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

The Great Audio Scam - Anything above $2 buys you features NOT Audio Quality

Discussion in 'Sound Science' started by Denon2010, Nov 6, 2019.
2
Next
 
Last
  1. Denon2010
    Anything Above $2 Buys More Features, Not Better Quality

    Really amazing article and shows just how we trick ourselves into buying nonsense marketing gimmick to pretend that we have better sounding DAC than others.


    What I have personally noticed myself is even $1000 equipment sounds just like my cheap $40 Asus Xonar DG sound card and I couldn't even reliably tell the difference

    https://www.tomshardware.com/reviews/high-end-pc-audio,3733-19.html

    This article really highlights our bias, objective tests will show expensive equipment performing better while actually listening to it in a blind test only proves the "superior devices" are usually never heard by the human ear. So we make up nonsense to justify expensive purchases. The science is clear on this

    Even my built in free $2 onboard DAC in my S7 Edge Samsung sounds identical $250 Asus STX Sound Card when I tested both in a blind test
     
    Last edited: Nov 6, 2019
    atahanuz likes this.
  2. atahanuz
    Hmm
     
  3. Kammerat Rebekka
    Probably not the best place for this thread as the majority of posters in this part of the forum seem uninterested in blindtesting and so forth.
    Personally I’ve come to the same conclusions as the link you posted. I have tried something like 15 dacs ranging from 20 bucks to 3-4000 and heard a multitude of differences sighted...yet once the merchandise was hidden from view? Sounds exactly the same.
    The kicker? Try this project with solid state amps:wink:
     
  4. DutchGFX
    While I do think there is some truth to this, I can't say I entirely agree. There is certainly some diminishing (and often non-existent) returns, but the fact is that a nicer DAC, such as a Chord, uses very fast FPGAs to perform the reconstruction in ways that simple $1 ICs from TI can't replicate. You are, factually, getting a higher quality reconstruction of the sampled audio in this case. Whether or not everyone can discern this is a different debate for sure, but to say "they are the same" is unfortunately not true lol.

    Now, does that mean paying $1000 more means it sounds better? Not necessarily. Consider the original audio scam, the Gaincard. This amplifier was literally just an OpAmp, yet cost something like $4700...
     
  5. bigshot
    I've done a blind controlled listening test between a $40 Walmart DVD player and an Oppo HA-1 with the Sabre chip and they sounded exactly the same. DACs are designed to be audibly transparent. Some have better specs than others, but all those differences are outside of the range of human hearing. If you want to upgrade, upgrade your ears to those of a bat first.
     
    digititus likes this.
  6. gregorio
    1. But is just doing it differently, actually better?
    1a. Your statement raises two questions: A. Presumably, the very fast FPGAs allow more taps to be executed in the reconstruction filter, how is more taps "higher quality reconstruction"? And somewhat related, B. What do you mean by "getting"? Even if we assume that the very fast FPGAs do in fact result in higher quality reconstruction, for us "to get" (receive) that higher quality requires that the combination of the analogue section of the DAC, amp, speakers/headphones and listening environment can actually reproduce that higher quality. If, for example, all that higher quality is down at say the -120dB level (or lower), then it's below the noise floor of the amp+transducer and cannot be reproduced, so we WON'T be "getting" a higher quality reconstruction.

    2. Agreed ... but what about if no one "can discern this", is it still a different debate?
    2a. Clearly, to state "they are the same" is NOT true. To start with, they look different plus they have different price points, etc. The statement "they are all the same" therefore takes this for granted and actually means (or is explicitly stated as) "they are audibly the same". While it's true that at some level all DACs have different audio performance (even the same DAC played twice does), for that difference to be audible requires that it's actually reproducible in the first place (as "B" above) and if it is, then requires it to be within the thresholds of audibility. Unless both of these conditions are met, the statement that they are audibly the same ("they are the same") IS true!

    G
     
  7. DutchGFX
    I should say that I am not an expert in this stuff. I do a decent amount of signal processing for work, but nothing in real time on an FPGA, so I don't have expertise in this area. I really don't have expertise in any area, but here is the "short" story as far as I know

    So, I assume that the filters used in the FPGA are FIR, not IIR. IIR filters are fast, recursive filters, but they don't have linear phase response. IIR filters have sharper frequency cutoffs. In real time, IIR filters can cause problems due to the nonlinear phase, so you can't easily get perfect reconstruction. As a result, I suspect that DACs use FIR filters when possible. Analog RC filters are IIR in nature, hence why people try to avoid coupling caps.

    Now, when talking about FIR filters, we can increase the number of taps in two ways - increasing the sampling rate, and increasing the number of cycles in the filter. Increasing the number of cycles makes the response longer and increases the rolloff rate. Increasing the sample rate allows us to interpolate more accurately.

