Purchase of usb I2S converter

I would need a recommendation for USB to I2S converter. I need the best possible sound quality. There are too many producers and sellers and it is not easy to find your way around. Is there any sound difference between xmos XU208 and XU216?

I like jl-sounds for its wide range of settings and options external masterclock, xmos from diyinhk has compared to that a more powerful XMOS chip. I don't need a flip flop reclock or isolated i2s. Diyinhk do not offer any datasheet for their product. It is not good to buy only according to the picture, even if iz has more powerful xmos. Masterclock from diyinhk is only 49,xxMhz? I have older DAC, i need also 24,xx MHz MCLK.

Thank you.

I2SoverUSB - I2S over USB Audio
Home
XMOS 768kHz DXD DSD512(DSD1024) high-quality USB Type-C to I2S/DSD PCB - DIYINHK
 
Best sound quality I have found is either I2SoverUSB or the latest Chinese USB board with Accusilicon clocks. The latter board can be found at: USB digital interface AS318B PCM1536 DSD1024 compatible with Amanero Italy XMOS to I2S|Digital-to-Analog Converter| - AliExpress

Also, JL Sounds makes an optional oscillator board that allow use of alternative clocks with I2SoverUSB. There are clock footprints that can work with Crystek 975 or the large case size Accusilicon AS318-B.

Regarding the Chinese board mentioned above, to get ASIO drivers for it you may want to inquire with the seller. They don't have their own ASIO drivers, but they did something funny that lets the USB board use other ASIO USB board drivers by use of some special configuration software the seller can provide (figures, I guess, considering where it comes from). Also, if the board is used in a system with other USB boards that have ASIO drivers it will try to steal the ASIO connection from them (as I found out when weird problems occurred a few times). The stealing happens when the other ASIO USB board is playing then drops out momentarily such as from the USB cable being briefly unplugged. However, when used with a computer that doesn't have other connected ASIO USB boards, then no problems were observed.

Something perhaps worth knowing about both I2SoverUSB, the optional oscillator board, and the Chinese USB board is none of them appear to leave both clocks running at all times. Rather they appear to use the disable pin on the clocks to turn one off when the other clock is in use. While that approach might prevent some potential problems from occurring, it can cause its own problem. Good clocks such as NDK SDA, Accusilicon, and or Crystek 957 all take roughly 3-days of continuous running for audible jitter to settle down to steady state level (at least I the settling with Topping D90 and with AK4499 evaluation board; that said, and IIRC, D90 leaves everything running all the time once plugged into the wall and the hard power switch on the back is turned on; the display turns off and the mute turns on when the soft power switch on the front panel is turned off is all).

Regarding diyinhk, I have tried their USB boards and do not find the jitter to be audibly minimized as much as the other USB boards mentioned above. Should not be a problem if you use their USB boards with your own external clocks and you reclock the I2S signals before they go into the dac.

Maybe worth also mentioning is that genuine Amanero boards purchased directly from the manufacturer in Italy can work quite well with external clocks, and they are the only USB board I know of that accepts 22/24MHz external clocks. The main trick with getting good sound out of Amanero boards is to use external clocks and to remove inductor L1, then use your own clean isolated 5v supply to power the USB board (rather than letting it use the default USB bus power).
 
Last edited:
Mr. Markw4, thank you very much. That's exactly how I plan to do it. Master clock connected to usb board and DAC, usb board in slave mode and reclock I2S. Secondary clock PCB for i2s over usb I will not buy, I will create my own clock in DAC layout.

It's probably solved, I'll buy of JLsounds ...
 
Good clocks such as NDK SDA, Accusilicon, and or Crystek 957 all take roughly 3-days of continuous running for audible jitter to settle down to steady state level.

I'm sorry, your product recommendations are appreciated, but this part is total BS. Maybe your brain needs 3 days of continuous running to settle down? Are you even sure the standby pins shut the oscillator off? Several I have used over the years just tri-state the output buffer.
 
Last edited:
Chris, I have not checked every clock I may mention from time to time to see if they all stay warmed up when disabled. Until proven otherwise I will assume they don't. I know NDK SDA should both be left on, same for Crystek 957 (I know they tri-state, but they don't sound warmed up after being disabled overnight).

Not sure about Accusilicon. However IIRC, Topping leaves both AS318-B enabled all the time in D90 and they are switched on/off to the dac chip in the CPLD (which also performs clock division when needed).

Regarding the warmup time after power is applied and after being left off for some period of time, they all take time to warm up again. Each brand/model of clock also has its own characteristic audible aberrations that gradually diminish over the warmup and stabilization time. Measurements of clocks by Andrea Mori and other people with the appropriate phase noise measurement gear confirm that warmup and stabilization can take a long time to fully settle. Andrea says it can be weeks. For audio I'm not sure that we humans will notice it too much after a few days of settling.
 
Last edited:
Disabled Account
Joined 2019
That's good infos, thanks for that. Never knew the xtals needed till 3 days of powering to reach their best level.



