Master Thread: Are measurements Everything or Nothing?

[RayDunzl]

What would you estimate the sharp rise would correspond to in kHz? (I would do the math but I don’t know what the axes are!)

The jump is from 0 to +max and back to 0 in successive samples, so it would correspond to a half cycle of a 11025Hz wave if continued to -max and back to 0 and repeat, at 44.1kHz sample rate.

 

[DonR]

I understand what you are saying, but let’s pose a scenario.

The eardrum is a physical system, perhaps like a drumhead drawn tight. Assume it is hit by a 22kHz standing pressure variation. It may be too stiff, or have too much back pressure, to transmit much (or any) of that wave.

Does that mean it is impossible for a steep wavefront to provide a transient acceleration in that range?

I don’t think we can simply perform a Fourier transform and state that components above 18kHz, or 20kHz, are simply discarded. Not until we have independently mapped both transient response and steady-state response of the ear, can we decide whether this math describes the physical system.

And if it doesn’t, the steady state numbers we measure are the wrong numbers. And assumptions we make about the Nyquist cutoff point are also wrong, and the design of a lot of digital reproduction is based on a false model of hearing, and are wrong as well.

I kind of think the design of the system starts with understanding the ear and nervous system, not with the electronics.
Then, we can measure meaningfully.

Our upper limit to hearing is mostly defined by the physiology of the middle ear acting as a low-pass filter. Note that Fletcher-Munson curves indicate that human hearing becomes exponentially insensitive above about 15KHz and this degrades with age from the late teens onwards. I am in my late 50's and hear nothing above 13.5K which is average for my age. I hear no difference in music that is rolled off at 14KHz vs music that retains all its upper harmonics. There is no other component to sound waves other than sinusoidal air pressure changes.

 

[Blumlein 88]

The system implemented in the Fried Studio V was known as m.a.r.s., for McShane Ambient Recovery System. The patent is listed at https://ppubs.uspto.gov/pubwebapp/static/pages/ppubsbasic.html. The PDF is complete with circuit diagrams for those who want to try in the lab.

US-4847904-A

Preview PDF

Ambient imaging loudspeaker system

McShane; Charles L.

1989-07-11

9

The design is meant to address the very relationships between loudness and phase that you cite above.

In keeping with the purpose of this thread, if a sound source does not stimulate the same mechanisms that interrelate loudness, phase, and frequency, could it really be said to be an accurate transducer? Perhaps technically accurate, but is it accurate at recreating the sound field that the listener needs?

This seems like something we can, indeed measure. Can we say a system has accurate bass if it does not embody these effects? Or can a digital room correction be correct if these wavefronts are not recreated so they match the original psychoacoustic stimulus?

This isn't about bass. He has a midrange on the side putting out a difference signal. Not so different than a speaker pair with a built in Hafler ambience decoder. It would put out a blob of ambience. I've actually done some experiments much like this not knowing there was a patent. Didn't work very well. You can get some of the effect with mid/side processing, but not as much effect without having a separate difference speaker. One that plays the difference between the channels.

Here is the Dynaco version. Much the same connection only the out of phase difference speakers are in the rear instead of being 90 degrees around the cabinet on the front speakers.

 

[Blumlein 88]

I understand what you are saying, but let’s pose a scenario.

The eardrum is a physical system, perhaps like a drumhead drawn tight. Assume it is hit by a 22kHz standing pressure variation. It may be too stiff, or have too much back pressure, to transmit much (or any) of that wave.

Does that mean it is impossible for a steep wavefront to provide a transient acceleration in that range?

I don’t think we can simply perform a Fourier transform and state that components above 18kHz, or 20kHz, are simply discarded. Not until we have independently mapped both transient response and steady-state response of the ear, can we decide whether this math describes the physical system.

And if it doesn’t, the steady state numbers we measure are the wrong numbers. And assumptions we make about the Nyquist cutoff point are also wrong, and the design of a lot of digital reproduction is based on a false model of hearing, and are wrong as well.

I kind of think the design of the system starts with understanding the ear and nervous system, not with the electronics.
Then, we can measure meaningfully.

