Originally Posted by narcoman ➡️

Okay - I've absolutely explained this enough and we're going round in circles. Take it easy.....

Originally Posted by psycho_monkey ➡️

Well of course there is. You agree a standard! Same way we all drive on one side of the road in a given country...I can take my session from one studio to another, and provided the converters are lined up the same, my gain staging will be the same.

You really are trolling now aren't you? It can't possibly be that you don't grasp this basic concept.

Originally Posted by narcoman ➡️

Well - for one thing , you can't just design output voltages against reference willy nilly. There are standards to adhere to.....

PROPORTIONALLY, the reference voltage at any given value will be over the nominal value. Read on and see why.....

That isn't what's being discussed!! The only thing that is being discussed is a sample doesn't represent peak code. THAT is all intersample peak means!!

The inter sample peak refers to any reconstruction value over the integral curve being higher than a sample value.

That is all it means. If you can't understand that then I've no idea what to say! It's unbelievable that you'd argue this. What else can anyone say? heh!!


What happens if you SRC a 44.1 set into a 96 set? What happens then?
You got it - some of the reconstructed and NEW samples will be above pairwise 44.1 samples. In other words there will be inter sample peaks with respect to the original 44.1 sample set. Now - the re-coinstruced analogue waveforms will be the same - but that isn't the point.



And nobody has said there are. What they have said is that the peaks in the analogue reconstructed waveform don't necessarily line up with the highest samples in sample code. How could they? We don't join dots!

HOWEVER - the respective sample points do not occur (most of the time) when the voltage is at a locally maximum point.



Once you set a reference - for example

-18dB(FS) = 0dB(VU)

in a 24bit sample set then the number 000111111111111111111111 (actually it won't be exactly that as 6dB is not 1 bit - but it's close enough for getting the point across) is equatable to 0dB(VU). That is core to the idea of calibration.


That is why those of us who do this for a living set standards either locally (on your own) or you obey international convention;

There are several. ITU, AES, EBV, BBC and quite a few more. We set standards and STICK to them year in year out.

Even within hardware there are accepted standards of what 0dB(VU) equates to in terms of voltage! For MOST hardware 0dB(VU) is +4 dBU (not all ).

+4dBU is nominally accepted internationally and across equipment asset at 1.23Vrms.

Now - there are a couple of points to your argument all based around corrupting the central point. It is important when discussing AD and DA that an equivalence is set. However, for ANY digital argument ALL of the points about reference level no matter what value is set as 0dBVU equivalent.

The AES define -18 (or -20 for some use). EBU define -20. The BBC have defined -17, -18 and -20 at various times......

Nobody is confusing analogue and digital because most of us in this discussion are savvy and well versed enough to understand that we all refer to analogue/digital comparisons with a defined reference. The point is - the arguments stand no matter what the reference is; the only thing that changes is the absolute value behind the argument, not the semantics themselves.


Okay - I've absolutely explained this enough and we're going round in circles. Take it easy.....

Originally Posted by dcollins ➡️

The existence of inter-sample peaking is not even remotely controversial, so I have no idea what OAG is on about. It's also trivial to demonstrate in D/A (or async. SRC for that matter).

Also, there are many D/A converter chips that have no internal headroom, so the claim that it's just a matter of proper design is also questionable. One would have to digitally attenuate the input to the chip by some number of dB to allow for peaking at the conversion stage................
DC

Originally Posted by doug hazelrigg ➡️

Yes, you did, and magisterially

Originally Posted by dcollins ➡️

Also, there are many D/A converter chips that have no internal headroom, so the claim that it's just a matter of proper design is also questionable. One would have to digitally attenuate the input to the chip by some number of dB to allow for peaking at the conversion stage................

Originally Posted by oldeanalogueguy ➡️

there are LOTS of standards
and many companies/people ignore them
so there is no way to know tht if you give -xxxdBFS to someone else that when they play it it will result in yyydBV or zzzdBSPL or any meaningful number

Originally Posted by doug hazelrigg ➡️

Although I disagree with (almost) everything OAG says, it's important I think to let him have his say, not for entertainment value, but because icons and shibboleths need to be questioned from time to time -- keeps us all honest, makes us think about what we claim to know or believe

Originally Posted by oldeanalogueguy ➡️

there are lots of standards
and you should not use them if they cause a problem

Originally Posted by oldeanalogueguy ➡️

you have to confuse digital with analog to make that peak statement

Originally Posted by oldeanalogueguy ➡️

there are no samples in analog domain
the only way to peak in analog is if the reconstructed signal
was bigger than the original sampled signal

Originally Posted by oldeanalogueguy ➡️

if you create a signal in the digital domain
then peaks in the analog domain are meaningless

Originally Posted by oldeanalogueguy ➡️

what you do with uprezzing in the digital domain has nothing to do with the analog domain
the new values will often be higher but so what !!!!!!

Originally Posted by duckoff ➡️

What kind of sorcery is this!!

Originally Posted by doug hazelrigg ➡️

Although I disagree with (almost) everything OAG says, it's important I think to let him have his say, not for entertainment value, but because icons and shibboleths need to be questioned from time to time -- keeps us all honest, makes us think about what we claim to know or believe

I mean, it wasn't all THAT long ago I subscribed to popular falsehoods like "bit-hogging" and "even multiple SRC" or the silly assertion that

HIGH SAMPLE RATES SOUND BETTER BECAUSE IT SOUNDS MORE LIKE ANALOGUE

Originally Posted by doug hazelrigg ➡️

Although I disagree with (almost) everything OAG says, it's important I think to let him have his say, not for entertainment value, but because icons and shibboleths need to be questioned from time to time -- keeps us all honest, makes us think about what we claim to know or believe

Originally Posted by narcoman ➡️

!!!!! that's a pretty odd one!

