Real audio science - exploring beyond the obvious and bringing back true science and exploration

Discussion in 'Audio Science' started by purr1n, Jul 7, 2023.

  1. Psalmanazar

    Psalmanazar Most improved member; A+

    Pyrate Slaytanic Cliff Clavin
    Joined:
    Sep 27, 2015
    Likes Received:
    5,345
    Trophy Points:
    113
    Compressors are not time invariant. Neither are many amplifiers subject to heat.
     
  2. caute

    caute Lana Del Gayer than you

    Pyrate Contributor
    Joined:
    Jul 12, 2022
    Likes Received:
    1,991
    Trophy Points:
    93
    Location:
    The Deep South
    ya, glad jason and mike just didnt give up on the ol' mother-in-law when it didn't sell like lim and og, and are actively working on it to make it sound better...

    edit: re: measurements—sorry i'm so off topic guys, but can anyone link me (in PM so it doesn't crap on this excellent thread) to the awful TDA1543 measurements?
     
    • heart heart x 1
    • Agreed, ditto, +1 Agreed, ditto, +1 x 1
    • List
  3. Garns

    Garns Friend

    Pyrate
    Joined:
    Jul 9, 2016
    Likes Received:
    2,484
    Trophy Points:
    93
    Location:
    Sydney, AUS
    In this context time invariant means something very weak. With your equipment in a known steady state, you can feed it an input signal and record the result. Returning the equipment to the same steady state, you now re-record the same signal with a delay on the input. The resulting output should be the same apart from the added delay.

    Any piece of gear with no modulation should meet this criterion including most compression. If it was linear then such a device would have to behave in a history free way but because it can be non linear this still allows for memory effects of various kinds.

    Now you would need extremely long and computationally impractical kernels to accurately capture a compressor via Volterra series, sort of like how you need very long kernels to capture a linear reverb, and in practice this does not reflect how our ears work. Any piece of music is in principle captured by a single FR and phase chart but our ears hear it in a windowed way with FR that changes over time because they are not pure Fourier analysers. Similarly Volterra kernels will not be efficient at detecting time domain biased nonlinearities like in compressors. However they should be good at detecting frequency dependent but history free or very short history nonlinearities. It's unclear if this is what is going on in amps but this would potentially be a way to diagnose this. An interesting question is whether there is a nice basis for the space of time invariant nonlinear transformations which is biased to some extent to time domain effects (sort of like higher-order wavelets).
     
  4. Psalmanazar

    Psalmanazar Most improved member; A+

    Pyrate Slaytanic Cliff Clavin
    Joined:
    Sep 27, 2015
    Likes Received:
    5,345
    Trophy Points:
    113
    Compressors are volume modulators by definition. The easiest way to build them since the 70s is to use voltage controlled amplifier ICs that were originally used in synths. They do not have a steady state with any non-continuous signal above threshold. The signal is always being turned up and down. What is going on internally in the DC sidechain cannot be captured. The process is too non-linear for an accurate Volterra series. The Volterra series convolution processors are awesome for modeling distortions but they cannot be as dirty as the real things or possible capture the degradation of many circuits over years!
     
  5. Pancakes

    Pancakes Friend

    Pyrate Contributor
    Joined:
    Aug 13, 2020
    Likes Received:
    1,420
    Trophy Points:
    93
    Location:
    Atl
    I believe the issue with noise beyond the hearing range is how it effects downstream components - ie, preamps and amps. I think the very high frequencies can tax the output devices and therefore degrade performance. So it's not the DAC itself that sounds different - it's that it makes your other components sound different. But, I may also be completely wrong lol.
     
    • Agreed, ditto, +1 Agreed, ditto, +1 x 1
    • List
  6. Polygonhell

    Polygonhell Acquaintance

    Joined:
    Nov 27, 2019
    Likes Received:
    73
    Trophy Points:
    18
    Location:
    Redmond, WA
    I think it's always interesting to go back to the language we use to try and describe the differences to see of there is anything there that might indicate what we could measure.
    comparing DAC's, While I clearly hear them, I've always found difficult to describe the differences.
    Resolution, Clarity, Weight, Separation, Control, Slam all get thrown around, and there is probably some agreement on what most of those imply perceptually.

