The only thing that’s a bigger and dumber waste of money than ultra precise DAC timing is $30,000 mains cables. A one dollar crystal typically has a jitter on the other of picoseconds. It’s definitely not audible and barely measurable unless you have access to a very sophisticated lab.
This is correct. Unless you have an FFT with a length that perfectly matches the period of the signal (which you don’t most of the time) you’re always going to see spectral leakage and there’s no easy way of telling whether that’s due to phase noise or FFT artifacts.
However, this kind of hammers home the point. The phase errors caused by clock jitter are orders of magnitude below the time between samples that it’s neither measurable, nor audible. Phase errors caused by clock inaccuracies noise is an issue in very high frequency RF applications. Audio is something like six orders or magnitude slower than those applications. The rest is just audiophile marketing…
Last edited:
It would be fun so play around with a track and shift the channels by a small number of samples here and there and see how big an error you need for it to be audible. Is one sample enough? I’m not convinced.
There are VST plugins to do sample delays. It may be possible to automate the delay using a DAW's automation track feature.
However, how would you assess soundstage cue precision? To do it right you need a well-treated room (not just DSP correction), good speakers, and a good enough dac. If the soundstage is already blurred by room treatment and or existing dac issues/limitations then there is already a problem with the test setup (aka experimental apparatus). Professional experimenters have to think of and account for such potential sources of errors.
However, how would you assess soundstage cue precision? To do it right you need a well-treated room (not just DSP correction), good speakers, and a good enough dac. If the soundstage is already blurred by room treatment and or existing dac issues/limitations then there is already a problem with the test setup (aka experimental apparatus). Professional experimenters have to think of and account for such potential sources of errors.
I would disagree. Its not individual samples, its how the dac reconstruction filter tries to connect the dots that are the samples into a continuous audio signal within the Nyquist limit. Noise is always a factor in accuracy of the points. Thus the output stage is left to connect noisy dots as best it can (including for both AN and PN errors). Individual channel phase deviation vs time, and channel differences in phase deviation at reconstructed audio frequencies where the ear is especially sensitive to ITD timing is what we would need to know. Nobody knows exactly how to measure that at this point in time, or if they do they haven't published how to do it.
Last edited:
This is gibberish. The anti-aliasing filter won’t and can’t amplify the miniscule effect a few picoseconds here and there would have. I would love to see the math behind how you think this would work.
Sensitivity to ITD is still the forementioned 10us. Can you show how phase noise can result in such high timing difference? Perhaps you are referring to IPD although IPD is more related to azimuth localization.
I am willing to bet you won’t be able hear the effect of an entire sample of worth of phase shift.
But if this is the hill you want to die on, I’m out. 😉
This is my entire point and you argued it better than I did: The effect that timing errors from a one dollar crystal are so many orders of magnitude lower than all the other errors introduced throughout the chain that they’re all but erased and definitely indistinguishable.
But if this is the hill you want to die on, I’m out. 😉
That's an assumption on your part. You may as well make the same type of argument about LM317 being just as good as LT3042 because there are plenty of other sources of noise in a system. In terms of what is being argued, the main difference between voltage regulator noise and clock jitter noise is that AN is easier to measure in a meaningful way that PN is.
Look, maybe I should just cut to the chase. I know what is possible because I have a diy dac that can accurately do what "standard" dacs can't do. Anyone is welcome to visit here and see for yourself. As far as proof by measurements go, no such measurements exist at this time that I am aware of. Listening tests could in theory be used, but they are hard to do to professional publication standards. Even then, many people will reject studies out of hand without reading them if the results are contrary to the person's preexisting beliefs. This has already happened in other areas of research; its just a part of human nature for people to want to protect their already well-established beliefs. (And I know I am human too.)
Last edited:
Likely it results from modulation by random noise at audio and lower frequencies. 1/f noise grows in amplitude at very low frequencies. Strong modulation by random 1/f noise could be, but is not proven to be, a causal factor in dac phase noise deviation at audio frequencies.
Also, the effects of clock phase noise has been studied extensively in the area of radar insensitivity to small targets with low return energy. Another area where phase noise is known to be a critical factor is for GPS satellite clocks. But no serious study of ultra low phase noise clocking in various audio dac architectures used for human entertainment has been made. All we can do at this point is try SOA clock technology and see what happens. As it turns out what happens is not presently easy to measure. However, nobody has come along yet who can't easily hear what it does for spatial imaging. The real question is not whether its audible or if people would like it if they heard it, the real question is more like whether or not its worth the cost. Probably its not worth the cost to most people; its a lot of money for nothing more than nicer sounding entertainment equipment. OTOH, MP3 or satellite radio is quite satisfactory to many people.
Last edited:
It’s not an assumption. It’s based on research that you need a time shift on the order of milliseconds for it to be audible, while even a crappy clock only has jitter in the picoseconds.