Bandwith and transient response are the same thing as Ray Dunzl and others are telling you. Did you know one method of testing newborns for hearing is a device that fires an impulse into the ear canal and records what comes back from which they can get the frequency response of the ear.

 

[antcollinet]

I understand what you are saying, but let’s pose a scenario.

The eardrum is a physical system, perhaps like a drumhead drawn tight. Assume it is hit by a 22kHz standing pressure variation. It may be too stiff, or have too much back pressure, to transmit much (or any) of that wave.

Does that mean it is impossible for a steep wavefront to provide a transient acceleration in that range?

I don’t think we can simply perform a Fourier transform and state that components above 18kHz, or 20kHz, are simply discarded. Not until we have independently mapped both transient response and steady-state response of the ear, can we decide whether this math describes the physical system.

And if it doesn’t, the steady state numbers we measure are the wrong numbers. And assumptions we make about the Nyquist cutoff point are also wrong, and the design of a lot of digital reproduction is based on a false model of hearing, and are wrong as well.

I kind of think the design of the system starts with understanding the ear and nervous system, not with the electronics.
Then, we can measure meaningfully.

No - it can't produce the transient acceleration.

If we just take the eardrum - but bearing in mind our hearing is a complex system of interacing mechanica/acoustic and electrical systems - it is a transducer after all:

The eardrum is essentially a mechanical system of mass/spring/dampers. It is analogous to a speaker cone in reverse - or more accurately a microphone diaphragm.

Bear in mind that mass/springs/dampers in a mechanical system are exactly analogous to inductance, capacitance, and resistance in an electrical system. Force, Velocity and Displacement are equivalent to Voltage, current and charge. You can apply the same maths to both, and come up with the same results.

It's mechanical properties prevent it from responding instantly. It takes time to accelerate so even if there were a square wave input into the ear (limited only by the bandiwdth of air itself) the eardrum COULD NOT MOVE with the same movemennt as the pressure wave. It would take time to accelerate at the start of the pressure wave, and it would take time to stop at the end of the pressure wave. The mass of the drum, interacting with the springiness and damping will act like an LCR filter, and create a ramp up to speed, a slope on the upside and an overshoot and slight oscillation at the transient as the rise turns into the flat top of the square wave. It would behave exactly as it would if the pressure front had been prefiltered to have the same frequency response as the ear drum. The movement of the eardrum will look like a band limited square wave voltage in an electrical circuit.

The same will apply to all of the other mechanical systems in the ear - right up to and including the electro mechanical part that converts the vibration to an electrical signal.

 

[Reynaldo]

@Holmz

Is it not a 32bit DAC?
Using a Mac OS with Audirvana.
I'm playing the files exactly as they are, without any UPsampling.

The first image shows the DAC turned on without playing any files.

In the second image playing a 32/192Khz file

In the third image playing a 24/96Khz file

 

[Thorsten Loesch]

Hi,

> In my mind… a DAC using a 24bit DAC chip,
> is not a 32 bit DAC.

Ok, then let me use another example.

We have a DAC with the ability to receive 768kHz/32Bit Samples.

An FPGA is used to turn the 32Bit PCM into 45/49MHz Delay Sigma modulated 1Bit audio.

This digital audio is converted to ananlogue through a discrete "Analogue FIR" filter, say an 8-Bit delay chain with non-equal weights.

The audio SNR is 114dB or 18 it Enob. This means there will be no measured or otherwise objectively observable difference feeding a 24 Bit or a 32 Bit signal.

There is no off the shelf chip for DAC, digital filter etc. There is no datasheet from some manufacturer stating "xx bits" or anything such. There are no datasheet's.

32 Bit Input
1Bit DAC
18 Bit ENOB

What kind of DAC do I have?

1 Bit?
18 Bit?
32 Bit?

Suffusion of yellow?

> In this case, the “Bonor de Jur”, includes
> 8 imaginary bits that go between the true
> 24 bits and the imaginary 32 bits…

What is the "true 24 Bits"?

