Pcm1792 has correct 1bitDSM converter in DSD mode, where it employes 67 taps analog FIR. But TI must convert the voltage output into the current one to be compatible with PCM mode. If it has +5v and -5V like pcm1704 and DSD only, there can be another solution. That's why I was interested in discrete 1bitDSM, where true 1bitDSM is possible.
I clearly stated that the underlying structure is sdm based and never (purely) pcm anymore.
It's not me underestimating dsd nor was it you stating benefits of dsd, so what made you change your mind all of a sudden, I don't know. Maybe you changed it when 1 bit became 6 bits? It's still sdm, no matter how many bits you make it, so it's easy not to understsnd what your point was.
Btw what was your point exactly?
I honestly have no clue what you are talking about. Going to leave this thread to the topic now.
Yes, it is a starting point. But Input buffer, high resistor tent to add noise, cap's tent to add THD and with the combination of the required opamp's tent to add THD too... it's not easy..
that's why I still favor the WADIA 27 with the used PCM1704 and SPLINE oversampling.. against AK4490 & ESS9039Pro
Also keep in mind, a high powered DAC with it's RF content (EMI wise) will always deal with the EMI source and sink rules
This means all will tell nice boards but in EMI terms, it do not stops after the DAC output and made the howl design as a night mare
Hp
To make one final off-topic remark, Lagadec and Stockham published a most interesting analysis of the relation between passband ripple and pre- and post-echoes in AES preprint 2097, "Dispersive Models for A-to-D and D-to-A Conversion Systems", way back in 1984. Quite small passband ripples in linear phase filters correspond to substantial pre- and post-echoes - and I mean echoes in the frequency range that is traditionally considered audible for humans.
This is an artifact of equiripple filter design, if I recall correctly. The sinusoidal ripple in the frequency domain becomes an impulse in the time domain and where depends on the frequency of the allowed passband ripple.
1984. Yet here we are discussing it, 35 years later.
Nice find Marcel!
P.S.
The intra-channel phase distortion was tested with a multi channel system, not a normal stereo one as has been for the inter-channel phase distortion.
No idea how this translates to stereo intra-phase distortion.
P.P.S. Personally liking the off topic subjects a lot also, what does xx3stksm think of all this clutter, interesting or unnecessary distraction?
We can always move to another thread if necessary.
necessary off-topic
No need to move to another thread. I don't think off-topic is unnecessary. This is the thread about noDAC with FPGA and transistors, but real noDAC includes at least conversion from PCM to 1bitDSM, OS with digital filters, accurate oscillator, and low noise analog amplifier. Without them, noDAC can't be a real PCB. Even PCB design know-how is relevant to noDAC for me. I appreciate "off-topic."
My off-topic. I'm basically against time-domain analysis because there are many chances to misunderstand, and human ears are based on frequency domain. The 1st pic has the same sound at least for my ears. The upper is in phase, and the lower has the fundamental with a 45-degree phase shift. Both have the same FFT plot. I ignore time-domain difference since I can't hear the difference between them.
The 2nd pic is another example. The upper is so-called ringing. The lower has some phase shift. The former is ringing and the latter is distortive? Both are the same FFT, too. I regard them as a waveform with three spectrums, where phase information can be ignorable if your target is a human ear. Ringing or echo is a little bit subjective. I like more objective definition like three spectrums.
No need to move to another thread.
I don't think off-topic is unnecessary. This is the thread about noDAC with FPGA and transistors, but real noDAC includes at least conversion from PCM to 1bitDSM, OS with digital filters, accurate oscillator, and low noise analog amplifier. Without them, noDAC can't be a real PCB. Even PCB design know-how is relevant to noDAC for me. I appreciate "off-topic." My off-topic.
I'm basically against time-domain analysis because there are many chances to misunderstand, and human ears are based on frequency domain. The 1st pic has the same sound at least for my ears. The upper is in phase, and the lower has the fundamental with a 45-degree phase shift. Both have the same FFT plot. I ignore time-domain difference since I can't hear the difference between them. The 2nd pic is another example. The upper is so-called ringing. The lower has some phase shift. The former is ringing and the latter is distortive? Both are the same FFT, too. I regard them as a waveform with three spectrums, where phase information can be ignorable if your target is a human ear. Ringing or echo is a little bit subjective. I like more objective definition like three spectrums.
