Distortion down to -300 dB, what exactly does that mean physically?

[LTig]

About the Hugo DAC:

As Rob explained, the inter-aural-delay neural network of our brain measures time delays between our ears. It operates at ~4µs for a biological 250kHz sample rate. That's far in excess of Redbook's own 22µs timing. Rob's contention is that a FIR filter akin to our brain's processing power would require 1'000'000 filter taps. That's still beyond current tech. But Hugo's WTA filter already uses 26'368 taps which rely on 16 paralleled 208MHz DSP cores. Hence Chord's refusal to work with commercial chips. Their 150-250 taps are far too low-rent to keep up with the bio DSP of our human brains.

Source http://6moons.com/audioreviews2/chord/2.html

So we need high res audio (high sample rates more specifically) not for the extended frequency response but to get the timing of transients right, which would improve the sense of depth of the stereo image?

No. Just google Monty's Digital audio show and tell on Youtube.

 

[TurbulentCalm]

I attended of his latest talks. At the end I asked him if he could hear such differences blind. He said he didn't believe in blind tests because they are too stressful. I let him be at that point.

Reminds me of Uri Gellar and Johnny Carson:

 

[TurbulentCalm]

300dB down seems preposterous to me.

I think we should call this the Homeopathy of the audio world.

 

[KSTR]

Are you serious about -60dB H3 to be ”easy to hear” on a pure sine? At what level and frequency? Shall I make a DBT for you? And with music, there is no chance in a DBT.

-60dB Hx is usually around or below the HD of the transducers, no way to reliably test harmonics audibility with that, IMHO. Any amount of cancelling may appear, only mic measurement can tell what's really happening with harmonic levels, within the distortion uncertainity of the mic itself.
OTOH, I clearly heard 0.1% THD when the harmonic spectrum was the fully spray to infinity with little attenuation (mediocre class-D).

For low orders testing, IMHO, it's better to use non-integer but still harmonically related "test harmonics", say like 1kHz fundamental + 2.5kHz@-60dBc. The more the tones are apart wrt ERB frequency scale the higher the chances for not getting masked out.

 

[skyfly]

I attended of his latest talks. At the end I asked him if he could hear such differences blind. He said he didn't believe in blind tests because they are too stressful. I let him be at that point.

What about making a blind test look like a non-blind test?

Instead of hiding the DACs, put each DAC in a beautiful hi-fi-DAC-like enclosure. Expose the enclosures to the listeners.

 

[JohnA]

I attended of his latest talks. At the end I asked him if he could hear such differences blind. He said he didn't believe in blind tests because they are too stressful. I let him be at that point.

No wonder I couldn't hear *any* difference with MScaler on or off.
Others have shared a similar view to me in private messages or in person.
I should have returned it days after I bought it, foolishly I was trying to convince myself that it was my fault I couldn't hear the transformational magic everyone else was claiming.
A week after the return window ended the dealer would buy it back at 2/3rds of what I paid them 5 weeks earlier!
Says it all I guess..

 

[Cbdb2]

About the Hugo DAC:

As Rob explained, the inter-aural-delay neural network of our brain measures time delays between our ears. It operates at ~4µs for a biological 250kHz sample rate. That's far in excess of Redbook's own 22µs timing. Rob's contention is that a FIR filter akin to our brain's processing power would require 1'000'000 filter taps. That's still beyond current tech. But Hugo's WTA filter already uses 26'368 taps which rely on 16 paralleled 208MHz DSP cores. Hence Chord's refusal to work with commercial chips. Their 150-250 taps are far too low-rent to keep up with the bio DSP of our human brains.

Source http://6moons.com/audioreviews2/chord/2.html

So we need high res audio (high sample rates more specifically) not for the extended frequency response but to get the timing of transients right, which would improve the sense of depth of the stereo image

Any research that confirms we are capable to notice ITD's as accurately as the speed our brain cell membranes operate?

Another case of taking a real effect (altough ive seen 10-20us more often) and using it wrong. If you delay one side 20us its like moving one speaker a few mm closer. So what. Only if the the delay between channels is freq. dependent (or changing in some way) will it be noticable. Why woul any DAC do that. If this mattered it might only show up while recording,( but I dont think it actualy does) once its digital your not going to add any information in the decoding. No matter how many millions of taps you use.