> And it potentially gets even worse, as if
> the 24 bits are assumed to be in the upper
> end of the 32 bit register, and the sending
> unit had them at the bottom, then we may
> end up with 16 bits.

Big endian / small endian is not an issue in digital audio. So we always end up actually with as many bits as the converter can render into the analogue domain.

By that standard there are few 20 Bit DAC's especially at the budget end (>= 122dB SINAD @ 0dBFS) and absolutely zero 24 Bit DAC's (>= 146dB SINAD @ 0dBFS).

So we either accept "32 Bit Input possible" as "32 Bit DAC" and the DSD1793 and most other BB/TI DAC's do this (even if their release predates the 32 Bit fad and thus doesn't mention this explicitly). Or we accept that there is no such thing as a 32 Bit DAC.

Incidentally, talking datasheet's, the DSD1793 Datasheet also states the Chip as 192kHz maximum sample rate, yet I routinely use it with a 768kHz sample rate input (and yes, it can pass a 384kHz signal, if significantly attenuated by the 80kHz analogue filter).

Also, DSD1793 Datasheet lists only 2.82Mhz DSD (aka DSD64) as DSD format, but I routinely pass 45/49MHz DSD through the Chip.

And no, I am not "overclocking" the chip nor am I doing anything "illegal". The IC is inherently capable of doing all this, explicitly per datasheet.

But as these "formats" simply did not exist when the Chip was introduced to the market they where never tested or charaterised or advertised.

So I tested instead.

And for whatever reasons, TI never updated their datasheet's and re-charsterised and relaunched / readvertised the part.

Thor

 

[Thorsten Loesch]

If I were to make a bet, I'd put my money on placebo.

You would loose, unless you select an ABX test (as opposed to a properly implemented double blind test that passes scientific standards) as proof, which I would in turn decline, with a long missive, deconstructing the "ABX" as either self deception or deliberate fraud on the part of the originators and Stockholm syndrome from adherents.

So we would be unlike to resolve your bet.

In fact, "ABX" deliberately and intentionally uses nocebo to make real differences disappear.

But they can, thankfully be restored. Try this:

Go to the ABX tests website, if it disappeared use the way back machine.

Take ALL negative outcome (no difference) tests and perform a meta analysis.

You will find reliable statistical evidence that in fact differences existed but the test was not able to confirm them, because it actually is subject to a strong bias towards what is called a "Type B Stastistical Error".

A type A Error is one where the "#different" hypothesis is confirmed in error. A type B error is one where the "#different" hypothesis is rejected in error.

The risk of both types should be equal for good statistics.

It is not for ABX. ABX seeks to avoid Type A errors at almost all cost, creating a high risk of Type B errors. That is, we have very confidence that any difference confirmed by ABX really exist, but we have in fact almost no confidence that if an ABX test rejects the existance of a difference, that this difference does not actually exist.

Secondly, the way ABX tests are usually conducted introduce a very strong randomising factor in the form of expectation bias.

It actually doesn't matter if you expect to hear a difference or expect to hear no difference, it works equally well.

As humans we like to have our bias's confirmed, so our brain and sensory system works overtime to make it so, or, as Nietzsche once rather charitably remarked: "We are all greater artists than we realize."

The result of combining these two problems is to make any rejection of the "#different" hypothesis by ABX highly unreliable, sufficiently in fact to simply declassify the "Audio ABX Tes" as unscientific parlor trick and confidence trick.

I'm open to discuss testing by listening further (certainly casual, non level matched, sighted tests are just as useless as ABX) but not in this threat and perhaps not in this venue.

Thor

 

[Thorsten Loesch]

Hi

> I was able to try several DACs and almost
> all of them that have the Burr-Brown chip
> have a different sound. Is it a placebo
> effect or is there something that really
> changes?

It is a quite serious question.

Based on my experiences it is not placebo. But often differences are subtle.

Further, when we say "Burr Brown DAC" (now TI) we are really talking about around a dozen different cores, filters, modulators etc. that all perform quite differently objectively, even though they do derive from only 3 different basic architectures.

