Ultimate USB to I2S interface

This thread is a result of non succeeded GB on M2TECH OEM I2S interface, that showed a great demand for a product like this. It’s purpose is to came up with good alternative and create new GB in the end.

The paramount things in product design we need:

- Asynchronous USB (device as clock master)
- Two separate low-jitter VCXO for 44.1k (22.579) and 48k (24.576) sampling derivation.
- Working bit-perfect (ks) drivers for Windows (XP,Vista,7).
- Good documentation/support from manufacturer
- 192 kHz 24 bit I2S output
- Pricing range ~ 50Eu (75$)

Preferable things we need:

- As currently no technical solution for 480 Mbit/s USB2.0 isolation exists (only upcoming isolators from AD?), I2S isolation (optical with high speed cmos optocouplers like HCPL-7723 or digital/inductive as ADuM3440 or better ISO7240 – 1ns eye pattern skew jitter declared)
- Very customized drivers that are managing to stay within 150Mbit/s USB limit, and isolating at USB side (it is doable, at least in theory)
- No onboard PSU, connection for external
- Slaving to external clock (clocks) on “clean” side via isolation (if that is reasonable, as any isolation will introduce it’s own jitter)
- Drivers support for Mac, Linux.
- I2S modes selection

Current candidates:

Musiland 01 MINI
+ Price ~60USD
+ Good manufacturer updates and support, HDCD decoding
+ Well documented mods
- Only one clock

XMOS USB Audio 2.0 Reference Design Kit
+ All paramount objectives satisfied
- Price

+ Well documented mods
+ All paramount objectives satisfied
- Price

XtremeUSB XD1014 EVB
- Only one clock ?

DIYAudio own design using XMOS XS1-L1 128TQFP
+ Fully open source project
+ Just 8$ for bare chip, all other money goes to PCB and top notch parts
+ Already available reference design that works
+ Onboard shunt supply’s could be added
+ Possibility to interface with any dac of your chose, without any glue logic
+ Opportunity to make things right, once and for all
+ Etc…..
- Time?

Let’s discuss our options here.
Wireless USB 2.0 480Mbit/s exists.

Yes, I'm aware of that. 480Mbit/s is on paper only and far from reality. IME using both wireless ends 802.11n compatible, will yield 130-150 Mbps at best in real world environment using 5 GHz band. Maybe it will be enough ? We still don't know real bandwidth demand for these products.

I agree, it probably would be better choice in terms of GND cleanness, then PC.
IMHO, removing the USB connection to add a (relatively) powerful RF digital RTX isn't such a great idea.

If you're looking for alternatives, ultimate isolation could be done using some optical link (e.g. Gigabit Ethernet on fiber) instead of USB. Unfortunately, that would require some supporting hardware which in the end will likely look a lot like another computer. So the noise we're trying to get out of the door will get back in by the window, so to speak... :)

Isolation on I2S (with clock on the "clean" side) should be just fine. And we can get away with standard technology (USB 2.0 async audio) that in the long run will be natively supported by all OSs, so our device will eventually become easily "plug & play".
Also "normal" 100Mb ethernet on TP is galvanically isolated with transformer

a solution could be to use the XC-2 Ethernet Kit from XMOS coming with example software available as source code for:
- 10/100 Ethernet MAC/MII driver code
- TCP/IP and UDP stack
then adding some i2s ports using this should be a simple task.
With relatively little effort we could have an interface that take a stream of n channel PCM audio via ethernet and outputs it as i2s ports.
The cons are that in this way we have something non standard (as an USB audio class) but maybe this could be overcome by implementing a standard streaming protocol like UPNP for which there are plenty of HW and free SW implementation of the server, controller and renderer components.

I think we should stay open minded here, so all alternatives are welcome.
Biggest drawback of ethernet based solution is functionality. If someone would came up with a wise software , that would allow this device to function as external sound card (thus supporting all players etc.) it would be perfect.

Some thoughts about optical isolation. There are more optical buses then just ethernet. One such an example could be MOST, that is used by AUDI MMI and other car manufacturers for their hi-fi audio-video systems. I have done some repairs to MOST based amplifers/MMI's and can tell that this interface is not another pc :) Basically only optical converter and programmable MOST interface chip. Obviously not a lot of most familiar programmers around here, but such options exists.

Btw, audiodesign posted XD1014EVB pinout, reposting it here. No more technical details is available for the moment as I understand.


  • I2S Signal Description.JPG
    I2S Signal Description.JPG
    67.1 KB · Views: 5,499
I personally do not think isolation is so important.

As long as one uses completely separated power supplies for each chip (e.g. batteries for laptop, uP, clock, DAC, .....) and the ground is only connected from the preceeding stage to the following stage at a single point, and none or at most one of them to Earth, then ground noise is IMHO not an issue.

And for the solution we are looking at, this is entirely feasible. Only when you go for tube stage for an AD1865 would you really need a Tx based PSU, and even then the rest can still be battery powered.

Had a quick overview of xmos forum. Source code for this I2S reference design is available only by signing license of NDA. Had not seen any additional price mentioning. This should be considered from legal point of view. IMHO redistributing compiled binary files for non profit purposes, based on modified source should not be considered as violation of agreement, but this aspect should be not ignored before making any decision.