NOS DAC idea...

[barrows]

Russ and Brian,

First thanks for all you guys do for us.

I have recently noticed, that there is now the beginning of what I suspect may become a trend for computer based music systems: the use of a high sample rate compatible NOS DAC, with computer playback software capable of applying oversampling/filter(s). With a computer interface capable of 24-32/384-768 sample rates, and an NOS DAC with equal capabilities, the possibilities seem really interesting, as one could try many different oversampling and filtering approaches, all implemented in the playback software. This approach provides many advantages, as the computer can have plenty of processing power to use complex oversampling/filtering, and it would be easy to apply many different approaches, while being able to keep the same NOS DAC hardware.
Now I know you guys probably have your hands full of projects-but I would love to see a NOS DAC module from you, and since you already have a high resolution USB interface in the works, it might make sense to pair it with a NOS, I2S input DAC (perhaps 1704 based, with a new version of the Legato tweaked to match).
Just something to ponder...

[jgazal]

May I ask some questions?

1. Why you say "computer playback software capable of applying oversampling/filter(s)"? I thought oversampling was a way of translating multi-bit tracks into 1 bit stream to use sigma-delta converters. If you increase sample rate before that step and feed a R-2R ladder converter with a 24 word bit track, I would call it up-sampling (although both means digital filtering). Am I thinking something wrong?

2. Which anti-imaging analog filter in the I/V stage would you use? I think you would need at least a low-pass filter at 768 kHz, right?

3. PCM1704 has 20-Bi or 24-Bi INPUT AUDIO DATA WORD. Suppose I have a 16 bit audio track in my hard-disk.

3.a) Do I need to convert it into 20 bit word?
3.b) Does that conversion demand a complex algorithm? Can we just do a word bit dislocation with no complex calculations or approximation?
3.c) Can the USB to i2s hardware do that? Does that introduce jitter?

4. Suppose I want to play a 16 bit and 44 kHz audio track without computer bases pre-processed up-sampling digital filter. If my anti-imaging analog filter cuts at 768 kHz, would the analog output present signal images in the 88 KHz to 768 kHz region? Would that images waste power on the amplifiers* or affect the transducers?

5. What is the cut-off frequency Legato 3 and IVY-III low-pass filters?

6. What do you have in mind related to the master clock? COD or Opus slaved approach (perhaps with the USB to i2s asynchronous converter as master clock)? Maybe a Secure Digital card transport?

On the one hand, I agree that would be an ultimate experimenter kit. No over-sampling digital filter for sigma-delta converter, up-sampling digital filter for R-2R ladder converter and perhaps no anti-imaging analog filter at all, all in the same hardware. It looks interesting.

On the other hand, you are asking Russ and Brian to build a Wadia 922 decoding computers... Okay, except Legato I/V stage differs from swift-current I/V stage...

* Suppose I do not have a class-A amplifier with shunt power supplies "always sinking" amplifier and that such amplifier has enough slew-rate to reproduce such high frequencies...

[barrows]

Originally Posted by jgazal

May I ask some questions?

1. Why you say "computer playback software capable of applying oversampling/filter(s)"? I thought oversampling was a way of translating multi-bit tracks into 1 bit stream to use sigma-delta converters. If you increase sample rate before that step and feed a R-2R ladder converter with a 24 word bit track, I would call it up-sampling (although both means digital filtering). Am I thinking something wrong?

Forget about word length: consider a 1704 DAC, typically, these are implemented with an oversampling filter (like the matching DF 1704 from TI), this filter does not change the word length, it oversamples the data and applies the digital filter. The 1704 DAC(s) still operate at 24 bits. I am proposing a DAC which does not include the oversampling/filtering, but is capable of handling high sample rate input (the 1704 can accept 24 bits at up to 768 KHz as I recall). Now you combine this DAC with a really good async USB interface capable of high sample rates, and then you apply the oversampling/filtering in the playback software in the computer-playback engines like Pure Music and Audirvana are already capable of doing this. The advantage here is that the oversampling can be done via the higher processing power of the computer, using better math, and upgrades to the oversampling/filter are much easier to try as they are all in software, rahter than in (limited processing power) hardware chips in the DAC. Do not get confused about up vs oversampling, these are just really marketing terms, and have no real technical meaning: typically, in marketing speak, upsamplijng refers to asynchronous sample rate conversion, and oversampling refers to synchronous sample rate conversion (like the 8x or 16 x typically done in delta sigma style DAC chips). My proposal is to use an R2R type DAC, so word lengths will stay the same (although we may pad 16 bit sources up to 24 bits in the computer software in order to better utilize the computer processing power for volume control, etc).

2. Which anti-imaging analog filter in the I/V stage would you use? I think you would need at least a low-pass filter at 768 kHz, right?

You do not have to have one necessarily... But some will want one. Because this proposal is for DIY, this could be ommitted on the board if one wants, or it can be included. All that would be needed for one who is nervous about very high (out of band) frequencies would be a simple 1st order filter, and the frequency of this could be adjusted by cap selection by the builder.

3. PCM1704 has 20-Bi or 24-Bi INPUT AUDIO DATA WORD. Suppose I have a 16 bit audio track in my hard-disk.
3.a) Do I need to convert it into 20 bit word?
3.b) Does that conversion demand a complex algorithm? Can we just do a word bit dislocation with no complex calculations or approximation?
3.c) Can the USB to i2s hardware do that? Does that introduce jitter?

No worries, the computer sends 24 bit data only. This is easy, no processing for this step, as we can just pad with zeros. We do this in the computer software where there is tons of processing power so we can do it perfectly.

4. Suppose I want to play a 16 bit and 44 kHz audio track without computer bases pre-processed up-sampling digital filter. If my anti-imaging analog filter cuts at 768 kHz, would the analog output present signal images in the 88 KHz to 768 kHz region? Would that images waste power on the amplifiers* or affect the transducers?

If you are worried about that, then you would not send a non oversampled/filtered data stream to the DAC. The beauty of this approach is that you could try it either way, and decide what you like. Some people like the sound of a bunch of alias artifacts in the output of their DACs (not me)! The point is that with this approach we have much more flexibility, rahter than being locked in to what hardware is built into the DAC. Plus, our oversampling filter can be much better, because it is not limited in processing power. I think you are missing the point-you get to choose with this set up.

5. What is the cut-off frequency Legato 3 and IVY-III low-pass filters?

These have changed over the course of the the Legato (I do not know for the IVY). Right now I have a Legato II, and mine is set up as a ~88.5 KHz corner, but I believe the Legato III is set up with a ~480 KHz filter. This filter is easily changeable by switching cap values. In any case, the filter used (if any) would be different, as the needs of the 1704 (R2R) would be totally different than the ESS (sigma delta).

6. What do you have in mind related to the master clock? COD or Opus slaved approach (perhaps with the USB to i2s asynchronous converter as master clock)? Maybe a Secure Digital card transport?

Since this would be specifically, and only, for computer use via async USB, I would put 2 masterclocks at the DAC chips, with clean power supplies (Trident) and then send them back over I2S to the USB receiver, this will result in lowest possible jitter. One would still want to have very short I2S lines, and the USB receiver close by. Or, to avoid additional developement, the forthcoming TPA async USB board could provide the masterclcok, via I2S. Russ has mentioned that this module is going to feature very good clocks.

On the one hand, I agree that would be an ultimate experimenter kit. No over-sampling digital filter for sigma-delta converter, up-sampling digital filter for R-2R ladder converter and perhaps no anti-imaging analog filter at all, all in the same hardware. It looks interesting.

On the other hand, you are asking Russ and Brian to build a Wadia 922 decoding computers... Okay, except Legato I/V stage differs from swift-current I/V stage...

Well... the Wadia has built in oversampling/filtering-the entire point of my proposal is that there is none, so this is very different.
Call it whatever-this is not a difficult thing to do. The real limitation would be expense, as I would want 2 1704 per channel, and the 1704 does not come cheap... Add a couple of Crystek CCHD-957 oscillators, and parts cost is getting up there... alternatively, it would take less development to use the forthcoming USB interface, and masterclock from the clocks on the USB interface (apparently Russ is speccing great clocks). And this would reduce the cost of the DAC module a little, this clock distribution might add a little jitter, but it would probably not be that consequential. As it is DIY, one could choose whatever output stage they prefer, as usual. TPA would likely want to build an output stage which ideally suits the loading requirements of the 1704. This could also be done with Delta Sigma chips like the 1792/1794 series, as they allow one to bypass the internal OSF, but the SDM still operates in a very complex manner, which is philosophically against the ideal of what I am proposing.

* Suppose I do not have a class-A amplifier with shunt power supplies "always sinking" amplifier and that such amplifier has enough slew-rate to reproduce such high frequencies...

[jgazal]

Originally Posted by barrows

1.
Forget about word length: consider a 1704 DAC, typically, these are implemented with an oversampling filter (like the matching DF 1704 from TI), this filter does not change the word length, it oversamples the data and applies the digital filter. The 1704 DAC(s) still operate at 24 bits. I am proposing a DAC which does not include the oversampling/filtering, but is capable of handling high sample rate input (the 1704 can accept 24 bits at up to 768 KHz as I recall). Now you combine this DAC with a really good async USB interface capable of high sample rates, and then you apply the oversampling/filtering in the playback software in the computer-playback engines like Pure Music and Audirvana are already capable of doing this. The advantage here is that the oversampling can be done via the higher processing power of the computer, using better math, and upgrades to the oversampling/filter are much easier to try as they are all in software, rahter than in (limited processing power) hardware chips in the DAC. Do not get confused about up vs oversampling, these are just really marketing terms, and have no real technical meaning: typically, in marketing speak, upsamplijng refers to asynchronous sample rate conversion, and oversampling refers to synchronous sample rate conversion (like the 8x or 16 x typically done in delta sigma style DAC chips). My proposal is to use an R2R type DAC, so word lengths will stay the same (although we may pad 16 bit sources up to 24 bits in the computer software in order to better utilize the computer processing power for volume control, etc).

2.
You do not have to have one necessarily... But some will want one. Because this proposal is for DIY, this could be ommitted on the board if one wants, or it can be included. All that would be needed for one who is nervous about very high (out of band) frequencies would be a simple 1st order filter, and the frequency of this could be adjusted by cap selection by the builder.

3.
No worries, the computer sends 24 bit data only. This is easy, no processing for this step, as we can just pad with zeros. We do this in the computer software where there is tons of processing power so we can do it perfectly.

4.
If you are worried about that, then you would not send a non oversampled/filtered data stream to the DAC. The beauty of this approach is that you could try it either way, and decide what you like. Some people like the sound of a bunch of alias artifacts in the output of their DACs (not me)! The point is that with this approach we have much more flexibility, rahter than being locked in to what hardware is built into the DAC. Plus, our oversampling filter can be much better, because it is not limited in processing power. I think you are missing the point-you get to choose with this set up.

5.
These have changed over the course of the the Legato (I do not know for the IVY). Right now I have a Legato II, and mine is set up as a ~88.5 KHz corner, but I believe the Legato III is set up with a ~480 KHz filter. This filter is easily changeable by switching cap values. In any case, the filter used (if any) would be different, as the needs of the 1704 (R2R) would be totally different than the ESS (sigma delta).

6.
Since this would be specifically, and only, for computer use via async USB, I would put 2 masterclocks at the DAC chips, with clean power supplies (Trident) and then send them back over I2S to the USB receiver, this will result in lowest possible jitter. One would still want to have very short I2S lines, and the USB receiver close by. Or, to avoid additional developement, the forthcoming TPA async USB board could provide the masterclcok, via I2S. Russ has mentioned that this module is going to feature very good clocks.


Well... the Wadia has built in oversampling/filtering-the entire point of my proposal is that there is none, so this is very different.
Call it whatever-this is not a difficult thing to do. The real limitation would be expense, as I would want 2 1704 per channel, and the 1704 does not come cheap... Add a couple of Crystek CCHD-957 oscillators, and parts cost is getting up there... alternatively, it would take less development to use the forthcoming USB interface, and masterclock from the clocks on the USB interface (apparently Russ is speccing great clocks). And this would reduce the cost of the DAC module a little, this clock distribution might add a little jitter, but it would probably not be that consequential. As it is DIY, one could choose whatever output stage they prefer, as usual. TPA would likely want to build an output stage which ideally suits the loading requirements of the 1704. This could also be done with Delta Sigma chips like the 1792/1794 series, as they allow one to bypass the internal OSF, but the SDM still operates in a very complex manner, which is philosophically against the ideal of what I am proposing.

Thank you for your response.

I wonder if the BUS bitrate does mean a significant restraint to a digital filter.

Certainly your computer allows different, new and perhaps more efficient algorithms, but it will be limited by your USB to i2S bus bitrate.

Suppose I have an inefficient algorithm and low processing DSP working synchronous at a higher "revolution" (sample rate) than your bus bitrate. Maybe that higher "loop" in your inefficient algorithm gives a better result (more precise approximation) than a more efficient algorithm processed asynchronous with lower loop (your bus bitrate).

What do you think?

p.s.: I suppose audio signal bandwidth is identical in both arrangements. I am thinking about the filter response above 22 kHz.

[barrows]

TPA's forthcoming USB-I2S converter is rumored to work up to 32/384, for the 1704 we would only then use 24/384.

"Certainly your computer allows different, new and perhaps more efficient algorithms, but it will be limited by your USB to i2S bus bitrate. "

A 352.8 sample rate allows for 8x oversampling filters to be applied to 44.1 sources. At this rate, it should be possible to do very good filtering, and push problems way up in frequency to where they will be inaudible. I believe the 1704 can handle up to 768 KHz in. Those who feel doing D/A at rates much higher than this are going to be happier with an ESS style DAC operating in the MHz range.

If someone wants higher rates (705.6 and 768), one will have to develop a USB-I2S interface capable of handling the hiher rates-certainly USB 2 HS has plenty of bandwidth to handle higher oversampling rates for only 2 channels.

Unfortunately, I have neither the knowledge, experience, or skills to develop a DAC board to do this myself, and this is why I suggested it here, and I just wanted to start a discussion on one approach which might suit computer based playback really well.

[jgazal]

Originally Posted by barrows

TPA's forthcoming USB-I2S converter is rumored to work up to 32/384, for the 1704 we would only then use 24/384.

Perhaps a R-2R ladder working at 23 bits (PCM1704) at each cycle without sigma-delta demodulators also allows to achieve better filtering even at a lower oversampling, since you can go to zero analog level in just one cycle, right?

As I see, with 1-bit sigma-delta demodulators you need a certain number of cycles to achieve zero analog level and that add out-of-band high-frequency noise.

Originally Posted by barrows

Unfortunately, I have neither the knowledge, experience, or skills to develop a DAC board to do this myself, and this is why I suggested it here, and I just wanted to start a discussion on one approach which might suit computer based playback really well.

Neither do I. Still, I endorse your suggestion.