Hi tbrowne,
Although I may sound sceptical about your project, but I am not, really. After all I don't know nothing about 0 & 1 (except that I have built a couple of TDA's dac as anyone else around here). Please, go on, you have my full attention.
BTW, did I managed somehow to design a 16-bit R2R differential dac?
Although I may sound sceptical about your project, but I am not, really. After all I don't know nothing about 0 & 1 (except that I have built a couple of TDA's dac as anyone else around here). Please, go on, you have my full attention.
BTW, did I managed somehow to design a 16-bit R2R differential dac?
Please, go on, you have my full attention.
There isn't much more I can say until someone makes a commitment to build. 🙄
If everyone is waiting for me to develop a "complete" schematic, they're going to be in for a long wait, and haven't been reading properly what I've been saying... there are plenty of people here who would be able to contribute to the interfacing issue if they so wished.
I spent a little time today looking for other potential donor players to hopefully inspire people. Unfortunately, it's hard to find a player with a good filter that doesn't cost the earth. A few Sony ES have switchable filtering, but no data is available on them, AFAIK... would need to reverse-engineer the pinouts, and the players are probably worth too much to modify.
Casting back my mind, I seem to remember that an old Theta Data Basic II on loan used a custom filter implemented on Motorola DSP, and did a lot of things right for an S/PDIF DAC... they aren't too expensive on the second hand market either.
Case is horrible, though... rings like a bell... clannnnnggg...
The simulation certainly shows a differential output, given perfect impedances. I don't think there is anything significant to be gained from differential operation in this case, though... the noise isn't an issue, and the even harmonics don't need attenuating.
Four separate ladders used randomly a la DEM would give greater measured performance benefits at similar parts cost... but I don't really want to go into that right now.
- Tom.
Measurement maybe is the most important requirement for a discrete DAC project.
If distortion is harmonic, it is still bad if you do not want to change original sound.
Listen to a 1 kHz -60dB signal via 14 bit DAC and compare to a clean DAC.
You should hear a difference ...
There is no speculation at all. All correlations are proofed by experience.
It is so funny how many people refuse to accept that the pattern of harmonics of a DAC are 90% responsible for the sonic signature.
On the other hand they blame unmeasureable artefacts for sonic degradation.
A 1540 spectrum will always look like a 1540 spectrum. Same for a 1541 or a BB chip. Only if performance is degraded by optocouplers or other design flaws it will look different.
Tom,
just looked at your -60 dB measurement, and would say you already got 14bit performance ( K2 -40dB, K3 -50dB ) which for sure is not easy.
What I see is a little noise, -50dB around 1 kHz, integrated 16 bit chips will give -60 to -63 dB
No. The most important requirement for an audio enthusiast is that it sounds good playing music. Measurement is just a tool which can sometimes assist in getting there.
Perhaps you should think on that some more.
You might want to think on that some more too.
Really? From here, it looks awfully like you've seen something on your analyser, and convinced yourself that you hear what you're seeing.
What I found particularly good was the "rich tube-like sound" emanating from harmonics that aren't even audible to mere humans at less than ear-destroying 0dBFS levels.
It is somewhat more amusing when people pull figures out of their a**e... don't you agree?
When you can provably identify audible differences in players with a -60dBFS 1K tone when 0dBFS is at a normal listening level (players level matched to at least 0.1dB, naturally), then you might have something to contribute.
Otherwise, all I see is academic onanism. That's the kind that leaves you wondering where the years went...
On the whole, "they" couldn't care less whether problems are measureable or not. They just want them to go away.
It's not an entirely unreasonable position for someone who cares more about results than how they get there - providing they are satisfied (some are more easily satisfied than others, of course), but I would recommend basic tests as a guide to save time in the longer run.
Hey - look at that... daddy, I can read a graph too!
There's only one noise problem in this thread, and it isn't the DAC...
- Tom.
tbrowne said:
All subjective. Playstation one is another good example.
tbrowne said:
It is untrue for you that adding distortion does change sound. 😀
tbrowne said:
I heard the differences long before I was owning an analyzer, bought the CD304 in 1987 and had it for about 15 years, beside that I have had about 25 other players, believe me, I know what I am talking about.
The differences between a CD304 and a 304mkII are very audible, friends of mine hear it too.
From here it looks like your approach is too academic, better stop calculating dBs in theory...
tbrowne said:
Again you are in error with your calculations.
Do you really believe that the harmonics suddenly show up at -60dB ? 😛
I do not want to say that I know everything, still I ask lots of questions in the forum, but you seem to lack the very basic understanding of the subject, which is that the distortion of a DAC is rising continiously with falling signal level.
If you still don't get it: The problem starts at -20dB.
The term "digital sound" is characterised by increased low level distortion and is similar to cross over distortion in bad class B amps.
tbrowne said:
See above.
tbrowne said:
Your hobby ?
You try to prove me wrong by presenting theoretical calculations.
Game over.
Really? I cannot seem to find any reference to such a statement, implied or otherwise. There is only one way you could reach such a conclusion.
Ray - sorry, I now see where you're coming from. Blame it on being relatively young and idealistic.
Nope.
Nope.
You seem to lack the very basic understanding of the subject, which is that a lower signal level is less audible than a louder one of the same frequency.
More of a fetish. I've cut back, preferring to build instead.
There's nothing theoretical about the acoustic noise floor of a real room.
Thank goodness electrons are so cheap.
- Tom.
Mod hat on.
Gentlemen, this discussion is becoming unpleasant, both for the participants and the viewers.
Please behave, forget about your egos and show just a tiny bit of respect for others opinions/beliefs.
You've been warned. Consider this an official hint .......
Mod hat off.
Gentlemen, this discussion is becoming unpleasant, both for the participants and the viewers. Please behave, forget about your egos and show just a tiny bit of respect for others opinions/beliefs.
You've been warned. Consider this an official hint .......
Mod hat off.
From my POV, a little bartering is still better than silence... slightly... it's probably quite clear I was getting a little frustrated trying to keep this thread on-topic, and it started to show... as such, the moderation could have been more constructive, IMO.
I do have strong ideas on what makes a good audio D/A, and it has little to do with mere harmonics... a particularly good example is putting an output buffer on the DAC... neither of the two showed significant difference in harmonics, but the sound was strongly degraded.
Or how about the AK4524? The DAC side of this can sound utterly appalling... the -60dB 1K plot can be completely clean, even using the nastiest support components with a poor power supply and clock. How helpful, eh? No soundstage depth, and a horrible, spitty, cheap sound. I have heard a much better implementation, but even that one was only talented at making bad recordings more listenable (could actually listen to WTSMG).
When it comes to the Discrete, to an extent, I can change the measured performance far more with DSP than I can through changing the analogue side (as I had already implied quite strongly when talking about the -60dB test), yet beyond changing the digital filter, the various techniques for dither and shaping give much smaller effects in actual subjective performance (to my listening criteria, anyway).
We've reached a bit of an impasse here, really, haven't we? The reason I released the information and results from many months of R&D is that I felt there was no real commercial application for such a design (much to my disappointment... I do try to do this stuff for a living) - at least for one that didn't "measure pretty" against some things that have only passing relation to the human ear.
The parts cost alone is considerable, even with careful sourcing of resistors, and the amount of labour to build just the published design to a high standard is only feasible for a serious enthusiast.
I have been here to offer advice to such people who want to take it on. The schematic is there for those who prefer to source their own bits, and plenty of information has been published with regard to interfacing.
If people are considering building, but have questions, they should ask them. If not, the thread has no future. If I really didn't think there was anything special about this project, I wouldn't be wasting my time or anyone elses', for that matter.
- Tom.
I do have strong ideas on what makes a good audio D/A, and it has little to do with mere harmonics... a particularly good example is putting an output buffer on the DAC... neither of the two showed significant difference in harmonics, but the sound was strongly degraded.
