How do DDCs improve audio? (Denafrips Hermes, Pi2AES, Schiit Eitr, ifi Igalvanic)

Comments · 575

• KI

  @kingjongun7295 · 5 years ago · pinned

  Respect my avatar

  121

• XE

  @xentiment6581 · 3 years ago

  We need a hidden camera inside Schiit HQ when they are naming their products. They know.

  8

• DA

  @damirskrjanec · 1 year ago

  Every half-decent DAC today has PLL based clock, which filers out all time domain errors.

  5

• JE

  @jensastrup1940 · 5 years ago

  I’m so glad I have found this channel. Looking forward to the Hollow May review.

  7

• OR

  @oryandunn · 5 years ago

  I thought by now any reasonable DAC would have it's own internal elastic buffer and just internally re-clock using it's own precision clocks.  Paul Lesso wrote an AES paper about S/PDIF design 15 years ago called "A High Performance S/PDIF Receiver".  A small buffer chip should be super cheap these days, and should be able to wipe out any jitter from a PC optical port.  I get the impression audiophiles like to make simple problems hard.

  160

• TH

  @TheVoicefrombelow · 5 years ago

  It sounds like some dacs would benefit a lot from a Pi2AES, where as others (the more expensive one's presumably) would benefit less if at all. I'd love to hear your recommendations for a good bang for buck combo with the Pi2AES if you end up doing a dedicated video on that device.

  6

• TE

  @teighan7829 · 4 years ago

  Love the little fun fact about the creator of USB audio thinking it sounds like doo-doo haha

  4

• JE

  @jeffhampton6972 · 4 years ago

  This is super helpful. Also, I just bought a May because of you, gosh darnit. So thanks.

  2

• ES

  @eskamobob8662 · 5 years ago

  I would love to see a video just flat out on time domain vs frequency domain and what each is used to look at

  8

• JA

  @jafizzle95 · 5 years ago

  I appreciate the actual graphs that prove what you're saying. I'm new to high-end audio, and for the most part I assume that the things that people tell me, like "this is better than this because this one costs 10x as much and it makes me feel good" is mostly subjective, but you've supplied receipts which actually puts weight behind your words.

  7

• DA

  @DarkMatterPartical · 5 years ago (edited)

  Awesome channel by the way finally someone who can measure and has an open eye for the subjective part of this hobby.

  4

• -S

  @-szega · 5 years ago (edited)

  <a href="https://www.youtube.com/watch?v=Jvu_doQfAI0&amp;t=400">6:40</a> for what it&apos;s worth an OCXO is still voltage-tuned (EFC) with a typical pulling range of something like 1-2e-7. The ovenized part is for short-term stability, not frequency adjustment. Essentially, quartz crystals have a characteristic curve of frequency change vs temperature change over temperature, which is very different for different &quot;cuts&quot; (the literal way the crystal is cut and polished). The standard is the AT cut, which is pretty much all quartz crystals you see everywhere. The AT cut crosses 0 Delta-Hz/Delta-K at around 25 °C. Which is just about perfect for normal ambient room temperatures. But the slope around that point is steep (relatively speaking, so let&apos;s say you&apos;re at 20 °C, you might be getting 5 ppm/K or so). Now the clever trick about OCXOs is that they use different cuts, often the SC cut. The SC cut has a completely different response to temperature changes; it crosses 0 at around 90 °C or so, and has a very gentle slope around this area. So OCXOs keep their crystal heated to a very <i>stable</i> 85-95 °C, using a control loop. The attenuation of ambient temperature changes by the control loop colludes with the shallow slope of the dHz/dT to give extremely low tempcos.<br><br>The <i>expense</i> of that is: SC cut crystals are much more expensive (tiny tiny tiny fraction of the market), OCXOs are a lot of hassle to build and even fast-stabilizing types will exhibit a pretty wild swing from turning on to becoming mildly stable (at least 15+ minutes), they have very high power consumption compared to normal oscillators (couple Watts steady-state vs. a few mW for a normal oscillator and maybe 50-100 mW for something fancy. (And some OCXOs tend to fail under repeated power cycling, like you&apos;d do in a residential setting -- they&apos;re <i>absolutely not</i> designed for that and <i>will not</i> hold their specs when treated like that).<br><br>Most (all?) OCXOs have their own reference voltage which is typically ovenized as well. Sort of like bootstrapping, the reference has a tempco (selected for ~0 tempco at the elevated operating temperature, another reason these are expensive), so it would impact the temperature controller, so it&apos;s ovenized itself. It&apos;s of course stable, the tempcos of precision references are tiny to begin with.<br><br>Now the reason people went to this trouble is that for, mostly physical experiments, extremely accurate and stable clocks are a useful tool. A good OCXO achieves short-term stability better than 1e-12, that&apos;s one part per billion, or for the typical 10 MHz reference, that&apos;s less than 0.00001 Hz variation over a few seconds. (Jitter and stability are not the same thing). These are mechanically vibrating devices: Vibration creates low-level spurs. Most physical phenomena influence oscillator output, temperature is just a big one. Gravity is another; angling or rotating a quartz changes the frequency (much more than 1e-12). So for optimal performance you&apos;d want to dampen vibrations of all frequencies. Sounds like something you could sell to audiophiles. Big fat granite rock on a bunch of thick steel rope springs.<br><br>Obviously, as you allude to at the end, the oscillator can&apos;t actually be used as a reference for anything (what&apos;s 10 MHz / 44.1 or 48 kHz?), so you&apos;re looking at a fractional-N PLL anyway, and at that point the quality of the output clock is liable to be swamped by the PLL&apos;s qualities (or lack thereof).<br><br>To be frank, a normal quartz is probably just dandy for audio, if the rest of the system design holds up. Audiophility seems to be especially liable to buzzword engineering instead of system-level engineering.

  29