I understand more why some boards with LiPoFe4 cells are comming with opions for the Xtals to be 24/7 switched on !



Could be a challenge in our diy dac to isolate the others stages and dac chips and lytics caps from a 24/7 switched on boards ? Many PCM chips can be hot like the tda1541a and what about the smmothing caps in front of diode bridges as well ?
 
Yes, to the separate power supply. ^

I2SoverUSb always requires a separate clean 5v supply since it is gavanically isolated (only the 'dirty' side runs on USB power).

The Chinese USB board with Accusilicon AS318-B clocks looks very similar to Amanero. It has a component labeled 'L1' in the same location, and the method for converting it to clean 5v power is generally the same as for Amanero.
 
Best sound quality I have found is either I2SoverUSB or the latest Chinese USB board with Accusilicon clocks. The latter board can be found at: USB digital interface AS318B PCM1536 DSD1024 compatible with Amanero Italy XMOS to I2S|Digital-to-Analog Converter| - AliExpress
How about this one (about $37 USD on eBay) ?
Xmos +Cpld U208 Dac Daughter Card Usb Digital Interface I2S Dsd Output Suitable Ak4497 Es9018 Es9028 Es9038 Dac Decoder Board|Cable Winder| - AliExpress


Also, there may be future here if Schiit were to drop the price a bit (from $150) , and sell it as a solitary unit:

unison%20board.jpg

Schiit Audio: Audio Products Designed and Built in California
 
How about this one (about $37 USD on eBay) ?

Would avoid that one myself. No ASIO support, clocks are not ultra-low phase noise at low offsets, no clear way to run on clean power, no galvanic isolation, etc.

Also, the Schiit board you mentioned to is probably not suitable for dacs other than the ones Schiit recommends it for. It is limited to 192kHz PCM maximum, and no DSD at all.
 
Please compare/contrast in detail

Would avoid that one myself. No ASIO support, clocks are not ultra-low phase noise at low offsets, no clear way to run on clean power, no galvanic isolation, etc.
.
(Let's stick to the Chinese models (yours and mine) )
.... how can you discover all that from either the provided item descriptions and/or images? In other words, how can you tell yours has "galvanic isolation" and "way to run on clean power" and "clocks are not ultra-low phase noise at low offsets."
What am I looking for / reading for?

Let's compare your's to mine:
==================
YOURS:
[Ali link]
Hbe8d76270bd741dda337cd2d00bb95536.jpg

Hfdf15f81e5ab4b1d96f65568764204e0o.jpg

===================
===================
MINE:
[Ali link]

HTB1m80mSOLaK1RjSZFxq6ymPFXaX.jpg

HTB1Jls6SrrpK1RjSZTEq6AWAVXao.jpg

===========
 
The difference is the clocks, and in the voltage regulation. The tiny silver color Accusilicon AS318-B are quite good. The larger black clocks on the other board don't even have quartz crystals in them, and the jitter (as it affects audio reproduction) is much worse. The first board also has separate voltage regulation for the clocks.

Neither of those above has galvanic isolation, takes more chips on the board to do that. I2SoverUSB by JL Sounds has it. It is a more complex looking board as I think can be seen: I2SoverUSB - I2S over USB Audio ...however, the stock clocks are NDK SDA, not quite as good as Accusilicon AS318-B.

By the way, the first board will be best with L1 removed and external 5v power applied. It takes about 3-days for clock jitter to fully stabilize (or come pretty close to it).

You see, I am not theorizing. I bought several (well, more than several) USB boards to evaluate. Same for dac boards. Reverse engineer most of them, sketching out circuits from following traces and ringing out connections with a meter. Many listening tests performed too. After enough time and effort invested, one gets to know a lot about how dacs, USB boards, etc. are designed just by looking at at pictures of them.
 
Last edited:
MEMS vs. xtal

My actual ($37) board is a bit different than the "Mine" I noted in my prev. post (somehow, the eBay seller has updated his orig offering). Mine has two of those SiTime MEMES-based oscillators, and a (third) metal-can quartz-based osc. at 24Mhz. I'm not sure whether traditional crystal is better than the newer MEMS oscillators. However, many of the manuf. datasheets and white papers claim (dollar-for-dollar) that MEMS outperfom quarts:
https://www.sitime.com/sites/default/files/mature-datasheets/SiT8008-datasheet.pdf
https://www5.epsondevice.com/en/information/technical_info/pdf/wp_e20140911_osc.pdf

On the other hand ...
https://q-tech.com/wp-content/uploads/Q-Tech-MEMS-vs-Crystal-Oscillators.pdf
 
On the other hand ...

That last document is one the most relevant to audio dacs.

The board you found with TXCOs is not good either. For best sound quality, dac clocks don't need to be temperature compensated, but they do need to be ultra-low phase noise at very low offset frequencies (among other offsets, depending).

Here's the thing: There are only a few manufacturers and models of good audio clocks. We all know about the ones that are commonly available, which aren't very many. At the lowest cost there is NDK SDA. Crystek 957 can be very good if powered correctly, but they are a bit too expensive for common USB boards.

Accusilicon is a fairly new player with its AS318-B clocks. They come in three physical sizes, but sound the same so far as I can tell. The high end AK4499 Chinese dacs from Topping, SMSL, Gustard are all using the large format AS318-B. Apparently they can get them at good prices in China where they are manufactured. The very small AS318-B clocks like you see on the USB board I linked to require special layout and a tiny bypass cap located within 1mm of them in order to not malfunction (apparently, the larger can versions have internal caps?).

Two above clocks are the best commonly available right now for less than $30/ea or so. The NDK SDA are less than $3/ea in quantities. Crystek 957 maybe $25 in small quantities.

Large can Accusilicon can be had in the Western world for more like $50/ea and up. They aren't worth it IMHO. The smallest ones purchased in bulk in China must be the cheapest for them to be appearing on a USB board.

So, the best one can find on low cost commodity USB boards are NDK SDA and now the one Chinese company making AS-318-B boards.

Most of the clocks you see on Chinese USB board cost well under $1/ea, and they aren't great for audio use.

All the above having been said, clocks are only one component of a good sounding dac. Other things matter too. Power supplies matter a lot, particularly for ESS and AK4499.

Output stages matter a lot and people are still learning how to design them well.

In addition, dac board layout matters a lot. Using a multi-layer board matters, even for 16-bit dacs.

Getting the best sound a dac chip is capable of requires doing things right in every area. I wouldn't sweat the clock thing too much if you are not going to sweat the other things too. You will get some improvement out of a good clock, but you won't hear all of the benefits of it until everything else is fixed.

On the topic of USB boards and clocks, you have my advice already. If you don't like it then you might want to buy a selection of recommended and cheap USB boards and decide for yourself what you think is best.
 
Last edited:
Burn-in

I'm sorry, your product recommendations are appreciated, but this part is total BS. Maybe your brain needs 3 days of continuous running to settle down? .
I have heard sonic improvements after several hours (days, longer) of burn-in (or break-in). Current-day science and metrology and even psycho-acoustics have not been able (yet) to authenticate (quantify) given "modern" instruments and standardized criteria (what's in texbooks, white papers, etc).
Nevertheless, there are far too many audiophiles and audio enthusiasts (that are not paid-for writers and pro reviewers) that claim/report similar burn-in-phenomena effects to dismiss them as "total BS", irrelevant or (by extension) not statistically significant.
 
One more thing about USB boards, it is good to look for one that has ASIO drivers available if the USB board will be used with Windows OS. ASIO allows for Native DSD and makes it possible to bypass poor quality real-time resampling by Windows Sound Engine (but only if audio device are configured so that the ASIO device is not a Windows default sound device nor default communication device). :)
 
Too many boards to choose from!

About which board I'll ultimately stick with. Well, I bought that $37 unit (1 crystal and 2 MEMES oscillators). But the $59 Chinese unit you suggested may be an option in the near future ;)

About ASIO ... I use Linux and Windows and Linux devices and Linux is such a huge jump up in sonics that I don't take any Windows device seriously anymore.
 
USB-to-I2S design variety

On the topic of USB boards and clocks, you have my advice already. If you don't like it then you might want to buy a selection of recommended and cheap USB boards and decide for yourself what you think is best.
Well, the HUGE variety of boards -- Chinese or otherwise -- is source of confusion.
I don't get it ....
Why do some boards use a single XMOS lsi (eBay, $45) ...

$_57.JPG


..... while some (like the Ali device you suggested) feature three lsi's?

Hbe8d76270bd741dda337cd2d00bb95536.jpg


Too much variety!

How much PROCESSING power -- via ASICs, FPGA's, etc -- does conversion "require"?
Is there a single design guide for USB-to-I2s boards? Maybe one of those datasheet "suggested" designs found in the back pages of the manuf-model PDF?
 
There are two basic version of USB board chipsets. XMOS is a big company that makes parallel processor chips, a subset of which are the most common for USB. The two chip boards you have found are generally based on the Amanero design that came out of Italy (CPLD and MCU).
There have been other players trying to get into the game, some still have problems with USB stacks of questionable reliability. There are also a few old dac chips that had built-in USB support, but not with ASIO and DSD support.

Modern USB boards run in asynchronous USB mode. That is, they are clocked asynchronously with the computer clock, and synchronous with the dac's clocks. That is the best way to go to minimize jitter. Also, very high sample rates in use today and protocols such as Native DSD can require a fair amount of processing on the USB end. Also for an OS like Windows, very sophisticated USB drivers at the PC end. Thesycon makes most if not all of the USB drivers for XMOS-based USB boards. Board manufactures pretty much have to license a driver if ASIO or any protocol or in/out channel support above basic stereo PCM playback is wanted for use with Windows. Its not that someone else couldn't write a driver stack, its that Thesycon are about the only ones who have figured out how to make it work reliably, given the complexity of Windows.
 
Last edited: