Except its not only all about people hearing what they expect to hear (or are temporally primed to hear). Its never so simple as that with humans. Sometimes some people notice when they don't hear what they expected to hear (or when they do hear something unexpected).
So, anyway, we have some people who can notice or at least feel side-effects of what is wrong with, say, for example, the sound of ESS Hyperstream II dac chips, or whatever. And that's not everyone. But it isn't so simple as to ABX a bunch of people and stress them out into not being able to score 95% in a quick series of trials on equipment they are unfamiliar with. So the guys that don't hear whatever it is then jump to the conclusion that all those people are imagining things that aren't real. At the same time, people certainly can imagine things that aren't real. So, the reality is complicated, not as simple as guys who don't hear the same things would like it to be. So then they are in denial of complexity of human perception. Its a huge mess that is holding back any further progress in the field of audio.
Then we have the people who have an honest subjective belief in the infallibility of harmonic analysis as the explanation for how everything sounds. It completely misses crest factor of combined harmonics, and it misses a lot of other things. Then it becomes too much like trying to read tea leaves to explain why this or that thing sounds the way it does. Again, its a huge mess, this time arising out of the WYSIATI human bias.
A few guys in a diyaudio forum are never going to be able to clean up the messes. It may never happen, certainly not in our lifetimes.
Therefore we just have to learn how to get along with each other. Always busting into someone else's threads every time subjective sound is described isn't going to be the solution. It can't be, too much about sound is subjective to demand it be eliminated from discussion.
So, anyway, we have some people who can notice or at least feel side-effects of what is wrong with, say, for example, the sound of ESS Hyperstream II dac chips, or whatever. And that's not everyone. But it isn't so simple as to ABX a bunch of people and stress them out into not being able to score 95% in a quick series of trials on equipment they are unfamiliar with. So the guys that don't hear whatever it is then jump to the conclusion that all those people are imagining things that aren't real. At the same time, people certainly can imagine things that aren't real. So, the reality is complicated, not as simple as guys who don't hear the same things would like it to be. So then they are in denial of complexity of human perception. Its a huge mess that is holding back any further progress in the field of audio.
Then we have the people who have an honest subjective belief in the infallibility of harmonic analysis as the explanation for how everything sounds. It completely misses crest factor of combined harmonics, and it misses a lot of other things. Then it becomes too much like trying to read tea leaves to explain why this or that thing sounds the way it does. Again, its a huge mess, this time arising out of the WYSIATI human bias.
A few guys in a diyaudio forum are never going to be able to clean up the messes. It may never happen, certainly not in our lifetimes.
Therefore we just have to learn how to get along with each other. Always busting into someone else's threads every time subjective sound is described isn't going to be the solution. It can't be, too much about sound is subjective to demand it be eliminated from discussion.
Last edited:
As I said the main point of that article is that our perception is heavily influenced by our subjective beliefs on the physical world. That is nothing new. You still haven't explained what in that article supports your claim that your audible fidelity is not based on subjective beliefs but is more scientific than that of ASR members.
Was that the precise claim? Didn't seem that way to me. Thought the claim had to do with the usefulness of a lot of the prior published research on hearing, maybe things like thresholds in untrained subjects, and or the degree of their general applicability. IOW, more recent evidence suggests there is likely a dearth of old listening test science that will hold up to serious modern scrutiny. Doesn't mean some people do or don't notice more or less small effects than other people. That was always known about to some extent.
The difficulty of measuring perceived differences in a meaningful was was not so much known in the prior, more early days of research, would seem to be more the issue at hand. Thus some mistaken conclusions were likely drawn; and therefore some conclusions might not have been absolute, indisputable facts after all.
Last edited:
First, I did not state that this article supports any claimes. I stated it was interesting.
Second, I made no claim about "my" audible fidelity.
I observed that audible fidelity is a well understand scientific concept and that signal fidelity and audible fidelity are unrelated.
I cited as evidence research into audible fidelity that underpins perceptual coding, specially MP3, which was based on large scale blind listening tests that were seeking to determine what levels of signal fidelity impairments could be tolerated without audible fidelity impairments and if audible fidelity impairment did occur as result of decreased signal fidelity, if such impairments could be made to not be objectionable.
As usual, I see no actual problem here at all, nor in stating that there is no demonstrated link between signal fidelity (how closely the electrical and acoustic signals resemble a given stimulus ) and audible fidelity (how humans precieve said signal fidelity impairments listening to music).
I should also think that there nothing contentious in stating that all electromechanical audio production/reproduction chains give rise to gross signal fidelity impairments.
Thor
Last edited:
Still though, didn't it turn out that people could learn to hear what was wrong with mp3s? IOW, people could be variously fooled by mp3, with some people noticing the differences between that and CD. And trained persons could tell even better.
My own experience with MP3 was that 256kbps or 320kbps mp3 could in some cases sound subjectively better than CD. To my ears it was because the mp3 encoder was removing some objectionable junk from the particular CDs.
My own experience with MP3 was that 256kbps or 320kbps mp3 could in some cases sound subjectively better than CD. To my ears it was because the mp3 encoder was removing some objectionable junk from the particular CDs.
According to the article from TU Dresden how humans percieve is based on their subjective beliefs. So that makes audible fidelity subjective, not something universal. Which is why I have to disagree with your statement that there is some science-based audible fidelity that applies to everybody. And your claim that there is no demonstrated link between signal fidelity and audible fidelity is also not true for everybody.
Not sure I have the right link to that, could you please post it or else provide a citation? The premise as stated sounds bit unlikely, although maybe it says that. Would seem to make people untrainable without changing their subjective beliefs, yet it is well known people can be trained in various listening skills, including among other things, as musicians are trained in college level music ear training and transposition classes. It can be hard for them to learn to deal with masking, but apparently to some extent they are able to do it well enough to pick out different instruments playing the same pitches, including for instruments playing at reduced volume levels along with other louder instruments.
Last edited:
I did not claim MP3 was perfect, nor was it designed to.
It was designed to allow acceptable audible fidelity degradation with extreme signal fidelity degradation due to data reduction from 1411kbps to 192kbps or less. To do that essentially almost 75% of the complex signal needs to be removed.
My point was that the underlying research demonstrated comprehensively and scientifically the fact that signal fidelity and audible fidelity are unrelated.
So debating signal fidelity in isolation is pointless and actually unscientific. Or more precisely, it is cargo cult science.
Thor
https://tu-dresden.de/tu-dresden/ne...as-was-sie-zu-hoeren-erwarten?set_language=en
The actual research used measurements of brain activity.
It showed that brain activity related to certain sounds happened, if the sounds were expected but actually were absent.
So literally people were hearing things that didn't actually exist because they expected to hear them.
What this means, for example, if you have an expectation as to what you will hear in a given listening test, blind or sighted, there is a high likelyhood that your expectation will override your hearing.
Say someone invited a person to a listening test of mains cables and that person held a strong belief that audible differences between mains cable do not exist and mains cables are a scam.
Now let's also postulate another person who greatly believes in mains cables making a huge difference in the same test.
Finally we have someone who was invited without being told what is being tested.
Now say the sneaky b@stard doing the test, instead of changing mains cables behind the curtain switched the polarity of one of the two stereo channels (literally just swap the speaker cable poles at the amp).
Would you venture a guess for an ABX style test of this actual grossly audible difference, who of our three testers would identify the difference correctly and who will fail? Bonus points for offering an explanation why this would be the case.
Thor
I think it would depend. Some cables sound quite a bit different from other cables (like they can make the bass blurry, or not, or something more or less like that), but that isn't always the case. Sometimes any differences can be pretty subtle.
Also, it would depend on someone's memory of the stereo imaging or the music itself to notice the channel switch for what it is (no surprise since its well known ABX requires memorization).
For instance, there is guy I know who NP said still has one of the best sets of ears in the business. I have seen the guy listen A/B to a song twice using two different power amps and rattle off a list of everything he heard that was different. Funny thing was I heard all those things too (there was once a time when I did listening and mixing professionally). As it happened, it was the day when I decided to stop using the AHB2 with the very difficult to power Sound Lab speakers.
However, neither of us noticed on another day when I played the same standard test music we normally use but with channels accidently swapped (without any A/B). After that I made it a point to memorize the instrument channel mapping in a particular recording which I now use to make sure the channels are not swapped before any listening tests.
Last edited:
Regarding the linked article, the open source paper is available in full at: https://elifesciences.org/articles/64501
Kind of looks to me like its more about short-term expectation (or priming, in the psychological parlance), rather than long term test subject subjective beliefs.
Kind of looks to me like its more about short-term expectation (or priming, in the psychological parlance), rather than long term test subject subjective beliefs.
Indeed. However the same mechanisms may be expected to be at work there.
We have plenty of evidence outside of audio/sound for learned brain responses, where the brain responds before the actual stimulus is observed.
So we have to put these two together.
Would you venture a guess on my little proposed experiment?
Short summary, swap the polarity in one channel, which has gross audible impacts, but claim you are switching mains cables, which you do not.
Set the test up under the ABX protocol.
Procure three subjects, one strongly biased to the "mains cables are BS" position, one strongly biased towards "mains cables make big differences" the third who is not biased towards either position (ideally kept in the dark about what change they will hear).
Have all of them take the test. Note, we are not testing if swapping the polarity in one channel is audible. It is. We are instead testing if the bias in the subject can influence the outcome of a double blind test.
And yes, I know the practical outcome, it is statistically highly significant as well, but as the numbers are so small statistical power is low.
Thor
Last edited:
Not really. Don't enjoy gambling. Also, don't know you would ask if anyone wants to guess unless the result was initially non-intuitive.
That said, if I move my head when listening to speakers, I usually notice right away if there is any phasiness in the center image. One channel swapped or some reflector problem in the room. A wider perceived soundstage is also a give-away. That type of thing has happened a few times when I was listening for other things. Then I have to stop and fix it.
Last edited:
Ok, reveal.
Only unbiased participants can "hear" the change. They turn in a 10/10 result.
The biased participants could not hear the change with any significance based on a x/10 score as used in ABX.
In other words, at least in my experiment expectation bias overrode actual acoustic stimulus.
It would need a lot more testing to be able to make this into something to publish, but seeing what we know generally and specifically how bias works I would have predicted (I did actually) the outcome as observed.
Upshot, listening tests in which the subject is aware of what they listen to, are almost invariably biased, even those that are double blind. This makes their outcome unusable and their use as evidence to support any position on what is not audible highly questionable.
Thor
Funny thing is, even heavy biased (stubborn like a mule), can hear the difference. My best man, heavy stubborn man, very much so biased into cables are not making a difference. I sat him, let him listen to one well known song, swapped power cables (regular rubber cable that comes with most equipment, and diy cable), have him listen to the same song/file, and he was flabbergasted. What's important is to keep an open mind, it is much easier to notice differences, as there are many objectively better or worse things we notice, and definitely something subjective. It is always a mix, never a black and white picture.
Okay. Do you ever try Descriptive Analysis listening tests? It requires some expertise on the part of listeners because they have to know the descriptive language, which means they need to be trained by listening as to what the words mean. Kind like pointing to a banana then a school bus and saying, "yellow," for one simplified example. Then test them on a carrot, a lemon, and cardboard box, something like that. Did they learn what yellow means? Once you have some trained listeners, IME it can be worth the training effort in terms of the amount of information you can then get from a small-ish group of people. When their reports correlate well, a lot of information can be gotten without a whole lot of ABX trials. While its not accepted for some publication purposes, IMHO it can be very useful for internal use. Its something already being done in other areas of perceptual research.
