• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Review and Measurements of Schiit Yggdrasil V2 DAC

A 120dB SNR re: 4V = a noise residual of .000004V = -108dBV. There are many in-ear monitors whose sensitivity meaningfully exceeds 120dBSPL/V, some are as high as into the 140s (which is, in my opinion, exceedingly bad design for exactly this reason, but it is what it is). With any such IEM, that noise floor will be audible (most likely as a "hiss", as it will likely be white/Johnson noise).
How will not "summing everything together" help with that? Seems like what's missing is additionally specifying the max output, but that's true regardless if everything is summed together or not.
 
How will not "summing everything together" help with that? Seems like what's missing is additionally specifying the max output, but that's true regardless if everything is summed together or not.
It doesn't help with noise having a vastly lower threshold of audibly than distortion (because it is generally uncorrelated with the signal, and there is no helping that), but it is one step on the path to measures of distortion that are actually predictive of audibility without requiring overshooting the actual required linearity by a factor of 1000.

It should be noted that THD alone is also an exceedingly poor predictor of audibility, which is why measures that account for masking effects and the actual dynamic behavior of audio content are useful. Doing this requires significantly more than just splitting up harmonics, but at bare minimum we can get a better general impression when all distortion is not summed to a single value, and then summed with noise as well.

It should also be noted that max output over noise floor is the measure of dynamic range, what would be beneficial in audio products is an equivalent of EIN, but this is impossible because of the highly varied sensitivity of headphones.
 
the threshold of audibility in quiet is approximately 0dBSPL
Another thing (just a nitpick :) ), AFAIK 0 dBSPL is the threshold of audibility for 1 kHz tone. Is the threshold for broadband noise also 0 dBSPL? I may be wrong but it is probably higher.
 
Another thing (just a nitpick :) ), AFAIK 0 dBSPL is the threshold of audibility for 1 kHz tone. Is the threshold for broadband noise also 0 dBSPL? I may be wrong but it is probably higher.
I actually don't know this off the top of my head (although the 1k level isn't 0dB, it's more like 2-4 per my memory of the ISO226 curves Edit: Just checked 226:2023, the threshold value given is 2.4dBSPL for 1khz in free field), but as usual, Louis Fielder to the rescue.
1726707242762.png

This would indicate that white noise is not that dissimilar to a 1khz sine in terms of threshold, and that even in environments with some environmental noise, it was possible for listeners to discern low level noise.

I will admit, I just took 20uPa/0dBSPL because it's a "probably in the same order of magnitude as the right answer" option, it wouldn't have overly surprised me if the threshold were +/-10dB around that area, but my point would remain here: it's very much possible to get audible noise on IEMs with DACs or amps that beat the SNR of the APX555 (on the order of 124-126dB at most "good" points in its ranging)...but of course that noise audibility has no relationship whatsoever to the audibility of any distortion components, and vice versa.
 
Last edited:
I - obviously - really disagree here. SINAD/THD+n is a measurement of convenience for a past era. There is quite literally no reason to sum together nonlinear distortion and noise, and doing so makes it functionally impossible to establish useful audibility thresholds - there are permutations of distortion where -40dB may be inaudible, and permutations of noise where -120dB may be audible. In the era where this measurement was a notch filter and a voltmeter, it was unavoidable, but nowadays we are performing FFTs of all of our data as par for the course! What is the point of summing everything together?
This post and all your subsequent ones seem totally ignore @tmtomh s point number one that you replied to.

One value of sinad (and only one is needed to negate your denial) is that it provides a pointer to all the other measures. It is vey rare for a device with excellent sinad to measure audibly poorly elsewhere. In his way it is useful for a quick and easy shortlisting: eg as i do “choose from the blue and green sections of the chart “
 
This post and all your subsequent ones seem totally ignore @tmtomh s point number one that you replied to.

One value of sinad (and only one is needed to negate your denial) is that it provides a pointer to all the other measures. It is vey rare for a device with excellent sinad to measure audibly poorly elsewhere. In his way it is useful for a quick and easy shortlisting: eg as i do “choose from the blue and green sections of the chart “
It is exceedingly rare for source equipment to measure with audible nonlinearity, full stop. It has been since before I was born. The edge cases are a few bad-by-design products (and i have to emphasize here, plenty of bad by design products still have almost certainly inaudible distortion), noise audibility, and defects/glitches, but you don't need to look at SINAD to know you won't hear the distortion on an amplifier or DAC, you can functionally assume that simply from their device class.

There are applications where errors in the parts per million range matter, but audio playback is thankfully not one of them - if it were, the gross nonlinearity of headphones and speakers would put those problems to shame.

This, indeed, is basically the problem with summing the noise and distortion terms (leaving aside summed distortion's very poor prediction of audibility for a moment): it moves the "threshold of audibility" line from "almost assuredly not below -60dB if the test stimulus is music" to "-120dB or even lower", because of the vastly larger potential for noise audibility vs nonlinear distortion.

Edit: While this thread is obviously quite old, this all is probably still too much off-topic, we're getting towards a full page here. My apologies to the moderators, I legitimately posted here with the original intent of shutting down a cheap attempt to bait with my article.
 
Last edited:
t is exceedingly rare for source equipment to measure with audible nonlinearity, full stop. It has been since before I was born. The edge cases are a few bad-by-design products (and i have to emphasize here, plenty of bad by design products still have almost certainly inaudible distortion), noise audibility, and defects/glitches, but you don't need to look at SINAD to know you won't hear the distortion on an amplifier or DAC, you can functionally assume that simply from their device class.
This is a point I've tried to make many times. As humans, we just have lousy sensitivity to distortion compared with our high sensitivity to frequency response, level, localization... I have seen countless assertions about "voicing" electronics and product preferences because of harmonic distribution, yet when I've asked for evidence... crickets.
 
This is a point I've tried to make many times. As humans, we just have lousy sensitivity to distortion compared with our high sensitivity to frequency response, level, localization... I have seen countless assertions about "voicing" electronics and product preferences because of harmonic distribution, yet when I've asked for evidence... crickets.
I think that at least some of the misconception there comes from guitar effects - people get used to the idea of distortion as a "musical element", ignoring that 1, it's incredibly large amounts of distortion, and 2, it's being applied to a single instrument, not the meaningfully more spectrally dense full mix.

... and, of course, some of the misconceptions certainly can't be hurt by the fact that they can reinforce sales :p
 
I think that at least some of the misconception there comes from guitar effects - people get used to the idea of distortion as a "musical element", ignoring that 1, it's incredibly large amounts of distortion, and 2, it's being applied to a single instrument, not the meaningfully more spectrally dense full mix.

... and, of course, some of the misconceptions certainly can't be hurt by the fact that they can reinforce sales :p
Of course, distortion thresholds are the most stringent with single tones, not program material.

That said, I do understand (but don't use) SINAD as a proxy for good engineering. Not everyone can read and understand the detailed measurement data. I would agree that as a stand-alone metric, it's not exactly comprehensive, but I will confess to not having a good alternative for (e.g.) someone who just wants to buy an amp for a specific use case.
 
Of course, distortion thresholds are the most stringent with single tones, not program material.
If one wants to be really spiteful, the most absurd thresholds are usually difference frequency intermodulation products, since the main tones can be far enough away to provide no masking at all :D but I take your meaning of course.

That said, I do understand (but don't use) SINAD as a proxy for good engineering. Not everyone can read and understand the detailed measurement data. I would agree that as a stand-alone metric, it's not exactly comprehensive, but I will confess to not having a good alternative for (e.g.) someone who just wants to buy an amp for a specific use case.
My big complaint there is that SINAD/THD+n isn't really a proxy for  good design so much as just low noise design. My default example here is buffered inputs: you can improve your input Z, your CMRR, and pay a noise penalty that has no impact on real world usage... but will look worse in THD+n.

Call me a philistine, but when someone asks me for an amp recommendation, unless the use case is specifically for precision metrology, I pretty much recommend the lowest cost device in the desired power/form factor that I don't have good reason to believe will catastrophically fail in use. I can see merit to having an audio figure of merit, but I'm really not super keen to get back to 20th century amp marketing of "ours is better because we can claim .01% lower distortion and two more dB of dynamic range".
 
Back
Top Bottom