Or how about the AK4524? The DAC side of this can sound utterly appalling... the -60dB 1K plot can be completely clean, even using the nastiest support components with a poor power supply and clock. How helpful, eh? No soundstage depth, and a horrible, spitty, cheap sound. I have heard a much better implementation, but even that one was only talented at making bad recordings more listenable (could actually listen to WTSMG).
When it comes to the Discrete, to an extent, I can change the measured performance far more with DSP than I can through changing the analogue side (as I had already implied quite strongly when talking about the -60dB test), yet beyond changing the digital filter, the various techniques for dither and shaping give much smaller effects in actual subjective performance (to my listening criteria, anyway).
We've reached a bit of an impasse here, really, haven't we? The reason I released the information and results from many months of R&D is that I felt there was no real commercial application for such a design (much to my disappointment... I do try to do this stuff for a living) - at least for one that didn't "measure pretty" against some things that have only passing relation to the human ear.
The parts cost alone is considerable, even with careful sourcing of resistors, and the amount of labour to build just the published design to a high standard is only feasible for a serious enthusiast.
I have been here to offer advice to such people who want to take it on. The schematic is there for those who prefer to source their own bits, and plenty of information has been published with regard to interfacing.
If people are considering building, but have questions, they should ask them. If not, the thread has no future. If I really didn't think there was anything special about this project, I wouldn't be wasting my time or anyone elses', for that matter.
- Tom.
tbrowne said:
Please don't try to have the last word in this, Tom. It is not upon moderators to moderate contructively, we do the housekeeping here if discussions go out of control and such. We leave constructive remarks to the participants of the discussion. You are one of them and as such you are responsible as well for the direction things are going. Please reread your own posts to see what I mean.PLEASE
Moderation was required, I think it could be toned down a little but PLEASE continue the discussion as I find it very interesting indeed.
I like to read everyones views but not when people are just disagreeing between themselves.
Nothing matters as long as you enjoy what you have been doing / listening to. Its our hobby right ?
Moderation was required, I think it could be toned down a little but PLEASE continue the discussion as I find it very interesting indeed.
I like to read everyones views but not when people are just disagreeing between themselves.
Nothing matters as long as you enjoy what you have been doing / listening to. Its our hobby right ?
Hi Tom,
I have a couple of questions about your simulation posted earlier.
I see the values you used have slight offsets from true R-2R.
I 'think' this is something to do with the 'switch impedance' and the amount of current pulled through each switch, but I'm a little stumped if I can work out how you go about calculating the values for each resistor.
My aim is to produce a 16bit ladder that uses similar values to yours, and uses the same logic devices. I have worked out the decoding logic I need (1st go ended up over-complex, 2nd go ended up simpler than I expected)
I "think" I can just trim off the 4 Least Significant Bits, but I'd like to have a bit more understanding behind what I am doing before I go away and build something.
I'd appreciate any guidance you can give.
Jim.
I have a couple of questions about your simulation posted earlier.
I see the values you used have slight offsets from true R-2R.
I 'think' this is something to do with the 'switch impedance' and the amount of current pulled through each switch, but I'm a little stumped if I can work out how you go about calculating the values for each resistor.
My aim is to produce a 16bit ladder that uses similar values to yours, and uses the same logic devices. I have worked out the decoding logic I need (1st go ended up over-complex, 2nd go ended up simpler than I expected)
I "think" I can just trim off the 4 Least Significant Bits, but I'd like to have a bit more understanding behind what I am doing before I go away and build something.
I'd appreciate any guidance you can give.
Jim.
Here's my 16bit version of the 20bit simulation originally posted by Tom.
Warning: It's based on the assumptions mentioned above, and so is probably wrong.
I dont yet know how to analyse it for "goodness", for 2 reasons:
-It's the first time i've tried to use switchercad.
-I need to go away and think about it - probably for a long time...
Jim.
Warning: It's based on the assumptions mentioned above, and so is probably wrong.
