The No.1 Website for Pro Audio
 Search This Thread  Search This Forum  Search Reviews  Search Gear Database  Search Gear for sale  Search Gearslutz Go Advanced
if higher sample rate doesnt matter then why .... Effects Pedals, Units & Accessories
Old 19th September 2011
  #361
Lives for gear
 

Quote:
Originally Posted by DaveDaveDave View Post
So, if the only relationship is "random", shouldn't all DAC's turn my samples into noise? Would that be white noise or pink noise?
wrong dimension for random

the max voltage is random
depends on the d/a design

the shape is not random
the d/a will give you the same shape every time
and you can twist the knob to the right/left to make
the resulting analog signal as big/small as you like
Old 19th September 2011
  #362
Gear Nut
 

Quote:
Originally Posted by oldeanalogueguy View Post
the picture is meaningless
you are mixing the digital and analog domains again
and doing it wrong
[/quote]

Oh, it's not over is it?
It may be meaningless to you, but it isn't to everyone arguing this with you. It seems that your understanding of DSP and analog digital conversion is - unique.

Quote:
Originally Posted by oldeanalogueguy View Post
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
if you create a signal in the digital domain
then peaks in the analog domain are meaningless

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 !!!!!!
meaningless in both domains except for the distortion added
How about if you're say, um, I dunno MIXING? You keep saying that anytime you manipulate a digital signal that it's "meaningless" If I sum two PCM streams, is that "meaningless", regardless of whether it causes overs or not?

How can you walk into a room full of audio engineers and say "meaningless... except for the distortion...." - it sounds like the punchline to a joke

You use the word "meaningless" so much that I don't think it means what you think it means.
Old 19th September 2011
  #363
Lives for gear
 

Quote:
Originally Posted by narcoman View Post
Let's put it another way. When a DA converts the stream of data into a smooth voltage variation WRT time some of those voltages will be higher than the preceding and following samples. If they weren't then you'd be implying that every single sample was at or above the peak of every waveform element at digitisation - which would be impossible as it would need an infinite sample rate. Not that I'm saying they join up (that'd be stupid), not that I'm saying they are a problem (might be if you decide to calibrate against VU from a poorly reported digital value); but those peaks MAY not get reported in digital meters and WILL be above the equivalent converted reference level.

Those voltages above your particular sample points are what everyone else (us, most of GS, all DSP books, Shannon et al) in the biz refer to as inter-sample peaks. Those same peaks in the digital domain are the very peaks that many meters do not report. How much of an issue these things are depend very much on DA design, DAW design and any hardware you interface in the analogue stage (although it'd have to be a stupidly designed piece of equipment that couldn't deal with the sort of peaks we're discussing).

If you choose not to accept that then that's okay. You RIGHTLY say that they would only be a problem with a poorly designed DA or 0dB(VU) references with NO headroom. Read that last bit carefully - wouldn't want you to think I was putting VU into the digital realm.

Is that any better for you? Or are we done here in an exhaustive "one man against the masses" sort of way? Doesn't matter to me either way!

Right - I've given what I hope is an explanation of what people are referring to. I'm no longer going to argue with you. Any other readers of this particular part - go read the textbooks yourself. Go read Shannon et al. Go read Dans paper.
thanks
read them all
nobody here repeats them correctly

you described it accurately

infinite sample rate is identical to analog !!

i agree that you cannot tell accurately in the digital domain what peak analog values may result from a given signal
but that is not any alleged intersample peak problem
that is nyquist in action and your not knowing how to determine analog maximum values while in the digital domain
Old 19th September 2011
  #364
Lives for gear
 
minister's Avatar
You guys, give up... Notorius O.A.G. has wandered into just about every forum and exclaimed that Inter-sample Peaks are the phlogiston of audio. All of the papers and links that you are putting before him were put before him before. His only refrain is, "you are confusing the analogue and digital worlds, there are no samples in analog and no peaks between the samples.". And he says it is all avoided if the DAC is designed well.

He makes assertions and provides no proof or sensical explanation and says all that the provided papers do nothing but make assertions and supply no proof.

To wit (not to be confused with Twit) :

Solutions for True Peak / Intersample Peak limiting that really work?

This includes a valiant effort by Alexy Lukin to spell_it_out. The recitations are remarkably consistent.
Old 19th September 2011
  #365
Gear Maniac
 

actually...Ocean's theory (or is that theorum) makes a hell of a lot more sense than saying peaks higher than the preceeding sample and before the next one occur between samples...if a line is drawn between two samples to connect the dots as it were, then why on earth would a machine decide to draw an arc as opposed to a straight line...and the four samples making a wave...is ridiculous...I could place four points on a piece of paper and draw a condo in between them...that serves no purpose ...place 44,100 dots on a sheet and tell me again why a piece of software would fill in the blank spots with arcs or partial triangles.

you guys oughtta try and get with what Ocean is saying...he could be wrong but is making some solid points...it's a little depressing to see how the witch hunt continues...if you disagree, be kind about it.
Old 19th September 2011
  #366
Lives for gear
 
Alexey Lukin's Avatar
 

There is a reason why machines use arcs, not line segments. It is stated by the Nyquist theorem, which is the basis of all sampling theory, that the proper way to reconstruct analogue signal from digital samples is to use a sinc interpolation (not a linear interpolation!) The sinc function is indeed smooth and oscillating, which results in all the "curves" during interpolation.
Most audio DACs approximate this sinc interpolation quite closely and may indeed reconstruct analog voltage that is higher than the level of digital samples (something that Mr. OAG calls "Nyquist in action" and the rest of the world — "intersample peaks"). As a side note, reconstruction filters in most DACs are approx. 50–100 samples long (not 4 or 2).
Old 19th September 2011
  #367
Lives for gear
 
dcollins's Avatar
Quote:
Originally Posted by oldeanalogueguy View Post
here are 4 samples
0 0 0 0
whatcha going to recreate ?
Your next post?


DC
Old 19th September 2011
  #368
Lives for gear
 
doug hazelrigg's Avatar
Quote:
Originally Posted by oldeanalogueguy View Post
rotflamo

not in a corner

nr of samples depends on teh bandwidth being limited

you cant reconstruct anything meaningful from 4 samples
even worse if the signal was not band limited

here are 4 samples
0 0 0 0
whatcha going to recreate ?
Per cycle dude, per cycle
Old 19th September 2011
  #369
Lives for gear
 
doug hazelrigg's Avatar
Quote:
Originally Posted by oldeanalogueguy View Post
they are the correct voltages per nyquist
I STILL don't know what this means
Old 19th September 2011
  #370
Lives for gear
 
doug hazelrigg's Avatar
Quote:
Originally Posted by MandyC View Post
and the four samples making a wave...is ridiculous...I could place four points on a piece of paper and draw a condo in between them...that serves no purpose ...place 44,100 dots on a sheet and tell me again why a piece of software would fill in the blank spots with arcs or partial triangles.

.
Because it's understood we're plotting a sine function, not a condo LOL

I suppose I should elaborate here. It's commonly asserted that all you need are 2 sample points per cycle to reconstruct a waveform, as per Nyquist-Shannon, and that's true, but this is just how it works out statistically over time... At a fs of 44.1khz for example, the potential errors in estimating peak values increase as we go up in frequency (above a certain point), but these errors gradually minimize if we increase higher sample rate, and are quite small at very high sample rates:

Worst case peak error in dB off FS

64K sample rate
@20K -5.105
@18K -3.953
@12K -1.603

96K sample rate
@20K -2.011
@18K -1.603
@12K -0.688

192K sample rate
@20K -0.4736
@18K -0.3823
@12K -0.1685


ATTRIBUTION: pj geerlings, gs member
Old 19th September 2011
  #371
Moderator
 
narcoman's Avatar
 

Quote:
Originally Posted by oldeanalogueguy View Post
sort of is like a little bit pregnant or almost dialing a phone nr.