 

[Cbdb2]

And when recording, the sub 22us delays show up as level differences and are replayed as time delays. Heres a stereo signal with the right side 4us behind. So looks like total BS.

 

[SIY]

Another case of taking a real effect (altough ive seen 10-20us more often) and using it wrong.

This was common in CD players pre-1985 or so. I haven't seen it in any player or DAC made after the Reagan administration.

 

[Jim T]

As some of you guys might know, the famous Rob Watts has mentioned on more than one occasion that he can hear and measure distortion down to -300 dB. Cited by @amirm here for instance. He also says something similar in this video:

Staring at roughly the 17min mark.

Now I was thinking about a thought experiment that you guys can probably help me with, especially the physics nuts. I'd like to explore what this actually would mean on a physical level? So how little is the voltage change? How many electrons would that actually need? And if we're talking, what about air pressure difference? But you guys can come up with other stuff. Probably the infamous shoutometer will come up, which I find rather uninteresting because it just not model the practical world very well.

So I had a crack at voltage. So let's see how small of a voltage difference -300 dBV is. Well, I had to resort to Wolfram Alpha because my go-to calculator gave me just zero So Wolfram was more helpful and gives 1*10^-15, which is 1 fV RMS (so femtovolt). To count electrons, we need to add some assumptions: we need current. So let's assume it's a DAC, and it's syncing into 20K input impedance of a pre. Now we can calculate the number of electrons needed for this 1 fV voltage. So 1A has 6.24 quintillion electrons (10^18). So, how many electrons does out -300dbV signal have into 20k? Well, it's not even 1 electron per second: it's 0.312 Already feeling silly Rob?

Hope you guys can fact check me on this. Would be fun to do the same for the air pressure difference.. Once again my go-to calc gives 0

I stopping worrying about pixey dust long ago. With noise floors at -120 and -130db, jitter so low, but yes filters do matter. I bought two of the Project Audio S2 DAC from the Stereophile review and I can hear the slight changes in filter choices. You have to listen hard and close...if it matters to anyone.

One with a headphone amp is off my computers for pcm and DSD files; the other improved an old DVD transport and improved the sound of CD s and my 2496 DVDs I record and burn.

I try and not follow smoke and mirrors. I remember that at nearly 74 I am more defective and my speakers not far behind. Look at most speaker frequency responses in-room and stop worrying about this stuff.

 

[Chester]

Blind tests can be stressful if someone is watching over your shoulder. In the case of Rob Watts, no one is looking at him so he can take his time and verify what he thinks he is hearing, he really does. I routinely do this in developing EQ for speakers and headphones. There is no stress involved other than realizing after the fact that you were wrong.

Although I have heard him state the same, he has also stated he does do blind listening tests when he can not explain what he is hearing. He wrote about it recently on another forum, see below:


Objectivity is absolutely essential; whenever we hear a change in sound quality (assuming that the listening test is correctly done and the results consistent and accurately characterised) then there must be a science based reason for the change - and this reason must be able to explain the subjective results too. The problem is that extremely small errors are very audible - recently I was testing the sample rate converter for the ADC project. This started off very simply with fractional Lagrange interpolators; a process of designing, simulating the code, then running listening tests. The first attempt had worst case aliasing at -200 dB - but you could easily hear the effect of turning the interpolator off (then getting -160 dB aliasing). Then I kept on improving the order of the interpolator and the accuracy of the time interval, and listening to the results. Over several months, I got better and better performance - eventually hitting below -350 dB (it was at -310dB or so aliasing at this point). Frankly I started to question my own sanity at this point, as these numbers are ridiculously low. So I did a blind listening test with my son and asked him to characterise the sound - listen to this and listen to that - and he said that that one had better clarity and focus than the other, and he could easily hear the difference. His assessment agreed perfectly with mine.

The problem with this is that if you were to overlay waveforms that have -310 dB aliasing against -350 dB it would be impossible to see the difference - but the amazing thing is that we can actually perceive these very, very small differences.

 

[roskodan]

"So I did a blind listening test with my son and asked him to characterize the sound - listen to this and listen to that -" That's totally not a blind test. XD

 

[Chester]

"So I did a blind listening test with my son and asked him to characterize the sound - listen to this and listen to that -" That's totally not a blind test. XD

I don’t know the criteria for something to be considered a blind test, but surely there has to be an element of someone instructing you on what to listen for?

 

[SIY]

I don’t know the criteria for something to be considered a blind test, but surely there has to be an element of someone instructing you on what to listen for?

That’s sometimes true in advance, but no one who knows the choice being played can be anywhere in the test area. The description here is almost classic coaching.

In my article Testing One, Two, Three, I related the classic example of the wife in the next room and how that tripped me up.

 

[amirm]

I don’t know the criteria for something to be considered a blind test, but surely there has to be an element of someone instructing you on what to listen for?

There are two types of tests: single and double blind tests. In double blind tests the person doing the switching doesn't know which is which. Usually this is a computer so no human is involved and can't telegraph the identity of the items verbally or non-verbally.

Single blind is the type he attempted when the person doing the switching knows the identity. Single blind can work if super care is taken in not communicating anything to the person taking the test. Clearly this is violated in the way Rob explained what he did. The assumption going into such a test should be that there can be no difference in addition to being one. So one can't ask, "which one do you think sounds better?" That supposes there is a difference and forces the tester to then think that way.

If on the other hand you were in another room switching gear with the tester having no ability to see you are not doing switching in any way that telegraphs the identify of the equipment, then I think it is good enough for informal testing.

This is beside the main point however. Any test like this needs to be repeated to make sure the person was not just guessing. If I asked you to predict a coin flip and you said heads, and I flip and get heads, does it mean you can predict a coin toss? The answer is no of course. If I repeated that test enough, you would start to miss some and eventually land at 50% chance of being right. Our goal in such testing is 95% being right. If you conducted the test 10 times, you would need to get 9 right to achieve that. I personally like to get 100% right or very close to it when the subject is controversial like this one is.

Another important tool is a control. You first test your son by giving him two identical samples and see if he can tell there is no difference. If he says one is better than the other, then you know something is wrong with the test or conclusion you are drawing from it. Repeating the test as stated above, catches this outcome.

Inversely, you can put a lossy MP3 in there and see if the person can tell the degradation there. If he can't, there is no sense in asking him if he can hear -300 dB artifacts.

So even if the test was double blind, Rob just asking his son once which sounded better doesn't count for anything.

I have explained a number of these topics in my video on blind testing:

 

[Geert]

The first attempt had worst case aliasing at -200 dB - but you could easily hear the effect of turning the interpolator off (then getting -160 dB aliasing). Then I kept on improving the order of the interpolator and the accuracy of the time interval, and listening to the results. Over several months, I got better and better performance - eventually hitting below -350 dB (it was at -310dB or so aliasing at this point).

How do you notice artifacts that are more than 40dB below the human hearing threshold, and more than 100dB below the noise floor of a DAC so more than perfectly masked?

 

[roskodan]

Sir Watts is selling future proof technology for when FPGA will be connected directly into your brain which, if you gonna be still around, will be a FPGA as well.

 

[voodooless]

How do you notice artifacts that are more than 40dB below the human hearing threshold, and more than 100dB below the noise floor of a DAC so more than perfectly masked?

You forgot that they are also ultrasonic

 

[preload]

As some of you guys might know, the famous Rob Watts has mentioned on more than one occasion that he can hear and measure distortion down to -300 dB.

I can only hear distortion down to -294dB, which means his hearing is twice as good.

 

[peterzuid]

Think I am missing the point here. Looks like everyone including Amirm is gathering to convince each other that Rob Watts cannot hear what he claims he can. How do we know what he can hear/perceive ? I'm convinced he is a well trained listener but I/we cannot know and what's the point ? In my opinion he is just someone who is trying to develop/design serious digital to analogue audio-gear by a thorough understanding of the mathematics involved and also by listening. What can be wrong with that ? In addition; aiming for the sinc impuls response, as he does, is the mathematical way to go as we were educated in University. In case you don't believe his experiences and conviction then leave it and don't listen to it or buy it. Anyway, judging the reactions here, Rob Watts caused a lot of doubt in this community and not without reason. All our brains are listening with two ears and doubting is part of the process.