OK Mike,

Intersample Peaks = invisible overs. 

If you take the image you posted above (Post #55) and you draw straight lines from one sample to another (connecting the dots if you will), you would get the picture of a waveform as seeing inside your DAW.  Notice that it doesn't go over 0dBFS, which is why it would show up as a legal signal in Sonar (or any DAW for that matter).   But the REAL waveform is shown as going over 0dBFS, which is what happens after it passes through the A/D process.  How long is it?  It is the distance between two samples.  It's all there in the picture you posted.  You just needed to take a closer look.

Does that help?

mike_mccue


In all honesty, the repeated suggestions that I read something (which I did) that you seemingly appreciate and honor but are not willing to discuss your understanding of makes me think someone else has got your login info today.

all the best,
mike

LOL, but what do you want me to do?  I am discussing the subject with you, and trying to explain my point as much as I can.  I just don't feel like posting pictures, that's all.  When you said this:
 
"I didn't know there were 'illegal" frequencies and always assumed everything between 22.5 and 44.1 was just described somewhat inaccurately but still existed as ripple.

So if "This filter removes all content above the Nyquist frequency (half the sample rate). " it can't be mysterious digital harshness."
 
You were obviously talking about the Anti-aliasing filter.  At least that's what they were talking about in that part of the article.
 
Now, how am I not keeping up?  You type faster, I'll give you that :-)

Yes, it does. And I just finished an illustration that I find interesting:

Its a 17 second animation.

Well, there you go.

I just know that DAWs don't show this kind of info.

"I just know that DAWs don't show this kind of info."

What the heck does that mean?

Are you suggesting that the processing is not sample accurate?

Are you suggesting that the peak meters in a DAW do not account for each sample?

Are you suggesting that the track peak flag markets in SONAR are somehow skipping samples?

I can't imagine another way for an over to go unreported if it lasts longer than a single sample.

You are acting like a wave form goes up above 0dBFS and proceeds unnoticed with, what did you say, "invisible" overs until it naturally descends along the curve. How does that happen, exactly, with out the DAW telling you its in OVER?

Now that we see that anything above 11025 is poorly interpolated at 44.1kHz it seems even less likely that one will ever encounter an intersample peak.

It's been a while for me. have you ever read the SSL papers about intersample peaks? I recall that they were far more detailed than those you linked too but still left me with the very questions I'm asking of you.

And I do know a bit about anti aliasing but not enough to act like I know what either of us are talking about.

mike_mccue

And I do know a bit about anti aliasing but not enough to act like I know what either of us are talking about.

Then I think this is where you should start, no offense Mike. 
 
If neather of those papers (the one from SSL and the one I linked) made sense to you, then nothing I say will.  The questions you ask imply very clearly that you need to understand how digital audio is captured first.  Once you do that, then everything else will become much clearer.  It might just be poor choice of words on my part.  That's also very possible.  But I'll leave you with these:
 

• Yes, Intersample Peaks are invisible to the regular digital meters found in DAWs.  That's why SSL and other plugin companies have specialized meters to detect them.
• Intersample Peaks, as the name implies, happen in-between samples.  Because of this they are not reported as overs even when the samples appear to be well below 0dBFS.
• A waveform is only fully reconstructed in the analog world.

 
 
HTH

"Intersample Peaks, as the name implies, happen in-between samples.  Because of this they are not reported as overs even when the samples appear to be well below 0dBFS."

So are you talking about oversampling? What happens when you get to the next discrete sample? It's either over or no right? wrong? Let's talk about that.

How does something that occurs in between samples last longer than a sample?

"A waveform is only fully reconstructed in the analog world."

I think transfer functions occur in many domains.

best,
mike

Does this help?

http://www.gearslutz.com/board/mastering-forum/185883-inter-sample-peaks.html

Jose7822


Does this help?

http://www.gearslutz.com/board/mastering-forum/185883-inter-sample-peaks.html

From that page:

It's another case of artistic license. This lower track is an illustration of a 22kHz (or higher if the sample rate of the project shown is higher than 44.1kHz) tone. One should share that fact with the reader before asking people to believe they will hear any of it:







Here's a 1/4 rate reconstruction described as "around 11025".

In this case you can see the intersample peaks are not overs. But if you did bump up the level to force a intersample peak to occur (for testing of course) some of the intersample peaks will only peak above 0dBFS within a sample and some will read an over on an adjacent sample. What else happens with those adjacent samples?

Don't you see that everyone is side stepping this issue and cleverly producing images that avoid revealing how obvious this question is?

Imagine if you will, that as the frequency get's lower it's more and more likely that you will get an over reported along with any intersample peaks that are not detected.






This image comes with the caveat: "Noisy square looking signals are often ripe with intersample peaks. I tried feeding a digital signal to the chain to look what peak levels occured. The result was this: ... Musical signals are never this bad,"



Why can't he show us a sample of music. My premise is that it's because MOST OF THE TIME when the "reconstructed meter" is banging red SO IS THE PEAK METER.

BTW why is the RMS reported as above the Peak in that image? It must something about the non-musicality of that "digital signal" He is using for the test?

That ain't music.



Here is SSL's illustration of a "High Frequency Sine wave" they do not bother showing it as an over but suggest the same shape could be. What they do not point out is that the particular example will result in either a one sample length over cased by an intersample peak OR  a intersample peak with an adjacent sample that should read as over. They don't point out that the over will last about 1/44,100th of a second.

They totally side step that issue by showing you a generic illustration of a sine wave.





So I have to say, this hasn't clarified anything for me.