I dont yet know how to analyse it for "goodness", for 2 reasons:
-It's the first time i've tried to use switchercad.
-I need to go away and think about it - probably for a long time...
Jim.
Hi Jim,
The current doesn't really come into it for these kinds of resistor values... if you click on the resistors in the simulation, you'll see that microamps are being drawn.
The desired offset is mostly dependent on the switch impedance, when sourcing and sinking current. Here is the simplified method to get a rough figure...
Turn the logic output on that you want to test (in the case of '595s, you can do this with a few bits of wire, morse code style, but it's slightly easier with a clock source), with a resistor of known value going from output to ground. The value isn't critical - make it low-ish so some accuracy is achievable, but not so low as to potentially damage the logic - I normally use a 332 ohm.
Now measure the voltage across the resistor, and also the rail voltage. With these three values you can now work out the resistance of the switch.
This is only a DC resistance measurement, and only tells you it for sourcing current, of course, not sinking. The two aren't guaranteed to be the same... but if you go with VHC family, apparently they are.
The resistances tend to be quite similar from output to output. You can measure every output if you want - this is quite time consuming to do, but it will help with matching later on.
Once you have an idea of switch Zs into your 2R positions, you need to subtract those values away in the 2R positions themselves, in order to rebalance the total 2R position to double the resistance of the Rs.
If your switch Z is quite low, you might not even need high ohmic trim resistors to perform this, if the range of resistors in your collection is wide enough in value distribution, and you have enough low ones.
Hopefully that all makes sense!
It depends a lot on your source format. 🙂
I've been looking at various possible donor DAC boxes (particularly the ones with custom digital filters like the Thetas), and a lot of them use 18-bit chips, using an 18-bit shifter... not great, but providing the source is just stopping the bitclock for some unused slots, and is actually using a 24 or 32 frame internally, it can be worked around.
Nothing wrong with reducing the ladder down like that. Probably stating the obvious, but aim to have all the Rs the same - I showed real values of a hand-built ladder. The total 2R position (as in switch Z + actual 2R resistor) ideally should be exactly double of the Rs.
If you were to go with plain HC logic from SGS Thomson, for example, the switch Z is meant to be around 45 ohms... so if R was 10000, 2R resistor would need to be ~19995, to give the desired 20000 in circuit.
A Discrete built without compensating for switch Z does not have to be unlistenable, IME... it's just a bit on the bright side (that'll be... eek... harmonics...). The importance of compensation varies with how high the ladder values are, and how low the switch Z is...
A simple way is to click on the simulation window, go to File, Export, and select V(out). This saves out a text file with the step values. The output voltages should ideally be in the scaling of 0, 1, 1+2, 1+2+4, 1+2+4+8, etc...
- Tom.
The current doesn't really come into it for these kinds of resistor values... if you click on the resistors in the simulation, you'll see that microamps are being drawn.
The desired offset is mostly dependent on the switch impedance, when sourcing and sinking current. Here is the simplified method to get a rough figure...
Turn the logic output on that you want to test (in the case of '595s, you can do this with a few bits of wire, morse code style, but it's slightly easier with a clock source), with a resistor of known value going from output to ground. The value isn't critical - make it low-ish so some accuracy is achievable, but not so low as to potentially damage the logic - I normally use a 332 ohm.
Now measure the voltage across the resistor, and also the rail voltage. With these three values you can now work out the resistance of the switch.
This is only a DC resistance measurement, and only tells you it for sourcing current, of course, not sinking. The two aren't guaranteed to be the same... but if you go with VHC family, apparently they are.
The resistances tend to be quite similar from output to output. You can measure every output if you want - this is quite time consuming to do, but it will help with matching later on.
Once you have an idea of switch Zs into your 2R positions, you need to subtract those values away in the 2R positions themselves, in order to rebalance the total 2R position to double the resistance of the Rs.
If your switch Z is quite low, you might not even need high ohmic trim resistors to perform this, if the range of resistors in your collection is wide enough in value distribution, and you have enough low ones.