    I'd probably start with resolution, because the perception of it almost certainly not related to the resolution of the signal, my current "if I had the time/gear I'd measure it to see" theory, is it's about transients at low input levels, I think we gloss over a lot about behavior changes at differing volumes.

    On the High frequency debate, I think this is a combination of possibilities, just because you can't hear a sine wave at 20KHz doesn't necessarily mean we can't perceive a difference signals with 50uS rise times and 10uS rise times.

    I also read a piece recently that suggested at frequencies under 4KHz, we are sensitive to absolute phase, because there are distinct signal paths for each direction the little hairs move in. This was obtained by measuring the signal, do it doesn't speak to relative importance, but perhaps changes in phase response are a part of the picture.
     
    Last edited: Jul 13, 2023
  7. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,888
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Haha. Don't need to have measurements because Philips even admits this chip is garbage in their datasheet "economy version". -75db THD spec.

    upload_2023-7-13_10-24-4.png

    --

    The vintage Philips chip to get is the TDA1541. Crown / double crown to up to -95db THD or so. The TDA1543 is garbage. It was garbage back then and garbage today. It's criminal that boutique vendors make DACs from this chip with freakshow implementations like tube rectifier and sell for several thousands of dollars. The only reason the x43 gets mention is because of the cachet of the x41, of which the high-spec versions of unobtainium. The x41 actually sounds very good, although nowhere as resolving as modern DACs.

    In general, and this is in respect to correlating measurements to what we hear and only in the context of DAC chips, THD has a moderate correlation to what I've described as microdetail or "plankton" (ability to resolve low level information), but only up to a certain point of which there are no audible benefits. Also some manufacturers / implementations are able to extra more "plankton" than others for a given chip.

    One thing I've found with steady-state measurements is that are more relatable / comparable / usable with taken within the context of a closely related group of gear type. That is THD is a good measurement, but best when used when comparing (that is distinguishing observable sonic differences) tube amps with tube amps, solid-state amps with solid-state amps, DAC with DACs (or even DS DACs with other DS DACs, and R2R DACS with other R2Rs).
     
    • Like Like x 4
    • Agreed, ditto, +1 Agreed, ditto, +1 x 2
    • List
    Last edited: Jul 13, 2023
  8. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,888
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    One of the reasons for SBAF was to establish a common vocabulary: https://www.superbestaudiofriends.o...logy-subjective-terms-used-on-superbaf.11665/

    The idea is that people would be going out and meeting each other, comparing notes, describing sound. When it comes to describing sound, a shared vocabulary is crucial. The idea was that a core group of people, those with experience, those who attended meets locally or across the country, would provide a foundation for such, especially for the smaller and rather tight community we have here. The understanding that it wouldn't be perfect, but it would be better than random words from random people.

    Many would be surprised at the consistency of what people hear when all variables are controlled for. I've never seen anyone disagree on attributes of sound when a large or small group of people hear a system at the same place and same time. Although personal preferences to those attributes will differ - but not as much we would think. It's typical that for any piece of gear, 60-75% of listeners will agree on a winner of four competitors. 90% of people will agree on two of the four being better than the rest.

    There is a reason why Schiit has been doing the "Thunderdomes". We should start doing more stuff like this in meets and document the results. There's no reason why subjective impressions cannot be used in science. Those who think subjective impressions, however imperfect they are, cannot be used in science are dumb. (How da F do we think painkiller effectiveness is measured? Duh!). I do think we can be more rigorous with data collection, higher number of samples, more studies, better documentation. And even then, who cares? This is audio, not medicine. We don't need 99% statisical confidence levels. I'd be happy with 85% confidence. And even then, some people will not able to hear the difference.
     
    • Like Like x 5
    • Agreed, ditto, +1 Agreed, ditto, +1 x 5
    • List
    Last edited: Jul 13, 2023
  9. Polygonhell

    Polygonhell Acquaintance

    Joined:
    Nov 27, 2019
    Likes Received:
    73
    Trophy Points:
    18
    Location:
    Redmond, WA
    Oh I agree, but it's somewhat orthogonal to my point.

    I understand the common vocab used to describe things, and yes I agree it's critical, I'm suggesting using those as a starting point to establish measurements that in someway map to them. I don't believe that when we talk about improved "resolution", we're discussing just how much information exists (in the information theoretic sense) in the output signal.
    But rather than we perceive more information in the signal, and it's that perceptual gap I think it's interesting to try and close.

    The terms we use often imply a mapping to specific technical measurements but they aren't there when we look at those measurements. As an example better DAC's tend to present without grain in the treble, but any DAC that's not actually broken will measure flat from 20-20KHz. We can subjectively agree that a DAC has better Bass control or even quantity, but again it's not in the graphs. I'm not suggesting that were not hearing the differences, but rather we map them to the language we have, which can be misleading in what we need to measure to see them in that measurement.
    So I'm suggesting picking something looking at the language used (since it was picked for a reason) and trying to ascertain what else it might be.

    In retrospect, perhaps this is obvious.
     
  10. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,888
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Yup and agreed. Frequency response has a huge correlation to what we hear. However, it's also extremely limited. Those familiar with me know that I've been harping on "Frequency Response being da man" movement in the past year or two. FR to me means jack shit (well other than it's boring). We can take the HD600, HD800 and LCD-2 - EQ them so they have the same frequency response, yet they will still sound totally different. The HD800 eq'd to have the same frequency response as the HD600 will still have oddities in the highs. An HD600 EQ'd to have the same FR as the HD800 would be a sonic disaster but still without the HD800's soundstage. The LCD-2 EQ'd to HD600 sounds horribly bright (and just plain wrong). These headphones just want to do their own thing. Trying to EQ them too much (a little bit is fine) is like a gay person trying to pray the gay away. It just don't work.

    With respect to measuring DACs' observed "treble Fuckery", I suspect moderate correlation to garbage in the ultrasonics (pre- analog filtering). As to what kinds of garbage, it's hard to say. Multibit DACs, which tend to sound less harsh in the highs have plenty of garbage in the ultrasonics, but very a specific kind of a shit. Delta-Sigma DACs have less ordered random shit. With respect switching power supplies, many a good audio gear designer I know have attributed bad sounding (specifically limp-dick, fuzzy) switching supplies to have switch frequencies too low, like in the hundreds of kHz instead of megahertz region. With respect to output transformers, I know Craig feels that bumps in the 50-60kHz region can result a sharper prickly highs. The list goes on and on and as I mentioned, all the decent audio designers have their own secrets or hunches that they have developed over time from measuring and listening to their designs.

    The idea going forward, if we really want to talk about science, should be more data, and even more importantly, more creative processing and better visualization / presentation of data to facilitate correlation of measurements to specific subjective aspects. Not an easy task and requires lots of tweaks with large datasets of gear where the sound aspects are known or largely agreed upon. And even then, expect to his a lot of dead ends.

    For such dead ends, we already got one: AmirNAD, where anything over 60-75db is a dead-end. This number doesn't tell us anything. I've got a 55-60db tube amp that sounds cleaner, more resolving, and less colored (that is probably more similar to a THX789) than a 120db AmirNAD Topping L30 or Magnius. Heck, AmirNAD tells us completely NOTHING about the MIL and MIL-B.

    Sound Science is not slapping shit on an APx555 and producing bar graphs of top performers' SINADs, all well over 100db, to the tenth of a decibel. This approach is dogma / religion.
     
    • Like Like x 6
    • Epic Epic x 1
    • List
    Last edited: Jul 13, 2023
  11. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,888
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    And then there's a lot we don't know (at least not yet) even where to start with respect to measurements. Like WTF measurement shows us how Nitsch Magni Piety sounds more tubey than the Vali+? Or why x2, x4, x8 oversampling sound DACs slightly different. Or why the heck does the Vali+ sound so close to the Magni+ (even closer than the Magni Heresy) despite SINAD / AmirNAD of the Vali+ being a magnitude or two worse than the Magni+.
     
  12. gepardcv

    gepardcv Almost "Made"

    Contributor
    Joined:
    Nov 20, 2015
    Likes Received:
    408
    Trophy Points:
    63
    Location:
    Terra, Sol System
    A few years ago, I built a switching box for comparing DACs. It was just 2 pairs of XLR inputs running to a pair of 4-gang volume pots into switches into a pair of XLR outputs. The idea was to do level matching by voltage using a multimeter behind the volume pots and be able to easily switch which DAC outputs into an amp. Mac OS supports coupling audio output devices, so feeding two DACs from one source was easy.

    Well, it turned out that I couldn’t find the right part to switch both channels from DAC A to DAC B at the same time, and instead put in two switches: individual right channel and left channel. And I had some crazy results with that. For some pairs of DACs, running one channel from DAC A and another channel from DAC B produced a weird sound effect I can’t even describe. It’s kind of like an out-of-phase sound, with pressure between the ears. One of the DACs I was testing was the RME ADI-2 and it has a polarity switch — so I’m pretty sure I used it to rule out different polarities on different channels.

    I’ll be the first to admit that my switching box is flawed. Volume pots, especially 4-gang units, do not perfectly match. I could get to within 0.01V on R+/R-/L+/L- signals, but not much better than that. Maybe I messed up the polarity switch on the RME or maybe it didn’t work. I no longer have the Soekris dac1541 that I compared with the RME. I also reused one of the volume pots in another project. So I can’t (easily) retry the experiment.

    Still, it’s something I have never heard of anyone trying: running the left channel through one DAC and the right channel through another DAC. With some DAC pairings, there was no difference, but some produced the crazy and very audible effect I described. Switching the signal to run from one DAC made the effect go away.

    It made me wonder how much of “DACs sound the same” is more a function of the way the brain interprets signals coming from the ears. When the signals are inconsistent from left and right, as they might have been with separate DACs feeding the two channels, the brain gets confused and differences between DACs become more apparent.

    Just thought I’d throw this out into the conversation. There’s probably a better way to try this experiment and see if it produces robust results.
     
  13. Josh Schor

    Josh Schor Friend

    Pyrate Contributor
    Joined:
    Jul 13, 2020
    Likes Received:
    931
    Trophy Points:
    93
    Location:
    Ann Arbor Mi
    I appreciate the science, even if it goes over my head. Its not a science question, wondering how the new dac compares to Gungnir MB a2
    best,
    Josh
    I appreciate the science, even if it goes over my head. Its not a science question, wondering how the MIL-2 dac compares to Gungnir MB a2
    best,
    Josh
     
  14. Armaegis

    Armaegis Friend

    Pyrate BWC
    Joined:
    Sep 27, 2015
    Likes Received:
    7,533
    Trophy Points:
    113
    Location:
    Winnipeg
    Years ago I used to listen to sine sweeps and freehand draw what I heard. The things I drew did not quite match FR curves or CSDs, which reshaped how I evaluate those things for myself.

    Then a decade later I tried again and things were so wildly different again. My ears got older. The things that used to bug me don't anymore, and some things that didn't before now do. My own data points are so transient.
     
    • Agreed, ditto, +1 Agreed, ditto, +1 x 1
    • List
  15. peef

    peef Friend

    Pyrate
    Joined:
    Oct 5, 2015
    Likes Received:
    311
    Trophy Points:
    63
    Location:
    Montreal
    This lines up with my experience, too, when designing equipment and measuring finished pieces. When comparing two iterations of the same amplifier, built around the same topology, with tweaks such as improved voltage/current sources, op points, etc., the better measuring amp will always sound better. I've not yet run across a sitation where an amp became insipid because it was too thoroughly optimized.

    On the flip side, I've had no luck with producing insightful measurements that compare different amp topologies. Douglas Self (Mr. Negative Feedback) writes a lot about the importance of linearizing a circuit before applying feedback. I think some of what we're hearing is differences in open-loop performance that feedback isn't effectively correcting, but I'm not sure how we'd go about testing it. I think this might be in play with the TI chip, too. Gain tends to fall with frequency, so perhaps the older generation was running low on feedback at the top end of the audio band.

    Is this something we want to start doing as a community? I've taken to building two of each amp with daughterboards for back and forth listening/measuring and found it to be a very insightful exercise. I'd be happy to design some thing simple for us to pass around, like what we did with the Pass H2 injector if there's interest.
     
    • Like Like x 7
    • Epic Epic x 1
    • List
    Last edited: Jul 14, 2023
  16. Grattle

    Grattle Friend

    Pyrate
    Joined:
    Feb 17, 2017
    Likes Received:
    1,018
    Trophy Points:
    93
    Location:
    USA
    I believe I’ve read that extraneous ultra high frequency sounds can damage tweeters.
     
  17. ergopower

    ergopower Friend

    Pyrate
    Joined:
    Mar 21, 2018
    Likes Received:
    821
    Trophy Points:
    93
    Location:
    South Central PA
    This is a very interesting article that maybe belongs in this thread. Current research on defining the components of sound.
     
  18. Pancakes

    Pancakes Friend

    Pyrate Contributor
    Joined:
    Aug 13, 2020
    Likes Received:
    1,420
    Trophy Points:
    93
    Location:
    Atl
    Don't play low frequencies through them and there's nothing to think about.
     
  19. yotacowboy

    yotacowboy McRibs Kind of Guy

    Pyrate Contributor
    Joined:
    Feb 23, 2016
    Likes Received:
    10,833
    Trophy Points:
    113
    Location:
    NOVA
    Home Page:
    I think it was on HTguide's DIY forum where a handful of folks had been experimenting with some of the newer Be dome tweeter that push their peak resonance up into the ultrasonic range. By implementing a low pass at a point just below the 24-26kHz-ish peak, distortion in the pass band fell quite considerably. I'll dig to find a link but it was a couple years ago around the time the Bliesma tweeters started showing up in the US. Makes me wonder what a 20-40k sweep on Utopia would look like and if an inline notch at the Focal Be peak would change the distinct twinge I felt in my molars while listening to Utopias on a few occasions...
     
  20. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,594
    Trophy Points:
    113
    Location:
    near Munich, Germany
    Hey, I feel this way about tweeters, even.

    I also find it kind of weird that we always use individual harmonics (D2/D3/D4,...) vs frequency plots for transducers. And sometimes at different levels like 70 to 100dB. At least that's what I do....
    Yet for DACs suddenly 1kHz THD at 0dBFS is all we need.

    I find D2 bumps far more benign than D3 bumps, for example.

    However I think in your example of an LCD-2 sounding different vs an HD600, I think that's only partially true, partially a wrong measurement and in part also a compromise because the right measurement depends on HRTF and varies by person. But up to 3-5kHz I think we should get some solid agreement, so most of the midrange can sound similar. Bass again not so much. Can't push an HD600 to do LCD-2 bass. Same as resolution, liveliness, texture in the mids. So overall yes, but tonality can get rather close.

    I find a better example might be you can't make a 10"+1" sound like a 5"+1". Totally different dispersion. You can't emulate OB sound, either. You can't make speakers sound like headphones unless you're some Princeton Professor selling a sound processor.


    Maybe I should use my cheapo oscilloscope on the output of DACs and amps. It correlated really well with my impressions of the power stations when including measurements of DC offset, etc.
     

Share This Page