the picture is meaningless
you are mixing the digital and analog domains again
and doing it wrong
hahahahah!
Old 19th September 2011
  #372
Moderator
 
narcoman's Avatar
 

Quote:
Originally Posted by oldeanalogueguy View Post
rotflamo

not in a corner

nr of samples depends on teh bandwidth being limited

you cant reconstruct anything meaningful from 4 samples
even worse if the signal was not band limited

here are 4 samples
0 0 0 0
whatcha going to recreate ?
You can reconstruct a fairly boring waveform from 4 samples alone. And baby - listen, in the science world once we're IN a debate all parameters are assumed. We've already said we're talking about digital audio - therefore it's 44.1, 48, 88.2, 96, 176.4 or 192. That also predefines what bandwidths we're talking about. It's a pointless argument to talk about unbounded PCM sampling.

As zero crossings? That would absolutely be the signal of EXACTLY half nyquist rate for any particular rate. It would cause you issues as you'd have no idea about amplitude as you'd soon find out on the way out !! In other words - 0 0 0 0 won't happen.

However - it wasn't the point of the diagram. It was to show you that the values captured by samples are NOT the waveform itself but a set of interpolation descriptors.
Old 19th September 2011
  #373
Moderator
 
narcoman's Avatar
 

Quote:
Originally Posted by MandyC View Post
actually...Ocean's theory (or is that theorum) makes a hell of a lot more sense than saying peaks higher than the preceeding sample and before the next one occur between samples...if a line is drawn between two samples to connect the dots as it were, then why on earth would a machine decide to draw an arc as opposed to a straight line...and the four samples making a wave...is ridiculous...I could place four points on a piece of paper and draw a condo in between them...that serves no purpose ...place 44,100 dots on a sheet and tell me again why a piece of software would fill in the blank spots with arcs or partial triangles..
No you couldn't. You don't join dots. There is only ONE outcome from sampled points in Nyquist-Shannon. The re-construction is a smooth curve exactly mirroring what was samples. All waveforms are sine waves (there is no such thing in the real world as a square wave or triangle wave - all harmonic motion is sinusoidal).


This isn't open for debate or interpretation. This is well understood and complete wave mechanics. If you think that you can draw arbitrary points or you join up samples then I'm afraid you don't understand how sampling works. No insult intended.
Old 19th September 2011
  #374
Quote:
Originally Posted by oldeanalogueguy View Post
finally
somebody stated the issue properly

yes that happens as described
but they are not intersample peaks

they are the correct voltages per nyquist
nothing has peaked inter-samples

i am basically saying that the sort of terms mixing digital and analog signals like saying intersample peak are erroneous
and if you think they are true then what they claim does not exist (because the terms are misused, andor meaningless)

you have to mix the analog and digital domains to even come up with intersample peak

now can you also get peaks or other problems due to badly designed d/a but that is not intersample peak problems either
I think with this post, I finally see what you're saying.

The reason everyone is arguing with you is because it's pedantry of the highest order, it hinders people's understanding of digital audio, and really goes against the whole idea of what digital audio is (digitising an analogue audio signal).

Technically, you are "right". If you read out a list of sample values of an AD conversion of an audio signal, such as the sine wave repeatedly posted, there will be no phantom "peak" reading in between any two actual samples, even if (as in narco's diagram above) the actual voltage level of the analogue waveform is higher than the samples captured. In the digital domain, it's just a list of numbers, and to actually "see" the wave, we need to plot this on a graph against time (which is what a DAW visual display does).

Then, we recreate the wave, the numbers go to the DAC and are changed into electrical signals and the wave is back in the analogue domain - where you are arguing there are "peaks" but no "samples". Correct again.

However - and this is a big however - you are completely missing the point that when plotting the wave out of the samples you've taken there will be values REPRESENTED in the digitised audio data that do NOT fall on the actual samples taken - there may be no ACTUAL peak reading, because as you've stated to death, there's no "intersample-sampling" going on, but the sample you've captured stores enough information to recreate this value upon reconversion to the analogue domain.

If (for example) we sampled the same audio signal at a higher sampling rate, these "gaps" in the samples will be filled in to some extent, and there may be (gasp) samples in between where the original samples (at the lower rate) would have been taken, which in some cases will be (again,gasp) higher than the previously recorded samples. In other words, the "phantom" peaks that others refer to as "inter-sample" have been captured exactly (although of course, there will be some analogue peaks that won't be represented by exact samples because we can't sample at an infinitely high rate).

So - as has been argued before, all that you're actually doing on here is doggedly sticking to your own specific terminology, that is not only contradictory to what the rest of the digital world understands, but you're also doing those at a more formative level of their development a disservice, because you're confusing them. Anyone paying for a lecture with you on this subject deserves their money back, because you're not teaching them any curriculum but your own.

Effectively you're a mechanic teaching someone how to repair an engine, but giving the parts your own names. When that same mechanic goes to work for a commercial enterprise, they'll have to spend a large amount of time relearning everything with the accepted names, and cursing their teacher for hindering their learning.

Except it's worse than that, because the theory behind the term "intersample peaks" helps with visualising the issue. If I explain the capture of a waveform using the term "intersample PEAKs", then I ask a student why there might be a problem if a sample is normalised to 0dBFs and played back, should be able to see that the reconstruction of the waveform could contain peaks in that might clip the DA conversion (the "phantom" peaks not represented by samples captured). If I doggedly insist there are no peaks in a digital domain, beyond the value of the actual sample captured, this is more confusing for a student.

Like I said - you're a crap teacher if this is how you think you should educate people. And thank f**k I didn't have the same sort of person teaching my degree!

As an addendum, on my bookshelf right now, I have a book called "fundamentals of acoustics" - we referred to it at uni as "Kinsler + Frey", the two primary authors - I think I've got the F. Alton Everest around somewhere too. Unless you've written a tome on digital audio that has the authority of these 2 books, please for f's sake either accept the existing terminology, or just bugger off and stop confusing people.
Old 19th September 2011
  #375
Moderator
 
narcoman's Avatar
 

Quote:
Originally Posted by oldeanalogueguy View Post
thanks
read them all
nobody here repeats them correctly

you described it accurately

infinite sample rate is identical to analog !!

i agree that you cannot tell accurately in the digital domain what peak analog values may result from a given signal
but that is not any alleged intersample peak problem
that is nyquist in action and your not knowing how to determine analog maximum values while in the digital domain
I'm not discussing a problem; I don't even see there is a problem. But inter sample peaks (and troughs) exist mathematically and are represented and identifiable in their analogue counterpart. As for the subject area, I understand this every bit as nicely as Dan. Shockingly - I'm higher qualified than he is, academically, in this area. He has the edge on ground breaking industry concepts and pushes though!! heh

To your point on telling peaks : Actually - you can, but there are conditions. As long as you have a supplied reference (for example -18dBFS calibrated to 0dBVU. If 0dBVU is referenced to +4dBU then we have an absolute voltage reference. We can then calculate the peak value will be using a lim->inf ). In practise obviously not possible but we could find it iteratively OR we could use a differential to find the max of a fourier model for those particular points. It'd be a bit pointless though as correctly designed analogue stages would render it a pointless exercise.... they'd never be an issue.

For the record : NOBODY has stated that the peaks themselves exist as samples in the digital domain, nor has anyone stated that samples fly around in the analogue domain. So you're argument is not just a little over pedantic BUT also not looking at the relationship we must maintain as the digital samples ARE supposed to be a stored version OF an analogue waveform set!

