Some have derided my observation that higher sample rates do improve quality.

So if higher sample rates do not matter then why would a vendor produce this new DAC?

Phasure NOS1 24/768 async USB :

Asynchronous USB with maximum input of 32 bit 768KHz.
Output : 24/768 max. All further sample rates supported.

So feel free to tell me why they andor I are idiots for believing that higher sample rates are better.




should this be in the new gear listing too ?
more info at
Phasure NOS1 24/768 async USB DAC

Seems to be a DAC aimed at audiophools rather than professional recording engineers ...

Reminds me of how car stereos can be rated in thousands of watts - more has to be better right? Conveniently not mentioning that peak watts have nothing to do with RMS watts, and in real terms they are probably 100W noisy POS.

The point is - this is a DAC. A DAC will play back whatever sample depth and rate you give it, assuming it's capable of handling those numbers. All they are saying is it, theoretically, can handle any conceivable sample rate.

Show me a 32 bit 768 kHz ADC and i'll be more excited ...

And I don't know any audio engineer who would dispute that higher sample rates sound better ... just whether it is necessary and/or practical given the state of computer technology right now ...

I don't understand this part of your post, given that even as much as 10 years ago a new computer could record at 96kHz with no problems.

Are you implying modern DAW's have a hard time tracking at 96kHz?

okay
where do they get 32/768 to play from if there is no a/d at those rates

they also say they do this (real soon now)
8 channel 24/384 operation.

who needs 8 channel output?
sounds to me like it goes both ways
or this whole thing is really stupid
but people are buying them as fast as they are made

maybe this is all hype for goldeneared stereophile moneybags

8 channels is for 7.1 surround. FWIW.

But, there are plenty of people HERE who will dispute it!

operative word engineer

there are plenty of slutz here who claim that you can get perfect results with just 2x+epsilon sampling rates.

and they arent shy about telling me that i am an ignorant slut for claiming that more resolution gives better d/a results

but then they ignore nyquist in the digital domain , merrily do all sorts of non linear processing, and ship infinite bandwith signals back through the d/a and wonder why the aliasing and foldback gives them peaks.

i guess

but does anybody actually use 7.1?
and where do they get 768 signals to use?

i guess that having a spare channel in case one dies is a plus

yes, there are plenty of people here who would dispute that more samples is better. and they do! very vigorously as if their religion had been insulted.

if they mentioned clock jitter they would be partly right. but they claim that 2x sampling is enough for perfect recovery via d/a.

Forgive my ignorance, but are you referring to what they call "Loopback"?

If not, would you please explain what you mean by this.

Thanks.

They don't have 768 signals. It's just to avoid having to rebuy a converter in the future...it's making it "future proof"...as long as the future is PCM, that is.

dont know what they call loopback

i am referring to doing a good a/d
then
they screw up teh digital signal with non linear processing
and fail to run it through a low pass digital filter
next they
ship the now infinite bandwidth (after non linear diddling)
into a d/a circuit
which
since the signal is not band limited
will cause aliasing in the analog domain

and the extra signals that fold back into the audio band
can add and cause peaks

Who exactly are you suggesting does this?

Just because something is technically feasible, doesn't mean there are practical, real-world applications for it.

Regarding the value of higher sample rates being disputed: the only somewhat similar argument that I'm aware of is that of - given the choice - a higher bitrate having a bigger impact than a higher samplerate.
More precisely: compared to 44.1kHz/16 bit, 44.1/24 is a bigger improvement than 48/16.

And regarding the device in question: high-end-audiophile- voodoo, methinks.

who?
this?

guessing at your refernces but:

everybody who claims that intersample peaks are a problem.
they are just nyquist in action.
most original samples will be below the analog signal peaks.
so when you recreate the original signal exactly it would appear, to those who confuse digital and analog, that the analog has now "peaked" somehow. nope. the d/a is just doing its job correctly.

worse, these people do not realise that i can design the d/a so that max digital level can be as low or as high as I choose back in the analog domain.
sensible operation would be to follow a low output d/a with an amp that you adjust to get whatever analog max you want.

secondly, most everyone who is trying for louder is doing non linear processing that will cause problems if they do not LP filter the digital before d/a so as to have a valid nyquist band limited digital signal.

who would spend money for something so iffy.
why not wait for 768 signals to happen first?
and for competition to bring the price of these down .
and for newer designs to improve them even more?

the application is there

for sure feasibility is an issue
but most people ignore the real world aspects
clock jitter, sampling resolution accuracy, linearity, yada yada all will limit how high a sample rate makes sense to do

AES some years back showed that higher sample rates are way more important than higher bit depth samples.
that was a subjective experiment, not proven mathematically.

44.1/24 vs 48/16 is a non issue.
the resampling errors from 48 to 44.1 would be bad.
88.1/16 would beat 44/24 AND DID in the AES experiement

oh, those guys

I'm not sure why you have the idea that people who mention intersample peaks or true peak levels are confusing analogue and digital. It seems like you might be confusing the implied meaning of "peak" with "clip". Probably not, but ... ?

Just a bit of pseudo devil's advocate here ... how many years ago? Could you ref the AES paper? How significantly have conversion designs improved since that test?

nope
they draw the original digital samples back on the analog image
and say: there! the analog is higher than the digital
which is total nonsense

of course the original analog signal when REcreated will be higher than the original digital samples were. that is just nyquist in action.

Worse they assume that the d/a is designed to duplicate the levels in the a/d. the d/a circuit is not a mirror of the a/d and can be designed to be eg 60 db higher or lower than the original signal if the designer wanted to do that. now that would be a peak problem if it were designed to be +77dbu peak out for max digital bits for example.

clearly they are confusing the digital and analog domains if they claim there are intersample peaks that are a problem because that is just nyquist working correctly.

i am not confusing peak or clip.
they merrily clip, compress, limit, and destroy the original valid samples with all sorts of non linear processing.
then they can and often do get peaks in the recreated analog signal because it is no longer nyquist compliant and the foldbacks/aliasing can add to create true peaks.

no valid nyquist signal that is run back through d/a has any peaks at all. it is just the original signal which was sampled and the samples have to be below the analog peaks in almost all cases of real signals.

i am guessing it was 8+ years ago.
and i do not have the paper.
i found it when googling on teh subject a couple of years ago.

Who screws up the digital signal with non-linear processing to make it infinite bandwidth? Plug-ins operate at a designated sample rate which follows the rate of the session you are working on. Yes they use internal oversampling, but the output is at the rate that you are running the converter. There are no "infinite bandwidth" audio signals in the box - if there were, you would have infinite CPU load problems. Besides, all digital signals are processed by the anti-aliasing filter on the way out to analogue, so you are never going to have a problem.

If you are going to sample at ever higher rates, like those in the OP, you need to follow that with higher resolution in order to make the most of it. The advantages come in the internal processing of the recorded sound, rather than the quality of the raw audio itself.

Intersample peaks have nothing to do with supposed infinite bandwidth digital signals.

The result of the higher voltages denoted in those phantom peaks is that it eats in to the headroom of the analogue part of the d/a circuit. If the designer of the D/A did not leave enough headroom and there is excessive "intersample peaks", analogue clipping will occur. This has nothing to do with the output level, but how much headroom the designer left in the solid state components.

Well, I don't design DACs, but I'm guessing by internally upsampling, or for use in the future. (which could obviously present more potential issues, depending on the implementation)

upsampling will not improve anything
bad upsampling is worse than not doing it

thank you
I have asserted that the only problem is with a bad d/a design.
the OPs claim that intersample peaks are an inherent problem.

compression and clipping and other fx are non linear and will create harmonics with infinite bandwidth. not all plugs or daws will filter them out.

you are confused
bandwidth has nothing to do with cpu use
only the number of bits you are crunching
and infinite bandwidth does not make more bits
just a different arrangement of their values

there are no phantom peaks
they do not eat up any headroom
you are contradicting your earlier statements

I would suggest that gear like this is made for the same reason that car companies make cars that are faster than people will drive them, stereos that go louder than people will listen to them, houses with more space than people need to live in, and any other number of over-built, "newest/biggest/bestest" -- and typically most expensive -- thing.

Hell, audiophile companies produce stones that you strategically place around your living room that are claimed to vastly improve the acoustics. Would you really put it past them to build a uselessly over-designed D to A?

Please show me an arrangement of bits that includes infinite bandwidth. There is no such thing, neither in the physical world nor in software. There are any number of bad things a plugin may do to the sampled signal, but injecting infinite bandwidth is not one of them.

Besides, your converters have low-pass anti-aliasing filters that should pretty much stop any frequencies that approach infinity.

Non-bandlimited theoretical digital square wave for example ... instantaneous jump from high to low would require infinite bandwidth.

A lot of folks don't understand digital.

a popular error:

-"my interface sounds soo much better at samplerate "X" versus samplerate "Y" - EVERY OTHER INTERFACE has to be like that.

I'm not sure that anyone has said higher sample rates aren't better, or can provide a noticeable difference. I think it's just been debated how much difference that makes. Is it the "night and day" that some claim, or is it just a small improvement, that could be offset by the workflow implications?

At any rate, the above listing is pure marketing BS. Human hearing dynamic range doesn't reach 144dB (the dynamic range of a perfect 24bit converter). To put it another way, if you turned something up until your ears were bleeding at full scale deflection, you wouldn't be able to hear the quantisation noise in the reverb tails at the end of the song. Since a perfect 24bit converter doesn't exist, we're looking more at 21, 22bit conversion in practice - which is still more than good enough.

This advert is aimed at people who think they do, but don't actually understand, digital audio.

FWIW I've seen you go on about "intersample peaks" before, I'm not truly convinced you know what you're talking about or why you think it's relevant. Intersample peaks happen because there's no guarantee that a particular sampling interval will be at the peak of a waveform. Given the sampling frequencies used (from 44.1 upwards), it's enough to make sure you never limit above -0.1dBFs, and you won't have any issues. Nothing to do with a ridiculously specced, BS hyped converter - the only thing which it has going for it is the ability to separate a fool from his money.

No money left for popcorn, gotta go...

768KHz sample rate, that'll take a snap shot of 384KHz. What note is that? Does it sound to my ears as infrared looks to my eyes? I don't see it... but it is there...

That's an interesting perspective I hadn't thought of.

Like a "new moon", you can't see the moon when it starts a new cycle, it's shadowed out. Yet it's still there as a gravitational influence. Still an astronomic co-controller of tides and currents (no small feat).

The effect of the moon is measurable, calculable, predictable, visibile and provable - and probably lots of other words ending in "ibble" (random Red Dwarf quote there).

I'm not sure if that's the same for supersonic frequencies. The 96k argument is valid because it allows for less steep aliasing filters, which could quite easily have an audible effect below 20k. Unlikely for there to be a difference once the slope of the filter is out of the way above 20k. Still, if anyone wants to take a blind test on this...I'm all ears.