Attachments
My off-topic.
Just for the record I have too strongly disagree with you here (just as we apparently do about Jitter / PN). I can clearly hear the effects of time domain performance.
Even in "simple systems" (such a Opamps / Power amplifiers etc.) the adjustment of the compensation capacitors (thereby effecting the phase margin) has a massive sonic impact on audio quality.
This effect of loop phase margin can also be clearly heard on the DAC array PSU regulator, to little phase margin and the DAC sounds brighter (such as enhancement of siblance with Female vocals) - over compensated and the DAC has a dull sound, its about finding the balance that suits the system.
Your typical regulator (LM317, LDO's etc.) will have poor phase margin when incorrectly loaded by output capacitors (and most will cross unity loop gain around 1KHz which is smack in the middle of your most critical audio region). To prevent this, we design the regulator circuits that power the DAC array to achieve "active" regulation across the whole audio bandwidth to reduce "Capacitor sound", the regulator has a full 20KHz B/W.
Not only does the regulator loop B/W span the full 20KHz B/W but we also design for flat output impedance across the full 20KHz - so typically a flat 5mohms across the 20KHz B/W can be achieved, Kelvin sensed feedback from the center of the DAC array is used.
When playing with the Phase margin, Steps of just a few degrees prove to be clearly audible in practice and it can be troublesome to find the correct balance with standard Film capacitor values (Capacitors in the compensation loop).
Granted, that while playing with compensation will also have an impact on your loop gain at any spot frequency, however the small differences in "tuning" values we find ourselves tweaking can only be explained by the rather larger impact PM has on the Q (Percentage of Overshoot / Undershoot) of the circuit.
A phase margin change of 5Degs from 70Deg to 75Deg goes from about 2% overshoot to undershoot, and yet the sonic impact is so CLEARLY audible, yet will only have a 1-2dB change in any spot frequency of the loop gain)...
It should be noted that typically, "Unity Gain stable op-amps" where considered "Stable" with a PM of just 45Deg!!! - as a result they will ring like a bell when used in unity gain circuits.
I put the typical complaint of "bright sounding op-amps" down to a few obvious factors :-
1. The effective Negative input impedance of the input stage devices (especially with ultra high speed opamps).
2. Lack of phase margin
3. Input stage RF rectification
Personally I dont like the sound of any Opamp we have listened to, the great leap in sound quality when they are replaced with a fully discrete design is always repeatable and unquestionable.
Its with great heartache that the DAC design (PCB) we have just issued uses opamps to reduce the designs complexity (component count)
- but atleast we have buffered the Opamps output - still I dont sleep well at night!We will go fully discrete on the next PCB spin.....
What I find interesting is how it would appear some people are more sensitive to the time domain then others...
Last edited:
I would say there are two ringings out there; one is oscillation caused by NFB with insufficient phase margin, the other is an artifact of the band-limited system caused by a brick-wall filter. Your comment is about the former one if I don't have a misunderstanding. They apparently look similar but ultimately have another meaning. I don't ignore insufficient phase margin of an opamp and a regulator. I don't care about the latter one. As long as digital data is "legal," no ringing occurs. What you need to do is to make legal digital data.
My ignorance of phase is a relation between harmonics. Musical instruments have many harmonics. I can't hear the phase difference between harmonics. Only amplitude is relevant. I mean you don't need to have time domain waveform in such case.
My ignorance of phase is a relation between harmonics. Musical instruments have many harmonics. I can't hear the phase difference between harmonics. Only amplitude is relevant. I mean you don't need to have time domain waveform in such case.
There is some evidence people tend to be more sensitive to phase at low frequencies and for percussive sounds. However, there is contradictory evidence too. Perceptual testing of humans is very difficult to do well and EEs that have published about it though AES do not seem to keep up well with advances in human perceptual research. Doing it well is costly and complicated.
What is not nearly so controversial is that humans are sensitive to dynamically changing phase.
It is also well known in the mixing and mastering business that ear training helps a lot for being able to notice small details that go by quickly in music. What JohnW describes hearing is likely the result of many years of carefully listening to small differences in dac designs.
