Clocks

[treitz3]

Is there a clock that you found to have the performance to cost ratio and will make a difference in most systems? Looks as if you recommend the Mutec?

Hello and good morning to you. While I am not Alex, I do have the Mutec REF 10 SE 120. I used to have a After Dark Emperor Signature ClayX Giesemann OCXO 10MHz Reference Clock and the difference between the two are considerable. I personally could not imagine the SE 120 not making quite the change in any system. It was worth every penny and then some to these ears and catapulted the sound out of an Esoteric P-02 transport and an ANK 4.1 Pro to new sonic heights. FWIW.

Tom

 

[Republicoftexas69]

Hello and good morning to you. While I am not Alex, I do have the Mutec REF 10 SE 120. I used to have a After Dark Emperor Signature ClayX Giesemann OCXO 10MHz Reference Clock and the difference between the two are considerable. I personally could not imagine the SE 120 not making quite the change in any system. It was worth every penny and then some to these ears and catapulted the sound out of an Esoteric P-02 transport and an ANK 4.1 Pro to new sonic heights. FWIW.

Tom

Thank you Tom.

 

[Superdad]

Is there a clock that you found to have the performance to cost ratio and will make a difference in most systems? Looks as if you recommend the Mutec?

Some comments:

While (actual measured at the output) phase-noise performance is very important, it is not everything.
a) How the output of the clock is driven, how it is isolated, and its local power networks are also extemely important.

b) While a sine wave is fine (and easier to produce; most all the Asian clocks have truly crappy square waves), a really good square wave clock will be better for all digital devices (including EtherREGEN, though with the eventual Gen2 we will use an expensive and amazing sine>square converter—to the benefit of the many users of sine wave clocks); Digital devices (chips) have an easier/more consistent time (pardon the pun) triggering off a square wave. The Mutec clocks are the only ones we have ever seen with a proper clean square wave. (There are plenty of independently made waveform measurements of various clock brands to back this up.)

c) Aside from the finely engineered output driver circuit and tiny transformer at each isolated output of the REF10 Nano (the very same circuits as in their flagship REF10/REF10 SE120), the phase-noise specs you see for the Mutec boxes are VERY conservative. We measured a Nano at -114/-145 (dBc/Hz at 1Hz/10Hz offsets).

d) The Cybershaft clocks are very fine sine wave units--and some are quite a good value since they opened their direct web shop and introduced more entry models. I have a nice relationship with the owner, Mr. Kenji Hasgawa, and enjoy one of their mid-models in my own system. But since last years introduction of Mutec's 4-output REF10 Nano (with built in PS and jack to optionally externally power), well let's just say that it is hard to compete with...

e) Despite a lot of emphasis being put on the 1Hz offset phase-noise offset number, our tests--both listening and measured--indicate that performance around 10Hz offset is actually more critical to audible enhancement. A lot of clocks will make it to -130 or -135dbC/Hz at 10Hz, but there is some magic that seems to happen once you truly obtain (at the end of the clock cable) -140 or better.

f) Clock cable considerations are quite different for square wave clocks than for sine wave clocks! And impedance matching--of clock/cable/device--matters for square wave clocks but does not matter at all for sine wave clocks.
We wrote a simplified paper about those topics.

 

[Superdad]

Other possible circuits like analog (non-crystal based) PLLs or Frequency Synthesizers are all much higher jitter than a simple average “jellybean” crystal oscillators so they are not worth getting into their technical intricacies. In my view, if a product supports external clocks it had better be for a really! good reason, because the inevitable drawbacks of supporting them.

Hi Dustin:
With great respects for your all your advanced design work at MSB (and we have met a few times at shows over the years when I co-owned Hovland Company--the first time was about 23 years ago--we were so young! ), I can point out that clock synthesizers have come a LONG way. The phase-noise/jitter added by the best chips from Skyworks and Analog Devices is quite negligible--especially when you look at low offsets.
We have been using such parts for a while, and they are fantastic at producing differential outputs (so much better than running SE clock lines all over a board) at various frequencies (we need a couple 250MHz lines in our current model and 1.25GHz in our forthcoming switch to run high-speed reclocking flip-flops).

And as for DACs:
I'm sure you know well the OEM cost ($300 to $600 at quantity) of an OCXO with truly outstanding PN performance (You mentioned VCXOs but I've yet to see one with PN better than a good $30 Crystek 957 XO.)
So while indeed a DAC with a pair of built-in "native" frequency top-caliber OCXOs would obviate the need/benefit of an external reference clock, it is rather the rare DAC that includes such.

Yet I fully agree with you that, if an external 10MHz reference clock is to be used, the means of frequency conversion, the cabling used, and the clock itself must all be outstanding. Otherwise there is little point and people are just wasting money.