Here’s what we’re going to do: I will prepare a few audio clips with the following errors:
1. A shift of one sample between the channels. That’s still orders of magnitude larger error than even a basic clock would cause.
2. Random inserted or removed samples a few thousand samples apart. Inserted samples will be an interpolation of their neighbors to avoid clicks and pops.
I will create three tracks of each kind, along with three reference tracks with no errors. If you can successfully tell all of them apart I will send you $100. If not, I get to say “told you so”. Up for some fun?
Nope. First we have to agree how how channel sample timing will be modulated individually and between channels. I would suggest to model the deviation for each individual channel as modulated by random 1/f noise in the 1Hz to 10Hz band. Stepped modulation intensities should be produced with the most intensely modulated variations being audible to anyone on any dac. Then see at what modulation intensity the effect becomes inaudible to an average person on a "standard" dac and on an ultra-low phase noise dac in a treated room, possibly with the option to try headphones too.
In addition, the music sample used must be recorded very simply with very high spatial resolution as heard on a very good dac in a very good room. That's to qualify the music as appropriate to see if there is any effect on, or degradation of perceived localization precision.
Also, hi res and dithered-down-to 16-bit music should be tried. For 16-bits, a suitable adaptive noise-shaped dither should be used.
At least that would come a little closer to a proper testing procedure.
In addition, the music sample used must be recorded very simply with very high spatial resolution as heard on a very good dac in a very good room. That's to qualify the music as appropriate to see if there is any effect on, or degradation of perceived localization precision.
Also, hi res and dithered-down-to 16-bit music should be tried. For 16-bits, a suitable adaptive noise-shaped dither should be used.
At least that would come a little closer to a proper testing procedure.
Last edited:
Radar is in the range of MHz to GHz and you’re dealing with the speed of light, so even very small errors in time can cause huge errors in distance. We’re talking tens of kHz here, so 3-6 orders if magnitude less.
Same thing with GPS. We’re talking signals from space that in an application where relativistic effects need to be taken into account.
If you fail to understand the vast difference between these applications and audio, I doubt that we can have a very fruitful discussion on this topic.
I’m going for a bike ride instead.
You obviously haven't read a book on phase noise is radars. Its about target masking by close-in phase noise skirts. Atmospheric EM wave propagation issues are separate factors.
You have to listen to your components one by one , then find which sounds balanced or which type of resonance it has, every thing resonates at one or more frequencies, try to place a big rubber on the dac chip or place wood blocks under the gears and turntable
Use a dac which has a low resonance , the pcm53-k has a palpability and the lowest resonance of all Dac it will have the subjective effect of the equalizer without frequency manipulation,
Build a wood diffuser for ceiling , walls then bass trap, place huge shelves full of books, double carpets and an acoustic screen behind the loudspeakers!!!
Use a dac which has a low resonance , the pcm53-k has a palpability and the lowest resonance of all Dac it will have the subjective effect of the equalizer without frequency manipulation,
Build a wood diffuser for ceiling , walls then bass trap, place huge shelves full of books, double carpets and an acoustic screen behind the loudspeakers!!!
Lets assume that is the case, it will still not be solved by playing on a 'better than 16 bit' dac.
Your statement was:
"People are now hearing what low level signals are actually encoded on a good CD recording for the first time"
Pure speculation. Sound localization in human hearing has been studied extensively and nothing in those studies supports your speculation. It is also good to realize that just by moving your head slightly will have a much larger impact on ITD or ILD. E.g. moving your head forward by about an inch will increase level by 0.1dB which is considered requirement for proper level matching in controlled listening tests.
Which reminds me: you still haven't shown the designs for the jig you use to keep your head steady when you listen to these alleged sound localization cues.
What I said is correct for certain "better" dacs of any bit depth (many dacs such as ESS and AKM are internally 5-bits or so, and oversampled; my dac is only 1-bit and oversampled). Improved low-level-signal reproduction accuracy comes about for the same reasons that localization cues become more precise. It appears to be because of reduced correlated noise, particularly PN. Such noise can be seen in noise skirts at the base high res FFT spectral lines. They apparently have a masking/smooth-blurring effect on low level details (which was also an observed effect seen here where I am with audio-signal-correlated resistor-current-noise). It may help to consider what dac AN and PN noise skirt components consist of as viewed in the time domain (which includes correlated noise-intermodulation products). We could talk more about that if it would be helpful.
-----------------------------------------------------------------------------------
Regarding a different issue, that of head movement:
"Moving the head while a sound is playing improves its localization in human listeners, in children and adults, with or without hearing problems."
https://pmc.ncbi.nlm.nih.gov/articles/PMC9609159/pdf/fnhum-16-1026056.pdf
Last edited: