DCS Vivaldi
- Thread starter MylesBAstor
- Start date
So the ring dac always applies 64X OS? Or does it reduce the 64X to 32X when the input is provided at 88k2?
I'm sorry but I don't follow this at all. You can't get LF sidebands out of the HF signal without mixing in a nonlinear system. The DS loop does do that, but the NTF suppresses it like everything else. It does not get directly translated back to baseband; the purpose of the feedback loop is to suppress that very effect. However, you bring up an excellent point in that the HF noise (and clock) must be dealt with. It can get into everything through any number of sneaky little paths, coupled and radiated, and if coupled back to the signal can wreak havoc. Perhaps that is your point? I agree there!
It looks to me like dCS has implemented a discrete DAC using switches and resistors. I suppose we could determine if it is a ppure binary R2R or segmented by counting them, but I'm a little too lazy for that. I would guess they have followed typical design practice by segmenting the DAC into a unary MSB section and binary lsb section to reduce glitches and improve linearity by reducing matching requirements. Clock distribution is a bear in something like that since any skew among all those discrete latches will induce jitter (if that's what they are doing). Impressive engineering. I wonder if part of the function of those Xilinx FPGAs (field programmable gate arrays, digital logic processors) is to compensate the DAC. There are a number of schemes that use DSP to compensate amplitude and timing errors among the bits.
Well, the ones I have worked on are primarily RF/mW designs so ears wouldn't help...
Dynamic spectral analysis works, though. FFTs and other transforms are available, along with various means of capturing and analyzing time-varying large-signal waveforms. If there were in-band spurs we'd have seen them, or our customers would (and we'd be in trouble).I've not met either personally, but last year Scott Wurcer (on DIYA) said he'd pass on my question to Bob. I didn't get a response though - must've gotten lost in the modulation noise
From dCS
Ring Dac is a discrete, proprietary, 5 bit 64 times oversampling architecture. 64 times oversampling means that the actual converter data rate is either 2.822MS/s or 3.072MS/s,
depending upon whether the input sample rate is a multiple of 44.1kS/s or 48kS/s. This allows the use of a very gentle and therefore very transparent analogue output filter.
The dCS Ring DAC avoids the limitations inherent in the conventional one bit and multibit off-the-shelf converter ICs that are the basis of most other D/A converters.
The dCS Ring DAC is a five bit unitary weighted converter. With this design, the problems usually associated with using resistor chains to accurately define extremely
small current values that occur in binary weighted multibit types, simply do not exist.
Ring Dac is a discrete, proprietary, 5 bit 64 times oversampling architecture. 64 times oversampling means that the actual converter data rate is either 2.822MS/s or 3.072MS/s,
depending upon whether the input sample rate is a multiple of 44.1kS/s or 48kS/s. This allows the use of a very gentle and therefore very transparent analogue output filter.
The dCS Ring DAC avoids the limitations inherent in the conventional one bit and multibit off-the-shelf converter ICs that are the basis of most other D/A converters.
The dCS Ring DAC is a five bit unitary weighted converter. With this design, the problems usually associated with using resistor chains to accurately define extremely
small current values that occur in binary weighted multibit types, simply do not exist.
I'll have to look up what a ring DAC is, but I am wondering now if they have implemented a fully-unary 5-bit DAC (~32 switches and resistors) and used it in an oversampling loop (don't know if delta-sigma, delta, or something else)? That is an interesting scheme...
My experience with audible improvements from improved clocking has been varied, but a couple of things I have noticed are:
I think a lot of things are improved as we move up the price/performance curve, and clocking is just one. I am not at all sure it is the most important, or even really significant after a base threshold is reached. Better isolation, better power decoupling, better analog output stages (buffers and filters), more advanced digital signal processing, etc. all have an impact on the measurements, and (one certainly hopes) the sound.
I am a skeptic from Missouri ("Show Me") but am always hesitant to use absolutes. Too many times in the past things that seem too small to matter become noticeable in the end system, be it an audio DAC, wideband communications system, radar, test instrument, whatever... The we (engineers) get to figure out why (let the fun begin!)
My experience with audible improvements from improved clocking has been varied, but a couple of things I have noticed are:
- Some DACs use minimal clock buffering and derive the clock from a noisy source (cheap PC and USB DACs come to mind). These tend to suffer from high noise (not just from clock jitter) and distortion (ditto) and going to a clean clock source can make a difference. It may be as much from better clock/signal isolation as from reduced clock noise and jitter.
- Better clock implementations provide lower noise and jitter but also less sensitivity to other things like power supply noise, loading, etc. Again, lower jitter is often just part of a better clock design, and may not be the most important in the audio world (or elsewhere).
I think a lot of things are improved as we move up the price/performance curve, and clocking is just one. I am not at all sure it is the most important, or even really significant after a base threshold is reached. Better isolation, better power decoupling, better analog output stages (buffers and filters), more advanced digital signal processing, etc. all have an impact on the measurements, and (one certainly hopes) the sound.
I am a skeptic from Missouri ("Show Me") but am always hesitant to use absolutes. Too many times in the past things that seem too small to matter become noticeable in the end system, be it an audio DAC, wideband communications system, radar, test instrument, whatever... The we (engineers) get to figure out why (let the fun begin!)
wizard posted whilst I was typing; nice to see my guess was correct about it being a unary DAC. That is great information wizard, thank you!
Binary 5-bit DAC = 5 cells, each must match to an lsb (1 part in 32, 3.125%) to have 5-bit accuracy
Unary 5-bit DAC = 32 cells, each must match to within 50% to have 5-bit accuracy
Their design must ultimately achieve 16-bit accuracy (1 part in 65,536, 0.0015%) or better, of course, through matching and processing. Even if a low-resolution DAC is used, it must still achieve N-bit precision for N-bit output.
Binary 5-bit DAC = 5 cells, each must match to an lsb (1 part in 32, 3.125%) to have 5-bit accuracy
Unary 5-bit DAC = 32 cells, each must match to within 50% to have 5-bit accuracy
Their design must ultimately achieve 16-bit accuracy (1 part in 65,536, 0.0015%) or better, of course, through matching and processing. Even if a low-resolution DAC is used, it must still achieve N-bit precision for N-bit output.
Found this:
The Ring DAC samples 5 bits at a time, oversampling at 64 times the base frequency, and uses noise shaping, much as a traditional bitstream converter. However, rather than merely having five current sources,
the Ring DAC uses those 5-bit samples to drive 24 separate current sources.
Not all 24 sources operate at once, however. The Ring DAC uses a proprietary algorithm to vary which current sources participate in each sample,
as though they were randomly picked from a circle of sources. This gives their design its snappy name.
The Ring DAC samples 5 bits at a time, oversampling at 64 times the base frequency, and uses noise shaping, much as a traditional bitstream converter. However, rather than merely having five current sources,
the Ring DAC uses those 5-bit samples to drive 24 separate current sources.
Not all 24 sources operate at once, however. The Ring DAC uses a proprietary algorithm to vary which current sources participate in each sample,
as though they were randomly picked from a circle of sources. This gives their design its snappy name.
Last edited:
I think that you're such a guru in noise shaping loops that you keep focussing inside the loop. I know very little about what goes on inside the loop so my attention is focussed further down the signal chain, once we've got to the DAC part itself. You admit to LF sidebands from 20kHz caused by jitter at the output of the DAC, right? So why not LF sidebands on the 1.5MHz noise also coming out of the same DAC? The DAC is always putting out the 1.5MHz noise, even when the wanted audio signal is very low in level. I suspect that the effects of jitter will be much more noticeable at lower levels than at higher - at higher its going to be masked. Multibit DACs simply do not have this problem because as the signal level goes down, so do the sidebands.
No, that was another point I was saving up for later on
I was too lazy too so I looked up the patent. It looks like its using 32 equal value current sources (resistors from constant voltage into virtual ground potentially).
@wizard: Thanks again. 24 implies a segmented design to reduce errors. "Shuffling" current sources spreads the errors among time samples, "breaking them up" so they do not generate single tones related to the signal. That is a well-known technique that can work really well but is a pain to implement in practice, espcially in a discrete approach; kudos to dCs for using it!
@opus111: I think we are on the same page now. I thought you were talking inside the loop and that was driving me nuts (it's a short drive ). Yes of course spurs can be passed on; hopefully filtering and decoupling knocks down the very HF stuff but it is indeed part of the art and magic of designing such beasts. And yah, sems like a highly segmented DAC, or some variation of it.
Thanks all! - Don
p.s. I am not a DS guru; working on a few designs does not a guru make. I do some knowledge based on painful real-world experience, however. Most of my data converter designs were conventional'ish, just wicked fast (well above audio).
@opus111: I think we are on the same page now. I thought you were talking inside the loop and that was driving me nuts (it's a short drive
). Yes of course spurs can be passed on; hopefully filtering and decoupling knocks down the very HF stuff but it is indeed part of the art and magic of designing such beasts. And yah, sems like a highly segmented DAC, or some variation of it.Thanks all! - Don
p.s. I am not a DS guru; working on a few designs does not a guru make. I do some knowledge based on painful real-world experience, however. Most of my data converter designs were conventional'ish, just wicked fast (well above audio).
a giant leap forward in a carefully connected dCS chain (this is somehow a challenge) is the usage of an excellent atomic clock like shown on Audio16.com. Might be the better step than investing in a Vivaldi system but... An excellent analogue chain will feel some heat from the dCS digital gurus.
Wizard has recently put up two pictures of complete internal boards of the Vivaldi here : http://cybwiz.blogspot.com/2012/09/dcs-vivaldi-dac-inside.html
From a first glance it appears to shed light on a question considered earlier - why does the external clock on Vivaldi sound better? The answer appears to be that the internal clocks are on a different PCB from the ring DAC. So even the 'internal' clock appears to be compromised by having to travel from one PCB to a (presumably) adjacent one. Or am I missing something? {Ring DAC in pic 1, two oscillator modules visible in pic 2).
From a first glance it appears to shed light on a question considered earlier - why does the external clock on Vivaldi sound better? The answer appears to be that the internal clocks are on a different PCB from the ring DAC. So even the 'internal' clock appears to be compromised by having to travel from one PCB to a (presumably) adjacent one. Or am I missing something? {Ring DAC in pic 1, two oscillator modules visible in pic 2).
Opus111 - thanks for that observation. From what i understand (and i understand very little on technicals)...the MSB Femto clock which they advertise so heavily is not sold in a separate box...it is sold as directly being input right next to the DAC chip on the board itself. I think it might even be that they just replace the standard clock which is next to the chip with the Femto clock.
Yes that's how I understand it too. But since MSB is using multibit technology, the requirement for such a precise clock is not there. I suspect this clock has more placebo effect than technical effect on the output waveform.
Yes that's how I understand it too. But since MSB is using multibit technology, the requirement for such a precise clock is not there. I suspect this clock has more placebo effect than technical effect on the output waveform.
always learning from you, Opus! Thanks
Oh Whatmore was asking about the DIN41612 96pin connector? Glad you were able to work that out
If that connector really is the only way for the clock to be connected to the DAC then its rather a loooong way for a sensitive signal to travel.
If that connector really is the only way for the clock to be connected to the DAC then its rather a loooong way for a sensitive signal to travel.
Similar threads
Staff online
-
treitz3Super Moderator
Members online
- thekong
- treitz3
- Mike Lavigne
- MusicFellow
- Downtheline
- Ampman
- dennis h
- J007B
- andrew1
- pdxaudio
- KeithR
- SlapEcho
- wcheng
- Anthony Ho
- directdriver
- PLiXiR Power James Soh
- Ron Resnick
- jhung168
- Mark212121
- Young Skywalker
- jbrrp1
- saturdayboy
- Gregadd
- AudioHR
- rob
- cal3713
- kyoya78
- Opabin
- marycz
- CKKeung
- MarkusBarkus
- Emorante
- audioquad
- viola
- MPW
- JFRMusic
- Scar11
- audiobomber
- Klonk
- Long Live Analog
- Johan K
- astrotoy
- Lee
- Wavetouch
 | Steve Williams Site Founder | Site Owner | Administrator |  | Ron Resnick Site Co-Owner | Administrator |  | Julian (The Fixer) Website Build | Marketing Managersing |