Cheers to you and Daniel,
--Alex Crespi

 

[Republicoftexas69]

Some comments:

While (actual measured at the output) phase-noise performance is very important, it is not everything.
a) How the output of the clock is driven, how it is isolated, and its local power networks are also extemely important;

b) While a sine wave is fine (and easier to produce; most all the Asian clocks have truly crappy square waves), a really good square wave clock will be better for all digital devices (including EtherREGEN, though with the eventual Gen2 we will use an expensive and amazing sine>square converter—to the benefit of the many users of sine wave clocks); Digital devices (chips) have an easier/more consistent time (pardon the pun) triggering off a square wave. The Mutec clocks are the only ones we have ever seen with a proper clean square wave. (There are plenty of independently made waveform measurements of various clock brands to back this up.)

c) Aside from the finely engineered output driver circuit and tiny transformer at each isolated output of the REF10 Nano (the very same circuits as in their flagship REF10/REF10 SE120), the phase-noise specs you see for the Mutec boxes are VERY conservative. We measured a Nano at -114/-145 (dBc/Hz at 1Hz/10Hz offsets).

d) The Cybershaft clocks are very fine sine wave units--and some are quite a good value since they opened their direct web shop and introduced are entry models. I have a nice relationship with the owner, Mr. Kenji Hasgawa, and enjoy one of their mid-models in my own system. But since last years introduction of Mutec's 4-output REF10 Nano (with built in PS and jack to optionally externally power), well let's just say that it is hard to compete with...

e) Despite a lot of emphasis being put on the 1Hz offset phase-noise offset number, our tests--both listening and measured--indicate that performance around 10Hz offset is actually more critical to audible enhancement. A lot of clocks will make it to -130 or -135dbC/Hz at 10Hz, but there is some magic that seems to happen once you truly obtain (at the end of the clock cable) -140 or better.

f) Clock cable considerations are quite different for square wave clocks than for sine wave clocks! And impedance matching--of clock/cable/device--matters for square wave clocks but does not matter at all for sine wave clocks.
We wrote a simplified paper about those topics.

Thank you so much for sharing

 

[tony22]

Yet I fully agree with you that, if an external 10MHz reference clock is to be used, the means of frequency conversion, the cabling used, and the clock itself must all be outstanding

Hi Alex. So (without trying to put words in your mouth), if a (let’s say in these cases a switch or a streamer) has a frequency conversion circuit whose output is really not as good as that of the onboard clock, then using an external clock may not really be the most effective approach.

 

[Dustin Symanski]

Hi Dustin:
With great respects for your all your advanced design work at MSB (and we have met a few times at shows over the years when I co-owned Hovland Company--the first time was about 23 years ago--we were so young! ), I can point out that clock synthesizers have come a LONG way. The phase-noise/jitter added by the best chips from Skyworks and Analog Devices is quite negligible--especially when you look at low offsets.
We have been using such parts for a while, and they are fantastic at producing differential outputs (so much better than running SE clock lines all over a board) at various frequencies (we need a couple 250MHz lines in our current model and 1.25GHz in our forthcoming switch to run high-speed reclocking flip-flops).

And as for DACs:
I'm sure you know well the OEM cost ($300 to $600 at quantity) of an OCXO with truly outstanding PN performance (You mentioned VCXOs but I've yet to see one with PN better than a good $30 Crystek 957 XO.)
So while indeed a DAC with a pair of built-in "native" frequency top-caliber OCXOs would obviate the need/benefit of an external reference clock, it is rather the rare DAC that includes such.

Yet I fully agree with you that, if an external 10MHz reference clock is to be used, the means of frequency conversion, the cabling used, and the clock itself must all be outstanding. Otherwise there is little point and people are just wasting money.

Cheers to you and Daniel,
--Alex Crespi

You are correct, I have added VCXO ability to many oscillator circuits and they always necessarily increase their phase noise, However it doesn't have to be by much, a db or two at the most. I measure clock circuit candidates all the time with the high end phase noise measurement system that lives on my desk (I like nice toys), and frequency synthesizers never reach amazing performance in circuit. Properly implemented VCXOs with software control are always better in practice. For frequency agile applications however synthesizers are an amazing solution, if that is required by the application, and for audio it is not. You can see by the following graph that Crystek oscillators beat their datasheet specifications easily, however they don't hold a candle to a properly implemented OCXO. I don't bother saving data from anything with higher phase noise than a Crystek because honestly for 24bit audio its the minimum that is required for high quality converters (in practice, not necessarily in theory, but often theory is lacking and that's where experience comes into play)and I have yet to measure a PLL or synthesizer that comes close in circuit (with the filters required for spur removal, synthesizers are terrible for sound quality if you don't remove the spurs).

 

[Superdad]

Hi Alex. So (without trying to put words in your mouth), if a (let’s say in these cases a switch or a streamer) has a frequency conversion circuit whose output is really not as good as that of the onboard clock, then using an external clock may not really be the most effective approach.