!!!!

Don't know how many times one has to tell you that the reconstructed signal does not peak at the samples!


eh? That doesn't even BEGIN to make any sense!


Correct . It's called an inter sample peak. Higher... higher than? higher than the neighbouring samples. Sort of a "peak" if you will, between the samples. Sort of "inter" yes?heh

And as an "inter sample peak" it addresses some of the issues that happen when we re-construct signals.

I'll find you a nice picture then I'll give up


Here we go.... I know it's not ideal but it'll suffice to illustrate what we're all talking about. I'm convinced that you're thinking of something else....

Look at the nice picture of a very small window of a waveform. Now look at the four sampled points (let's ignore quantisation for now but assume it's nicely dealt with). I draw your attention to the two middle points. Look at the peak between the points. It's a specific voltage not sampled by the time domain samples BUT it doesn't matter. The waveform is correctly sampled EVEN though there is a peak in the waveform between the sample points. This peak is completely and 100% re calculable within the digital domain if we so wish. It's also correctly reconstructed when we go DA. It's a peak that happened between the sample points, it's a peak that is mathematically maintained such that a proportionally correct waveform results from the sample points.

Originally Posted by bdenton ➡️

Let me state outright that I have absolutely zero knowledge of this subject (which is why I'm reading this thread, to learn something).

But you don't have to be very smart to look at the drawing and wonder: Could not the "inter sample peak" just as easily be an "inter sample dip"?

After all, it occurred between samples, i.e. it was not sampled.

While it would admittedly be highly unlikely that a dip of any significance could occur in such a small "slice" of time, there must be some interpolation going on...

Originally Posted by don4777 ➡️

I am almost ashamed to admit that I have read every post in this thread.

It seems that the primary disconnect is that OAG insists on using a different definition of the word "peak" (among others) than every other person on the planet. To OAG it appears to only be associated with clipping and not just the highest value reached during a cycle. There have been several excellent posts to clarify the definition and attempt to get OAG on the same page as everyone else. They have all failed. Not because they weren't very clear but because for some reason OAG won't acknowledge the accepted definition of the terms he bandies about.

I feel sorry for the students OAG must have confused in his past and feel fortunate that I didn't have professors of his ilk during my engineering program at Case Western. They were content to use definitions that were consistent with the rest of the engineering community.

Although I know this thread is a complete waste of time I can't seem to ignore it. It's like driving by a car accident. You don't really want to see the carnage but you have to look. On with the carnage...

Don

Originally Posted by bdenton ➡️

Let me state outright that I have absolutely zero knowledge of this subject (which is why I'm reading this thread, to learn something).

But you don't have to be very smart to look at the drawing and wonder: Could not the "inter sample peak" just as easily be an "inter sample dip"?

After all, it occurred between samples, i.e. it was not sampled.

While it would admittedly be highly unlikely that a dip of any significance could occur in such a small "slice" of time, there must be some interpolation going on...

Originally Posted by narcoman ➡️

!!!!! that's a pretty odd one!

!!!!

Don't know how many times one has to tell you that the reconstructed signal does not peak at the samples!


eh? That doesn't even BEGIN to make any sense!


Correct . It's called an inter sample peak. Higher... higher than? higher than the neighbouring samples. Sort of a "peak" if you will, between the samples. Sort of "inter" yes?heh

And as an "inter sample peak" it addresses some of the issues that happen when we re-construct signals.

I'll find you a nice picture then I'll give up


Here we go.... I know it's not ideal but it'll suffice to illustrate what we're all talking about. I'm convinced that you're thinking of something else....

Look at the nice picture of a very small window of a waveform. Now look at the four sampled points (let's ignore quantisation for now but assume it's nicely dealt with). I draw your attention to the two middle points. Look at the peak between the points. It's a specific voltage not sampled by the time domain samples BUT it doesn't matter. The waveform is correctly sampled EVEN though there is a peak in the waveform between the sample points. This peak is completely and 100% re calculable within the digital domain if we so wish. It's also correctly reconstructed when we go DA. It's a peak that happened between the sample points, it's a peak that is mathematically maintained such that a proportionally correct waveform results from the sample points.

Originally Posted by doug hazelrigg ➡️

Maybe I shouldn't psycho-analyze OAG, but it appears he's backed himself into a corner of an untenable position and at this point has no choice but to fight, perhaps "to the death." Funny, I've been there myself. I don't think he's a troll, but I DO think we should stop feeding him

I'm just glad somebody has asserted that 4 samples are required to reconstruct a wave, instead of just 2, as is popularly asserted. ask anybody who has designed a plug-in that employs oversampling about this!

Originally Posted by oldeanalogueguy ➡️

it should be because it is flat out wrong
any chip without headroom was badly designed
been saying that all along

Originally Posted by psycho_monkey ➡️

Well of course there is. You agree a standard! Same way we all drive on one side of the road in a given country...I can take my session from one studio to another, and provided the converters are lined up the same, my gain staging will be the same.

You really are trolling now aren't you? It can't possibly be that you don't grasp this basic concept.

Originally Posted by theblue1 ➡️

[bold added to reflect the bolding that OAG did when he quoted this in his post below]


Would it have made you feel better if the bolded bit had instead read, When a DA converts the stream of data into a smooth voltage variation WRT time some of those voltages will be higher than the [voltages generated from] preceding and following samples... ?


You know, I've read a lot of your writing on this issue and I'm still not entirely sure -- do you acknowledge the phenomenon but are quarreling with the descriptive terms of art used? Or are you suggesting that the scientists and design engineers describing this phenomenon are imagining it?