Hopefully that all makes sense!
It depends a lot on your source format. 🙂
I've been looking at various possible donor DAC boxes (particularly the ones with custom digital filters like the Thetas), and a lot of them use 18-bit chips, using an 18-bit shifter... not great, but providing the source is just stopping the bitclock for some unused slots, and is actually using a 24 or 32 frame internally, it can be worked around.
Nothing wrong with reducing the ladder down like that. Probably stating the obvious, but aim to have all the Rs the same - I showed real values of a hand-built ladder. The total 2R position (as in switch Z + actual 2R resistor) ideally should be exactly double of the Rs.
If you were to go with plain HC logic from SGS Thomson, for example, the switch Z is meant to be around 45 ohms... so if R was 10000, 2R resistor would need to be ~19995, to give the desired 20000 in circuit.
A Discrete built without compensating for switch Z does not have to be unlistenable, IME... it's just a bit on the bright side (that'll be... eek... harmonics...). The importance of compensation varies with how high the ladder values are, and how low the switch Z is...
A simple way is to click on the simulation window, go to File, Export, and select V(out). This saves out a text file with the step values. The output voltages should ideally be in the scaling of 0, 1, 1+2, 1+2+4, 1+2+4+8, etc...
- Tom.
AHA!
Thanks Tom!
I thought all the R's should be the same and all the 2R's should be the same.
I had looked at the values for the resistors you had entered on your sim, and was drawing incorrect conclusions from the way the that the R and 2R values increased the further one got from the MSB.
I 'think' I know enough now to go away and build something.
No doubt more questions will occur....
Jim.
Thanks Tom!
I thought all the R's should be the same and all the 2R's should be the same.
I had looked at the values for the resistors you had entered on your sim, and was drawing incorrect conclusions from the way the that the R and 2R values increased the further one got from the MSB.
I 'think' I know enough now to go away and build something.
No doubt more questions will occur....
Jim.
Good luck with the build. 🙂
For those who might be puzzled about the 19995 instead of 19955 in the example given, that's a typo.
- Tom.
For those who might be puzzled about the 19995 instead of 19955 in the example given, that's a typo.
- Tom.
Hi,
Found this thread an interesting read. 🙂
Also the links below provided me with an introduction.
1. http://www.planetanalog.com/printableArticle.jhtml?articleID=171203226
2. http://www.maxim-ic.com/appnotes.cfm/appnote_number/641
(for a glossary - DNL)
3. http://en.wikipedia.org/wiki/Digital_to_analog_converter
4.http://www.chinaecnet.com/tool/md4.pdf
So as i understand from the article 1. to get the best 16bit dac from that method would need 65,536 equally weighted current sources with logic, and for a 24 bit dac would require 16,777,216 current sources, and associated logic. Not really feasible so have to compromise, and use R2R / scaled current sources.
I'm not too sure what ratio would be needed as getting both bits right seems difficult for a diy'er. Seems it's either thousands of precision current sources or laser trimmed resistor ladder.
Kind regards,
Ashley.
Found this thread an interesting read. 🙂
Also the links below provided me with an introduction.
1. http://www.planetanalog.com/printableArticle.jhtml?articleID=171203226
2. http://www.maxim-ic.com/appnotes.cfm/appnote_number/641
(for a glossary - DNL)
3. http://en.wikipedia.org/wiki/Digital_to_analog_converter
4.http://www.chinaecnet.com/tool/md4.pdf
So as i understand from the article 1. to get the best 16bit dac from that method would need 65,536 equally weighted current sources with logic, and for a 24 bit dac would require 16,777,216 current sources, and associated logic. Not really feasible so have to compromise, and use R2R / scaled current sources.
I'm not too sure what ratio would be needed as getting both bits right seems difficult for a diy'er. Seems it's either thousands of precision current sources or laser trimmed resistor ladder.
Kind regards,
Ashley.
It is difficult, but those who have made the effort have been rewarded.
I have yet to build a Discrete DAC which is an accurate 16 bits, and to be honest, I don't lose any sleep over it.
What I am missing quite badly is my Rev.2 boards which are now part of a commissioned DAC modification for a member of this forum.
It's hard to say exactly why they sound more relaxed... it could be a whole host of things, but I am getting withdrawal symptoms...
As I am now up in Edinburgh working full-time, I don't have the facility (or much time, really) to do any more building for a while.
- Tom.
I have yet to build a Discrete DAC which is an accurate 16 bits, and to be honest, I don't lose any sleep over it.
What I am missing quite badly is my Rev.2 boards which are now part of a commissioned DAC modification for a member of this forum.
It's hard to say exactly why they sound more relaxed... it could be a whole host of things, but I am getting withdrawal symptoms...
As I am now up in Edinburgh working full-time, I don't have the facility (or much time, really) to do any more building for a while.
- Tom.
Hi Tom,
I know that was the last you posted on this forum, so I am not sure if you are reading this or not, but question for you or anyone who'd like to chime in on this -
I have always wondered why its not possible to use any of the chips mentioned in the first post of this thread to build a discrete dac ? I mean if it can be done using hand picked resistors and mosfet switches, why not with a chip that already has all that?!
I know that was the last you posted on this forum, so I am not sure if you are reading this or not, but question for you or anyone who'd like to chime in on this -
I have always wondered why its not possible to use any of the chips mentioned in the first post of this thread to build a discrete dac ? I mean if it can be done using hand picked resistors and mosfet switches, why not with a chip that already has all that?!
Hi Percy,
After all these years, I got an e-mail that someone had made a post, so I dropped by to have a look.
There's nothing to stop you using something like a DAC8871 to build a good audio DAC from, other than getting the data/clock formatting right. On the face of it, talking to something like a PMD100 doesn't look too hard. You'd need to sit down and compare all the input timings to be sure.
It does look like a very nice chip... probably the closest you'll get to the Discrete I (and my enthusiast pals) built in a chip... but by no stretch of the definition could the result be called a Discrete DAC... it's a single chip solution with no discrete resistor ladder. So who knows what it'll sound like. That would require building something.
The wide reference voltage range makes it perfect for feeding into a valve amp and might mean you can drop a valve, which is nice. Either that or follow it with a step down TX to get a nice low impedance to use into a passive attenuator.
After all these years, I got an e-mail that someone had made a post, so I dropped by to have a look.
There's nothing to stop you using something like a DAC8871 to build a good audio DAC from, other than getting the data/clock formatting right. On the face of it, talking to something like a PMD100 doesn't look too hard. You'd need to sit down and compare all the input timings to be sure.
It does look like a very nice chip... probably the closest you'll get to the Discrete I (and my enthusiast pals) built in a chip... but by no stretch of the definition could the result be called a Discrete DAC... it's a single chip solution with no discrete resistor ladder. So who knows what it'll sound like. That would require building something.
The wide reference voltage range makes it perfect for feeding into a valve amp and might mean you can drop a valve, which is nice. Either that or follow it with a step down TX to get a nice low impedance to use into a passive attenuator.
Hi Tom,
Thanks for responding!
I appreciate your sharing thoughts on the subject.
The idea of using the chip appeals to me because it rules out having to mess around with serial to parallel conversion and trying to make 16 switches for each of the ladder rungs, besides hand picking the resistors (which would require some highly accurate and reference resistance measurement equipment in the first place).
btw, what resistor chip did you use in the Discreete DAC ? I couldn't quite find that info on the webpage.
Thanks!
Thanks for responding!
I appreciate your sharing thoughts on the subject.
The idea of using the chip appeals to me because it rules out having to mess around with serial to parallel conversion and trying to make 16 switches for each of the ladder rungs, besides hand picking the resistors (which would require some highly accurate and reference resistance measurement equipment in the first place).
btw, what resistor chip did you use in the Discreete DAC ? I couldn't quite find that info on the webpage.
Thanks!
- Status
- Not open for further replies.