Also, of course you have to interview listeners separately and so forth so they don't have a chance to influence each other.
Last edited:
This thread needs to get back on track discussing what I will present as my approach to getting the best or close to the best subjective sound from MarcelvdG's RTZ dac, including in relation to clocking, reclocking, and other factors external to the dac board itself. For the purposes of on-topic discussion and to provide some relevant context, a brief review of the dac board will be given and an opportunity will be given for readers to ask questions.
First, a dac schematic, version 10, is attached below. The schematic consists of 5 pages, the first page of which shows I/O connections, and interconnections of dac circuitry shown on separate schematic pages. Thus the boxes shown on the first page do not represent different physical boards. IOW, all 5 pages of the schematic refer to one single board.
As can be seen on the first page there are a few inputs shown in the left, including the 3 I2S bus inputs. The lower right shows power inputs, and a connector to share dac power with a separate output stage board (shared power being a not so good idea, IMHO).
The upper right corner of the first page shows the differential outputs for both channels which are intended to go to the inputs of a balanced output stage board. We can talk about the original output stage board and some alternate output stage designs later if there is interest.
Sheet 2 of the schematic shows a low noise voltage reference composed of long strings of transistors on the left. On the lower right are amplifiers and output stages to supply low noise power to certain parts of the dac requiring such power. Logic output signals are also provided to indicate when the power supply is up and running.
Sheets 3 and 4 show the the "shift register" architecture where digital audio signal are actually converted to analog via a FIRDAC structure.
Sheet 5 shows signal conditioning of the I2S input signals in order to prepare them to drive the "shift register" FIRDAC structures. Sheet 5 also shows mute circuitry at the top to mute the dac in the absence of DSD input signals, and or when dac power is not fully ready.
IMHO, a lot of the cleverness of this particular dac design is in the way the "shift registers" are used. Actually, the chips themselves are D-flip flops, the same logic function we use for reclocking. However, in the dac the chips are wired up to work like shift registers. The left and right channel data that is pre-conditioned on pages 3 and 4 of the schematic are shifted in the shift registers in a way that cancels out as much as possible the effects of ground bounce and Vdd bondwire inductance, and that cancels out as much as possible data-dependent variations on Vref (AVCC) power supply loading (which refers to the low noise power rails which power the "shift registers"). In brief, it is desired that the current flow from Vref supplies through the shift register outputs and into the output stage is overall always constant (except maybe for a little non-cancelation during switching transients and or because of other small, unavoidable nonidealities).
This is all well and good, and quite clever so far, except...
...to be continued.
First, a dac schematic, version 10, is attached below. The schematic consists of 5 pages, the first page of which shows I/O connections, and interconnections of dac circuitry shown on separate schematic pages. Thus the boxes shown on the first page do not represent different physical boards. IOW, all 5 pages of the schematic refer to one single board.
As can be seen on the first page there are a few inputs shown in the left, including the 3 I2S bus inputs. The lower right shows power inputs, and a connector to share dac power with a separate output stage board (shared power being a not so good idea, IMHO).
The upper right corner of the first page shows the differential outputs for both channels which are intended to go to the inputs of a balanced output stage board. We can talk about the original output stage board and some alternate output stage designs later if there is interest.
Sheet 2 of the schematic shows a low noise voltage reference composed of long strings of transistors on the left. On the lower right are amplifiers and output stages to supply low noise power to certain parts of the dac requiring such power. Logic output signals are also provided to indicate when the power supply is up and running.
Sheets 3 and 4 show the the "shift register" architecture where digital audio signal are actually converted to analog via a FIRDAC structure.
Sheet 5 shows signal conditioning of the I2S input signals in order to prepare them to drive the "shift register" FIRDAC structures. Sheet 5 also shows mute circuitry at the top to mute the dac in the absence of DSD input signals, and or when dac power is not fully ready.
IMHO, a lot of the cleverness of this particular dac design is in the way the "shift registers" are used. Actually, the chips themselves are D-flip flops, the same logic function we use for reclocking. However, in the dac the chips are wired up to work like shift registers. The left and right channel data that is pre-conditioned on pages 3 and 4 of the schematic are shifted in the shift registers in a way that cancels out as much as possible the effects of ground bounce and Vdd bondwire inductance, and that cancels out as much as possible data-dependent variations on Vref (AVCC) power supply loading (which refers to the low noise power rails which power the "shift registers"). In brief, it is desired that the current flow from Vref supplies through the shift register outputs and into the output stage is overall always constant (except maybe for a little non-cancelation during switching transients and or because of other small, unavoidable nonidealities).
This is all well and good, and quite clever so far, except...
...to be continued.
Last edited:
Moving along then, one thing conspicuously absent from recommendations by me has been an output stage design. I have tried a few different variations of output stages, including the original one designed by Marcel (and some variations on it). There are some other approaches which can be found in Marcel's RTZ FIRDAC thread at: https://www.diyaudio.com/community/threads/return-to-zero-shift-register-firdac.379406/post-6847540 Most if not all of them are of the differential input type, which is what the RTZ dac was intended to work with.
This is were it starting getting to be more subjective: However, I and a few other people like the sound of the dac better if using an SE output stage. My particular preference is for a single-ended transformer output stage with DC blocking before the transformer. I also tried using the transformer in differential input mode, both with and without a grounded center-tap. In the case of a ground center-tap version, DC blocking was used at the dac differential outputs. Overall, I keep coming back to the SE transformer version as the one I like best. Problem with that is the particular transformer is custom version developed for sole use in a private customer's audio products. It isn't for sale to the public at all. I have the transformers here for evaluation purposes only.
Therefore, a few of us Marcel RTZ dac users have been looking at other transformers and other discrete output stage designs in order to try and find one we either like best or maybe 2nd best. So far there is not a clear 2nd place winner but we're still working on it.
One thing that should be mentioned in the context of using an SE output stage is that the dac is still designed to perform best if inverting and noninverting outputs for each channel are both loaded equally, even if only one output is used to drive the rest of the reproduction system. My personal guess is that equally loading is likely to be more of an issue if the output stage is a current mode type with I/V inputs. OTOH, less current should be drawn from the dac outputs if voltage mode bridging-input output stages are used. The transformer I use represents about 10k of loading in the audio band, and up into the ultrasonic region (not that I can hear anything up there, I can't).
That's more or less what I have to say about output stages for now.
--------------------------------------------------------------------------
The other thing I wanted to talk about is effects of using various S-cut crystal oscillators/clocks. Three types have been tried by me. They are Acko, Andrea Mori DRIXO (without doublers), and SC Pure. .
What I will say next is going to be purely subjective, so be forewarned: The three types of clocks all sound a little bit different from other, at least to me and the other listeners here using the reproduction system with Sound Lab electrostatic speakers. The most detailed and probably the most accurate sounding clock oscillators are Acko but there are some possible downsides. First, they are the most expensive by far; second, they are very revealing of recording's SQ. Great recordings sound excellent, but flaws in many recording are easily audible. Next are the DRIXO clocks with 22/24MHz crystals (as the RTZ dac would be sensitive to the use of doublers). The DRIXO oscillators are to my ears, at least if using my squaring board, somewhere between Acko and SC Pure. SC Pure are next in line. I would say they sound excellent/superior compared the best consumer market clocks I have listened to with this dac (Crystek 957, and NDK SDA). To me SC Pure are great for casual listening and much more forgiving of less than nearly perfect recordings (as compared to Acko). They are also the most affordable of SC-cut crystal clocks. The downside is that they lack some of the soundstage depth and openness of the Acko clocks (where open sound is kind of like having some "empty or black space" separating instrument sounds from each other at low levels, and also having some more ability to provide clear low level reverberation tail fadeouts and little reflection details). At least some of what I subjectively report hearing in this paragraph are things which I don't believe there is any good, meaningful, practical way to measure at this point in time. Again, please take note this clock paragraph is entirely subjective and in my own opinion.