To argue about the term inter sample peak is to deny the science community a short term to reference the location of the peaks on waveforms that occur on reconstruction after interpolation using the points in the sample set. It also denies them/us a reference term to talk about each peak locally and how it relates to the digital sampled version. It's information and reference needed many times over!

Now - I've two papers on the very subject (although for satellite transmission not audio - but the principles are the same) which I wrote after my doctoral thesis on modelling wave propagation in virtual media (a three dimensional view of sampled audio). I used the term in the paper. I presented one of the papers at SIGGRAPH back in the day and it was very well accepted by the electrical engineering, computer graphics AND maths community at large. The point being, if that community uses the term then I don't see why you can't, unless you don't want to! In other words there is no scientific objection to the term and what it represents. Translating between media (digital and analogue) is crucial in what we all do for fun, a living or for research.
Old 19th September 2011
  #376
Gear Nut
 

Quote:
Originally Posted by narcoman View Post
So you're argument is not just a little over pedantic BUT also not looking at the relationship we must maintain as the digital samples ARE supposed to be a stored version OF an analogue waveform set!
Quote:
Originally Posted by psycho_monkey View Post
The reason everyone is arguing with you is because it's pedantry of the highest order, it hinders people's understanding of digital audio, and really goes against the whole idea of what digital audio is (digitising an analogue audio signal).
narcoman, psycho_monkey - I admire your patience!
Old 19th September 2011
  #377
Gear Nut
 

Quote:
Originally Posted by minister View Post
You guys, give up... Notorius O.A.G. has wandered into just about every forum and exclaimed that Inter-sample Peaks are the phlogiston of audio. All of the papers and links that you are putting before him were put before him before. His only refrain is, "you are confusing the analogue and digital worlds, there are no samples in analog and no peaks between the samples.". And he says it is all avoided if the DAC is designed well.

He makes assertions and provides no proof or sensical explanation and says all that the provided papers do nothing but make assertions and supply no proof.

To wit (not to be confused with Twit) :

Solutions for True Peak / Intersample Peak limiting that really work?

This includes a valiant effort by Alexy Lukin to spell_it_out. The recitations are remarkably consistent.
Wow - I was curious and read this thread too - unbelievable.

Guys, oldanalogueguy excepted, I really do appreciate all the time you guys put into communicating on this board, sharing your professional insights and real world experience are a boon to novice engineers like myself, as I'm sure it is to the teaming thousands who'll peruse these threads over time.

I'm out. Thanks again!
Old 19th September 2011
  #378
Lives for gear
 
DaveUK's Avatar
Quote:
Originally Posted by DaveDaveDave View Post
Wow - I was curious and read this thread too - unbelievable.

Guys, oldanalogueguy excepted, I really do appreciate all the time you guys put into communicating on this board, sharing your professional insights and real world experience are a boon to novice engineers like myself, as I'm sure it is to the teaming thousands who'll peruse these threads over time.

I'm out. Thanks again!
Seconded!
Old 19th September 2011
  #379
Lives for gear
 

+1 Daves!
Old 19th September 2011
  #380
Gear Nut
 

Quote:
Originally Posted by oldeanalogueguy View Post
no
i use teh word peak as all dictionaries and 99% of the world does

the few who erroneously use it as audio jargon are the ones claiming everybody else is wrong
Most assuredly NOT 99% of the people posting here. You are on your own. I'll agree that the group here may not be a true representation "of the world". We're much better looking. heh