And I'm not as dumb as you are treating me. :-)

Thanks for the link.

best regards,
mike

Does this help? http://www.gearslutz.com/...nter-sample-peaks.html

Second paragraph of the first response: "So those transients that exhibit intersample peaks will sound squared off (with lots of distortion) compared to peaks that stay under, say, -0.3dbFS."
This is what I'm saying. If you exercise any restraint at all with peak levels of your master, you don't have to worry about intersample peaks. The term itself makes them sound more extreme than they are. Inter-sample bulges is more like it. 

brundlefly

Does this help? http://www.gearslutz.com/...nter-sample-peaks.html

Second paragraph of the first response: "So those transients that exhibit intersample peaks will sound squared off (with lots of distortion) compared to peaks that stay under, say, -0.3dbFS."
This is what I'm saying. If you exercise any restraint at all with peak levels of your master, you don't have to worry about intersample peaks. The term itself makes them sound more extreme than they are. Inter-sample bulges is more like it. 

Yes, but that quote is from Peeder, whom everyone over at Gear Slutz knows is Jose's alter ego.  :-)  

Peeder is a lot more like me... he's always looking through the smoke to see if it's worth driving to the other side.

It doesn't surprise me that Mr Peeder and I were in agreement. We usually are.

At this point I feel like I did learn a few things yesterday... there's is even less chance of an unnoticed intersample peak passing through than I had previously thought.

That's my premise; There are very few circumstances where an intersample peak is not accompanied by an over and a ugly sound caused by the over. There are very few times that an intersample peak goes unnoticed.

A corrallary to my thesis is that EVERY diagram I have seen published uses a super sonic frequency to illustrate their point and make the issue seem compelling. My diagram is the only example I know of that illustrates the more likely circumstance that the source content has a lower frequency (something in an audible range) and that the intersample peak will be accompanied with a nasty hard limit or over.

I'm not trying to brag about "my diagram" but it certainly illustrates the concept "to draw is to see"... that drawing is the result of my desire to think thru the question "how do you illustrate a intersample peak on an audible tone and keep the adjacent samples from reporting an over?"

I hope it's clear by my repeated explanations... all the other charts side step the issue. They are illustrating supersonic frequencies. I guess very few people notice that fact. :-(    I find that disappointing.... a picture tells a thousand words or something like that?

I had previously thought the window for opurtunity was between upper mids like 8kHz thru 44.1kHz. Now I realize that there is nothing above 22049kHz in the stream... so the most likely frequencies to cause an unnoticed intersample peak are not even a factor.

An unnoticed intersample peak might occur between 8+/-KhZ (I'm still trying to think thru the lower limit... maybe it's fs/8?) and 22,049kHZ with a bias leaning towards the upper frequencies.

I refute Jose's assertion that anti aliasing has nothing to do with intersample peaks.

Anti aliasing eliminates the most likely source of intersample peaks.

I have been very confused by Jose's forum tai chi moves whereby he keeps deflecting my comments and questions to stuff written by other people... stuff that I read and then think... "Did Jose actually read this stuff?"

hmmm... I wonder If Jose ever bothered reading anything I wrote or ASKED?

Jose never responded to a single question with a direct response and kept referring me to those articles that I feel lack detail. When I was honest that I didn't have anti aliasing at the top of my mind Jose could have used his knowledge of the idea to help me understand that clearly the most likely frequencies to cause an intersample peak are not present. Instead of engaging the actual question he choose to suggest I learn more about anti aliasing. Well, I did and I'm back. :-)

We could have both agreed that the likely hood of encountering an intersample peak was from a narrow window of frequencies. 

When I explained that the white papers I have read suggest that a practical source of a unnoticed intersample peak is the first sample after a hard limit function begins... Jose assured me that this is not an issue that occurs with well written plugins. So it seems as if once again we can agree that the likely hood of encountering an unnoticed intersample peak is from a ever narrowing range of circumstances.

I finally found some real articles published by AES and they clearly state how rare and unlikely it is one will have an unnoticed intersample peak.

This link:

http://www.tcelectronic.c...lund_1999_level_co.pdf

Elaborates on the idea that this is a high frequency phenomena, provides specific examples with scale drawings, and discusses the very ideas that I have been trying to clarify.

for the record... here's my premise:

Intersample peaks are very unlikely to occur on musical content without being reported as an over or being accompanied by a legal and audible hard limit. If they do occur unnoticed it is most likely to happen at a supersonic frequency so you have to decide for yourself if it is something you or your listeners can hear.

If this worries you, simply use the SSL meter to confirm that what I say is correct.

Now, I think I will go back to reading the Divine Comedy... it's not higher math... but it keeps the mind sharp.

best regards,
mike


Jose, You know I think of you as great person, and you have been a great teacher to me on many occasions. We've discussed intersample peaks in groups many times... but never directly in debate. I apologize for putting you on the spot but you had every opportunity to answer my questions, tell me where I'm wrong, correct my misunderstandings, and spin this discussion to a consensus of understanding... even if we disagree on what is and isn't audible.

The idea that this phenomena only occurs in a very small set of circumstances should be discussed or debated in a way that results in a realistic assessment of the circumstances.

I'm mildly disappointed that you refused to help in that regard. It seems as if maintaining a level of misunderstanding in the discussion allowed you the advantage of presenting the idea of an intersample peak as being more common and problematic than it is.

I extend all my best to you, my friend.

crazy forum... mistaken double post.

I don't want to win... I'm just trying to learn.

all the best,
mike