Aside from undermining clockmonger business, why is the subject of settling-time ignored here? The diyAudio experts dismiss it as inconsequential yet they are obsessed with jitter.
Here’s a collage of traces looking at the DAC-AH’s left channel reproducing a 10KHz, dithered sine wave. The ‘scope is triggered on the falling edge of FSYNC. The right channel is mute. There is a 220ns delay between the trigger point and when the data starts to change. Of course, there is some jitter in the timing of that change time. Every clocked circuit has jitter. Jocko and the clockmongers are convinced they can hear that jitter, which in this picture amounts to less than the width of the orange cursor line. They say those tiny timing errors result in audible amplitude errors. What Jocko and the clockmongers don’t seem to hear is the very long period of time the DAC’s outputs are at the wrong level. In the case of the DAC-AH, it can take over 5us to settle a relatively small step. That’s one-quarter of the total sample period! If I was concerned about amplitude errors in the DAC output I would work on reducing DAC settling time, measured in microseconds, long before I bothered with clock jitter, measured in picosecond. It’s a tough problem to solve but that’s what makes it interesting. There are at least three approaches that are within the reach of serious DIY.
Here’s another view with 10022.727273Hz. The signal frequency beating with the sample rate makes the stationary pattern shown.
Here’s a collage of traces looking at the DAC-AH’s left channel reproducing a 10KHz, dithered sine wave. The ‘scope is triggered on the falling edge of FSYNC. The right channel is mute. There is a 220ns delay between the trigger point and when the data starts to change. Of course, there is some jitter in the timing of that change time. Every clocked circuit has jitter. Jocko and the clockmongers are convinced they can hear that jitter, which in this picture amounts to less than the width of the orange cursor line. They say those tiny timing errors result in audible amplitude errors. What Jocko and the clockmongers don’t seem to hear is the very long period of time the DAC’s outputs are at the wrong level. In the case of the DAC-AH, it can take over 5us to settle a relatively small step. That’s one-quarter of the total sample period! If I was concerned about amplitude errors in the DAC output I would work on reducing DAC settling time, measured in microseconds, long before I bothered with clock jitter, measured in picosecond. It’s a tough problem to solve but that’s what makes it interesting. There are at least three approaches that are within the reach of serious DIY.
Here’s another view with 10022.727273Hz. The signal frequency beating with the sample rate makes the stationary pattern shown.
Nice post. Very enlightening.
But isn't that the history of high-price audio? Somebody picks something everybody can understand, like solid-core wire, and runs with it. People spend big money on the Philips CD-Pro2, which just HAS to be better, while Meridian, perhaps the world's premier CD player manufacturer, uses €20 CD ROM drives!
This jitter thing is beyond my competence so I have nothing to say in that regard. But do I believe in it? Hell, no. How can I believe anything in an industry where 90 pct is lies and superstitious belief? (My lack of competence means I can't discard it either.) I apply Occam's razor here, and, in particular, that early audiophile, Buddha: "Believe nothing, no matter where you read it, or who said it, no matter if I have said it, unless it agrees with your own reason and your own common sense."
But isn't that the history of high-price audio? Somebody picks something everybody can understand, like solid-core wire, and runs with it. People spend big money on the Philips CD-Pro2, which just HAS to be better, while Meridian, perhaps the world's premier CD player manufacturer, uses €20 CD ROM drives!
This jitter thing is beyond my competence so I have nothing to say in that regard. But do I believe in it? Hell, no. How can I believe anything in an industry where 90 pct is lies and superstitious belief? (My lack of competence means I can't discard it either.) I apply Occam's razor here, and, in particular, that early audiophile, Buddha: "Believe nothing, no matter where you read it, or who said it, no matter if I have said it, unless it agrees with your own reason and your own common sense."
interesting!
with my limited knowledge of dac's i have a question on settling time.
does the dac resync during data transfer?
if so how often and why?
I thought settling time was on startup.
