Jean-Paul, you just hit the nose with the shovel. 🙂
* I was hoping to get Dietmar himself to participate here, and share some friendship and education with us.
I hope there is still hope because my intention is purely of genuine knowledge, and love, and friendship.
** And I don't have a clue what they're talking about over there; only from friend's transmissions over private communications with full disclosures.
- You cannot privatize the public sector, or destroy people's reputation on false assumptions, gossips, and dictatorship.
Last edited:
Adam has posted a very helpful graphic and lots of explaining text over on WBF. Its mostly correct but has one or two minor errors, perhaps because he's swallowed what Dietmar has told him rather than studied for himself.
One very interesting point I'd like to pick up is this - he says :
Dietmar's idea was to keep what makes the NOS DACs great, but find a way to get rid of its problems.
Its a subject of fairly hot debate as to what makes NOS DACs sound great. It seems Dietmar thinks its the absence of both digital and analog filters which is responsible for the 'NOS sound', hence he's engineered Trinity to do away with these. Certainly Adam's correct that filters introduce some kind of signature, the issue is though whether it really is the absence of both digital and analog filters which is the key element. If Dietmar's right then the Trinity is indeed an excellent technical accomplishment - with some caveats. But what if he's wrong ? What if NOS sound is only incidentally correlated with absence of those filters and not caused by their absence?
One very interesting point I'd like to pick up is this - he says :
Dietmar's idea was to keep what makes the NOS DACs great, but find a way to get rid of its problems.
Its a subject of fairly hot debate as to what makes NOS DACs sound great. It seems Dietmar thinks its the absence of both digital and analog filters which is responsible for the 'NOS sound', hence he's engineered Trinity to do away with these. Certainly Adam's correct that filters introduce some kind of signature, the issue is though whether it really is the absence of both digital and analog filters which is the key element. If Dietmar's right then the Trinity is indeed an excellent technical accomplishment - with some caveats. But what if he's wrong ? What if NOS sound is only incidentally correlated with absence of those filters and not caused by their absence?
Richard, is it possible to reproduce that post from Adam in full here (without a link)?
I would love to read it and discuss it.
I would love to read it and discuss it.
Its possible, but I am not sure that cutting and pasting a whole post from a thread constitutes 'fair use' under copyright law. So I'm not going to do it.
Certainly its considered fair use to extract portions of a text for the purposes of education, which is what I intend to do here.
Certainly its considered fair use to extract portions of a text for the purposes of education, which is what I intend to do here.
By the way, who is the person who told (emailed) Dietmar about this diy's thread here?
That person should be thanked I think, for making more accessible this brand new thread to a larger segment of the audio community.
And now Dietmar knows about it. 😎
- I don't have to email him myself; the good work is already done. 🙂
That person should be thanked I think, for making more accessible this brand new thread to a larger segment of the audio community.
And now Dietmar knows about it. 😎
- I don't have to email him myself; the good work is already done. 🙂
Last edited:
Adam (Elberoth) is not already a member here @ diyAudio?
Also, you are 100% free to ask any mod here if it is legal to reproduce a reproduction. 🙂
* The graphs are certainly not his property, right?
Here's a claim of Adam's which does look to be rather misleading. He says -
The LIANOTEC gives an equivalent of 8x oversampling rate, without using 8x oversampling digital filter.
He goes on later to explain that the additional points needed to go from 1X up to 8X oversampling are 'situated linearly between' the original sampling points.
So then that means additional points are indeed calculated mathematically between the original points and therefore there is a digital filter.
The only way to avoid having a digital filter that I know of is my 'LAID' architecture. This uses delays (as has been claimed for Trinity) but also needs an analog FIR filter to produce the extra sampling points. I see no sign of the analog FIR in either the pictures or the writing about the DAC. Therefore I conclude that there is indeed digital processing going on to calculate the right values for the interpolated points. In engineering speak (rather than marketing parlance) this does indeed constitute a digital filter.
The LIANOTEC gives an equivalent of 8x oversampling rate, without using 8x oversampling digital filter.
He goes on later to explain that the additional points needed to go from 1X up to 8X oversampling are 'situated linearly between' the original sampling points.
So then that means additional points are indeed calculated mathematically between the original points and therefore there is a digital filter.
The only way to avoid having a digital filter that I know of is my 'LAID' architecture. This uses delays (as has been claimed for Trinity) but also needs an analog FIR filter to produce the extra sampling points. I see no sign of the analog FIR in either the pictures or the writing about the DAC. Therefore I conclude that there is indeed digital processing going on to calculate the right values for the interpolated points. In engineering speak (rather than marketing parlance) this does indeed constitute a digital filter.