At least in my experience these architectural differences are quite audible and BB chips with similar architecture sound more similar than different.

> The impression I have (I could be wrong)
> that somehow devices like iFi and Musical
> Fidelity, the engineers and designers were
> very concerned with the sound that was
> closer to that of an analog device.

Yes, you may be right. I know a few of the people at Musical Fidelity in person and their priorities match what you describe.

For iFi the Engineer is me.

For much of my tenure at AMR & iFi I had access to an excellent listening space, high grade equipment for both technical (objective) evaluation of audio gear (including AP) and for subjective listening evaluation, AS WELL as a group of dedicated and patient listeners who would spend whole days in blind listening tests (not ABX), which often rejected my technocratic "this cannot possibly make a difference" bias.

And yes, my personal "sound preference" could be called "analogue" though I prefer "realistic" or perhaps "old master" as opposed to abstract, cubist, impressionist etc. modern styles. Not that I do not appreciate them in art, but are they appropriate for audio?

> This is a more technical forum where
> people look at test numbers, this is
> important but it seems to me that from
> a certain level it will not change anything
> in the sound quality.

Yes. I think the issue is that just because certain technical quantities are commonly measured, people make the logical jump to presume that these measurements somehow reliably correlate with "good sound".

In fact it may best to view "sound" and "measurements" as discrete non-overlapping magisteria (NOMA), though in reality there is substantial overlap, but not necessarily as expected from those subscribed to a specific magisteria as the relevant one.

To me, reconciling what I objectively measure with what I myself and others hear has been, well not quite an obsession, but certainly a quest, one that is still far from complete.

Bit in most public debate declaring "NOMA" keeps the peace better, bit naturally leads nowhere.

Thor

 

[Killingbeans]

I'm open to discuss testing by listening further (certainly casual, non level matched, sighted tests are just as useless as ABX) but not in this threat and perhaps not in this venue.

Audio Blind Testing - You Are Doing It Wrong! (Video)

Just uploaded a video on the basics of performing controlled tests in audio. It was motivated by saving myself text in having to write all of this down when telling someone who to do these tests right. And as a counter to a few online personalities to keep saying they do "blind" tests yet when...

www.audiosciencereview.com

What to do about the ABX test?

Seems increased commentary in recent weeks about ABX tests. Much of it stemming from people who come to ASR to set us straight about trusting our ears. I do agree with some who have said that calls for ABX or it didn’t happen have become almost like a club to beat people over the head with, and...

www.audiosciencereview.com

 

[Thorsten Loesch]

Audio Blind Testing - You Are Doing It Wrong! (Video)

Just uploaded a video on the basics of performing controlled tests in audio. It was motivated by saving myself text in having to write all of this down when telling someone who to do these tests right. And as a counter to a few online personalities to keep saying they do "blind" tests yet when...

www.audiosciencereview.com

What to do about the ABX test?

Seems increased commentary in recent weeks about ABX tests. Much of it stemming from people who come to ASR to set us straight about trusting our ears. I do agree with some who have said that calls for ABX or it didn’t happen have become almost like a club to beat people over the head with, and...

www.audiosciencereview.com

Ipse Dixit.

I will not particularly criticise Amir, but I'm looking forward to the day when he finally fixed the groundloops/pin1 problem plaguing his AP2 measurements, something really basic.

I generally use the methods propsed and endorsed by the VDT (Verein Deutscher Tonmeister) for subjective testing.

And I use deliberately a protocol where it is known to the listening panel (I am way to biased to just trust myself) only that we test if a specific difference in hardware, that is unknown to the panel in nature or specifics until AFTER the test is completed.

Plus, they are made aware that I might run a Bavarian fire drill (which I sometimes did), that is introduce deliberately a clearly audible difference (e.g. 1dB louder), just to keep everyone on their toes.

This eliminates any BIAS, because if you do not know the difference (and there may either be no difference or a clearly audible one) you must drop bias and actually listen.

I always include two identical units and two identically different ones. They are otherwise (excluding the change being evaluated) identical and confirmed by AP2 to measure identical sufficiently to at least pass the minimum requirements set by the ABX crowd.

