To Be Determined

Streaming Digital Audio Via the I²S Connection

Issue 141

I’ve talked a lot in recent articles about ditching the digital disc playback model and moving towards a strictly streaming-based setup for all my digital playback. I haven’t gotten rid of any of my discs (CD, SACD, DVD-Audio, Blu-ray), but I have ripped them all to either lossless FLAC (CD/DVD/BD) or DSF (SACD) files – which means that I’ve eliminated the need for any suboptimal playback from my disc player’s internal DAC. I simply stream them now via a USB connection from my music player/streamer directly to my DAC, using Roon as my library organizer, and the sound quality is outstanding, to say the least.

But a recent, out-of-the-blue occurrence had me rethinking my setup, when I was contacted by an artist rep with an offer to review an upcoming new album that was being released in a multitude of high-resolution formats. No physical media was yet currently available, but I was given a variety of download formats to choose from, and I chose the DXD (Digital Extreme Definition) 352.8 kHz, 32-bit PCM digital file. Which was a monstrous download (7.5 gigabytes!) that took freaking forever – and of course, this would be my very first experience with DXD files. I never gave the first thought as to whether my system would (or could) provide playback for the files.

Of course, the answer was “no,” found out immediately when I contacted PS Audio to confirm my Stellar GainCell DAC’s capabilities via the USB interface. It was limited to 24-bit via USB; the USB connection also restricted my DSD playback to DOP (DSD-over-PCM). The DOP part didn’t bother me too much, as it’s very commonplace. That is, until Dalibor Kasac, my contact at my music player/streamer manufacturer Euphony Audio informed me that they really didn’t care for DOP, and that any file conversion had losses. Native DSD was always preferable to DOP. On the positive side, he reassured me that the Euphony equipment was perfectly capable of handling the 32-bit DXD files. He also told me they were working with another European company on an outboard USB-to-I²S interface to add I²S functionality to the Euphony Summus and Endpoint equipment, but that it was still in the developmental stage and somewhere down the road.

The saving grace for me in the whole album review bit was that the release date wasn’t until mid-August, so I had some time to regroup and figure things out. I started digging around on the internet, and found several different Chinese-made USB-to-I²S converters available, but some of them looked really flimsy at best. I ended up settling on this model, the Douk Audio U2 Pro USB Digital Interface; they apparently rebadge their products for a number of Chinese resellers, and I’ve seen this same product under about at least a half-dozen different brand names online (and at a wide range of price points!). Douk Audio calls the device a USB Digital interface, and that is technically correct. There is no digital conversion going on, the device simply extracts and re-clocks the digital signal from the USB input, then presents the signal to your choice of I²S, coax, or toslink digital outputs. The I²S output  is essentially an HDMI connector, though there are varying viewpoints on the oversimplification of the differences between I²S and HDMI. HDMI seems to not really have anything to do with I²S, it just offers a really useful connection interface for the signal.


The Douk Audio U2 Pro isn't a converter, it's an interface that simply re-clocks the digital signal.

The Douk Audio U2 Pro isn’t a converter, it’s an interface that simply re-clocks the digital signal.


I²S apparently has been around for decades; I’ve included a link to a really informative video from PS Audio’s own Paul McGowan that talks at length about the I²S interface and why PS Audio has chosen it for all their DACs and transports. Paul’s explanation really helps to demystify the whole I²S connection thing.


The Douk Audio Digital Interface is actually a very useful device; it extracts and re-clocks the digital signal using a pretty impressive selection of high-end parts that you wouldn’t expect in a product that retails for just over $50 USD. And you have a choice of outputs available; you can use the interface for not only I²S, but also for coaxial and optical digital connections. All outputs are simultaneously active, so you can feed several different digital devices at the same time with no problems. The I²S output delivers a signal that is capable of PCM replay up to 384 kHz and 32-bit, and offers DSD playback natively in both DSD 64 and DSD 128. In my internet searches, I frequently came across an Italian-made device for this kind of interface that was more than double the price, but had frequent complaints from users. The Douk Audio unit actually had information on the site concerning the Italian device, and it appears to me that they probably reverse-engineered it to create their own. There’s also verbiage explaining how they improved upon the Italian model with better parts selection and implementation. For $56 and free shipping, I didn’t hesitate to pull the trigger; at the very worst, I’d just box it back up and return it to Amazon; no big deal, right?



