show up with RMAA, its likely the new op that is rolled in won't be stable.
  15. Head Injury

    If it were an ABX test (which tests the ability to tell a difference, not the ability to identify the test products) I expect the pros would have succeeded. When people know what they're supposed to be comparing, they'll make up plenty of crap. And if they fail, it might just be because their assumptions about certain kinds of gear were false. It doesn't mean they didn't hear a difference that is clearly there, like between two completely different violins.
    If you're attempting to use that as a defense against ABX and blind testing in general, your reasoning is flawed.
    So many people try to cite that article and those similar as proof that blind testing is useless [​IMG]
1 2 3 4 5 6
8 9

Share This Page