Food for thought: How does jitter affect DSD DACs?

[jabbr]

Let me resummarize the question incorporating my understanding of what @Miska is saying:

 

Suppose the DSD frequency is 22 Mhz (DSD512) and this frequency varies +/- 100 Hz, slowly:

 

Dividing +/- 100 Hz (200 Hz) / 22Mhz = 0.00000909

 

Is this degree of frequency variation audible? What degree of error (is this about - 100 dB?)

 

[Now realize that the common DAC clocks are - 160 dBc/Hz @ 100 Hz]

 

Is there another mechanism by which jitter on the BCLK results in error in the audio signal reproduction?

[jabbr]

So assuming 0.000009 frequency variation / Hz at worst case 20,000 Hz = 0.18

 

Perhaps an 18% frequency variation is audible but it’s -160 dBc/Hz down. 

1.8% frequency variation at -120 dBc/Hz down (10 Hz) and

0.18% variation (1Hz) down at -80 dBc/Hz

[Miska]

I would say best approach is just see how things really actually perform.

 

For the Pro-Ject Pre Box S2 Digital you can compare 44.1k and DSD512 inputs from jitter perspective.

 

44.1k input:

 

DSD512 input:

 

A bit less side bands. This difference has nothing to do with clock module phase noise, but all to do with the increase in DSP activity inside the DAC chip. In both cases, actual conversion stage is running at similar conversion rate anyway.

 

 

Another case, almost filter-less (just output transformer) R2R NOS DAC...

 

44.1k input:

 

352.8k input:

 

These are also not phase noise differences, but correlated jitter...

 

 

So overall, I'd say it is much more important to worry about many other things before worrying about clock's natural phase noise.

 

[jabbr]

42 minutes ago, Miska said:

So overall, I'd say it is much more important to worry about many other things before worrying about clock's natural phase noise.

 

At first glance, power supply noise goes right to DAC ouput essentially without attenuation, and clock phase noise is less than the circuitry it supplies as well as number of other issues — so yes of course. 

 

I’m trying to get some mathematical bounds on the effects of clocks — like is there a theoretical reason to go to DSD2048 —- don’t necessarily want to build if no reason it might be better.

 

As you recall many people thought DSD512 might be worse than DSD256 because of the clocking issues.

[jabbr]

@Miska, also googling will find all sorts of statements that "DSD is incredibly sensitive to jitter" -- which quote papers describing the jitter sensitivity within the ADC SDM loop, yet I can't find anything which addresses the particular jitter sensitivity of SDM in the DAC itself. So I'm curious to looking at this from a mathematical perspective.

 

The best theories I can find seem to suggest that higher BCLK rates lead to more IM distortion which could "blow back" on audio range -- but that would affect PCM equal to DSD, and 100 Mhz isn't very high speed clock these days...

[Ralf11]

I vote to dismiss any effects that are <= 160 dBc/Hz down - or 140 for that matter.

 

I get the take home msg from this thread that I should upsample...

[Miska]

1 hour ago, jabbr said:

I’m trying to get some mathematical bounds on the effects of clocks — like is there a theoretical reason to go to DSD2048 —- don’t necessarily want to build if no reason it might be better.

 

For clocks no, as I said if you keep increasing the clock rate, the low frequency phase noise proportionally increases too, even more. So I'd say it is more about finding a balance. But what I'm seeing, phase noise is not the primary problem anyway. So you can largely ignore the whole topic of clock module's phase noise and focus more on the actual challenges...

 

Increasing modulator output rates has more to do with reaching good solution for the analog filter. At what point you can reach flat noise floor and desired flat pass-band with some analog filter you consider good. Putting fc around 100 kHz like I did in DSC1 gives very small phase shift at 20 kHz with the filter I used, and very good transient response with virtually no overshoot. That combined with the D/A section filtering and modulator noise characteristics is what I'm after. So a DAC that has flat frequency response, flat phase response, no step response overshoot and flat noise floor while giving good THD+N and IMD figures. Sure, having nice jitter figures is also part of that, but just one part, but you need to take all the other aspects into account while dealing with that. I'm not the kind of person who focuses only on one performance parameter at cost of all the others. I want all aspects to be good, in a balanced way.

 

1 hour ago, jabbr said:

As you recall many people thought DSD512 might be worse than DSD256 because of the clocking issues.

 

No, clocking is not the problem. Dealing with conversion stage settling and transition times gets more challenging higher the switching rate becomes. Same goes for R2R ladders too; you cannot keep increasing sampling rate because you keep losing bits out due to settling time problems. If your R2R cannot settle within +-½LSB for majority of sample period you don't have that LSB at all...

 

14 minutes ago, jabbr said:

yet I can't find anything which addresses the particular jitter sensitivity of SDM in the DAC itself

 

Because SDM "in DAC" is purely digital domain, unlike ADC side and thus doesn't have any jitter aspect in first place! That's why you can run SDM even offline for a DAC. But you cannot do that for ADC. DAC side SDM is digital domain creature while ADC side is link between analog and digital worlds.

 

17 minutes ago, jabbr said:

The best theories I can find seem to suggest that higher BCLK rates lead to more IM distortion which could "blow back" on audio range -- but that would affect PCM equal to DSD, and 100 Mhz isn't very high speed clock these days...

 

Now here's the tricky part in picture; SDMs have BCLK, but DAC chips run out of MCLK and BCLK is only for data I/O but not for anything else. While "chip-less" DSD DACs like DSC1 don't have MCLK at all, but run out of BCLK. While again R2R DACs don't use BCLK for anything else than clocking data into a shift register and WCLK is what drives the configuration (and that's part of the problem when everything originated from high frequency MCLK divided down to WCLK).

 

[beerandmusic]

5 minutes ago, Miska said:

So a DAC that has flat frequency response, flat phase response, no step response overshoot and flat noise floor while giving good THD+N and IMD figures. Sure, having nice jitter figures is also part of that, but just one part, but you need to take all the other aspects into account while dealing with that. I'm not the kind of person who focuses only on one performance parameter at cost of all the others. I want all aspects to be good, in a balanced way.

 

 

so does the pro-ject do that well?

[Miska]

26 minutes ago, beerandmusic said:

so does the pro-ject do that well?

 

For some parts it is good, but certainly not all. But for the money it is incredibly good (assuming correctly configured through the front-panel menu).

 

[beerandmusic]

11 minutes ago, Miska said:

 

For some parts it is good, but certainly not all. But for the money it is incredibly good (assuming correctly configured through the front-panel menu).

 

i recall you mentioned it's spdif sucks...anything else to look out for?

 

curious if you have tried or plan to try the pro-ject streamer s2 ultra (streamer decrapifier in one)...relatively new.

I know you had the NT503 as well as I did...did 503 and pro-ject compare similarly? anything around $1K you have tried that measures better?

[Miska]

29 minutes ago, beerandmusic said:

i recall you mentioned it's spdif sucks...anything else to look out for?

 

No, just the settings...

 

For comparison, S/PDIF input...

 

At 44.1k:

 

And at 192k:

 

29 minutes ago, beerandmusic said:

curious if you have tried or plan to try the pro-ject streamer s2 ultra (streamer decrapifier in one)...relatively new.

 

That is just streamer without DAC, but doesn't have my software modules (NAA) inside, so I'm not really interested...

 

29 minutes ago, beerandmusic said:

I know you had the NT503 as well as I did...did 503 and pro-ject compare similarly? anything around $1K you have tried that measures better?

 

OK, now the big question is what is "measures better"; on big picture, or just Jtest results?

 

I still have NT503 and planning to have UD505 too.

 

I would put RME ADI-2 DAC FS on serious contenders list. Holo Audio Spring DAC is probably close to that price bracket too while being totally different since it is a discrete design. Note that there are probably others, but this is just from my sample set.

 

With RME you need to pay some attention to front panel menu config too, just like Pro-Ject.