Hi:
Just now seeing your questions.
The trouble is, one can not generalize in the answer, because as with all things, it depends upon actual implementation.
Permit me to explain some variables, which I'll pose as questions:
a) Does the device with external clock input actually have internal clocks at native frequencies (22.xxx/24.xxxMHz or 44.xxx/49.xxxMHz for example for a DAC; or 25.0MHz for a switch or 24.0MHz for a USB processor)--or is the box synthesizing what it needs from some other frequency clock? Perhaps a 10MHz internal clock.

b) If it has internal clock(s) native to what the device's chips actually run from, are those clocks truly ultra-low-phase noise OCXOs--at the level of what is available in an external?
[And one must be careful in comparing because of frequency differences; rule-of-thumb is about 6dBc/Hz derating per octave--in other words a --100dBc/Hz @10Hz 80MHz clock is about equal in performance to a -118dBc/Hz @10Hz 10MHz clock.]
We see a lot of unknown spec OCXOs tossed in equipment--just to be able to say they have an OCXO--which perform worse than a nice $10-$30 Crystek CCHD-575 or 957 XO.

While the output of a clock synthesizer can never (except for certain exceptions with particular "jitter attenuating" synths) be better than the clock it uses as reference (be it internal or external), the amount of phase-noise added does depend entirely upon the circuit--as well as how the clock lines are handled internally.

I could go on, but this truly is a case-by-case, implementation-specific subject that defies generalizations.
So sure, for some devices an external clock input may be nothing more than a gimmick, and for other units the use of a certified performance clock may be significant.

Hope that helps.
Best,
--Alex C.

P.S. The original reason that the EtherREGEN ended up with an external clock input was because since we already had to use a clock synth (a really good one) to create the 250.0MHz clocks (not 25.0MHz, though we use that those also) for our ultra-low-jitter 10GHz-capable differential data reclocking flip-flops, and since the synth has 2 inputs and 4 outputs, it was easy enough for us to add a BNC jack, a little switch (which the clock signal of course does not actually pass through--a tiny microprocessor just looks at the position of that switch at the moment power is applied to the ER and loads the synth with code telling it whether to use the internal 25MHz Crystek XO or an external 10MHz as reference) thus making use of both available inputs. We honestly did not know at the time of development if people would use or hear a difference with an external clock into EtherREGEN. Now, 3,500 units later, we know that there are possibly a 1,000 people or more doing just that!

 

[tony22]

GREAT ANSWER! Thanks!

 

[antigrunge]

Being one of the 1000 that Alex mentions I can attest that my Etherregen benefits significantly from my Antelope 10m Audiophile clock fed by a Sean Jacobs DC3/4 PSU. I cannot overemphasize the importance of short, highest quality BNC cables. Alex’s point on the absence of highest quality Ocxo clocks in most DACs for price reasons is equally pertinent

 

[tony22]

While a sine wave is fine (and easier to produce; most all the Asian clocks have truly crappy square waves), a really good square wave clock will be better for all digital devices

Hi Alex! I know this is an older post, but I had another question - since the really good clocks typically seem to have square wave outputs, what happens when trying to use one with a device that can (as a manufacturer might say) use either sine wave or square wave clocks but is "optimized" for sine wave? Is there any point in using a square wave clcok in this instance?

 

[Superdad]

Hi Alex! I know this is an older post, but I had another question - since the really good clocks typically seem to have square wave outputs, what happens when trying to use one with a device that can (as a manufacturer might say) use either sine wave or square wave clocks but is "optimized" for sine wave? Is there any point in using a square wave clcok in this instance?

I'd be skeptical.
The super-short answer is that there really isn't any such thing as "optimized for sine" (well there is, but it's highly uncommon and never done in the audio market to my knowledge). The steep rising edges of a square wave provide for less ambiguity for the chips detecting clock transitions.
With a sine wave, the low slope means the time variations are large. But for the same amplitude modulation amount a square wave’s time variation will be MUCH less. In other words, an input receiver is much more sensitive to AM on a sine wave than with a square wave.

In addition, any 10MHz clock is going to be fed through some variety of clock synthesizer circuit (there are good ones, great ones, bad ones, and truly dreadful ones out there) to produce the frequencies actually required by the chips or DAC circuits. The input receivers of most clock synthesizers are very sensitive to AM noise--and this is much more prevalent with sine wave clocks.

One way to correct for this is to use a good low-pass filter on the clock line (such as Mini-Circuits BLP-10.7+) to clean the signal of harmonics before the clock input/synthesizer.
Another method, and one will employ at the clock input of EtherREGEN Gen2 is to use an extremely low-phase (<45femtoseconds) noise buffer/driver logic converter as sine>square wave converter (we will also run our standard internal Cystek 575 though this expensive chip as it will be a great way to squash any extremely high harmonics).
So our case is a rare example of "optimizing" for a sine wave, but that does neither means a sine wave will be better than a square nor that a square wave won't be as good as without that special chip (which we have never seen ANYONE in audio deploy).

The above answer is about the shortest I can give on the subject. We did write a short paper on the matter.

 

[tony22]

Awesome and clear answer as always! Thanks!

 

[Kingrex]

@Superdad
What do you think of your switch compared to the high $ ones out now.

Is there anything your doing to change yours. Do you feel the others reach higher with the big $$ outlay?

Sorry if this puts you on the spot.