    I would assume all DACs boil down to, at the final output, a "sample and hold" device. This means that the DAC holds the sample value on the output. You get something that looks like this, with the red being the DAC output.

    [​IMG]


    I would say that aliasing effects are negligible, so you won't benefit much from adding more cycles to the filter. The FPGA can handle many more FLOPS (floating point operations) so we can use that to interpolate more. If we interpolate more, we get a much better reconstruction of the signal.

    [​IMG]

    Note, I did not actually do the FIR filtering, I just took the actual signal and sampled it at the new rate. If we assume the signal is band-limited (which is is), and we assume the FIR filter is reasonably long, this is an OK assumption. The actual distortion from the filter will be small compared to the error in the sample-and-hold output.

    That is correct. If the rest of the chain has a high noise floor, you are totally hosed lol


    Right, my point is that they aren't the same. People saying a TI chip is the same as an FPGA based DAC are kind of claiming that DSD is the same as FLAC. Maybe they sound the same, which is a different debate, but they are inherently different. FLAC is a traditionally sampled signal, while DSD is sampled at a really high rate but only the transition is sampled (I believe).

    Again, this is from a merely technical standpoint. You might not hear a difference, but there is a difference.

    Hope that helps clarify :)

    If you have any follow-up questions please do let me know
     
    SoundAndMotion likes this.
  8. bigshot
    DACs are designed to be used in home audio components. For that purpose, sounding the same is functionally equivalent to being the same. No one is looking at waveforms when they play a recording of a Mozart concerto. They are listening to music. If it sounds exactly the way it's supposed to, then it's fair to call it perfect sound.
     
  9. DutchGFX
    Yes, but this is the sound science section lol. The science says it isn't the same, so it's not the same.

    In the listening impressions section they might be the same, but in the science section they are not :)

    But yes, I would 100% agree that this does not matter. I'm only here to enlighten about the science haha. I didn't like the Stax SR-009 as much as I liked the Ether Flow, which costs much less and doesn't have the technology to compete with the high frequency resolution of electrostatics.
     
  10. bigshot
    If the line level matched, direct A/B switched, DBT says it sounds the same, for all intents and purposes it is the same. If your ears can't hear it, it doesn't matter what the equipment can reproduce. It's a waste of energy to worry about inaudible stuff.
     
  11. SoundAndMotion
    I think this thread is interesting because, like many, I'm interested in the question: is there an audible difference between the "typical" expensive DAC (> $1000, e.g. the Benchmark DAC 2) and a "typical" inexpensive DAC (<= ~ $100, e.g. Topping D30)? This translates into: does more $$$ mean better sound in DACs, in general. I realize many think it's already answered, but then another article is superfluous.

    If you prefer measurements to "audibility" tests, there are measurements of many DACs on ASR.

    I'm not very interested in extrapolating a similar, but different, question: does one particular expensive DAC sound different, even "worse", than a particular cheap DAC. The particular choices for the DACs may be very special cases, i.e. outliers.

    The interesting question of audibility breaks out into 3 sub-questions for me. Is the difference audible to:
    a. me? Obviously interesting to me for purchasing decisions, etc. Maybe others?
    b. the average, typical human [reader, person...]? Interesting to many, including me.
    c. any real human, rare genetic freak or whatever? Interesting to many, including me.

    The article in the OP falls short in answering any of my sub-questions for 2 reasons.

    First, they compare apples to oranges. They test one chip (or chip on a particular motherboard, really), one PCI card, and 2 standalone DACs. If I decide "Gosh, I'd like to use that chip", I'd have to spend much more than $2. I can't just order it from Farnell or Mouser and start listening.

    Second, they tested 2 people. If they tested one who could reliably tell the difference, that would at least be my sub-question c, and maybe hint at further testing to answer b. But 2 subjects with a negative result, answers none of them.

    @DutchGFX Nice graphs, but... With your background in signal processing, you should realize there are several ways you can show yourself that applying an appropriate reconstruction LPF to the red sampled data will reconstruct the blue data exactly. You can do it mathematically (see Shannons work), empirically-both numerically and electronically (or watch someone else do it - link), or with a suitable listening test.
     
    DutchGFX likes this.
  12. gregorio
    @DutchGFX I don't dispute the theory of what you've stated but I do wonder how directly applicable it is in modern DACs. For example your section on filters: In practice even the very first generation of consumer CD players had DACs which x2 or x4 oversampled and in 1992 my first pro ADC/DAC x128 oversampled. At what point are interpolation errors below the level of reproducibility (EG. Are below the noise floor or outside the reproducible frequency range)? Following on ...
    That's the point, "you are totally hosed" because even the DAC itself has a (relatively) high noise floor. A DAC must obviously have an analogue section and therefore, even a theoretically perfect DAC will always have a noise floor higher than the theoretical (24bit) digital noise floor, due (at least) to thermal noise. Most DACs top out at the equivalent of 20bits (or lower), although a few manage around ~21bits, however, that's not reproducible because just the thermal noise in amps and transducers is much higher than the thermal noise in the analogue section of even relatively cheap DACs. So again, this brings us back to what is meant by "getting" higher resolution/performance: If we measure the output of a DAC then sure but of course as human beings we can't "get" the output of a DAC, we can only "get" the output of transducers in a listening environment. And of course, as we're talking about reproducing music recordings, then we have the noise floor of the recording itself to consider: Mics, mic pre-amps, the mixing and mastering processes and of course the noise floor of the recording venue/s, all of which combined adds up to a noise floor that's typically (and very roughly) about 1,000 times higher than the noise floor of most DACs!

    1. Even on a purely technical level, in practice this is a more complex argument. DAC chips are complicated bits of engineering and for many years the vast majority have hybrid topologies. So as far as the actual conversion inside the chip is concerned, DSD or Flac may effectively be the same thing (as most DAC chips use a D-S stage). I'm not of course arguing that the DSD and flac file formats are the same, just asking if at the end of the DAC process there is a reproducible difference.

    2. Firstly, you'll have noticed that my above points all refer to a "reproducible" difference. So it hasn't got anything to do with whether or not "you might hear a difference" because obviously, if it's not reproducible in the first place, then it can't be audible anyway. Secondly, I've already stated that of course there are actual differences, I take that as a given and I don't believe anyone here is contradicting or disputing that fact/the science. The statement being made (either overtly or by implication) is that all DACs are audibly the same, within a set of fairly self-evident (consumer) conditions.

    G
     
  13. DutchGFX
    Of course, but sinc interpolation is non-causal and I don't believe it can be written as a recursive difference equation, so it can't be implemented in real-time. Maybe if you allow for the same number of taps as your signal length you could do the convolution in real-time, but that would be a ridiculous filter order.

    But yes, you are 100% correct - if you sample above the Nyquist rate, the Shannon-Nyquist Sampling Theorem states that you can perfectly reconstruct the signal. This is absolutely a true statement and you are right, I just don't think it can be implemented in real-time.

    My background is in DSP but I haven't been in industry that long, so I definitely don't know everything. I write DSP and navigation algorithms, so I don't deal with audio at all, so maybe I'm wrong about some of this stuff

    This is an excellent point. The thermal noise can definitely be higher than the DAC quantization noise.

    And, I also agree that once you are getting past x128 upsampling you are really probably not going to reap any rewards. Maybe the limit isn't 128, but there is surely a point above which we can no longer discern the difference. The thermal noise, and anything up the chain, are definitely more important at that point.

    I'm interested in how one can go about comparing the DACs. As we said, if we use a rubbish amplifier, obviously the difference in the DACs won't be as noticeable. When people test these DACs, do they use a ridiculously nice amplifier?

    I appreciate that we can have a civil debate here :)
     
    SoundAndMotion likes this.
  14. SoundAndMotion
    Hey Tyler,

    You're the DSP guy and I'm not, so I'd be grateful for any corrections.

    If you want to work in the theoretical/math domain, real-time kind of falls away and non-causality is not a problem. Follow Shannon, show x(t) = x(t), QED, have a beer.:beerchug:

    If you deal with real values (as in real-world, not real vs.complex), whether numerical in a computer or voltages in a circuit, you have to take into account the precision/resolution/signal-to-noise of your values. When each [reconstructed x(n) - original x(n) < epsilon], where epsilon is the smallest value you can represent, then you again have a "perfect" reconstruction.

    Indeed you can use a recursive (IIR filter) or non-recursive (FIR filter) difference equation, and either can be implemented in real-time. Because of the precision limit of all the values, many of the FIR sinc kernel end values (taps) will be zero and therefore can be ignored, making the kernel possibly much smaller. Also, you don't have to use a sinc kernel, since any LPF that does what you need in terms of phase, amplitude and stop-band attenuation, will work. Don't forget real world analog filters are reproduced as IIR digital filters, and technically you don't even need a digital filter if your analog filter (typically an elliptical or Bessel filter) is good enough.

    As for the Nyquist rate, I was talking about your red data, which clearly has more than 2 points/cycle, so all of the above applies.

    BTW, I also designed and built a sensor->Arduino->Pi system, but I used an IMU, which senses its own motion, not an ultrasonic or PIR sensor for remote motion. Also the Pi didn't send email. Cool stuff!
     
    Last edited: Nov 28, 2019
    DutchGFX likes this.
  15. DutchGFX
    Yeah IMUs are great (we use them a ton at work). I haven't written on my blog in a while, maybe I should start doing that again! How did you integrate the IMU data? Or did you just use the accelerometer to detect if it was moved?
     
    Last edited: Nov 28, 2019
    SoundAndMotion likes this.
2
Next
 
Last

Share This Page