Most untrained/unskilled consumers are not likely to notice such things directly, although the presence or absence of such effects may influence how much some people enjoy listening to reproduced music, without them being consciously aware of the underlying reasons.
What is not nearly so controversial is that humans are sensitive to dynamically changing phase.
It is also well known in the mixing and mastering business that ear training helps a lot for being able to notice small details that go by quickly in music. What JohnW describes hearing is likely the result of many years of carefully listening to small differences in dac designs.
Most untrained/unskilled consumers are not likely to notice such things directly, although the presence or absence of such effects may influence how much some people enjoy listening to reproduced music, without them being consciously aware of the underlying reasons.
My point it they are BOTH time domain effects, and I can hear both - when we optimize the step response of either a "simple" analogue system - or the step response of the digital replay system.
Its important to actually listen rather then throwing a whole load of reasons why it cannot be audible - if there is an audible effect then its left to us to try and understand the cause... it might take years, but its important to recognizes and trust in the results of listening tests otherwise we will never move forwards - PCM digital has been a step backwards in ultimate audio quality - SDM (DSD) bridges this gap IMO.
I'm lucky that my designs are used in the recording industry so I get to listen first hand to studio session, master tape / live events etc. and get to work with recording engineers / musicians, so I really have a perspective of "audio quality" not just based on work within the closed confines of a lab...
When I see s studio Mastering with a Pro-tools setup then I'm pretty quick out of there as theirs no hope...
Last edited:
The way recording and mastering engineers are trained has changed drastically. It used to be an apprentice system at the best of the big studios which takes many years to learn. Now people learn what they do from internet forums and buy soundcards online to get started. The big studios are mostly gone. Technology and economics cause what they do. It will take a long time to get to great digital audio at affordable prices. Maybe another 20-years, especially given there is no money for public research. What people do learn is often kept as trade secret information. What is to be expected, then?
Last edited:
Its indeed such a sad sate of affairs, how can we get young people into this industry when the music they listen too is so badly recorded - the quality is simply not there... Everything these days is being "Mastered for iTunes" - say no more...
One of my interest is design an affordable quality solution to allow archiving of vinyl in highrate SDM - so atleast you can have the quality of vinyl, but with the convenience of digital... Lets be honest, its a pain to keep a vinyl system in top form, and having to clean records etc. Forget skipping a track or Cueing Backwards / Forwards... Vinyl sounds great, but its a real pain to use.
A few things:
Jussi, the developer of HQPlayer says it's primarily a personal preference what one likes or not but when asked he does tell certain filtering should/could match better with certain specific music, based on the type of instruments used. Basically differences in attack and delay of e.g. acoustic instruments with big dynamic range vs the current loud pop music.
On top of that there's the claim that certain ad converter characteristics could match better with certain filtering and modulator dac settings. Maybe that's so they don't amplifying eachother's weak points?
Learning to listen is the same as developing a taste for wine or food.
At first there's just sour, sweet etc. Basic stuff. This develops into recognizing specifics that is learned best and fastest when in a group of knowledgable people, discussion, pointing out and sharing is key.
Knowing musicians or going to live venues clearly help.This all takes time.
Once learned it's hard to undo, up to the point that all you can do is try to get to another better version of what you previously built and felt was among the best, otherwise one could lose the joy. Junky alert!
I totally agree, somewhat annoyed even, there's much lost by not reading between the lines, which is easy to do, and that getting those specifics right is key to enjoying music, indeed not just for trained listeners.
Most listening sessions we do are in a rather fast setting. You know the system, let it warm up, then your ears then the changing of the gear. Most of the times the change is very easy to hear and one's opinion is easily made, sometimes you just have to let it play for a few hours while not really paying attention and other things than with just quick A-B ing start to get revealed. Those are things not yet learned to recognize in order to be able to conciously look for but do get to you when given some time.
From then on one hopes in the future it will be easier to recognize.
I'm assuming being able to recognize certain characteristics like phase differences one needs to put on the trained engineer's ears and sometimes the consumer's ignorant ears. It's best if you're able to switch. Listening with closed eyes makes both easier sometimes.
I also suspect the phase response of the loudspeaker plays part in showing or masking the perceivability.
Edit: by no means am I trying to say I heard clear changes in phase variations per se. Never known to have specifically listened for that myself.
Last edited:
Expectation bias is a possible cause for such differences, or unintended non-verbal communication when you do a single-blind test, or simple level differences when the levels are not matched within 0.1 dB. Have you done any tests that exclude these basic issues and if so, how?
By the way, when comparing digital filters or sigma-delta algorithms, it is fairly easy to add a circuit that randomly chooses a filter/SDM and only indicates afterwards what you were listening to. With some non-volatile memory you can even use that unknown filter or sigma-delta over several listening sessions and switch off the equipment in between.
Regarding phase issues, the question is not whether they are audible, but how bad they have to be to become audible. To give an extreme example, with DSP techniques and lots of memory you could make a filter with perfectly flat steady-state sine wave response that delays the bass by a few milliseconds and the treble by an hour. I don't think anyone would believe that to be inaudible.
Last edited:
There are so many instances where there has been truly blind conditions - an engineer changes something without my knowledge, or the configuration registers are swapped by accident in software - or my engineers try to test me and each other behind our backs (such as swapping filter names) - I don't need to play silly games with setting levels to 0.1dB - that's just being childish and stupid. (Filter and modulators are normally always always normalised to within 0.5dB anyway).
There will be times when we have to decide between two options in listening tests (such as testing modulator designs / digital filters etc.) and we will have 3 options (one option is duplicated), only the software guy will know the combinations. In most case its pretty easy to identify the options and choose a preference, other times we get confused between the options with no clear preference - but that in itself is the answer.
When testing a large number of filters it can take a week or two to get comfortable with the differences, and after awhile one is able to reliably choose preferences. Quick A/B comparisons are worthless to me, its often that one sound initially impressive, but after a little more listening its weakness become more apparent.
A while back we had to work with 7 digital filter options on a new IC design, a few where quickly discounted, that's was the easy part, then comes the more intense listening to optimize the remaining few - to understand there impact on SQ - this can be very hard and tiring work...
I guess one trains oneself to identify the differences, once trained its very easy to differentiate them - and I appreciate the times when engineers have "rename" the filters as it gives me a sanity check, I clearly recall a recent conversation (Thanks Dom):-
"I dont understand, yesterday I was certain I preferred filter X, but today I prefer filter Y ..." and then the truth comes out (filter name swap)... So when I've been in such situations it DOES reconfirm my faith in our listening tests.
Other times someones played with your settings, your not expecting any change, than ask yourself why the system sounds so bad and then upon investigation see the settings have been changed, Bias? Non!!how could there be as I was NOT expecting any change...
I'd try not to pass judgement in an unknown system or with unknown music as I'd have no reference based on past experience. Over time we have developed our routines, we know the weaknesses and strengths of the system. Importantly we establish a baseline, so when the units are used in a different system atleast we have some confidence with the design, what it does well and more importantly its weakness. We always have a few different units on hand to serve as references to get ones mind in gear.
There have been a few times when I wanted to remove any risk of bias (such as when its something I've spent a long time developing and important to me) - I will ask my assistant to flip a switch behind my back and only afterwards show me the switch position I preferred...
I do feel sad for you when you have to ask such questions as it shows you have really not had experiences to gain your own confidence, it truly say more about you - that you had to ask then anything else.
I'll say no more on the subject because you either believe or you don't and there will be no change in opinion, I learnt that life is too short to convince people like yourselves - I'm nobody's teacher or preacher - nor fool! I just dont wish to go down this rabbit hole
Last edited:
Actually it's standard procedure in controlled listening tests. Small amplitude differences, say 0.2 dB to 1 dB, are often perceived as a difference in sound quality rather than volume.
There is no scientific basis I am aware of for a claim of existence of something called 'Expectation Bias' as the term seems to be commonly used in audio forums. However, the term does have a very specific meaning in the field of cognitive psychology.
Expectation Bias:
"The tendency for experimenters to believe, certify, and publish data that agree with their expectations for the outcome of an experiment, and to disbelieve, discard, or downgrade the corresponding weightings for data that appear to conflict with those expectations."
Please refer to the follow publication for some more info on the subject: Jeng M (2006). "A selected history of expectation bias in physics". American Journal of Physics. 74 (7): 578–583.
List of cognitive biases - Wikipedia
Last edited: