Air gaps

January 27, 2019
 by Paul McGowan

The long-awaited PS Audio music server, Octave, is inching closer to making its debut in late 2019. Since I’ve not written much about it recently, I thought you’d enjoy an update about one of its innovations, a new invention called the Air Gap Audio Interface or AGAI.

The AGAI solves a big problem in digital audio servers, contamination by noise and electrical detritus.

Inside a server, you have a noisy computer. As it chugs away at its tasks it jitters and pollutes the output signal feeding the DAC. This is why, we believe, FLAC sounds differently than WAV even though the bits are identical. A FLAC file (for example) requires far more bit crunching to extract the original bits than does a WAV file. Those crunched bits contaminate the final output signal through mutually shared power, ground and physical signal traces.

How can we fix that?

Imagine that the noisy computer inside the server was not in the server box and was instead far away (as we are doing in the upcoming Ted Smith Signature DAC). Its noise and ground contamination would not be a problem as long as we took its distant output and regenerated it in a Digital Lens.

Since our goal is to build a one-box server, the next best thing is to physically isolate the two systems within a single chassis. To do that we need separate power supplies, ground planes, physical boards, and at the end of the proverbial day, a physically isolated connection between the internal computer and the output Digital Lens. That’s where the AGAI comes into play. By bridging the gap between the internal noisy computer and Digital Lens by nothing more than light traveling through air, we get excellent isolation.

(In the upcoming TSS DAC the problem is solved with two chassis: a digital and analog separated too by light using a fiber optic cable between the two.)

We’re all familiar with digital data transmitted through lightwaves using a TOSLINK cable but that won’t work for either of our applications because of TOSLINK’s bandwidth limitations. But that isn’t a show stopper. It just means we have to step up our use of technology. Some of the highest speed data in the world travel on beams of light.

Whichever method is used, AGAI or high-speed fiber, transmission of digital data over lightwaves offers the possibility of getting noise and jitter out of the signal and gets us that much closer to musical perfection.

Subscribe to Paul's Posts

39 comments on “Air gaps”

  1. Paul, is there a new transport design in the works to compliment the PS Reference Digital components, Octave Server -> TSS DAC?

    What is the distance between transmission and receiver of the AGAI and how exactly is that accomplished?

  2. Is your take home message, Paul, that the weakest design part is the DA-conversion based on an analog output module being prone to collect every kind of electrical contamination and noise? Thus why not keeping the signal pure digital in the chain as TacT Audio did it designing a PWM digital power amp (Equibit technology) fed by the digital output of a CD-transport or a music server or a AD-converter. As far as I remember the biggest advantage of digital signal transmission was claimed to be totally immune to any noise!

    1. I am a great fan, in principle, of direct digital amplifiers. Now that some use ADC at the speaker output to provide negative feedback the performance measurements are becoming impressive. There is still a long way to go with them but I believe they are the future for audio.

  3. All this makes sense now to me and what I heard from my tests so far, so for those playing wav/aif the difference between a „normal“ streaming unit and the Octave server might be less than for those playing flac.

    As for many with external storage solutions, an internal storage inside Octave will be redundant…will you offer Octave without an internal storage optionally?

    And will the Octave server chassis match the current DS design or the TSS design?

    While Bitperfect offers a sophisticated software to get over the flac/aif/wav matter (and probably more), it’s just available for Mac. I could also let Jriver do the conversion prior to sending the stream to the Bridge, but I decided to store aif directly instead.

    I understood that on the one hand a Bridge/external PC scenario has the advantage that one can isolate the server SW computing on the PC from the rendering which is still done by the Bridge maybe even more efficiently by using dedicated power circuits and galvanic isolation (which in an Octave scenario are both inside the unit and have to be isolated there), but in a Bridge scenario it’s not possible to isolate the rendering/computing done inside the Bridge from the same HW sending the result to the DAC, which seems more important. The latter can only be optimized there by using uncompressed lossless PCM format like aif/wav.

  4. It’s taking me a long time to decide on a new server. I’ver been at it on and off for at least there days now.

    So if your streamer is doing lots of digital signal processing such as upsampling or applying a digital lens, you need lots of processing power. You need a processor with several cores to do different things. You have to run a decent operating system and possibly also rip discs. Do you run the whole thing on SSD, or do key tasks in RAM, including the music buffer, and store in more reliable SATA drives designed for servers like WD Red? It is a juggling act between electrical noise, power and heat. Whatever it does, you need a really good user front end.

    Both my music supplier and music player partner with Roon for music management. Roon has a minimum i3 specification that allows 8 streams at once with DSP, including upsampling to DSD512, and to do all that requires a very hungry i7 processor.

    There are streamers around that do more or less at different price points. The amount of progress in the last 10 years appears to have been massive. Anything to get lowest possible noise seems to be the aim, although in my decision that is likely to be done by choosing a device with a low power and high efficiency processor (Intel N4200) as there are only 2 of us in the house and my player does the DSP so I don’t need the streamer to do it. I may re-rip to WAV rather than FLAC, for the reasons Paul mentions.

    It does seem to me that it helps to offer a range of streamers because users will have a wide range of processing and storage requirements that have a big impact on price. That is certainly how it works with commercial servers.

  5. I have done the processor experiment before on my computer, which runs the Roon core, as well as JRiver and other server software such as DivX movie server. Even when processing a FLAC or MP3 file to play, the processors (dual Xeon multi core) only run at around 3% of capacity. At what point do you feel the crunched bits as you call them pollute the signal? I’ve tried browsing the web, running Outlook, playing music and ripping a CD all at the same time and mine runs at about 40% and I really don’t hear differences. If I add in trying to re-render a video file that’s a different story, more like 90% or more but I am not sure I hear differences other than I get stuttering from time to time.

    1. I agree, Larry. Unzipping a FLAC or ALAC file is a trivially easy process. But just to be on the safe side I use a “Memory Play” feature in my playback software. If the FLAC or ALAC file is unzipped first, loaded completely into RAM, then sent out to the DAC uncompressed from RAM, there is no mechanism by which FLAC/ALAC can sound worse than WAV/AIFF.

      1. eatapc I would personally agree with your statements regarding memory play and raw data to the DAC but……now we invoke the argument that many have that “bits are not necessarily bits”. Oh boy, we all know that is a can of worms.

        Don’t forget about the type of connection used to connect to the DAC as well, that can distrupt the bits as well for many. I do not count myself in that camp but know many feel strongly about it.

        I do believe power isolation is a good thing. I am still smiling over Bernd’s comment the other day that got him thrown out of a demo about why one couldn’t use a standard power strip and just insert a 1 foot power cord in the mix instead of buying a very expensive multi-meter cord (the person demo’ing the cords made the point that the last one foot was the most important part).

        1. I didn’t mean to imply that “bits is bits” to dismiss audible differences between various pieces of hardware in the digital playback chain. But I do maintain that once the uncompressed data from FLAC/ALAC is loaded into RAM, it cannot possible sound different than a WAV/AIFF file loaded into RAM. The bits in RAM are indisputably the same; there is no debate about that. What happens to the bits after that is another subject and a different debate.

          1. I am on board with your comment. Implementation of the bits in the DAC is certainly a very key component! I have had many a yikes moment listening to various DAC units in the past few years and cannot understand how they can sound so different.

    2. I’ll let you pros laugh at me. Paul got me going digital, finally. I use the DS DAC and the DS memory player. Best digital I’ve ever used. The LP to digital argument is simply how good the record or digital file is, not the record or digital file technology. Done right both are good.

      Here is the laugh. I stream from a 2600X AMD PC using a X470 ASROCK taichi MB. I RIP only to WAV file format stored on a WD 860 GB 7200 RPM HDD. This goes out the LAN Port through Ethernet over power line adapters. Yep, you read that right. Downstairs, I pull that from the wall with another Ethernet over power line adapter into the bridge I I.

      I use minimserver on the PC and bubbleup upnp on an old Motorola cell phone as the controller.

      It all works great. Playing the memory player into the direct stream DAC and switching the line from i2s (memory player) to bridge II (streaming), I can’t say it is different.

      This seems to say the bridge II is pretty darn good. I have the “worst” digital stream with good hardware but it sounds great.

      1. No comment from me here about the Powerline adapter as I have only limited experience with them after they were first introduced. I understand they are greatly improved and use is widespread now. As for signal crosstalk or noise I will let those with more experience than I with this technology weigh in. I still prefer hardwired cable and it has been quite a feat in my home trying to find places to snake wire.

        1. Yes, this darn house was new in 2007, and it has ZERO 5e in the walls. We have coaxial cable, and no cable service available external to hook to!

          Getting from the basement to the server upstairs is impossible after the fact. But, another PC downstairs, maybe. More to keep $$$ up to date.

          I have 30 mbps so not bad. But noise?

          Playing old Moody Blues records today, wow, what a band.

          1. “Impossible after the Fact”? Oh sheesh, dude. Drill some holes in your house : ). But I used to do that for a living at one point.

            And +1 for the Bloody Moos (as one Limey here disparagingly put it recently). Sometimes, I think it is that their veneration for our indigenous music is equalled by ours for theirs. Or at least, our veneration for their interpretation of ours.

            I would maybe never have known about a lot of American Roots music when I did, had it not been for the British Invasions’s love for same.

            1. Badbeef,

              YES, there are tricks to this trade of wiring after the fact.

              The Moody Blue, or anything like it, may never be heard again. The music covers every age group and is a technical marvel. For people who haven’t ever tried playing in a band, I was a poor clarinetist, the raw talent of that band is mind boggling.

          2. At one prior house I managed to find a chase from the basement to the attic for a Cat5 and then it was easy to get them into a room or two from the top wall. In my last house I had to drill a hole in my heat duct in the basement and run it up through the register upstairs. Fairly jury-rigged :-). My current place had been prewired to key rooms, although there are haphazard cables all through the basement ceiling! I sure do like the hard wire.

      2. I used powerline adapters as a quick fix some years ago to give my PVR internet access for catch-up TV. They worked very well indeed. When more recently I came to stream music the obvious first step was to use the powerline Ethernet for that as well, and it seems to give no problems that I can detect. This may be due to cloth ears and coarse sensibilities. I would prefer an Ethernet cable on principle but installing one would be a major undertaking since the internet router is on another floor. Since nothing seems to be broken at the moment I am in no hurry to fix it.

  6. I understand that FLAC files need to be uncompressed for playback meaning more processing power and possible ‘effects’ therefore becoming noticeable to some people in comparison to AIFF/WAV equivalents but I thought the DAC itself did this process (or can do depending on your set-up) At any rate doesn’t this add to the argument of just storing files in the device itself and running them straight into the DAC?

    1. I don’t know of many DACs that can perform the conversion process of FLAC to WAV internally. Ours does but that happens in the Bridge, not the actual DAC.

      There’s no easy solution that’s perfect because all of them involve placing some sort of computer with its power supply and noise somewhere in the chain. The trick is to either separate that or keep it quiet if in the same box.

      1. Hi Paul – thanks, I’m pretty ignorant on this and assumed in fact that all DAC’s that can handle the main file types ‘unsquished’ FLAC/ALAC files as part of their operation.

  7. I know what Paul means, but fiber optics is SLOW. About a 66% VP in a glass medium. Going up the electromagnetic spectrum increases bandwidth, though.

    Look at it like this, fiber is like a line of greyhound buses moving at 66 MPH. Each bus carries 120 people. If we compare that situation to a car with even 5 people, the number of cars have to go mighty fast to move the same number of people per unit of time. Bandwidth is the key, not speed, with fiber.

    If you do not need the bandwidth, like high speed serial, than speed is important.

    The optical to electrical conversion process can be noisy, too, so those boxes need good isolation.

    Paul’s comment on PC and noise is so true. I have a buzz in my right PC speaker sitting on the PC. Pick it up and like magic, the buzz goes away. Looking at the frequencies a PC runs at, it is amazing they work as a PC is the classic lowest bidder product.

    1. I don’t understand what you are saying here. Fiber is a serial medium. There are only two ways to increase bandwidth, faster speed or multiplexed multiple fibers. And since we are talking about real time audio, where’s the speed or bandwidth problem? We can easily move 60fps 4K digital video in real time at 12gbs no problem.

      1. Hi Glimmie,

        Fiber moves data at the SPEED of the medium the light is in, a piece of doped silica glass @ 66% light in a vacuum. The shorter WAVELENGTH gets you the higher bandwidth. Moving from 850 nm to 1330 nm to 1550 nm increases the bandwidth as the wavelength gets shorter and shorter. This would be faster speed. You can WDM (wavelength division multiplex) slightly “around” the low water absorption peak of those specified frequencies, yes. This would be your multiplexing

        Fiber is high speed serial. My example may be too weird but here goes…

        A bus with a BUNCH of tightly packed people (shorter wavelength) moving at 66% the speed of light will still move more “data” than…

        cars (long wavelength is few people in the car) moving physically faster (in foamed dielectric polypropylene at 80% the speed of light).

        More people / data move per unit of time with fiber. Bandwidth is a speed distance value (MHz-Km).

        But, 50-micron multimode fiber offers nearly three times more bandwidth (500 MHz-km) than 62.5-micron fiber (160 MHz-km) at 850 nm. The same wavelength but faster? Yes, this is because the multimode fiber “spreads” the modes as they travel down the glass, messing up the signal, called chromatic (wavelength) dispersion. A given wavelength multi-mode fiber bandwidth has to take into account the fiber core size, too.

        I believe Paul wants to remove the NOISE between end points, not so much go faster as there isn’t the requirement, like you point out. But, sometimes the optical transceivers can be noisy and add as much noise at the ends as the signal can encounter on the trip with copper. Fiber isn’t the end all answer as the conversion electronics are after all, electronics.

        Do you settle for a “known” amount of interference in the box (fiber collects none on the trip), or possibly variable but lower, or higher, interference between the boxes with copper? Don’t know the answer to that one. Each solution can get the bulk of interference in opposite places.

        1. OK I got. I didn’t know you were referring to the wavelength as bandwidth.

          The multiplexing I speak of is using more than one fiber to spread the bandwidth across which could of course entail WDM or DWDM over a single strand.

  8. I’ve been using a one box solution for digital playback for the last decade+ from Mark Porzili (formerly as Melos, Nova Physics, Pipedreams design engineer) now called the Laufer Teknik Memory Player. It’s a direct amp to player (transport-dig/analog) player solution that has served me well over the years. I’m really anxious to learn more about your server and hope it will be ready by RMAF!

  9. I am using a network distribution system based on Rasberry Pi with SPDIF transports. I use the Toslink on one and coaxial on the other. The server is connected over ethernet using OpenSSH server. This obviously an inexpensive solution, which was what I could afford.

    I read that digital music began at about the same time as film gave way to digital sensors. My first Nikon D-1 was almost a dinosaur as soon as Canon offered CMOS technology. Sony (The “S” in SPDIF) developed a newer sensor technology, and at 16 megapixels, the quality equaled or surpassed the film alternative.

    Only retro-seeking Hipsters are rediscovering film, like vinyl’s resurgence in popularity. Professionals may still use a silver based emulsion on View cameras, but many pros have not shot the film in this Century.

    I am now discovering Tubes and Headphones, and I doubt if anyone can compare a solid-state to a tube amp and not appreciate the difference.

    Has “digital” music improved along the same timeline as photography? Many will point to less expensive DACs to improve upon the sound chips in computers, and further down the line, Class-D with ICE edge modules rival Class AB amps with both power and SQ, with advantages of “green” power supplies, and less heat. With those two examples, the answer is “Yes”.

    If a turntable is a film camera, and a Sprout100 is a digital camera, you can have your snap, crackle, and pops, and I will enjoy digital files that don’t include them. If you really want a film negative, remember scratches, dust, and enlarger errors will compromise the quality of your prints.

    We have become used to not having to spot prints, and we can control shadows and highlights that is far beyond the dynamic range of film. Same goes for digital music vs. vinyl.

    This is a hobby of choices–same as photography. I am choosing an Ethernet distribution system, a single board computer, and PCB transport, and the Music going out of Headphones sounds accurate, without distortion or noise.

    Sometimes it is best to recall, KISS, “Keep it Simple, Stupid.”

  10. None of the computer hardware protocols (USB, TOSLINK, etc) know what time it is. That’s a problem with music. Early high resolution attempts with music that involved these transport layers were acoustical failures. As computer circuitry has become cheaper, and audio engineering has developed proprietary methods of dealing with the lack of time in the transport layers, the price has come down. When someone adds time to the hardware layer, many of the issues with music will become more manageable. Until then, processing will remain digitally challenging and the best digital audio will remain expensive because if the need to develop proprietary digital transport layers.

Leave a Reply

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