Clock specifications have traditionally defined jitter from a two perspectives, in the time domain, and in the frequency domain. Even within the time domain, there are multiple ways that jitter can be measured, for example, period jitter, cycle-to-cycle jitter, time-interval-error (TIE), long time jitter, etc.
HiFi Clock vendors tend to mostly gloss over all this and just focus on the clock accuracy and the period jitter specifications (based around a standard deviation of 10,000 measured clock cycles).
Typical Crystal Oscillators have this standard deviation performance at the picoseconds level (10^-12), but it’s increasingly common to see oscillators with even better performance at the femtoseconds level (10^-15).
Theres anecdotal evidence that human hearing thresholds for jitter are down to the 10s of picoseconds levels, but it’s (currently) unknown whether anything better than that is actually even audible (let alone 1,000 times better which is what femtoseconds would mean!! ).
There’s a lot of typical audiophile hype around clock specifications, most just fall for the numbers games based on posts they see on various audiophile forums . dCS have not specified the jitter performance of their clocks, only the accuracy/stability (in parts-per-million).
Thanks Anup. Such an advantage to have smart people on this forum. From my perspective, I don’t know what my MSB Select II DAC would sound like with less than the 33 Femto Clock I bought it with. My guess is that this is as good as MSB can make it sound. And it sounds very very good. I do think that my Vivaldi DAC/Upsampler, as good as the combination sounds, is even better with the Vivaldi Clock, and even better, though by a lesser margin, with the Cybershaft OP21A reference clock. But that could be my wallet talking too, which might well be fed up with the pain I have imposed on it.
It is not possible to connect a 10mHz clock to a Rossini clock. This requires a reference clock input which only the Vivaldi clock has in the current dCS product range.
Adding a reference clock may or may not improve the outcome in sound quality terms. As Anup so rightly says :
It is also very worthwhile reading dCS’ ( in the person of James) own comments. in particular his final paragraph which directly concerns the issue that you raise:
Yes, both of these points are correct. If you are connecting a reference clock to the Vivaldi Clock, it has to be more accurate than the Vivaldi Clock in the long term for it to have any positive impact. So, the accuracy needs to be higher than +/- 0.1 parts per million over time.
Jitter (and subsequently phase noise) on a reference clock going into the Vivaldi Clock is largely irrelevant as in essence the Vivaldi Clock doesn’t respond to fast changes on the reference input - it has a slow acting PLL (also known as a narrow bandwidth) so any high frequency changes, i.e. jitter, are filtered out. Within limits of course - if the jitter is extreme enough it may cause the clock to fail to lock to the incoming signal, although one would like to think this would be rare on an external master clock!
It is also worth noting that the kind of clocks likely to produce this level of accuracy are likely to be atomic clocks, and these are usually a bad idea to connect directly to audio equipment such as a DAC as despite being incredibly accurate over time, in the short term there can be a fair bit of variation as the system tunes around the target frequency - jitter. Not great in terms of the DAC, but when fed to another master clock with a slow acting PLL the high long-term accuracy reference can have a positive effect.
I used to watch a lot of videos on YouTube published by PSAudio, who made my last DAC… the DirectStream … the guy behind it, Ted Smith explains this relationship between short term accuracy and long term, with atomic clocks, etc well here. Funny side note, he talks a lot about the cost of doing something in his various videos, and you get the impression that PSAudio (meaning Paul McGowan) lean hard on him to keep costs down… I heard him say the retail to cost ratio has to be 5 or 6 to 1, so that he’s not able to use the kind of transformers and clocks he necessarily would like to … which might explain their latest forree into the 2-box DAC that’s going to be known as the “Ted Smith Signature DAC”… I think a guy like this might thrive better at a company like dCS who tend to focus more on “Spare little expense” build, and less on the cost… as obviously exemplified by the Vivaldi system’s cost.
I like the idea of Ted at dCS, though I suspect it’s unlikely. On a bang-for-buck basis, I think PS Audio produces excellently performing and high value equipment. I’ve owned a lot of their stuff. Quite musical.
Hmmm. Before getting the Cybershaft OP-21A, I had pulled the trigger on both an SRS Perf10 and an Antelope 10MX rubidium clock. The latter’s studio capabilities scared me off a bit, and I knew less then than I do now. Among other things, I had “over-interpreted” the notion that atomic clocks are generally bad for audio devices. I am inclined to give both a try, and see how they compare to the Cybershaft. Their long-term specs seem to me to meet James’ threshold requirement, although I may be interpreting them incorrectly.
