Best Audiophile USB Card ?

[Superdad]

2 hours ago, Savolax said:

In my case I think the biggest culprit on "whole I2S" would be distance between my PC and DAC.

Not to mention the clocking being at the wrong end of the cable.  Virtually nobody does external I2S correctly--with DAC feeding its master clock back up to source... 

[Superdad]

1 hour ago, Mr Morris said:

That is expensive!

Especially if the board will have the Connor-Winfield OH300-50503CF as picture in the posted photo.  That OCXO is just $33 an its phase-noise specifications are extremely average, with just -115dBc/Hz at 10Hz offset for a 10MHz version (so probably -109dBc/Hz for the 20MHz version they are using.  We routinely measure $10 Crystek CCHD-575 XOs at that level.

https://www.micro-semiconductor.com/products/Connor-Winfield/OH300-50503CF-020.0M

 

https://www.micro-semiconductor.com/datasheet/0e-OH300-61005CV-030.72M.pdf

[Superdad]

1 hour ago, Gavin1977 said:

But, cheaper than some options: https://www.pinkfaun.com/shop/clock/67-4586-pink-faun-ultra-ocxo.html

 

I’ve heard this clock and it was better than their standard clock, but the price was hard to swallow - but I always remember how it sounded. Staccato on piano and realism was amazing.

 

Well Pink Faun claims -130dBc/Hz at 10Hz for their 20MHz clock, so quite a bit better than any Connor-Winfield.

https://www.pinkfaun.com/shop/sheets/PinkFaunUltra20MHz1.pdf

 

1 hour ago, Gavin1977 said:

If JCAT can do the same, but at lower cost then it crosses the cost/benefit boundary for me.

 

It is hard, because as discussed at length elsewhere, new OCXOs--not new-old-stock or surplus or used as many Asian clock-makers use--with really good phase-noise performance (the only parameter that really matters for a clock in the audio world) in the range of -138dBc/Hz to -147dBc/Hz (@10Hz offset) cost $330~$375, and that's if you order 250 pieces! 

[Superdad]

11 minutes ago, Marcin_gps said:

Regarding Crystek, the numbers you mentioned are hard to believe, because even Crystek claims -90dBc /Hz @ 10 Hz in their marketing materials: 

 

That's their spec for a 100MHz version.  Look again and you will see that for a 24.576MHz CCHD-575 they guarantee -100.6dBc/Hz at 10Hz. 

 

But the graphs Crystek sent us years ago with 3 samples of 25.0MHz CCHD-575 showed -108, -110, and -112 for the three samples. We immediately phoned them to get clarification. Were these selected or random samples?  They said random. Then why the big discrepancy between their published graphs and what they sent us? They said it is because they are a very conservative company, and because while they had greatly improved their production processes for the CCHD-575, they still have a lot of firms paying more for their big CCHD-957 (about $32 vesus $10 for CCH-575, qty. 250) and they dd not want to alienate them (or lose revenue I guess) by letting the world know that their lowly 575 was a bit of a giant killer among production XOs. 

[Superdad]

2 minutes ago, Dev said:

...for a 22.5Mhz which is the lowest they carry in that series. 

 

Other than a couple of stock (audio sample rate related) frequencies, everything is custom with Crystek, 14 weeks. They would make you 500 pieces of a 10.0MHz CCHD-575; $5,000. 

[Superdad]

3 hours ago, Marcin_gps said:

PS - testing an ultra-low-jitter clock is tricky. It requires a very expensive instrument. We are talking over 50k EUR.

It does take care. But not as much money as you state. You just need one of these: https://www.microsemi.com/product-directory/phase-noise-and-allan-deviation-testers/5565-53100a#overview. Plus a couple of good external OCXOs to connect for cross-correlation.

It is an industry standard which has been built and sold under several names (TimePod, Symmetricom 5120A, Jackson Labs PhaseStation) for many years. 

Measures down to about -140dBc/Hz at 1Hz, and -178dBc/Hz at 1KHz

About $23K.  It is what we use.

 

 

[Superdad]

3 hours ago, Marcin_gps said:

Still, Crystek sounds poor in comparison to SiTime 5711 that we use on XE cards.

 

Certainly the ruggedness of the 5711 would seem desirable in the vibration-prone environment of PCIe card in a computer case.  So I guess the $100-$130 cost of those buys you that--since the -109dBc/Hz (@10Hz for a 10MHz carrier) is about what we get with the Crystek 575.

[Superdad]

5 hours ago, wittao said:

No, unfortunately that equipments aren't enough to measure the exact phase noise datas. 

We are using to measure our masterclocks and cards Rohde&Schwarz FSUP and R&S FSWP analyzers. Totaly different level. Of course the price range too...

 

The Rohde & Schwarz analyzers are incredibly well built and beautiful machines with a great many fancy features and functions--most not relevant to simple clock phase-noise analysis in the range of interest to audio. And as you say, the prices to match, probably over $100K for the SFWP, and that might not even include the costly -B61 cross-correlation option for low phase-noise measurements.  

 However, it is not correct that the Rohde & Schwarz analyzers are more sensitive than the John Miles-designed TimePod/Symmetricom/Jackson-Labs/Microsemi 5120A/53100A series units.  Please refer to the specification tables for noise floor of each at 1Hz and 10Hz offsets for clocks in range of 5~10MHz. 

As evidenced below, the Miles unit (now sold by Microsemi for about $23K) with a noise floor of -135dBc/Hz at 1Hz and -145dBc/Hz at 10Hz (both typically even better but these are their guaranteed specifications) is significantly lower than either the R&S FSUP or FSWP units--even when those are equipped with the cross-correlation options.

The Rohde & Schwarz are of course terrific if you are measuring Gigahertz clocks and doing other very advanced analysis as required for microwave communications industry.