0,097 ps clock from MSB Technology

[ferenc]

Seems absolutely crazy if it is true, and why it would not be true?

 

http://www.msbtech.com/products/galaxy.php

 

MSB Tech proudcts are generally very nice products, it was absolutely the best sound at my home when I could have their Platinum DAC, Data Transport and big monoblocks at home for a few days.

 

[1audio]

The published numbers don't add up.

 

The calculated jitter 1Hz to 10 KHz is more like 600 fS. I'm attaching the calculation report from an online service.

 

Sometimes the number is too good to be true. Perhaps someone didn't enter the numbers carefully enough.The curve looks odd as well. And I'm not sure why it needs a space shuttle tile, a single oven OXCO in a proper Dewar will work better and use less power. (Sorry, BS detector on overdrive. . .)

 

When I started researching oscillators that could meet those numbers I discovered that they are considered controlled commodities and need special approval to export. They also take years to develop and many years of experience to optimize.

 

I'm sure its very good and contributes to good overall performance, but when the claims are over the top it can tarnish the credibilty of the rest of the presentation.

 

[PeterSt]

What seems wrong to me (from another angle) is the curve;

No oscillator with those specs will carry a 120dBc/Hz at 1KHz. What I use is better than 150dBc/Hz and still it is around 0.8ps RMS only.

Figures for the same curve points and a level of -10dBm (between brackets are the MSB numbers) :

1Hz: -70 [-85]

10Hz: -90 [-110]

1KHz: -152 [-120]

10KHz: -152 [-150]

 

Of course the real difference is in the lower frequencies, and maybe it can be done. However ...

 

I didn't perform the math myself, but to me it looks JitterTimes may be wrongish;

At the right side 3 cells show where I expect 4 of them. Ok, maybe that is correct. However, start changing the individual (4) base cells, and see *what* changes in that rightmost column. That does not look right to me at all.

 

Furthermore, maybe the oscillator doesn't run on your filled in 24576000. That matters (use a double speed and the math says RMS jitter halves. Do that one time again and you will be close to the speced numbers. Also it matters how many of the low frequency numbers are added to the table. The more lower frequency numbers added, the lower RMS jitter will "become".

 

All 'n all, I don't know ...

Peter

 

[1audio]

Peter:

You can enter the data yourself here: http://www.jittertime.com/resources/pncalc.shtml and it will do the calculations for you. Try a number of variations and it will all get back to the same places. The low frequencies will contribute the most. Numbers below -150 won't make a big difference. The noise below 10 Hz will be the biggest contributor. AES specifies phase noise from 50 Hz to 20 KHz only. There are valid arguments that the low frequency noise won't affect the audio much at all. The low frequency noise is similar to wow and flutter. Any digital system should have way lower wow and flutter than even the best turntable or tape system.

 

The usual curve for a crystal oscillator has a steadily increasing noise as the frequency descends. Here is a quick intro that describes the phenomena: http://www.ieee.li/pdf/essay/phase_noise_basics.pdf

 

I chose 24.576 MHz because its a common frequency for digital audio. You can try other frequencies and it scales. Real world parts and circuits don't cooperate quite the same way with higher frequency oscillators being noisier.

 

[PeterSt]

You can enter the data yourself here: http://www.jittertime.com/resources/pncalc.shtml

 

Demian, I hoped it was clear I just did that ...

 

The low frequency noise is similar to wow and flutter. Any digital system should have way lower wow and flutter than even the best turntable or tape system.

 

... which I would like to turn into SLOWER.

FWIW of course ...

 

Peter

 

 

 

[57gold]

About pricing!

 

[hiro100]

Yowsa, 10k for the clock!

 

[sq225917]

They should use polystyrene, not space shuttle tiles. Costs less and insulates better upto 200 degrees. Unless of course the clock runs closer to a thousand degrees...... lol

 

[The Computer Audiophile]

Maybe Space Shuttle tiles are on sale now that the Shuttle program is over :~)

 

 

 

[Panelhead]

Somehow I have to expect the clock for the ADC is as important as the the one in the dac. We are trying to reproduce an analog waveform.

I also expect a 0.097 ps jitter clock will be playing back data with maybe 200 - 300 ps jitter imbedded in the samples. But it makes sense that a dac clocked to this level of accuracy is better than one with an additional 200 - 300 ps jitter.

I really liked this clock before seeing price.

 

 

George

 

[ar-t]

Two things:

 

1.) If their numbers at 1 Hz and 10 Hz are right, the 1 kHz number does not fit the curve.

 

2.) They do not specify the clock frequency. If it is a 45 MHz clock, then using their numbers, then jitter is around 350 fSec. If I adjust the curve, using the 1 Hz, 10 Hz, and 100 kHz numbers, it gets close to 100 fSec. Lower the clock frequency to 11 MHz, the jitter rises to 400 fSec.

 

-85 dBc, @ 1 Hz is good. I would have to measure one to believe it. It is obtainable, but not with cheap components.

 

[1audio]

They updated their web page to show how they measured the phase noise. I think the measurement is bogus since they are measuring the sidebands of the test signal playback. It gives impressive numbers but probably is not the correct math. I just checked my reference and I can get -135 dB at 1 Hz measured this way. That would be attosecond jitter. The actual source is very good, in the 50 fS range but not that good. Not possible in the real world. . .

 

Also anything that is that good becomes a controlled commodity and you need DOD clearance to sell it.

 

[ar-t]

They have a table below the phase noise plot. They claim -99 dBc @ 1 Hz, with a 24.576 MHz clock. Also mentions it is at CMOS signal levels. Looks to me they measure stuff the same way we do. If they do, the numbers seem right.

 

[1audio]

That is a playback plot made using an audio track. Its not a close in measurement of the oscillators phase noise. If you look at the plot it was made using a 48 KHz sample rate and the center tone is 12 KHz. The Jitter time calculation won't be valid for that measurement, there are a number of pieces in the middle that affect the measurement.

 

I created a spreadsheet with correction values for a 44.1 KHz sample rate at 11.025 KHz tone. The following would apply to deterministic jitter tones. In effect, these are approximate values, the ones I needed for something:

-112 dB = 50 pS

-120 dB = 20 pS

-134 dB = 4 pS

-146 dB = 1 pS

 

Doing the integration for the random jitter is more than I can get my head around. I'll leave that for the math types.

 

The absolute floor for 24 bits is -144 dB but averaging and dither will get below that limit especially with narrow band (large fft) measurements.

 

 

 

[ar-t]

I don't think we are looking at the same page.

 

[1audio]

Its the plot on the page linked. If you look at the plot it lists the details on how it was made and its marked phase noise. The +/- 1 Hz lines are at -99 dB, like their table data.

 

[Pacific.Digital]

Demian is certainly correct re the low frequencies being where most of the phase noise accumulates. Let's see if we can get MSB to comment. The engineer doing the OXCO design certainly knows what he is doing.

 

[ar-t]

You are assuming they designed it. They probably bought one from the same place TAD gets theirs, for the D600 CD player. It also boasts -99 dBc @ 1 Hz. Supposedly, it cost them $500.

 

[PeterSt]

If you look at the plot it was made using a 48 KHz sample rate and the center tone is 12 KHz.

 

I have been looking and looking ... but where do you derive that 12KHz from ?

 

Regards,

Peter

 

[PeterSt]

I just checked my reference and I can get -135 dB at 1 Hz measured this way. That would be attosecond jitter. The actual source is very good, in the 50 fS range but not that good. Not possible in the real world. . .

 

Demian, you are saying that this is playing a 12KHz signal (normal sine) on to a 48KHz based clock, right ?

Well, if so, no way I saw something coming up like that plot shows, hence (without looking now) I think I would be in that -135dB range as well. BUT :

 

They used a Blackman window and I sure never to that. Did you ? this matters ...

 

Peter

 

 

 

 

[PeterSt]

From my earlier post :

 

Figures for the same curve points and a level of -10dBm (between brackets are the MSB numbers) :

1Hz: -70 [-85]

10Hz: -90 [-110]

1KHz: -152 [-120]

10KHz: -152 [-150]

 

Wich now turned into these number ? (between brackets the old MSB data) :

 

0.1Hz: -67dB [n/a]

1Hz: -99 [-85]

10Hz: -134 [-110]

1KHz: -157 [-120]

10KHz: -157 [-150]

100KHz: -157 [n/a]

 

I must say that the curve looks more normal to me now.

But what changed in a few months time ? Well, it got better again.

?

 

[1audio]

You are right, there is no indication that it was at 12 KHz. It could be at any frequency below 24 KHz with a 48 KHz sample rate. The "standard" would be to use Fs/4 or 12 KHz.

 

The Blackman window is what Stereophile uses I believe along with 100 averages of a 65K FFT. They use 1 average of a 262K FFT.

 

The numbers in their chart seem to match the displayed plot, which is the giveaway. There may be a way to scale that plot and then use the standard calculators to translate back to jitter. I am not sure I would have a lot of confidence in measuring that way. Further the plot does not follow Leeson's curve suggesting it may not be a real facsimile of the phase noise plot. PeterST, since you seem to be obsessive enough (maybe? I'm not.), here is an analysis of oscillators you can lose a few hours to: http://www.adret-electronique.fr/articles_scientifiques/phase_noise_in_oscillators.pdf

I think those hours are better spent listening to music.

 

There is some stuff here: http://www.fidelix.jp/technology/ on measuring jitter and how to match the Stereophile tests. Windowing is not a simple subject but for those making meaningful measurements it's important. They suggest a Blackman-Harris 7 term window to match the Stereophile measurements.

 

[jeroenvtn]

Well, it all looks spectacular of course, but this one

 

http://www.wickeddigital.com.au/index.php/2011-09-04-22-32-08/news/122-calyx-super-dac-at-ces-2012

 

is only half the cost of just the space shuttle tiled clock!

 

 

Greetings from Rotterdam,

Jeroen

 

[CG]

Oscillators with this level of phase noise performance are available from a number of sources - har, har.

 

Here's but a couple:

 

http://www.wenzel.com/pdffiles1/Oscillators/BTULN.pdf

 

http://www.vectron.com/products/ocxo/ox-042.pdf

 

http://www.greenrayindustries.com/library/yh1322.pdf

 

 

BTW, this is good reading for those so inclined.

 

http://www.greenrayindustries.com/library/PhaseNoise08.pdf

 

 

I do have to wonder what happens to these when applied to your typical DAC chip in a typical audio environment...

 

[1audio]

I just did a round of tests of Jitter in this manner. Depending on the windowing and the resolution I can get pretty remarkable numbers, as low as -125 dB at +/- 1 Hz. (I measured using a 16M point FFT at 48 KHz sample rate using a Hodie 7 window, for those that want to know). The oscillator under test is the one in the AB1.1 USB DAC ($138 here: http://www.qnktc.com/ab_11/). Its an async dac and the oscillators on board are certainly OK (Golledge) but hardly in the ultra class, however I can make them seem so.

 

I don't think I can measure real phase noise this way. I believe there may be a way based on the noise floor (-150 dB in this measurement) but that is very suspect since random thermal noise from many things will affect the measurement.

 

And to CG's point, getting that really clean clock to a DAC chip is difficult. Once its there the internal silicon probably will degrade whatever it gets by a significant percentage as well. And I'm not sure better clocking will actually improve things past a certain limit.