Webster Dictionary: Peak - "the highest level".
Old 19th September 2011
  #381
Moderator
 
narcoman's Avatar
 

on a not un related note - I've just had a cup of tea on Beachy Head. 560 feet of windy cliff! Awesome
Old 19th September 2011
  #382
Lives for gear
 
minister's Avatar
Quote:
Originally Posted by narcoman View Post
on a not un related note - I've just had a cup of tea on Beachy Head. 560 feet of windy cliff! Awesome
Peaks must be amazing!
Old 19th September 2011
  #383
Quote:
Originally Posted by oldeanalogueguy View Post
finally
somebody stated the issue properly

yes that happens as described
but they are not intersample peaks

they are the correct voltages per nyquist
nothing has peaked inter-samples

i am basically saying that the sort of terms mixing digital and analog signals like saying intersample peak are erroneous
and if you think they are true then what they claim does not exist (because the terms are misused, andor meaningless)

you have to mix the analog and digital domains to even come up with intersample peak

now can you also get peaks or other problems due to badly designed d/a but that is not intersample peak problems either
In other words, you are not arguing about what actually happens -- only what people call it?


Now, you headlined that post "this guy understands why intersample peaks is meaningless jargon" -- referring to me -- but it would be more accurate to say that I seem to have pinned down your complaint as one of terminology -- which is actually how I first interpreted the disjunct way back in the first few score posts -- but so much of what you said seemed [at the time, at any rate] to reflect a misunderstanding of the basic issues, I lost track of that.

Me, I completely understand why this phenomenon is called intersample peaks.

The peaks actually seldom coincide with sampled values -- but they still exist as peaks in the signal.

To make sense of that, we have to think about what signal is. (Signal theory.)

Signal does not stop at the AD nor does it miraculously reappear at DA. Signal is represented in the digital domain by both the measured values of the samples and the mathematical processes used to 'reconstruct' an analog version of that signal.

But, as I understand it, you seem to be suggesting that the peaks in the signal stop existing in the digital domain, simply because we don't have a specific sample coinciding with that peak. However, Nyquist-Shannon demonstrates that signal and the peaks it contains (subject to bandlimiting), continues to exist even in the digital domain and will be 'reconstructed' as a continuous analog signal stream on DA.


Speaking of bandlimiting, I pointed out above that the question of what would happen if the signal peak (or whatever we want to call it) between the 'top' two samples in narcoman's graphic was a dip instead is moot, since such a dip could only result from a frequency above the necessary frequency bandlimit.
Old 19th September 2011
  #384
Lives for gear
 
doug hazelrigg's Avatar
Quote:
Originally Posted by psycho_monkey View Post
... a book called "fundamentals of acoustics" - we referred to it at uni as "Kinsler + Frey"
Ah! I remember it well
Old 19th September 2011
  #385
Quote:
Originally Posted by narcoman View Post
on a not un related note - I've just had a cup of tea on Beachy Head. 560 feet of windy cliff! Awesome
This place?

It does look pretty awesome... even in the digital representation.
Old 19th September 2011
  #386
Lives for gear
 

Quote:
Originally Posted by doug hazelrigg View Post
Ah! I remember it well
Kinsler and Frey is a pretty hardcore text. Did you guys use this in audio classes or physics/acoustics classes?

Wish there were more acoustics programs in the states, only a handful of universities offer it.
Old 19th September 2011
  #387
Lives for gear
 
DaveUK's Avatar
Did you see a lambretta in the sea? At beachy head .
For the blue1
Old 19th September 2011
  #388
Lives for gear
 
doug hazelrigg's Avatar
Quote:
Originally Posted by johnnyc View Post
Kinsler and Frey is a pretty hardcore text. Did you guys use this in audio classes or physics/acoustics classes?

Wish there were more acoustics programs in the states, only a handful of universities offer it.
In my case it was part of a non-degree certificate program in audio engineering
Old 19th September 2011
  #389
Gear Guru
 
UnderTow's Avatar
Quote:
Originally Posted by oldeanalogueguy View Post
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
http://tinyurl.com/buowp2

Alistair
Old 19th September 2011
  #390
Lives for gear
 
DaveUK's Avatar
Nice birds ha ha!
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump
Forum Jump