Listening itself is single blind, that is listeners are faced with four outwardly identical boxes, identified by symbols, to avoid numerical or alphabetical bias.

The listening actually mostly resembles sighted listening as common among audiophiles, to avoid adding stress. Though we commonly have staff swap out units and do the legwork. The point being, I want to know what listeners actually hear, not what they expect.

In one test we laced enough same colour units, so two were one cour and two another, naturally differences were set across the colours.

I found that for most of our experienced listeners colour preference overrode listening! Rerunning the same test with all colours the same a few weeks later showed preferences that with good statistical confidence were based on the other physical differences (not colour).

This was actually one of the tests that caused me to do deeper research into preference (in general) and go from alphabet ID to symbols and take many more precautions to really "blind" the panel to the differences, going as far as covering serial numbers and swapping symbols during lunch break etc.

Listeners are asked to give scores for preference, not to determine same/different. Simply rate the unit with a square for how you like it, then a unit with a different symbol and so on.

Statistical analysis looks how likely the individual and overall preference ratings were to be random or if a preference for an individual is likely to relate to the actual physical differences.

Ultimately my objective with my tests was never to prove the existence or absence of a difference, but to determine what kind of "sound" was found appealing by listeners, in order to make products that appealed to a market segment we wanted to have sales in.

And yes, gender, age and cultural background actually have marked correlations as well, though my sample size for that was rather more limited.

As said, it's a subject that is well off topic for this thread and again, possibly for this venue. I greatly appreciate the work Sean Olive & Todd Welti are doing at Harman regarding frequency response preferences.

But sadly the same approach is not extended to even perceptual coding to improve it. About the only real example I am aware of is MP3 where "JJ" (who was influential to my approach) applied a similar system, which I guess is why MP3 passed the test of time. It is not transparent, but where it fails it does so sounding "good", so much that some listeners prefer 128k VBR MP3 to the CD source, as sounding better.

Anyway, I will leave it here.

Thor

 

[Thorsten Loesch]

Audio Blind Testing - You Are Doing It Wrong! (Video)

Just uploaded a video on the basics of performing controlled tests in audio. It was motivated by saving myself text in having to write all of this down when telling someone who to do these tests right. And as a counter to a few online personalities to keep saying they do "blind" tests yet when...

www.audiosciencereview.com

What to do about the ABX test?

Seems increased commentary in recent weeks about ABX tests. Much of it stemming from people who come to ASR to set us straight about trusting our ears. I do agree with some who have said that calls for ABX or it didn’t happen have become almost like a club to beat people over the head with, and...

www.audiosciencereview.com

I would like to recommend some reading to you.

It is a commencement address at CalTech for 1974, by noted latin percussion player Richard Feynman, who had also a minor sideline in Physics and the least common of the senses.
https://calteches.library.caltech.edu/51/2/CargoCult.htm

Especially interesting for us in audio is the section where talks about experiments of a certain Mr. Young in rat psychology.

Every time we conduct a listening test we are in effect performing an experiment in psychology.

Of course, most people just want a go/no go indication or a confirmation off what they believe, not to actually understand anything.

Thor

 

[Killingbeans]

As said, it's a subject that is well off topic for this thread and again, possibly for this venue.

I don't think it's unfitting for the venue. That's why I posted the two links as a couple of options for continuing the discussion on-topic.

Personally I don't believe in audible differences between well designed op-amps, but it's not a hill I'm ready to die on. I'm open to any kind of constructive debate.

You are sure to encounter some ASR attack dogs, but good arguments will get the tails wagging again in no time.

 

[Reynaldo]

@Thorsten Loesch
Every post you make for me has been a lesson.
Thank you very much

 

[Thorsten Loesch]

I don't think it's unfitting for the venue. That's why I posted the two links as a couple of options for continuing the discussion on-topic.

Personally I don't believe in audible differences between well designed op-amps, but it's not a hill I'm ready to die on. I'm open to any kind of constructive debate.

You are sure to encounter some ASR attack dogs, but good arguments will get the tails wagging again in no time.

Let's start with the premise that hearing, human hearing and the mechanics, learned response etc. are highly complex.

I don't expect creationists here, but realistically in this case it matters zero if our hearing evolved to avoid being eaten by a sabertooth tiger and to hear our baby crying across long distances because it is about to be eaten by a sabertooth tiger or it was explicitly designed by an intelligent creator.

If the human hearing system was a microphone and was tested by Amir, it would get worse than a headless Panther. Actually what is worse than a headless Panther?

At the same time human hearing can do amazing feats of discrimination and get a reliable signal read, where morse code, pll's etc. and learning AI algorithm fail.

So clearly, human hearing is not linear, low distortion etc. and there absolutely is a learned response to sound.

On the other hand we have technical equipment, designed generally according to abstract requirements that we think will give a specific result, tested using specific objective means.

Clearly these systems, while nonlinear, ate generally made as linear as possible, because intuitively it makes sense. Surely removing as much as possible any change between the original acoustic event and the Soundwave reaching the ear will give great fidelity in itself.

But is it really so? (Quoting Sporting Life from Porgy & Bass).

That is why I am invoking NOMA.

It is, in my experience it ain't necessarily so.

Now this is before I am even addressing the internal contradiction of calling a headphone or speaker with 60dB SINAD at 105dB SPL excellent (in case anyone came across such a rare beast, I never did) but calling a DAC or headphone amplifier with 90dB SINAD poor.

Clearly, there is no referencing here to what is audible, how and what makes good sound.

As long as we understand this (back to NOMA) I can take intellectual satisfaction in making a super low THD Headphone Amplifier, with five zeros THD (~140dB SINAD) at rated output, then listen to the machine and decide it sounds like poo and pull out one by one all the circuit tweaks I applied to get super low THD (or high SINAD), settle for 40dB less SINAD and enjoy the way this machine makes music sound.

If it was possible to make it switchable (in this particular case it was way too much switching) I would even include the 140dB SINAD mode to show "I can, but I thinks it's not a good idea, try yourself).

Then I might add a circuit tweak that degrades SINAD to 60dB (user selectable) and enjoy the way this machine makes music sound.

I actually do enjoy the 60dB SINAD mode, in this case the machine is the iFi iCan Pro which Amir measured and wished I had left the tube and 60dB SINAD switch out. I am sure if there had a 4th position for 140dB SINAD, Amir would have decided that is the only mode needed.

Instead I decided it was the only mode NOT needed.

Once we establish and accept the NOMA concept, we can progress to look if there truly is no overlap. That is next level stuff.

Thor

 

[Holmz]

@Holmz

Is it not a 32bit DAC?
Using a Mac OS with Audirvana.
I'm playing the files exactly as they are, without any UPsampling.
…

I dunno.
My question was that the spec sheet says it is a 24bit DAC.
And there are 28 legs on the chip, so I assume we have power, ground, a clock…

Hi,

> In my mind… a DAC using a 24bit DAC chip,
> is not a 32 bit DAC.

Ok, then let me use another example.

We have a DAC with the ability to receive 768kHz/32Bit Samples.

An FPGA is used to turn the 32Bit PCM into 45/49MHz Delay Sigma modulated 1Bit audio.

This digital audio is converted to ananlogue through a discrete "Analogue FIR" filter, say an 8-Bit delay chain with non-equal weights.

The audio SNR is 114dB or 18 it Enob. This means there will be no measured or otherwise objectively observable difference feeding a 24 Bit or a 32 Bit signal.

There is no off the shelf chip for DAC, digital filter etc. There is no datasheet from some manufacturer stating "xx bits" or anything such. There are no datasheet's.

32 Bit Input
1Bit DAC
18 Bit ENOB

What kind of DAC do I have?

1 Bit?
18 Bit?
32 Bit?

Suffusion of yellow?

> In this case, the “Bonor de Jur”, includes
> 8 imaginary bits that go between the true
> 24 bits and the imaginary 32 bits…

What is the "true 24 Bits"?

> And it potentially gets even worse, as if
> the 24 bits are assumed to be in the upper
> end of the 32 bit register, and the sending
> unit had them at the bottom, then we may
> end up with 16 bits.

Big endian / small endian is not an issue in digital audio. So we always end up actually with as many bits as the converter can render into the analogue domain.

By that standard there are few 20 Bit DAC's especially at the budget end (>= 122dB SINAD @ 0dBFS) and absolutely zero 24 Bit DAC's (>= 146dB SINAD @ 0dBFS).

So we either accept "32 Bit Input possible" as "32 Bit DAC" and the DSD1793 and most other BB/TI DAC's do this (even if their release predates the 32 Bit fad and thus doesn't mention this explicitly). Or we accept that there is no such thing as a 32 Bit DAC.

Incidentally, talking datasheet's, the DSD1793 Datasheet also states the Chip as 192kHz maximum sample rate, yet I routinely use it with a 768kHz sample rate input (and yes, it can pass a 384kHz signal, if significantly attenuated by the 80kHz analogue filter).

Also, DSD1793 Datasheet lists only 2.82Mhz DSD (aka DSD64) as DSD format, but I routinely pass 45/49MHz DSD through the Chip.

And no, I am not "overclocking" the chip nor am I doing anything "illegal". The IC is inherently capable of doing all this, explicitly per datasheet.

But as these "formats" simply did not exist when the Chip was introduced to the market they where never tested or charaterised or advertised.

So I tested instead.

And for whatever reasons, TI never updated their datasheet's and re-charsterised and relaunched / readvertised the part.

Thor

^That is a lot…^

I have no argument… 24 bits is enough and 32 bits will not result in anything “more revealing”.
My main question is whether it is a 24 bit DAC like the spec sheet says.

And then the observation that people will rush out to buy a 32bit DAC, which is a 24 bit DAC, and believe that it somehow is technically better than their current 24 bit DAC.
And the measurements do not really give us much insight as to whether it is better, the same, or otherwise… it looks like it the same as other DACs

 

[Sal1950]

specifically because I was required to make sure it did not sound as good as the micro iDAC 2.

So how did you purposely design the product not to sound good?

People can write detailed paragraph after paragraph on how product X sounds.
But the facts are that without having exacting scientific procedures in place while listening to the component,
anything said may as well be said by a monkey. They have as much ability to react to the sensory input as
anyone and the same ability to be in complete error.
Only in Hi Fi are unsubstantiated claims accepted as fact without any of the statements being held to some verifiable standard.
Yes, rigidly controlled DBT's are still the gold standard for the presentation of sound quality evidence while combined with
measurements done by today's SOTA measurement equipment.

 

[Thorsten Loesch]

I dunno.
My question was that the spec sheet says it is a 24bit DAC.
And there are 28 legs on the chip, so I assume we have power, ground, a clock…

My main question is whether it is a 24 bit DAC like the spec sheet says.

And then the observation that people will rush out to buy a 32bit DAC, which is a 24 bit DAC, and believe that it somehow is technically better than their current 24 bit DAC.
And the measurements do not really give us much insight as to whether it is better, the same, or otherwise… it looks like it the same as other DACs

Ok, let's try this again.

The spec sheet says it is a 192kHz/24 Bit DAC with support for 64FS DSD and it also states that it includes a mode in which the digital filter is bypassed and we directly input data into the DS Modulator (the system is a bit more complex, as it combines a thermometer code multibit DAC for the upper bit's, but that's of no real consequence in the discussion).

Under those conditions we need 32 Bit Clock's (that is 32 Bit's per wordclock) per wordclock, that is the "native" format the DAC receives if the 24 Bit/192kHz digital filter is bypassed is 32 Bit for PCM.

Finally the datasheet states that minimum BCK & MCK cycle time is 20nS or 50MHz.

In other words, the Chip will accept 32 Bit's per channel with 50MHz/32Bit = 1.5625 MHz.

And for a single-bit format (that is DSD) we get 1024FS DSD.

Now that is not explicitly spelled out in the Datasheet, however careful reading of the datasheet shows all of this as possible.

As for 28 Legs. what has the number of leg's to do with the number of Bit's the DAC uses? At a minimum a DAC can have 8 Pin's and still be 16/24/32 Bit depending on the internal design.

As to what the datasheet says, the front page says it is marketed as 24Bit/192kHz + DSD64 DAC and is formally tested and characterised for these sample rates etc.

Inside the Datasheet in the detail it says INDIRECTLY that the DAC can be fed with 32 Bit PCM at 1.5625MHz and Delta-Sigma (DSD) sign als up to 50MHz.

As the datasheet was originally written around Y2K the only formats that actually existed were up to 192kHz/24 Bit audio for DVD-A and 2.828MHz Single Bit (DSD_ for SACD.

As no other formats existed they were not mentioned or characterised, even though the DAC Chip is in principle capable of accepting and converting these formats.

As to "is a 32 Bit DAC technically better than a 24 Bit DAC", at least in any objective way.

For example, when TI started marketing the PCM1795, they decided to market it as 32 Bit Chip, even though it is basically the same underlying engine as other TI Chip's with a minor tweak to the digital filter. But in terms of SNR/SINAD etc. it performs worse than the headline "24-Bit" Chip from TI, which incidentally accept's the same formats (including 32Bit/1.5625MHz) as the DSD1793.

So again, 32 Bit is strictly a marketing thing and if TI could be bothered, they could instantly upgrade all their DAC's to 32 Bit by rewriting the datasheets as most of their higher grade DAC's can accept 32 Bit input.

Thor

 

[Thorsten Loesch]

But that is not the same as a FDAC chip doing a real 32 bits of resolution.

Please define "real 32 Bit of resolution".

It either is the ability to receive a 32 Bit word and output the converted analogue equivalent at however many BIT ENOB are available.

Or it is a DAC with an ENOB of 32 Bit which is impossible, just for now we do not see DAC's with 24 Bit ENOB.

Or is your definition of "real 32 Bit of resolution" not actually the physical capabilities of the DAC itself, but the way the Chip manufacturer markets the DAC?

Thor

 

[Thorsten Loesch]

So how did you purposely design the product not to sound good?

People can write detailed paragraph after paragraph on how product X sounds.
But the facts are that without having exacting scientific procedures in place while listening to the component,
anything said may as well be said by a monkey. They have as much ability to react to the sensory input as
anyone and the same ability to be in complete error.
Only in Hi Fi are unsubstantiated claims accepted as fact without any of the statements being held to some verifiable standard.
Yes, rigidly controlled DBT's are still the gold standard for the presentation of sound quality evidence while combined with
measurements done by today's SOTA measurement equipment.

I did not purposely design a product that did not sound good.

My purpose was to design a product that had minimal possible THD & N (or maximum SINAD, depending how you want to express it) and sounded good.

Once I was finished with the electronic design and fine tuning of compensation, stage by stage distortion compensation etc. I had better than -140dB THD&N or 140dB SINAD.

Once listening compared to other products both from our own company and some competing products, everyone who listened felt that the new product sounded underwhelming. Boring, nothing really wrong, but just no involving and engaging. It did not make you want to listen to more music, but instead turn it off and go down the pub for a brew.

After reversing the distortion cancellation tweaks, the extra loop gain stage by stage and scaling the circuit back to something a little more ordinary it had better than -110dB THD & N (110dB SINAD) and it sounded much better. In fact after "making it way worse" (30dB more H2/H3 distortion) it started to actually be preferred by listeners to the other options.

I am all favour of rigidly controlled DB Testing, where I take exception is when someone starts claiming the ABX Test as designed and popularised by the ABX company in Troy, Michigan as a commercial enterprise is a good example of such a test.

DB Testing is the gold standard for all perceptual testing, including Wine tasting, evaluating the efficacy of medicines etc. Non of the tests you find there however remotely resemble ABX. ABX is the equivalent of the medieval Ducking Stool of the Witchfinder in "Dr Terrible's House of Horror - Scream Satan Scream" Episode...

Thor