Exactly. Many of the comments in this thread are pretty entertaining I gotta say. If Roon were doing something in 1.8 that made it sound noticeably better/worse/different than previous versions, something has been dramatically broken. The idea that packets of intact data can sound different depending on how they are transferred etc just shows how much bias and our brains can fool us.
1 Like
Let’s be realistic here.
Comparing DACs at 140dB vs. 150 dB noise level is like upgrading your Ferrari to another model that goes 211 mph instead of 208.
I read one review of the Benchmark AHB2 amplifier which had a noise level about -131 dB, or 21 bits, the best results they had ever seen in an amplifier.
So 140 dB is ten times better than what the best amplifier can do, probably much, much better than what your room and your ears can do.
7 Likes
Same remark as yesterday for Anders. Or perhaps: if you cannot measure it via a standard measurement protocol, then it does not exist ? This is a production engineer’s reasoning. For example, those measurements are they remade substractively in musical regime instead on a pure 1 kHz sine wave at constant level ?
Yep, pretty much. There are thousands of people who lie awake every night trying to dream up new things to measure.
This assertion is wrong.
Just look up J. Atkinson’s measurements in Stereophile while also reading the explanation accompanying them, and you’ll have to admit that the midrange polarity in these designs is electrically and acoustically inverted.
Now, that would be really hard to accomplish by eye and feel.
I agree that’s what the measurements suggest.
OOOOOOOHHHHHHHHHHHHHH
It looks like the ■■■■ emoji! 
v
1 Like
The big advantage of higher sampling frequencies is not that we are able to record higher frequencies that we can’t hear, but to have more samples per second results in more refinement in the time domain. Also the digital filtering and noise shaping is easier than at 16b/44.1khz
3 Likes
For this reason I wrote we do not need more then 96 kHz. I did not say 44/16 is enough. My statement is also, most HiRes files are not HiRes recording.
1 Like
I guess you must be a production engineer rather than a scientist then 
Only 192khz is high enough in the time resolution domain, that a human cannot hear better. Sorry, bad explanation, I try it with an example:
44.1Khz = 1 sample each 2,26 microseconds
96 Khz = 1 sample each 1,04 ms
192 Khz = 1 sample each 0,52 ms
The average listener is able to recognize events as small as ca. 1,2ms
Professional musicians or conductors are even able to go down to 0,8ms
That means that the average listener is fine with 96khz and can’t make out any difference between this and 192khz. But then some trained listeners or musicians will benefit of 192khz.
Therefore producing music in 192khz makes sense, however if we double this again to 384khz the benefit is questionable (until new findings are made).
1 Like
Hi Marin;
You’d have to be more specific and point out the link, maybe we are in fact talking about the same thing just that you did not like my wording of it.
I am pretty sure of this point, talking of a 2 sides (low-cut and high-cut) filter 12 dB/octave one exits at crossover frequency at 2 x 90 degrees from the filtering cells and therefore the midrange receives a 180° signal phase shift to the woofer. To avoid cancellation and a dip in the response over normal overlap, it is common practice to invert the polarity of the midrange so as to get a flatter response.
Now it was a side remark and a warning, as also it is not the unique way to flatten the amplitude response though. By shifting the frequency of the low-cut less low, and the high-cut less high, one increases the overlap, that can be used to bridge the gap. And this strictly applies to 12 dB/octave.
About this being fairly frequent practice, see for example:
https://www.diyaudio.com/forums/multi-way/101489-12db-oct-tweeters-reversed.html
And well, the point is not to discuss crossover theory, the point is just to make sure that in the bass, the phase of the speaker is correctly defined. The rest can be labelled speaker/crossover design. If one can access or see the bass membrane, one don’t need this discussion at all.
My turn to correct others:
1/44.1K =22.7 us, not 2.27
1/96K = 10.4 us not 1.04
1/192K = 5.2 us not 0.52
Then the numbers you cite are from past scientific studies, that are not the most recent ones and these numbers have been revised since with finer protocols and perhaps a different sample of persons (again between our ears we have a trained network and some are better trained than others).
The 2 microseconds interaural limit cited by the MQA paper corresponds to a more recent study.
To have a sampling rate of 2 us you are at 500 kHz. The highest frequency to be reproduced at this sampling has not 1 but 2 samples per wavelength, that is 250 kHz - indeed this looks like extreme oversampling for human audition that hears at most 20 kHz on sine waves (but our hearing is non linear and we are very accurate on transients timing and content). Another example of one simple measurement not summing it all…
Then using an accurate interpolation function, you can recreate the waveform and reconstruct the analog signal shape accurately, enabling to measure a shift smaller than the sampling interval. I guess this is why even 16/44, when perfectly reinterpolated at for example 384 kHz with increased amplitude resolution, can be excellent - better in terms of spatial imaging than coarser, more approximate resampling at lower resolution.
Hope this helps,
CD
2 Likes
Like I said, and they were quicker here than I thought 
So it seems that 384khz is the new king or DSD of course.
The choice of sampling frequency dictates the bandwidth limit for frequencies that can be perfectly reconstructed. If we concern ourselves with the audible frequency range, let’s say up 22kHz then sampling at rates higher than 44kHz won’t allow better reconstruction of the signal up to 22kHz. 44kHz sampling gives all the information required to recreate the bandwidth limited signal perfectly. Selecting a higher sample rate allows you to increase the bandwidth range but doesn’t provide extra accuracy in the original range. Put another way, if I only care about recreating a signal accurately up to 22kHz then higher sample rates make the reconstruction easier but not more accurate. You have all you need for a perfect reconstruction already. Higher sample rates are only required if you can hear frequencies higher than 22kHz, not an easy task.
3 Likes
I completely disagree with that statement.
1 Like
And you need to use the corrected numbers. Only by a factor 10.
There’s a formally proven theory that you’re taking issue with there, that is what sampling theory says regardless. A more precise statement:
- Larger-than-necessary values of fs (smaller values of T), called oversampling, have no effect on the outcome of the reconstruction and have the benefit of leaving room for a transition band in which H(f) is free to take intermediate values.
Put another way, if 44kHz sampling provides all the information needed to reconstruct a 22kHz bandwidth limited signal perfectly, how can oversampling make this better. It’s already perfect.
3 Likes
The issue is to be found primarily in the antialiasing filter effects and in Gibbs phenomenons.
I happen to have just watched this morning a video that Sean (dabassgoesbombom) sent me yesterday.
Video of a talk by Claude Cellier, founder and still chief Engineer at Merging Technologies. Nagra specialist.
Claude Cellier - Trust your ears! The quest for the optimum (digital) audio representation - YouTube
Enjoy! For audio, that one is a true gem needle in the haystack of talks on YouTube…
1 Like
you make me really curious about what loudspeakers you are listening to at home.
1 Like