It actually turns out that the Douk Audio device is a very big deal; upon its arrival, I unboxed it and marveled that it actually appeared to be very well made. However, expecting nothing but trouble right out of the gate, I took it downstairs anyway and inserted it into my system. The I²S interface requires power from either a 5V wall wart power supply (not included), or it can draw power from a USB connection (I wouldn’t recommend that setup). I have several linear power supplies in my system, and one of them, a Keces P8, has an available 5V USB power port, which has worked perfectly for my needs.

I had recently contacted Stephen Mejias at AudioQuest to ask him about HDMI cables and how they related to the I²S connection in various digital devices. While helping me wade through the technical details, he also essentially told me that virtually every HDMI cable made would work with I²S, but of course, you want to get a decent-quality HDMI cable. Stephen very generously sent me a couple of the latest AudioQuest 48G HDMI cables, a Cinnamon and a Carbon; I chose the Carbon because of its higher silver content. I connected everything to the Douk Audio interface, and then to the Stellar GainCell DAC, and to my great surprise, it immediately worked perfectly!


AudioQuest offers a huge selection of well-made cables; I’m currently using the Carbon 48 HDMI in my system.


I’ve read for years that the USB connection is an inferior one, and tends to really muck things up, but in general, it’s a good compromise for digital audio. And there are many recent advances in USB-related technology, and many ancillary devices are available to help remove noise from USB connections. I have USB isolators and a galvanic isolation device in my USB signal chain, but with the Douk Audio interface in place, those devices no longer offer any benefit. And are completely unnecessary; the sound quality without them in place is unaffected by their absence. The Douk Audio interface has actually simplified my playback significantly, by allowing me to eliminate extraneous devices, and by simultaneously eliminating unnecessary signal conversions.

I’ve only experienced one hiccup during the process. When playing through the Douk Audio interface to my Stellar GCD’s I²S input, the GCD doesn’t display any signal information. Again I contacted PS Audio about this, and they told me that it’s a quirk of the GCD; in certain circumstances, it won’t display the correct (or any) bit or sample rate information. Well, that’s unfortunate! As a fail safe, I contacted Euphony Audio in Croatia; my streaming system is network connected, and they can remote into my system at any time to observe playback protocols. Zeljko Vranic at Euphony was able to confirm for me that the throughput from my setup was presenting both 24-to-32 bit PCM and native DSD signals to the Gain Cell DAC, and that they were playing as delivered. Sweet!

PS Audio Stellar GainCell DAC.

PS Audio Stellar GainCell DAC.


So how does it sound? In a word, astonishing! If I had any doubts about the street cred of my PS Audio Stellar GainCell DAC, they’ve been completely eliminated by this experience, and the sound quality of the GCD has been totally legitimized. I’m now experiencing the cleanest, clearest, least-gimmicked and most musical digital sound my system has ever delivered; the addition of a $56 gadget from Amazon and China has been nothing short of transformational. All the DSD files I recently ripped from my SACD collection sound more impressive than ever, and that DXD file I mentioned earlier – it’s easily the best sounding digital file to ever play across my system. It takes the illusion of reality to an entirely new level of believability– it’s really that good. DXD was developed to allow for easy editing of SACDs as the one-bit DSD format wasn’t easily editable. As good as DSD is, DXD is on another level altogether. And believe it or not, my FLAC rips of 16-bit/44.1 kHz CD’s have taken on an improved level of clarity and transparency– via the I²S interface, they sound better than ever.

The I²S interface has its detractors, one of the principals being a guy named Amir who runs the Audio Science Review website. Trust me, you don’t want to get me started about what I consider the pseudo-science employed at ASR. I encourage you to make up your own mind. I²S connections are rapidly becoming the standard, especially in the excellent new crop of affordable DACS from the likes of Topping and Gustard. As far as I’m concerned, I²S is the ultimate connection for digital audio.

Header image courtesy of Wikimedia Commons/wdwd. Other images courtesy of the author and PS Audio.

13 comments on “Streaming Digital Audio Via the I²S Connection”

  1. Your article was pleasantly received at home this morning. Specifically I have had problems gaining info on the problems with the PS Audio gain cell which I have owned since 2020. My major problem was that the sample rate was frozen at 44.1 irrespective of the input sample rate. I have spent over one year trying to resolve the sample rate problem and even called ESS the manufacturer of the chip and they stated that their chip works perfectly in other venues. Thus the chip was not the problem in display difficulties.
    Finally I was told by PS Audio that there is no resolve to this issue! recall, the gain cell manual does state that sample rates over 300 are possible but intact the display does not reveal such levels it is difficult to resolve as to whether the unit is only functioning at 44.1 or is it just the display that is limited and the device is actually functioning at higher sample rate.

    It is very disconcerting to accept a design flaw with no resolve. Similar problems with their 300S amp were encountered with the 12 v trigger mechanism.

    I am eager try the unit mentioned above in my system. If my read is correct, it appears that unit you are discussing can be used with all inputs.

    Any further information on the use of this device would be appreciated.


    1. Hi, Larry,

      The Douk Audio Digital Interface is designed to reclock the signal from a USB input and present it to the three outputs, which are I2S, coax, and toslink. My only experience is with the I2S, as it’s the only one of the three choices that would pass native DSD and 32-bit, 352.8 kHz DXD files to the Gain Cell DAC. I’m not completely certain about the capabilities of the coax and toslink in general, but I do know that they won’t work for my intended purpose with the Gain Cell DAC.

      I too am not particularly happy that the GCD will not show the correct sample rate with anything other than a USB input. And unfortunately, I know of no way to verify the input if your DAC doesn’t properly display the specs on the file type your system is playing. In my particular instance, my music server/player/streamer set up from Euphony Audio is network connected. The Euphony Summus shows me the bit and sample rate of what’s playing, and the tech guys at Euphony can remote into my system and confirm that what’s being displayed is what is actually playing in the system. That it’s being presented to the GCD, and is not being downconverted — the packets are delivered from the Euphony equipment and are accepted at the GCD with no processing or conversion.

      I wish I could offer you more information; I’m very happy with the sound I’m getting from the Gain Cell DAC, but I have to be honest — I do wish it had a better, more functional display.

      Thanks for reading!!


  2. Wow Tom; you really put your foot in it. I’m a long time reader and supporter of ASR. As such, I’ve never observed Amir having anything negative to say about I2S. Or have I observed anybody on the forum likewise. Just a modest modicum of smarts says its an ideal method for digital transmission.

    On the other hand, you seem to be deeply into pseudo science. You can reclock, regenerate, isolate, fiber optic until the cows come home and it will make NO difference to the output of an asynchronous DAC. Science says so. And Amir has demonstrated that many times over testing the computer audiophile BS gadgets.

    Furthermore, get a better DAC. The PS Audio GS DAC is one of the worst measuring DACs on ASR. As a mater of fact, any PS Audio product using the gain cell technology measures poorly. While the technology is good, PS Audio’s implementation is very poor. Note, they don’t use gain cell in the BK DAC.

    Also, get rid of the BS Audioquest HDMI cables. Get a nice Blue Jeans cable for $10-$15 for an equivalent result. And stop shilling for PS Audio a major ASR detractor (because all their stuff measures poorly).


    1. To each his own, but I’ve never been one who believes that great measurements always equals great audio. Listening is what your ears hear, not what you read on a spec sheet or a measurement chart. My main thrust here was to supply my DAC with a signal that would allow it to perform to its maximum capabilities, and a USB connection isn’t quite up to snuff for that.

      My bottom line: the PS Audio Gain Cell DAC is a superb piece of kit; it’s an excellent analog preamplifier, and the sound quality of the DAC is off the charts great for its price point. And I don’t “shill” for PS Audio — I have no connection to them, I’m just a satisfied customer for many years.

      Thanks for reading!


      1. it looks like the spirit of Julian Hersch (numbers are all that matter) is alive and well in the comments from Russ. and as far as i know, there is no BK DAC. I think he believes that BHK was involved in the DAC design when it was Ted Smith. Carry on Dude.

  3. Interesting article. I have been using I2S internally in a custom streaming system I have been developing. It is a very logical and well developed standard. The fact that it uses two synchronous clocks instead of deriving the clock from the data means it is more or less “lossless” I put that term in quotes, because even with digital media, there are elements that can corrupt the signal and influence the end result.

    If you think about what digital media is, it is simply a numerical representation of the audio signal sampled in time. According to the Nyquist theorem, if the samples are at least double the highest frequency, the original non-sampled result can be recovered. However, that’s in theory. In my experience there are three primary ways this concept can go wrong. 1) Clock stability; 2) DAC fidelity; 3) Sample rate.

    Sample ratę and DAC fidelity are topics on their own, but for the most part high sample rate, high bit depth DACs are available, so if you use them, those problems are more-or-less gone.

    Clocks are a major issue, however, and I2S is very sensitive to clock issues (PCM is also, but there the clock is embedded in the data). It seems so simple.. Just clock the digital words into the DAC and let it convert those digital words into analog signal equivalents. That’s the theory. However, if the clock is slightly off in time, then the samples will be slightly off in time, and this will show through in the analog signal. The issue is that even the best clocks are not perfect in this regard. They all exhibit what is known as “phase noise”, which is a measure of how accurately they keep the beat. More importantly, most processors (including those generating USB streams!) don’t necessarily operate with internal clocks that are some multiple of 96 kHz (e.g. 24.576 or 49.152 MHz).

    We discovered this issue when adapting a Raspberry Pi to serve as the core of a NAS streaming system. The issue here (and it is an issue with any non-dedicated streaming computer), is that the fundamental clock running the processor is a) not necessarily synchronous with the sample clock, and b) is not designed to have super low phase noise. In the case of the Pi, the internal clock is something like 19 MHz, so, while it spits out audio in I2S format, it generates the BCK and LRCK clocks by dividing the processor clock. The fundamentals for 48, 96, 192 and 384 kHz are multiples of 24.576 MHz, so to get these signals from a 19 MHz clock they have to dither the divider ratios. To understand this, consider trying to get a 96 kHz LRCK and a corresponding 6.144 MHz BCK from a 19 MHz clock. You need to divide the 19 MHz clock by 3.09. But digital clock dividers only divide by multiples of 2. You can sort of finesse this by dividing it by one multiple of 2 and then another, varying the number of times you divide by one and then the other so that “on average” the result is the right frequency (take 38 divided by 8 plus 38 divided by 4 and average them. You get 7.125. Do the divide by 8 a few times (about 6/5ths) more often than the divide by 8 and you can get 6.144…more or less). This is known by the odd term “Fractional N” frequency division. The problem is that it introduces a lot of jitter (phase noise) in the resulting clock, and that means the samples are not really being sampled at the right rate.

    In the case of the Pi, it meant that every now and then it would also drop a bit. For MP3 files or other low res media, this doesn’t really have a big impact. For high quality systems is is glaringly and obviously bad. The solution is to clock the digital data into a “first-in-first-out” memory (called a FIFO) using the processor clock, and then clock it out of the FIFO and into the I2S stream using a really low noise clock at the correct fundamental frequency. Think of this like an assembly line. If you need to have widgets come off the line at precise intervals, but they are being delivered at the end of the line at imprecise intervals, you need to store a bunch them on the shelf as they arrive (whenever they arrive), and then pull them off the shelf at exactly the rate you want them to be delivered. Hardly anyone does this, but, assuming you have a good quality high rate deep bit depth DAC, this is THE signal best thing you can do for the system. Forget $400 digital cables..Their contribution to noise is negligible, and they can do nothing to fix an already noisy clock. Getting the bits off of the data stream at exactly the right rate is the key, and the only way to do this perfectly is to use a FIFO and a super low noise clock at the right frequency.

    1. Wow, that was an impressively technical response! My bottom line: with an I2S input, I’m able to get the highest level of PCM and native DSD playback, that wasn’t possible via USB. And I can’t argue with the amazing improvement in sound quality — that alone speaks volumes to me.

      Thanks for reading!


      1. The really good thing about I2S is that because it uses separate bit (BCK) and word (LRCK) clocks the entire thing is synchronous, so technically, if you have a really low noise master clock, then the data bits will all be clocked into the DAC synchronously, and all will be well (true “bit perfect” reproduction). Unfortunately PCM signal based devices (e.g. TOS and S/PDIF) recover the clock from the data using a phase locked loop, and this introduces a lot of jitter and phase noise. I suspect USB is similar. Many of the TI chips used for rate conversion and source multiplexing do the same thing. These devices have an I2S output, which is substantially better than what you would get from a raw computer (mac, PC or RaspberryPi, which, if used directly, will NOT be bit perfect), but it is still somewhat noisy, so the sample outputs from the DAC will be at very slightly different time intervals, thereby introducing other frequency components not in the original digital source.

        To fully understand this, consider what the DAC is doing. Each digital word it gets generates a voltage level corresponding to the numerical value associated with the digital word. It produces this voltage for one sample period, and then when it gets the next digital word, it changes the voltage level to correspond to the next digital word. So the output of the DAC is actually a series of voltage steps. Each step is assumed to be as long as the sample duration (at 192 KHz this would be 5.208333 microseconds). However, if the word clock (LRCK) jitters the duration of the clock will vary. Imagine if it varied non-randomly at, say 1KHz. This would effectively introduce a low level pulse width modulation on the analog signal coming from the DAC(!!), and depending on the amplitude, this would be audible. In practice the noise is fairly random, so it shows up more as noise. Sadly, however, the closer to the fundamental clock, the higher the noise (it follows a 1/f function, f being the offset from the fundamental, so the greatest noise is in the audio band..grrr)

        So really the only way to avoid this, and obtain the true bit perfect behavior promised by I2S is to buffer the data that comes in via the PCM stream (or from a noisy source like a PC or RaspberryPi), and then use a really clean LRCK to clock it out of the buffer and into the DAC.

        Try buffering and re-clocking the I2S recovered from USB. My guess is that you will see a further improvement.
        Here’s a link to the FIFO project.

        1. Tried to edit this in, but, ironically(!!) ran off the end of the edit clock!

          The FIFO essentially puts the received digital audio data in suspended animation. It gets written into he FIFO is whatever herky jerky way the source delivers it. But once it is in the FIFO it is separated from time (at least briefly). It is just samples sitting in a memory. If you then read it out using a very low noise stable clock, and feed that to the DAC, then the samples are as stable as possible, and there is no new noise introduced.

  4. Wow. Quite a lot comments already. While I’m not an ASR fan by any means. I did see the interview he did with Joe’N’Tell a few months ago on YouTube. Nice Guy but his insists measurements are everything (?).

    As too the clocking woes. Without an oscilloscope or a Logic Analyzer, I wouldn’t know for sure what’s going internally across the entire digital chain.

    My I2S chain is my Sony UBP-X1000ES that feeds a Chinese HDMI Video to HDMI I2S adapter boards (main “black” PCB for PCM playback/a secondary “blue I2S only to HDMI I2S PCB for DSD only) which then goes into my PS Audio DirectStream DAC Sr. The I2S input on the DAC displays DSD 64 Bit when I spin an SACD disc or playback 64fs (2.8MHz) files. With Sunlight, it only played back 256fs for about 5 min. and I’ve never got it to work at the higher DSD rates since (RMA is already in the works – coax input died on it as well).

    I did have the SGCD/Preamp and loved it including it’s PCM playback. I never tested it’s DSD capabilities via I2S but I think it played DSD 64fs over USB (it’s been a while – I don’t remember).

    Excellent article Tom.

  5. With these Chinese gadgets, you never know whether they’re going to work or not in your particular setup. I probably researched this for months before I pulled the trigger on a $56 purchase, and was absolutely certain it would fail right out of the gate. Imagine my surprise when it worked first time, and has continued to work perfectly for about six weeks now without a hitch. The only issue has been, with DXD files, when played via the SGCD, you have to set the filter to “1”, otherwise you get glitchy performance. Everything else, no problems.

    And I agree with your comments about the SGCD — it’s a really great preamp and DAC, and the sound quality is great for the money. And with I2S, it will playback 32 bit PCM and native DSD, both 64 and 128 without problem. I don’t have any DSD files any higher than DSD 128, but I have about 400 DSD 64, and I think the sound quality is pretty magnificent, especially via I2S.

    Thanks for reading!


Leave a Reply

Also From This Issue

Notable Analog Recordings. Part One

I installed a new cartridge on my record player about…

Radios, Radios

I bought a new radio last week. I did not…

Amped Up

Art Dudley’s personal, well-worn copy of The Tube Amp Book,…

Eye of the…Donkey?

This donkey, Mister D, lived next door to where I…
Subscribe to Copper Magazine and never miss an issue.
Stop by for a tour:
Mon-Fri, 8:30am-5pm MST

4865 Sterling Dr.
Boulder, CO 80301

Join the hi-fi family

Stop by for a tour:
4865 Sterling Dr.
Boulder, CO 80301

Join the hi-fi family

linkedin facebook pinterest youtube rss twitter instagram facebook-blank rss-blank linkedin-blank pinterest youtube twitter instagram