We know already that there is: "Enhanced filtering" applied, and differently than your normal DAC.
Mentioned too is: "Impulse-optimized and Frequency Response-optimized".
And what there's not is: "No analog low-pass filter".
Mentioned too is: "Impulse-optimized and Frequency Response-optimized".
And what there's not is: "No analog low-pass filter".
Last edited:
Linear interpolation has been used before in place of FIR (call it polynomial) interpolation. It gives the benefit of less phase distortion at the expense of non-linear distortion (i.e. harmonic distortion). Since music is made up of superimposed sine waves and no sinewave has a straight line anywhere its bound to introduce harmonic distortion if you 'fill in' the missing points by linear interpolation.
I see distortion components on the spectrum analyser plot shown on Trinity's web page. There are just so many of them they look like an increase in the noise floor.
DAC
Here are some distortion plots. Mostly below -110 with one harmonic just over that mark. This is the case at all sample rates and inputs shown. Doubtful any of this is audible.
I see one plot of IMD on that page which looks pretty poor on the upper side of the 20khz tone. Images at only -20 db or so. Wonder if this contributes to it sounding different?
Here are some distortion plots. Mostly below -110 with one harmonic just over that mark. This is the case at all sample rates and inputs shown. Doubtful any of this is audible.
I see one plot of IMD on that page which looks pretty poor on the upper side of the 20khz tone. Images at only -20 db or so. Wonder if this contributes to it sounding different?
Last edited:
Different? ...Better?
* I'll talk to Adam and ask him if he can share some with us here. ...Adam is a good man.
His vision is not polluted; actually he can see and hear pretty clear. ...His view at home (on the lake I believe) is gorgeous; helps to clear the mind.
And thank you to the great member who suggested to me to start this thread in the first place; awesome Richard! 😎
And thank you too, to the person who made Dietmar aware of this thread; now he is free to join anytime, he knows exactly where we are.
And Dietmar, I hold no grudge against anyone (not even you) who made a wrong judgement about me in the past.
You were probably influenced by an incognito anyway.
And thank you to all the people who have a clear vision, and can see the good in others.
And that's exactly what I want to see; the good in the Trinity DAC (LIANOTEC) and its talented designer.
Free discussion, no cover charge.
It's not everyday we're talking about eight DACs per channel (B-B PCM-1704K), and which gives enough current as to get rid of the analog output stage normally necessary (99% of the time) to amplify the signal.
Two channels (balanced) means of course sixteen DACs all together, and hand-picked at that too.
Do you know the price of the 1704K (each) nowadays? ...About fifty bucks.
Eight of those per channel gives an effective oversampling rate of 8x; according to my readings, and permits you to get rid of the oversampling block.
A good link? ... By Ryohei Kusunoki ...Sakura Systems | Non-oversampling Digital Filter-less DAC Concept
Thx Adam.
* I'll talk to Adam and ask him if he can share some with us here. ...Adam is a good man.
His vision is not polluted; actually he can see and hear pretty clear. ...His view at home (on the lake I believe) is gorgeous; helps to clear the mind.
And thank you to the great member who suggested to me to start this thread in the first place; awesome Richard! 😎
And thank you too, to the person who made Dietmar aware of this thread; now he is free to join anytime, he knows exactly where we are.
And Dietmar, I hold no grudge against anyone (not even you) who made a wrong judgement about me in the past.
You were probably influenced by an incognito anyway.
And thank you to all the people who have a clear vision, and can see the good in others.
And that's exactly what I want to see; the good in the Trinity DAC (LIANOTEC) and its talented designer.
Free discussion, no cover charge.
It's not everyday we're talking about eight DACs per channel (B-B PCM-1704K), and which gives enough current as to get rid of the analog output stage normally necessary (99% of the time) to amplify the signal.
Two channels (balanced) means of course sixteen DACs all together, and hand-picked at that too.
Do you know the price of the 1704K (each) nowadays? ...About fifty bucks.
Eight of those per channel gives an effective oversampling rate of 8x; according to my readings, and permits you to get rid of the oversampling block.
A good link? ... By Ryohei Kusunoki ...Sakura Systems | Non-oversampling Digital Filter-less DAC Concept
Thx Adam.
Last edited:
It probably should be mentioned that the application of direct analog interpolation to audio conversion is an old technique. I believe that Krell has before utilized it in the DAC section of their top CD player of more than a decade ago, and that Wadia also apply it. in addition, one our own contributors, -ecdesigns-, has long had such a DIY project thread posted here.
http://www.diyaudio.com/forums/digi...building-ultimate-nos-dac-using-tda1541a.html
Also, I had independently developed such a design about a decade ago. However, two interrelated issues, one technical and the other practical, dissuaded me from actually constructing my own design. The main performance problem of linear interpolation is that it introduces an in-band response roll-off (I've got a plot of the curves stored somewhere in my computer). The good news on this issue is that the in-band roll-off becomes increasingly flat as the interpolation ratio is increased. If I correctly recall, decent in-band flatness is not obtained until the interpolation ratio is 8x or higher (I think Trinity mentions they use 16x). The bad news is that because analog linear interpolation is implemented via the time interleaving of D/A chips, the hardware complexity increases by the same interpolation ratio.
For example, an 8x analog linear interpolation requires 8 D/A units per channel. not to meantion, that the implications for management of system jitter similarly increases. However, I do believe that with the interleaving control logic neatly located in an FPGA, and with great pains taken to distribute time staggered low jitter clocks to each D/A unit, the benefits of analog linear interpolation can be had. The implications for commercial product cost should be obvious.
It does seem to me that implementing linear interpolation digitally via a high sample rate single D/A unit (per channel), rather than in analog form via 8 or 16 D/A units would be preferable. Certainly, from an system cost standpoint. However, such a single D/A unit would need to be able to support at least 8x oversampled input rates. The PCM1704 comes quickly to mind.
http://www.diyaudio.com/forums/digi...building-ultimate-nos-dac-using-tda1541a.html
Also, I had independently developed such a design about a decade ago. However, two interrelated issues, one technical and the other practical, dissuaded me from actually constructing my own design. The main performance problem of linear interpolation is that it introduces an in-band response roll-off (I've got a plot of the curves stored somewhere in my computer). The good news on this issue is that the in-band roll-off becomes increasingly flat as the interpolation ratio is increased. If I correctly recall, decent in-band flatness is not obtained until the interpolation ratio is 8x or higher (I think Trinity mentions they use 16x). The bad news is that because analog linear interpolation is implemented via the time interleaving of D/A chips, the hardware complexity increases by the same interpolation ratio.
For example, an 8x analog linear interpolation requires 8 D/A units per channel. not to meantion, that the implications for management of system jitter similarly increases. However, I do believe that with the interleaving control logic neatly located in an FPGA, and with great pains taken to distribute time staggered low jitter clocks to each D/A unit, the benefits of analog linear interpolation can be had. The implications for commercial product cost should be obvious.
It does seem to me that implementing linear interpolation digitally via a high sample rate single D/A unit (per channel), rather than in analog form via 8 or 16 D/A units would be preferable. Certainly, from an system cost standpoint. However, such a single D/A unit would need to be able to support at least 8x oversampled input rates. The PCM1704 comes quickly to mind.
Last edited:
I see things are hotting up nicely over on the WBF thread - the Phasure designer is demonstrating that his PCM1704s too are pre-selected.
The spectrum plots I was referring to where there's clearly more grass on one than the other are to be found here - Publikationen Lianotec
Notice that on the right where the test signal is 10kHz, the grass has grown above the second 10dB mark up from the baseline, whereas for the 20kHz the grass is significantly lower. So it seems that distortion has been turned into noise.
The spectrum plots I was referring to where there's clearly more grass on one than the other are to be found here - Publikationen Lianotec
Notice that on the right where the test signal is 10kHz, the grass has grown above the second 10dB mark up from the baseline, whereas for the 20kHz the grass is significantly lower. So it seems that distortion has been turned into noise.
Hey,
I am no party in this although it has my interest. So, I don't want to be seen by anyone as being in one camp or the other. But with Ken's post in mind I had already been looking for this :
One can have a patent, but it assumes that nothing of the technique applied has been to the public or one single person outside the legal entity (company) for that matter. This isn't necesserily about it being stolen, but just about the sheer fact that I can't get a patent because I am able to play music through a computer - so to speak. Because many can do this already, I won't be able to obtain the patent because it would back reverse all what's all happening in the world today (and sorry for my poor English).
Now this little subject (LIANOTEC) is interesting because it seems to be on the edges of "who was first". So yes, we all know about ecdesigns (nothing can be more to the public than his "little project" I'd say but he was way later than the 2002 of the patent), but what about others which at least I don't know about. Krell ? Wadia ?
Now it is not so easy for the patent judging party to find out whether the finding is really genuine hence not out there already;
So what I wondered is what would actually happen with this particular patent when afterwards it is found that at the time concerned (2002) the whole thing already existed ? I just don't know ...
So see ? I just turned it upside down; No DIY guy will go to court for this (like ecdesigns and followers). I (with commercial product) will or should.
But again the other way around : can Trinity now sue Krell and such (when later than 2002) ? theoretically Yes, because the patent is en public. So nobody can claim afterwards "I did not know". Worse, it can have been the example and it can not be proven that it has not been (even ecdesigns can have gotten it from there - just saying but *without* the suggestion please !).
I can make it more complicated I think;
Just suppose Krell was earlier (than that 2002). Patent judges thus made a small mistake. Krell obviously didn't put a patent for it. Outside of earlier implied complexities, now I have that commercial product and claim I copied it from Krell. The patent is worthless now. If I were Trinity I would sue that patent judging party from back then.
Remember, I don't draw party here; it just intrigues me.
Peter
(Phasure)
I am no party in this although it has my interest. So, I don't want to be seen by anyone as being in one camp or the other. But with Ken's post in mind I had already been looking for this :
One can have a patent, but it assumes that nothing of the technique applied has been to the public or one single person outside the legal entity (company) for that matter. This isn't necesserily about it being stolen, but just about the sheer fact that I can't get a patent because I am able to play music through a computer - so to speak. Because many can do this already, I won't be able to obtain the patent because it would back reverse all what's all happening in the world today (and sorry for my poor English).
Now this little subject (LIANOTEC) is interesting because it seems to be on the edges of "who was first". So yes, we all know about ecdesigns (nothing can be more to the public than his "little project" I'd say but he was way later than the 2002 of the patent), but what about others which at least I don't know about. Krell ? Wadia ?
Now it is not so easy for the patent judging party to find out whether the finding is really genuine hence not out there already;
So what I wondered is what would actually happen with this particular patent when afterwards it is found that at the time concerned (2002) the whole thing already existed ? I just don't know ...
So see ? I just turned it upside down; No DIY guy will go to court for this (like ecdesigns and followers). I (with commercial product) will or should.
But again the other way around : can Trinity now sue Krell and such (when later than 2002) ? theoretically Yes, because the patent is en public. So nobody can claim afterwards "I did not know". Worse, it can have been the example and it can not be proven that it has not been (even ecdesigns can have gotten it from there - just saying but *without* the suggestion please !).
I can make it more complicated I think;
Just suppose Krell was earlier (than that 2002). Patent judges thus made a small mistake. Krell obviously didn't put a patent for it. Outside of earlier implied complexities, now I have that commercial product and claim I copied it from Krell. The patent is worthless now. If I were Trinity I would sue that patent judging party from back then.
Remember, I don't draw party here; it just intrigues me.
Peter
(Phasure)
The old Krell unit of which I was referring is the KPS-25s. First let me say, that I have not seen a schematic of this unit. What I have is a product review published by the long defunct 'Fi' magazine, volume 3, issue 9, dated September 1998. In that review, there is a description of how Krell implemented a custom filter operating two DAC units, which, and I quote, "...slightly delays the audio samples feeding the second DAC." While I'm certainly reading between-the-lines here, and could be completely wrong in my assessment, that sounds like a description of time-staggered hardware based 2x analog linear interpolation to me.
From what I can tell, this Krell player also utilized an SINC function digital interpolation filter to first obtain 8x oversampling, then, obtained a final 16x oversampling ratio by time interleaving the two DACs to achieve a 2x analog linear interpolation of the 8x digitally interpolated signal. Having only two DACs, it could not delivered a fully analog 16x linear interpolation.
From what I can tell, this Krell player also utilized an SINC function digital interpolation filter to first obtain 8x oversampling, then, obtained a final 16x oversampling ratio by time interleaving the two DACs to achieve a 2x analog linear interpolation of the 8x digitally interpolated signal. Having only two DACs, it could not delivered a fully analog 16x linear interpolation.
Last edited:
Interesting Ken. But well, I don't think I would be able to obtain a patent because I'm glueing 16 1704's together like this (per channel) (and Trinioty holding the patent for 8x). It will be about the principe / the technique. Two of them seems enough to me.
Also, nobody says that 3,072 MHz is enough to be out of all amp's bandwidth. So the 8x doesn't do much either.
Edit : I meant it would hold back my patent for 16x (virtually of course).
Also, nobody says that 3,072 MHz is enough to be out of all amp's bandwidth. So the 8x doesn't do much either.
Edit : I meant it would hold back my patent for 16x (virtually of course).
Last edited:
- Status
- Not open for further replies.