 

 

[fas42]

1 hour ago, Ralf11 said:

I vote to dismiss any effects that are <= 160 dBc/Hz down - or 140 for that matter.

 

I get the take home msg from this thread that I should upsample...

 

Upsampling, offline, would be an easy upgrade to SQ for many systems. I did some experiments several years ago to sound files on a desktop, upsampling to the highest reasonable rates that didn't create monster sized files, and the benefits were obvious just using the motheboard's DAC playback.

 

If I were to seriously tackle getting best sound from a particular, writable storage rig I would do some experiments to find which sample rate gave the best subjective results - and with that as a fixed parameter then resample all the owned music that was important. Storage is cheap, it's a no-brainer ...

[beerandmusic]

18 minutes ago, Miska said:

 

No, just the settings...

 

For comparison, S/PDIF input...

 

At 44.1k:

 

And at 192k:

 

 

That is just streamer without DAC, but doesn't have my software modules (NAA) inside, so I'm not really interested...

 

 

OK, now the big question is what is "measures better"; on big picture, or just Jtest results?

 

I still have NT503 and planning to have UD505 too.

 

I would put RME ADI-2 DAC FS on serious contenders list. Holo Audio Spring DAC is probably close to that price bracket too while being totally different since it is a discrete design. Note that there are probably others, but this is just from my sample set.

 

With RME you need to pay some attention to front panel menu config too, just like Pro-Ject.

 

 

 

Holo spring like 3K, but i did give serious consideration when they came out with cyan dsd...

i have also heard good things about the RME, so i will look at that one closer too.

Anyway, thanks...i will stop bugging you for a week...have to finish with some projects around the house....smile

[Miska]

9 minutes ago, fas42 said:

Upsampling, offline, would be an easy upgrade to SQ for many systems. I did some experiments several years ago to sound files on a desktop, upsampling to the highest reasonable rates that didn't create monster sized files, and the benefits were obvious just using the motheboard's DAC playback.

 

If I were to seriously tackle getting best sound from a particular, writable storage rig I would do some experiments to find which sample rate gave the best subjective results - and with that as a fixed parameter then resample all the owned music that was important. Storage is cheap, it's a no-brainer ...

 

I don't bother storing such files... I have too many DACs and I update algorithms too frequently to bother and conversion would take too much time and I change my mind about settings too often. So I just let my player do the conversion on the fly...  Storage is certainly not much problem anymore though. But same goes for processing power.

 

[fas42]

49 minutes ago, Miska said:

 

I don't bother storing such files... I have too many DACs and I update algorithms too frequently to bother and conversion would take too much time and I change my mind about settings too often. So I just let my player do the conversion on the fly...  Storage is certainly not much problem anymore though. But same goes for processing power.

 

 

Yes, you're in a very dynamic environment - makes sense! Trouble with using processing power, in real time, is that the CPU workings may impact the analogue areas - which is the raison d'etre for the offline handling. So, if the user is going to be sticking with a fixed arrangement of hardware for a while, then it may be worthwhile doing some checking, to see if there are any benefits.

[jabbr]

2 hours ago, Miska said:

So you can largely ignore the whole topic of clock module's phase noise and focus more on the actual challenges...

 

Yes, I entirely agree about balancing design parameters. There has been a recent focus among some on reducing jitter in network switches and motherboards etc etc. As I’ve said elsewhere, modern professional network switches have also solved this issue as well. 

 

Given the degree of difficulty relating phase noise to the DAC itself, I imagine that trying to relate hardware that is once or twice or three times removed would be that much harder  

 

I do think that many other factors such as power supply noise, logic switching noise, EMI/RF, board layout, 3D design/volumetric SI etc are far far more important. 

 

Thanks!

[jabbr]

2 hours ago, Miska said:

I said if you keep increasing the clock rate, the low frequency phase noise proportionally increases too, even more. 

 

This was the original question.

 

Assuming the effects of phase noise in the audible range are inversely proportional to the clock frequency — your above equation agrees with my impression — and then according to Leeson’s equation where phase error is proportional to log(f^2) then limit of log(f^2)/f as f approaches infinity = 0 ...

 

so on balance upsampling decreases the effects of phase noise all else being equal 

[beerandmusic]

40 minutes ago, jabbr said:

 

Yes, I entirely agree about balancing design parameters. There has been a recent focus among some on reducing jitter in network switches and motherboards etc etc. As I’ve said elsewhere, modern professional network switches have also solved this issue as well. 

 

Given the degree of difficulty relating phase noise to the DAC itself, I imagine that trying to relate hardware that is once or twice or three times removed would be that much harder  

 

I do think that many other factors such as power supply noise, logic switching noise, EMI/RF, board layout, 3D design/volumetric SI etc are far far more important. 

 

Thanks!

you think emi/rf is an issue with modern dacs?  I don't know much, but i certainly wouldn't think that is an issue in these days??

what defines a switch as pro in your opinion?  You can buy cisco switches that were used in enterprise networks  used for dirt cheap.

It seems netgear calls all their "smart" switches "pro"...?

 

[fas42]

3 minutes ago, beerandmusic said:

you think emi/rf is an issue with modern dacs?  I don't know much, but i certainly wouldn't think that is an issue in these days??

what defines a switch as pro in your opinion?  You can buy cisco switches that were used in enterprise networks  used for dirt cheap.

 

 

 

The most effective approach is to always think in a system wide context - so, don't ask whether emi/rf is relevant to "modern dacs"; rather, is emi/rf impacting my rig? Easily tested by deliberately introducing excessive amounts of such in the environment, and seeing if it's audible; then, narrow down to where the the chain is most sensitive, and do what's necessary to make the circuitry more robust.

[jabbr]

23 minutes ago, beerandmusic said:

you think emi/rf is an issue with modern dacs?  I don't know much, but i certainly wouldn't think that is an issue in these days??

 

There is differential mode noise and common mode noise. Of these noise can be voltage, current & phase all as a function of frequency. (simplifying things for discussion)

 

Need to look at each.

 

EMI/RF is often common mode and can be pesky.

 

Quote

what defines a switch as pro in your opinion?  You can buy cisco switches that were used in enterprise networks  used for dirt cheap.

 

 

 

I use 10Gbe fiber. I’ve obtained Brocade switches on eBay for a few hundred $$.

[beerandmusic]

3 minutes ago, jabbr said:

 

There is differential mode noise and common mode noise. Of these noise can be voltage, current & phase all as a function of frequency. (simplifying things for discussion)

 

Need to look at each.

 

EMI/RF is often common mode and can be tricky to reduce. 

 

 

I use 10Gbe fiber. I’ve obtained Brocade switches on eBay for a few hundred $$.

ok, thanks for clarification on both points (more reading for me)....inre fiber, what are you connecting to what ?  do you have a streamer or a dac that takes fiber or do you use some adapter?  either way you only need 2 ports right?  I know there have to be lots of used enterprise cisco stuff with a couple fiber ports?

[Ralf11]

15 minutes ago, jabbr said:

 

There is ... noise and ... noise.

 

Need to look at each.

 

* * *

 

 

[jabbr]

13 minutes ago, beerandmusic said:

ok, thanks for clarification on both points (more reading for me)....inre fiber, what are you connecting to what ?  do you have a streamer or a dac that takes fiber or do you use some adapter?  either way you only need 2 ports right?  I know there have to be lots of used enterprise cisco stuff with a couple fiber ports?

 

Look at the

”ClearFog as NAA”

and

”Optical Network Configuration” 

threads

[beerandmusic]

12 minutes ago, jabbr said:

 

Look at the

”ClearFog as NAA”

and

”Optical Network Configuration” 

threads

this?  would connecting a pc directly to dac w/enet via fiber and adapters could be as good ....it seems adding the NAA into the mix complicates things, and I would prefer simpler configuration like the original poster, if you could get clean isolation that way too?

 

EDIT TO ADD:, i just read thread through more, but can't yhou buy a nic for a pc that can have an sfp?

 

[jabbr]

 

Should have lots of options if you read through