Being once to dac, decocder are sync'd they should stay that way unless something interrupts the signal?
allan
ps it's quite audible on some interruptions, eg weak laser.
with my limited knowledge of dac's i have a question on settling time.
does the dac resync during data transfer?
if so how often and why?
I thought settling time was on startup.
Being once to dac, decocder are sync'd they should stay that way unless something interrupts the signal?
allan
ps it's quite audible on some interruptions, eg weak laser.
I have suggested many times use of very low input bias current / fast settling OP's (8066) for I/V.
Extreme_Boky
awpagan said:
The digital samples come off the CD at the rate of 44,100 per second. Each sample is a number that represents the signal level for each sample period. It is the DAC's job to convert of those numbers into an electic current that is proportional to each value. Ideally, the current should change instantaneously at precisely the right moment. The moment is defined by one of the DAC clocks. If there is jitter in the clock, the current will change a little earlier or a little later than is should. This timing error, usually measured in picoseconds or 0.000000000001 seconds, is what Jocko and the clockmongers worry about.
Regardless of any error in the timing of the clock, DACs do not change the output current instantaneously: Far from it. The time it takes the DAC to change the output, from where it is to where it should be, is called settling time. It is usually measured in micro- or nanoseconds. The larger the difference in value between one sample and the next, the longer the settling time. Also, the closer the output gets to its desired value, the slower it changes. In other words, the output changes very quickly, at first, but then slows down and may even wander around before slowly creeping up on the correct destination. (You can see this behavior in the photos.) For that reason, settling time is usually specified for a given range, often one-half of full scale, and a given “nearness” to the correct final value, often ½ LSB. For example: the settling time for the TDA1541 is typically 1us to reach +/- 1 LSB.
In the case of the DAC-AH, it can take over ¼ of the sample interval for the DAC to reach the correct output value. During that time, the output is the wrong value. If the DAC has the wrong value for 5us, does it really matter if it has the wrong value for 0.000005000001 or 0.000004999999 seconds? Wouldn’t it be better to fix the big error, settling time, before trying to fix the tiny error, jitter, which is a million times smaller?
What about S/H ?
Used to eliminate spikes, it should also correct for settling time errors.
The settling time issue and jitter depends on the oversampling rate.
I have measured a few and some rather old CD players that use BB DACs with 1 kHz 0 dB signal.
All have clean spectrum without any harmonics down to below -93 dB.
Oh jitter & settling time where are you ?
Big differences are in low level linearity...
Used to eliminate spikes, it should also correct for settling time errors.
The settling time issue and jitter depends on the oversampling rate.
I have measured a few and some rather old CD players that use BB DACs with 1 kHz 0 dB signal.
All have clean spectrum without any harmonics down to below -93 dB.
Oh jitter & settling time where are you ?
Big differences are in low level linearity...
Very fast DAC outputs result in HF noise. All those step changes and all that ringing is just added noise.
Ulas said:
Far from being an expert I just want to add that only these 2 first sentences are enough to disattract from any further reading of the thread.
Why do some people have to provoke to proove some point ?
I have played a lot with 24/96 dual differential stereo (4 X 1704) DAC configuration.
I think jitter / clock accuracy should be considered separately from DAC’s settling time / resolution. There’s no point in repeating all over again about the jitter and its influence to sampling / reconstruction of the original data (time domain).
If we consider 24bit word length and 96kHz or higher sampling rates, then the settling time and I/V conversion are very important to accurately follow all those fast step-changes coming out of DAC’s. I have tried many OP configurations and types. It appears that the best results were obtained with extremely low input bias current / fast settling time OP’s. The fastest dual OP I used was 8066 which sounded MUCH better as an I/V conversion OP than 627 or 825.
However, even highly modified DAC / IV / analog filtering and buffering stage sounded very artificial compared to NOS DAC. In my opinion, the simplicity of NOS DAC is to blame for such a nice sound. It makes you wonder why oversample / increase word length…. when all the details are there even with 16bit DAC’s. It’s a different story with high resolution sources like DVD-A and SACD, but we are talking here about trusted 16bit audio CD.
I will have a nice NOS DAC design of my own (it has proven to be extremely popular and good sounding configuration and many have shown a huge interest), which details I am keeping for myself at the moment – my suggestions on forums have been copied already and copyrighted by others – so no more comments about this one…
Extreme_Boky
I think jitter / clock accuracy should be considered separately from DAC’s settling time / resolution. There’s no point in repeating all over again about the jitter and its influence to sampling / reconstruction of the original data (time domain).
If we consider 24bit word length and 96kHz or higher sampling rates, then the settling time and I/V conversion are very important to accurately follow all those fast step-changes coming out of DAC’s. I have tried many OP configurations and types. It appears that the best results were obtained with extremely low input bias current / fast settling time OP’s. The fastest dual OP I used was 8066 which sounded MUCH better as an I/V conversion OP than 627 or 825.
However, even highly modified DAC / IV / analog filtering and buffering stage sounded very artificial compared to NOS DAC. In my opinion, the simplicity of NOS DAC is to blame for such a nice sound. It makes you wonder why oversample / increase word length…. when all the details are there even with 16bit DAC’s. It’s a different story with high resolution sources like DVD-A and SACD, but we are talking here about trusted 16bit audio CD.
I will have a nice NOS DAC design of my own (it has proven to be extremely popular and good sounding configuration and many have shown a huge interest), which details I am keeping for myself at the moment – my suggestions on forums have been copied already and copyrighted by others – so no more comments about this one…
Extreme_Boky
How fast would you expect?
I come to these forums to learn something or share something for others' benefit. I wish there weren't so much posturing and needless bickering but this topic is very interesting and I hope we can learn something more about it.
Ignoring the output filter, can we estimate any constraints on the settling time for this DAC? For example, would we need anything faster than needed to reproduce 20kHz full scale? Wouldn't anything much faster than that be a problem for the output filter?
The basic question of how jitter in the psec range can be signficant compared to the output dynamics in the microsecond range is very very interesting. It seems like such a good question that it must lead to a real lesson in how these things work.
I come to these forums to learn something or share something for others' benefit. I wish there weren't so much posturing and needless bickering but this topic is very interesting and I hope we can learn something more about it.
Ignoring the output filter, can we estimate any constraints on the settling time for this DAC? For example, would we need anything faster than needed to reproduce 20kHz full scale? Wouldn't anything much faster than that be a problem for the output filter?
The basic question of how jitter in the psec range can be signficant compared to the output dynamics in the microsecond range is very very interesting. It seems like such a good question that it must lead to a real lesson in how these things work.
BlackCatSound said:
Regardless of settling time, an unfiltered DAC will have stair steps. If you don’t want stair steps you have to use a LP filter. For non-oversampling, you can use a brickwall, like the original Perfect Sound Forever or a lower order filter, which will leave some remnants of the steps. The following shows the normally unfiltered DAC-AH compared to the AN DAC 1.2, which is NOS with a second order LP filter.
10KHz
1KHz
Incidentally, the AD1865 in the AN DAC settles much faster than the TDA1543 in the DAC-AH. And in all cases, the ’scope is looking at the analog output of the DACs. This is the same signal that gets feed to the audio amplifier.
The area under the curve determines the energy contributed by each sample. That is, the width of the sample multiplied by the height or amplitude. Timing errors, such as jitter, cause errors in the width. Amplitude errors come from non-linearity and settling time. Non-linearity is when the amplitude of the sample is not uniformly proportional to the sample value. For example, the sample says 32000 but the DAC says 28500 is as far as I want to go. It is most prevalent at the extremes of the DAC’s range and with values close to zero. The DAC-AH’s linearity errors are obvious in the 1KHz picture above. While the DAC is settling, the amplitude has not reached the desired value. Depending on the direction the output is moving; the sample will get either too much or too little energy. Every error in height or width contributes to distortion. The DAC-AH has all three kinds of errors in abundance.
Re: How fast would you expect?
Where were the two of you when I brought this topic up some time ago and was told by the diyAudio experts that I didn't know what I was talking about? And where are the two of you whenever the diyAudio experts ridicule someone who doesn't agree with them that the single most important determination of a DAC performance is clock jitter? This has been going on for as long as I have been visiting this forum and I detest it. It's time for payback. If you don't like it, you can kick me out of the forum.
jean-paul said:
RFScheer said:
Where were the two of you when I brought this topic up some time ago and was told by the diyAudio experts that I didn't know what I was talking about? And where are the two of you whenever the diyAudio experts ridicule someone who doesn't agree with them that the single most important determination of a DAC performance is clock jitter? This has been going on for as long as I have been visiting this forum and I detest it. It's time for payback. If you don't like it, you can kick me out of the forum.
Bernhard said:
Bernhard, you are so right. Digital audio got off to a bad start with Perfect Sound Forever and it’s been going down hill ever since. The problem is the bean counters. We lost parallel DACs with their simple architecture and we lost sample and hold, all to cut costs. In return we got the serial DACs with clocks, shift registers, and other noise-producing garbage inside, DACs with settling times approaching ¼ of a sample time, up sampling, digital filters, delta-sigma, and the list goes on. After 30 years of digital, analog is still king. After all, what is the highest compliment a reviewer can give a DAC? “It sounds very analog-like.”
Nice plots
Ulas,
What I really like about this thread is the plots that go along with the point. Probably most of us don't have the opportunity to see this kind of data directly and then get a chance to talk about it.
I'm not sure about your point though. Can we over-communicate here? Is the AN DAC the yellow trace?
Typo? You did mean the 10kHz plot right?
Ok, so if I didn't totally misunderstand, how do you interpret which one has the better sound? The stairstepped AN DAC plot looks ghastly but of course its signal is LP filtered before it goes to the amp. So it seems like the appropriate point to measure distortion is at the input to the amp.
Why does the AN DAC have such a flat peak in the 1kHz plot? Wouldn't that be serious distortion?
At this point, I'm really not sure which one should sound better based on the plots.
Ulas,
What I really like about this thread is the plots that go along with the point. Probably most of us don't have the opportunity to see this kind of data directly and then get a chance to talk about it.
I'm not sure about your point though. Can we over-communicate here? Is the AN DAC the yellow trace?
Typo? You did mean the 10kHz plot right?
Ok, so if I didn't totally misunderstand, how do you interpret which one has the better sound? The stairstepped AN DAC plot looks ghastly but of course its signal is LP filtered before it goes to the amp. So it seems like the appropriate point to measure distortion is at the input to the amp.
Why does the AN DAC have such a flat peak in the 1kHz plot? Wouldn't that be serious distortion?
At this point, I'm really not sure which one should sound better based on the plots.
Re: Re: How fast would you expect?
Ulas, please forget the hostile attitude. "We" have no obligation whatsoever to check all that is being said here. That is practically impossible.
Besides that I see absolutely no point in an unfriendly attitude towards people who did not participate in some discussion you had.
Please behave, we are all ears to what you have to say. If you wish to pay people back please consider the possible consequences.
Ulas said:
Ulas, please forget the hostile attitude. "We" have no obligation whatsoever to check all that is being said here. That is practically impossible.Besides that I see absolutely no point in an unfriendly attitude towards people who did not participate in some discussion you had.
Please behave, we are all ears to what you have to say. If you wish to pay people back please consider the possible consequences.
Hi.
An interesting tread .. but .... please, not one more backbiting word [ even in quoted sections !].........you have our attention !!!
Correct me if I'm wrong....
The data that comes off the CD, does so in serial format. So, even if the earlier DACs were parallel, they would still require shift-registers, clocks etc to work.
I think that what you are trying to say is that the earlier DACs were better made - made to a standard rather than a price?
Andy
An interesting tread .. but .... please, not one more backbiting word [ even in quoted sections !].........you have our attention !!!
Ulas said:
Correct me if I'm wrong....
The data that comes off the CD, does so in serial format. So, even if the earlier DACs were parallel, they would still require shift-registers, clocks etc to work.
I think that what you are trying to say is that the earlier DACs were better made - made to a standard rather than a price?
Andy
Ulas said:
And the faster the settling time the more you have to filter out.
Note that 44.1k sampling is incapable of correctly reconstructing a 10kHz sine wave. The peaks are all different heights. And people worry about jitter 😉
The linearity issues you're seeing on the 1kHz are not to do with settling time.
Parallel DACs have issues, this being one of them. A very high speed well implemented 2nd order sigmadelta would give better results, but for some reason people run away screaming from '1 bit' DACs as they think 'more mumbers must be better'.
BlackCatSound said:
I think that was the point Ulas made in his original post.
Personally I will start worry about the 10kHz sine wave the day they start making loudspeakers that can reproduce more than one tone at the time.
OK, here we go.... probably my head will be chopepd off, but I feel the need to bring one element in which seems we loose a bit (not only in this thread).... Very simply: enjoying your music reproduction... even if it is 90% perception....
If low jitter clocks produce music which is much more liked by a whole lot of people, well fine, who cares ????
If silver, gold, oil, whatever capacitors produce music more people like better, who cares????
If parallel dacs (I mean now paralleing more dacs) sounds better for most, who cares ????
Cares on what ? On all this theory babble I mean....
If in a nice recepy for a good steak, most people like better some salt on it, who cares if Chemics start discussiong the flow of Ions in my meat and try to explain it cannot taste good anymore? The only thing we want is a great taste. As in hearing music this is perception.....
Fine, if we understand what technical metric can influence sound positively, you have a way to try to improve the sound. GO AHEAD ! That is what most DIY actually does, don't they?
If you start talking techno stuff, why something impossibly can sound better, you are on the wrong path imho and I think this is pittyfull......
A shame Ulas, as I though you wanted to bring something positive to the table with the settling time topic (I admit an interesting thought), but in staed of using it to do some research beyound looking at scoops and datasheets (I mean, built something, try something, listen to it !!) you use this to fight your war against people who have been bringing more joy in the music experience of many, many people. What have you done so far to bring more enjoyment to many other peoples music life ? Even if it is only perception ?
Techno is good, and very interesting, but should be used for the goals behind it, not a subject on its own.....
Just my 2 cents....
If low jitter clocks produce music which is much more liked by a whole lot of people, well fine, who cares ????
If silver, gold, oil, whatever capacitors produce music more people like better, who cares????
If parallel dacs (I mean now paralleing more dacs) sounds better for most, who cares ????
Cares on what ? On all this theory babble I mean....
If in a nice recepy for a good steak, most people like better some salt on it, who cares if Chemics start discussiong the flow of Ions in my meat and try to explain it cannot taste good anymore? The only thing we want is a great taste. As in hearing music this is perception.....
Fine, if we understand what technical metric can influence sound positively, you have a way to try to improve the sound. GO AHEAD ! That is what most DIY actually does, don't they?
If you start talking techno stuff, why something impossibly can sound better, you are on the wrong path imho and I think this is pittyfull......
A shame Ulas, as I though you wanted to bring something positive to the table with the settling time topic (I admit an interesting thought), but in staed of using it to do some research beyound looking at scoops and datasheets (I mean, built something, try something, listen to it !!) you use this to fight your war against people who have been bringing more joy in the music experience of many, many people. What have you done so far to bring more enjoyment to many other peoples music life ? Even if it is only perception ?
Techno is good, and very interesting, but should be used for the goals behind it, not a subject on its own.....
Just my 2 cents....
- Status
- Not open for further replies.