That's all I have to say about clock comparisons at this point.
This is were it starting getting to be more subjective: However, I and a few other people like the sound of the dac better if using an SE output stage. My particular preference is for a single-ended transformer output stage with DC blocking before the transformer. I also tried using the transformer in differential input mode, both with and without a grounded center-tap. In the case of a ground center-tap version, DC blocking was used at the dac differential outputs. Overall, I keep coming back to the SE transformer version as the one I like best. Problem with that is the particular transformer is custom version developed for sole use in a private customer's audio products. It isn't for sale to the public at all. I have the transformers here for evaluation purposes only.
Therefore, a few of us Marcel RTZ dac users have been looking at other transformers and other discrete output stage designs in order to try and find one we either like best or maybe 2nd best. So far there is not a clear 2nd place winner but we're still working on it.
One thing that should be mentioned in the context of using an SE output stage is that the dac is still designed to perform best if inverting and noninverting outputs for each channel are both loaded equally, even if only one output is used to drive the rest of the reproduction system. My personal guess is that equally loading is likely to be more of an issue if the output stage is a current mode type with I/V inputs. OTOH, less current should be drawn from the dac outputs if voltage mode bridging-input output stages are used. The transformer I use represents about 10k of loading in the audio band, and up into the ultrasonic region (not that I can hear anything up there, I can't).
That's more or less what I have to say about output stages for now.
--------------------------------------------------------------------------
The other thing I wanted to talk about is effects of using various S-cut crystal oscillators/clocks. Three types have been tried by me. They are Acko, Andrea Mori DRIXO (without doublers), and SC Pure. .
What I will say next is going to be purely subjective, so be forewarned: The three types of clocks all sound a little bit different from other, at least to me and the other listeners here using the reproduction system with Sound Lab electrostatic speakers. The most detailed and probably the most accurate sounding clock oscillators are Acko but there are some possible downsides. First, they are the most expensive by far; second, they are very revealing of recording's SQ. Great recordings sound excellent, but flaws in many recording are easily audible. Next are the DRIXO clocks with 22/24MHz crystals (as the RTZ dac would be sensitive to the use of doublers). The DRIXO oscillators are to my ears, at least if using my squaring board, somewhere between Acko and SC Pure. SC Pure are next in line. I would say they sound excellent/superior compared the best consumer market clocks I have listened to with this dac (Crystek 957, and NDK SDA). To me SC Pure are great for casual listening and much more forgiving of less than nearly perfect recordings (as compared to Acko). They are also the most affordable of SC-cut crystal clocks. The downside is that they lack some of the soundstage depth and openness of the Acko clocks (where open sound is kind of like having some "empty or black space" separating instrument sounds from each other at low levels, and also having some more ability to provide clear low level reverberation tail fadeouts and little reflection details). At least some of what I subjectively report hearing in this paragraph are things which I don't believe there is any good, meaningful, practical way to measure at this point in time. Again, please take note this clock paragraph is entirely subjective and in my own opinion.
That's all I have to say about clock comparisons at this point.
Last edited:
Two comments from my side:
The latest and greatest schematic version is not version 10, but version 10_holdtimefix. The only difference is that R124, R127, R129 and R131 have been increased from 39 ohm to 270 ohm. Measurements from Mark showed that the data input signal to the first flip-flop of each shift register changed a little too close for comfort after the clock, this change fixes that by slowing down the charging and discharging of the PCB traces and flip-flop inputs. The latest PCB layout is still version 10.
Regarding the constant current drawn from the reference, the idea is to keep it constant with data. That is, the current varies over each clock cycle, but this does no harm as long as it varies in exactly the same way in each and every clock cycle.