The idea that atomic clocks are bad for audio does have some basis, but it depends heavily on where they are used. Because the short term accuracy (the area that is important for audio) of an atomic clock is pretty low comparatively (jitter is created while the clock tunes itself around the target frequency), it would generally be a bad idea to connect it directly to a DAC for example. It will essentially be a jittery signal, at least by high-end audio standards.
However, while the signal is jittery, in the long term it will be incredibly accurate as all the tuning around the target frequency will average out over time. So, feeding it into a second Master Clock with a quartz oscillator (which have inherently low levels of phase noise and jitter) provides this second Master Clock with a long-term stable reference to adjust to.
This of course depends on how the PLL in that second Master Clock has been designed - if it has a wide bandwidth (is fast acting) the jitter may well be passed on to the DAC. If it has a lower bandwidth PLL (slower acting) the jitter should be filtered out before it reaches the DAC, mitigating the issue. For reference, the PLLs in a dCS DAC or Clock are the latter, being slower acting.
So, are they bad for audio? Really depends on where in the chain they are used, and how the products it connects to use the clock signal.
Thanks for that @James. Very helpful. And consistent with the earlier info you and @Anupc have posted here and elsewhere. I’ve got an SRS Perf10 on the way, and I am in the process of trying to borrow an Antelope 10MX. The Perf10 claims a long term accuracy of ±0.05 ppb ( ±5 × 10-11), and they also assert they have optimized short term phase noise for audio purposes (though, as you note, the Vivaldi Master Clock is less sensitive to this noise in a reference clock). Esoteric sources the same oscillator from SRS for their Grandioso G1 Clock. It will be interesting to see what, if anything, I hear. My experience with the Vivaldi Clock and the Cybershaft is that the changes are holistic, and not identifiable in traditional audiophool terms. But the change was readily apparent.
Maybe a cesium clock in the Vivaldi successor stack?
This last paragraph reminded me of something you wrote in that post that Pete linked above:
From the perspective of a reference clock connected to the dCS Master Clock on a 10mHz input, is this rate multiplication a factor, and does it occur in the reference clock or the dCS Master? Forgive my ignorance; I am just trying to gain a layman’s appreciation for all this. Let’s posit a reference clock of cesium accuracy [I know, not likely for any of us here, except maybe @Anupc ], with acceptable short term specs as well. Presumably, such a reference clock can “enhance” the dCS Master Clock’s accuracy, yes? No? Maybe?
Yes, in so much as the 10MHz signal still needs to be synthesized into something the dCS Master Clock can actually use for its own purposes (audio) which involves going between mathematically unrelated rates. Asynchronous rate converters typically equal a dirtier clock spectrum, which means jitter.
In the dCS Master. Using your example of a cesium clock, this will only operates in the 10MHz domain, so it is up to whatever it attaches to (in this case a clock for audio) to synthesize that into the relevant rates.
If the long-term accuracy was higher than ±0.1 PPM, so with the clock above you mentioned with the long-term accuracy of ±0.05 PPB yes you would likely hear an improvement. Note that this doesn’t have any impact on jitter performance, which is short-term accuracy (think looking at one rising edge of a clock signal next to another on a 'scope, whereas long-term accuracy could be over a period of an hour seeing how far one clock signal has drifted from the other).
Again just to reiterate because of the fact the PLL in the dCS Master Clock is slow acting, the short-term accuracy of the reference clock isn’t too important (if it was, an atomic clock would degrade the sound), it’s mainly the long-term accuracy that is the key factor.
@James, can you elaborate a bit more on how long term clock accuracy matters for audio? Most of the material I’ve read indicates that it’s not very important, as for the duration of say one song the clock skew between even a cheap quartz osciallator and an atomic clock is barely measureable. Intuitively this makes a lot of sense, so what’s being missed here?
I do understand what you’re saying about short term stability mattering less for the slower acting PLL in the Vivaldi. This does however seem to indicate that spending big bucks on a Cybershaft clock that’s optimized for excellent short term stability may not achieve what buyers hope for with a Vivaldi!
Jeff, just the point I made over at WBF over the weekend, and why I am eager to get the SRS Perf10 in to see if I can discern any difference. I am looking forward to @James‘ response, perhaps with some additional detail on the role of dither. Suppose, let’s say, the Cybershaft on paper is more accurate/stable in the short term than the Vivaldi. Is it possible this improves the Vivaldi, and if so, is it likely to be audible? We know our hearing is sensitive to timing, but are we now at a point where incremental improvement may not be discernible? Even if something like the Cybershaft could enhance the Vivaldi’s short term accuracy, does the dither option negate or work at cross purposes? Given the design of the Vivaldi, does adding a more short-term accurate clock simply not do anything? I.e., for practical purposes, is there truly no way to improve the Vivaldi’s timing accuracy in an audibly meaningful way?
Scientifically, I am out of my depth here, but I would love to get to a better cognitive understanding of why and how the Vivaldi might be improved other than the sense of openness and space emanating from the cavernous hole in my wallet.
One other variable that’s not well specified is how good the Cybershaft long term stability is. It’s possible that it’s also an improvement over the Vivialdi clock, so if that does indeed matter that could be what people are hearing. Cybershaft doesn’t specify this, but they do use aged clocks that have supposedly stabilized.
This also ties into the idea of removing the Vivaldi clock from the chain entirely and replacing it with something like the Sound Warrior that converts the Cybershaft to 44/48 Hz with minimal phase noise degredation. Of course if the DAC/Upsampler also use the lower bandwidth PLL, then maybe we’re back to square one again!
Well, the Vivaldi has to do this as well, correct? I mean, there is no avoiding this step, is there? I see the economic advantages of the Sound Warrior [!], but do we have user reports of an SQ comparison? Other than economics, is there a reason to believe the Sound Warrior is a better match for the Vivaldi DAC than the Vivaldi Clock?
P.S. I will dig up my OP21A spec sheet and see if it says anything about long term stability. I don’t think it does.
That’s actually a really good question. You’re right, absolute accuracy has much less of an impact on sound quality than jitter performance. Put another way, I believe the discrepancy on middle C worldwide is around 60,000 PPM, so we’re working at accuracies well above that as it is. Being perfectly frank, given this personally I do not have a solid explanation for why an incredibly accurate long-term clock would provide a sound quality increase.
With that being said, we’ve had reports from customers over the years who have done tests with different reference clocks going into a Vivaldi Clock. Invariably from a few who (I believe) were doing blind tests, they all independently concluded that an atomic clock with a very high long-term accuracy offered an improvement in sound quality. Obviously one clock’s long term accuracy being higher than another’s shouldn’t in itself offer a substantial increase in sound quality given our low psychoacoustic sensitivity to long term accuracy (as a whole the human ear and brain is much more sensitive to short term timing errors than long term frequency drift caused by long term inaccuracy).
I must admit that running an atomic clock into a Vivaldi Clock isn’t something I have had the opportunity to try personally, but the consensus seems to be it does offer an improvement. I should note that unlike some other “does it doesn’t it” topics floating around the audiophile world, this one is whether the human ear is sensitive enough to hear the difference, not whether the difference exists.
A clock directly feeding a DAC has a very different requirement to a clock feeding another master clock though, so obviously the relevant specs in each and how they should be taken into consideration across the whole system will differ. Absolute accuracy into a DAC is a different ball game to absolute accuracy into a Master Clock.
I try to stay away from inter-manufacturer debates and instead discuss the technology and engineering principals. I would say though that we have been in the Master Clock game for several decades now, and go to some impressive lengths to get the best jitter performance possible. We also have pretty much free-reign to create an audio system which is the highest performance possible, and with that over three decades we have settled on the FPGA enhanced VCXO system, with some incredibly capable PLLs throughout the system. The whole Vivaldi system has been designed to sound as good as is possible, period, with all the trade-offs that entails (there’s no such thing as “compromise free” in engineering, engineering is compromise), as opposed to something that’s good on paper.
In short, no, dither won’t impact an external reference clock’s effect on a Vivaldi Clock. The key point it impacts is the DAC’s ability to lock to the incoming Word Clock signal from the attached Master Clock when the phase difference between the two is low (somewhere PLLs naturally exhibit a dead-zone). It’s one of the ways the dCS system as a whole minimises jitter, but operates only within the dCS ecosystem and wouldn’t be impacted by an external reference clock.
. dCS have not specified the jitter performance of their clocks, only the accuracy/stability (in parts-per-million).
], with acceptable short term specs as well. Presumably, such a reference clock can “enhance” the dCS Master Clock’s accuracy, yes? No? Maybe?