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
Truth in metering? Utility Plugins
Old 4th September 2007
  #1
Gear Head
 

Truth in metering?

I recently picked up an older version of Elemental Audio's Inspector XL metering suite and tested it (and two other meters) out together on a source file. I would've expected similar results from such high end meters but no, I compared the same file on three different meters and got three different results.

In this case I used TL Master Meter, PAZ (by Waves), and the EA's Inspector MultiMeter. Waves and Inspector were close, roughly within 2 tenths of a dB but TL Master caught several "clips" that the others did not.


So my question is who is telling me the truth? Is the TL Master Meter a better meter because of its abilty to catch "oversampled" clip events that the other meters won't catch? Is measuring oversamples in metering just hype? What is the general consensus when it comes to plug in metering these days?

Thanks in advance,
Dusk
Old 4th September 2007
  #2
Lives for gear
measuring oversamples is not hype...think of it this way...when the audio hits the D/A it will be converted from a connect-the-dots type wave to a full on sine wave...which leaves room for additional overs. i.e., just because none of the samples go over, doesn't mean that the sine wave doesn't. Imagine a wave where the two "highest" samples are on either side of the peak.
Old 4th September 2007
  #3
Gear Head
 

Quote:
Originally Posted by stellar View Post
measuring oversamples is not hype...think of it this way...when the audio hits the D/A it will be converted from a connect-the-dots type wave to a full on sine wave...which leaves room for additional overs. i.e., just because none of the samples go over, doesn't mean that the sine wave doesn't. Imagine a wave where the two "highest" samples are on either side of the peak.

Stellar,

I assumed this was the case. So what I'm trying to ascertain is what good is a digital meter if it doesn't catch all that "inter-sample" stuff? Certainly there is a lot of valuable information in there that gets lost. How are we supposed to track it, especially while traveling a dB or two near the digital edge?


Just reading between the lines here (no pun intended) the TL Master Meter must be a superior meter compared to the others because it has the abilty to read some over samples (8). 8 times is nowhere near the resolution of 256. What meter tracks near infinte resolutions? Is it even possible (or neccesarry) to know whats going on in between the samples? How anal is too anal?

Judging from the several pop rock files I measured it doesn't look like anyone cares a whole lot about distorted oversamples anyways. I tracked over 300 clipped events before the meter stopped counting. I never made it to the first chorus of some well known rock band....even material I have had mastered rivaled that.

I'm just trying to get a feel for how other mastering engineers out there handle inter-sample distortion. Lord knows the loudness war is in full swing now but I'd like to be educated enough to communicate my needs to a mastering engineer before my stuff gets lodged into digital oblivion.
Dusk
Old 4th September 2007
  #4
Lives for gear
 
Darius van H's Avatar
 

Verified Member
Quote:
Originally Posted by duskb View Post
I'm just trying to get a feel for how other mastering engineers out there handle inter-sample distortion.
By listening.........sometimes i need to clip stuff to compete in the loudness jihad, and i have not the faintest idea how many samples i'm clipping. I don't care about the numbers. If i can't hear it, it's ok. Sometimes I can hear it, and it's still OK!.....however you have to keep in mind that distortion will certainly be added down the (consumers) chain.

Do some experiments.......see how far you can clip distorted guitar music before you hear the sizzle sound of square-top distortion. Compare this to a solo piano piece!

good luck!
Old 4th September 2007
  #5
Lives for gear
 
Nordenstam's Avatar
 

Verified Member
Quote:
Originally Posted by duskb View Post
So what I'm trying to ascertain is what good is a digital meter if it doesn't catch all that "inter-sample" stuff?
IMO, the false/sample-dot metering being used in 99,999% of digital equipment can take most of the blame for the current state of typically unecessarilly bad digital sound. In recent years, there's been a lot of talk about mixing and working at lower levels in the digital domain. This would all had been avoided in the first place if people used true peak metering.

Quote:
Just reading between the lines here (no pun intended) the TL Master Meter must be a superior meter compared to the others because it has the abilty to read some over samples (8). 8 times is nowhere near the resolution of 256. What meter tracks near infinte resolutions? Is it even possible (or neccesarry) to know whats going on in between the samples? How anal is too anal?
8x is probably more than enough. The figure that comes first to mind is that 8x should be true enough to about two decimals. I dunno how accurate that is though, that was just something that sprang to mind from testing these things some years ago. It will probably be at least accurate enough that leaving a tenth or two of headroom in the final master is enough to avoid end user equipment distortion. It's not only the oversampling that does it, the filtering is important too.

Quote:
I'm just trying to get a feel for how other mastering engineers out there handle inter-sample distortion.
I'm aware of it through metering. My playback system does not reveal this as much of a flaw. Good DACs, such as the ones used in mastering, will handle the intersample peaks without any problems. They will show some distortion as the waveshape is altered, but it will not show the distortion of the analogue parts in the DAC crapping totally out - because it doesn't - unlike the end user equipment. The cheaper DACs often go into havoc as it can't handle the excessive signal level. Straining some cheap transistor circuits, like a cheap mixer preamp, can give you an idea of the effect.

I think this explains most of why it is such a popular thing to do these days. By pushing the signal above the digital ceiling(0dBFS), the ME can cram in some extra level where there is not supposed to be any. Problem is of course that this requires quite formidable equipment to play back without unpredictable distortion. The ME does have that formidable equipment, end user does not. This is why I've been calling this the antithesis of translation.

Using a meter to watch how far above zero the signal goes tells a lot about how the processing is faring with the signal. Some limiters and processing is more prone to create intersample peaks than others. Most of the records laden with such peaks only have a wee bit of intersample peaks and these have a bigger chance of surviving end user equipment, MP3'ing, broadcast and so forth. I've seen a consistency in the output of some mastering houses that tells me that these either don't know about intersample peaks or don't care at all. Some have more than 3dB of intersample peaks and that's quite a feat to create! Often, such processes are accompanied by aliasing distortion. Some effects, like pure digital clipping, creates excess amounts of high frequencies. These ultrasonics have no where to go in the bandwidth limited digital signal, and they rear their ugly heads as aliasing distiortion in addition to the intersample peaks. This sounds quite unpleasant and it always puzzles me why the ME doesn't use plain old fashioned harmonic distortion instead to achieve the same loudness effect. (more on clipping in an old post here)


Regards,

Andreas Nordenstam
Old 4th September 2007
  #6
Gear Head
 

Thank you Andreas and Darius. (BTW the link you included helped answer a few questions but prompted even more. Forgive me as I push the discussion onwards).

A few things still puzzle me. It seems that if the meters dont display some sort of oversampling in the process they are useless, toys at best. Yet folks still sell and buy them. I cant imagine it would be all that hard to incorporate these instruments (over sampling meters) into software. In fact while re-metering old pre-mastered files originally mixed using Waves PAZ I caught several peaks on the TL that I didnt even know were there. Granted, I never heard the clips but that's not the point, after they are amplifed and crushed its possible _I might_, and then in mastering its too late to deal with it. My question is the EA and PAZ didnt capture events identically, if at all, that the TL did. What does this mean? Why don't all basic digital meters catch static information consistently?

Also Darius pointed out an interesting fact. Ultimately you have to listen for distortion in a master to capture it. This is what has been puzzling me for years. Fact is everytime I walk out of mastering I hear distortion on my mixes that didn't exist there beforehand, yet when I meter it using my old tools (PAZ for instance) I don't see any clipping. I just knew everytime I could hear it I zoomed into the waveform and I could see a massive kick drum that looked like it had lost at LEAST 7dB of its peak. I understand this is why we go into mastering and all but for god's sake...if _I_ can consistently pick out clipping over these flat tops you'd think that the ME and his metering would as well.

Lastly, you mentioned shoving crap into the digital ceiling over 0.00dB. How is this even possible? Both the TL and the EA have a display that shows a red peak over 0.00dB that tops out at +6dBfs. HOW THE HELL do you get gear to produce signals beyond its limits? Last I heard 0.00dB is the theoretical limit of digital. How can I go 6dB over that? This was never covered in my discussions in college years ago...is this a "new" thing?

Thanks for your insight....I appreciate the dialogue.
Dusk
Old 4th September 2007
  #7
Gear Maniac
 

Quote:
Originally Posted by duskb View Post
A few things still puzzle me. It seems that if the meters dont display some sort of oversampling in the process they are useless, toys at best. Yet folks still sell and buy them.
I don't know, people managed for years with only VU meters, and today's "digital zero" is typically 12 to 15 dB above the needle when it's pinned to the right side.

If your success hinges entirely on whether or not an intersample peak exists, those meters may be useless. If not, then useless may be an overstatement.

Considering the released records you measured that included countless hard clips, I don't think an oversampling meter would have told those engineers anything new or useful.

Quote:
Both the TL and the EA have a display that shows a red peak over 0.00dB that tops out at +6dBfs. HOW THE HELL do you get gear to produce signals beyond its limits? Last I heard 0.00dB is the theoretical limit of digital.
Actually, I think that's the physical limit. Theoretically the display could look at the slope of the last few samples and project where the peak might have been if it were allowed to exceed 0dBFS.
Old 4th September 2007
  #8
Gear Guru
 
UnderTow's Avatar
 

Verified Member
Quote:
Originally Posted by duskb View Post
Lastly, you mentioned shoving crap into the digital ceiling over 0.00dB. How is this even possible? Both the TL and the EA have a display that shows a red peak over 0.00dB that tops out at +6dBfs. HOW THE HELL do you get gear to produce signals beyond its limits? Last I heard 0.00dB is the theoretical limit of digital. How can I go 6dB over that? This was never covered in my discussions in college years ago...is this a "new" thing?

Thanks for your insight....I appreciate the dialogue.
Dusk

This is still the same inter-sample peak stuff we are talking about. You might already know this but many people don't: The way waveforms are represented in most DAWs doesn't represent the signal that comes out of the converters. It just represents the sample values.

A simple 20Khz sine wave might look like this in a typical DAW:




This doesn't resemble a sine wave. That is because the software just draws lines between the sample value when creating the image.
(This bad representation of signals as "steps" is also one of the causes of the most fundamental misunderstandings of how digital audio works pushing the crazy sample rate race but that's another discussion)

A more correct representation of the same 20Khz sine wave after reconstruction as it comes out of your converters would look like this:



This shows a proper sine wave. (Thanks to Audition/Cool Edit).


Now here is another waveform (created artificially and zoomed in) that only has sample values up to 0 dB FS:





Although the highest sample values are 0 dB, the actual reconstructed waveform as it comes out of a converter peaks just above +2.5 dB FS! (Check the dB scale on the right of the image).

When you use digital clipping and extreme limiting you get these kind of inter-sample peaks above 0 dB FS because the clipper or limiter is only looking at the actual sample values, not at the signal that will eventually come out of your converters.

Some limiters (and even clippers) are better designed and take the reconstructed signal into account but they are in the minority. Also oversampling limiters/clippers that don't actually check for inter sample-peaks will in practise cause less overs due to working at higher sample rates internally and due to the anti-aliasing filters built into the over-sampling engine.

Alistair
Old 4th September 2007
  #9
Gear Guru
 
UnderTow's Avatar
 

Verified Member
Oops. Maybe I should resize those images. Lol. heh

EDIT: Images cropped.

Alistair
Old 4th September 2007
  #10
Gear Guru
 
UnderTow's Avatar
 

Verified Member
Back to digital meters. The inter-sample peak aspect has been mentioned but there are two other important things about digital meters and that can give different (RMS) output values on different meters:

1) What is the 0 dB reference? Is it a sine wave or a square wave? In theory it should really always be a sine wave as square waves at 0 dB FS are technically illegal signals that go over 0 dB FS when reconstructed by the converters. Not all meters use a sine wave 0 dB FS reference. Using square waves as a reference can give a value that is roughly 3 dB lower.

You can test this quite easily: Load white noise into your DAW and insert a steep low-pass filter into the signal. Set the filter's cutoff frequency as high as possible (20Khz). Play the white noise and turn the filter on/off. Although you probably won't hear any difference in volume, in most DAWs and meters you will see the meter's level jump up by about 3 dB when you turn the filter on. That is because it is acting a bit like a reconstruction filter in a DAC. If the meter jumps like this, it is looking at sample values and not at the reconstructed signal.

2) Integration time (I think that's the correct term). How much time does the RMS calculating algorithm look at when doing it's maths? This will affect the values shown. In some meters this is user adjustable. I tend to use 300 ms.

I hope this helps,

Alistair
Old 4th September 2007
  #11
Lives for gear
 
robot gigante's Avatar
Quote:
Originally Posted by Darius van H View Post
By listening.........sometimes i need to clip stuff to compete in the loudness jihad, and i have not the faintest idea how many samples i'm clipping. I don't care about the numbers. If i can't hear it, it's ok. Sometimes I can hear it, and it's still OK!.....however you have to keep in mind that distortion will certainly be added down the (consumers) chain.

Do some experiments.......see how far you can clip distorted guitar music before you hear the sizzle sound of square-top distortion. Compare this to a solo piano piece!

good luck!
Well yeah, but consumer decks will clip and distort waay sooner than my DA- so I still think oversampling meters save me a few headaches.

What I like about the TL Labs meter is that it show how many db over 0 the 'clip' is is, which is important, I think. I read their white paper on oversampling and they measured several consumer systems and the amount of distortion that happens depending on how high over digital 0 the 'clip' is. From what I remember, the distortion increases exponentially the farther above 0db fs you get.

So if you have a bunch of intersample peaks at .05 above 0dbfs, the consumer DA is probably going to handle that (that is to say, distort less) a lot better than with, say, fewer 'clips' that are 1 or 2 db above 0dbfs. Imagine you have one 'clip' that registers 2 or 3 db above 0dbfs during a quiet section a song- pop!

And since your DA is probably not going to have any problems at all with it, you'd never know.
Old 4th September 2007
  #12
Lives for gear
 
Darius van H's Avatar
 

Verified Member
Quote:
Originally Posted by robot gigante View Post
Well yeah, but consumer decks will clip and distort waay sooner than my DA- so I still think oversampling meters save me a few headaches.
Interesting.......I've mastered loads of stuff with very hot levels and never had any complaints.....i'm guess i'm just not thinking technically enough.
Old 5th September 2007
  #13
Lives for gear
 
robot gigante's Avatar
Quote:
Originally Posted by Darius van H View Post
Interesting.......I've mastered loads of stuff with very hot levels and never had a complaints.....i'm guess i'm just not thinking technically enough.
Well, I have some nitpicky clients sometimes. So it's just better to avoid potential problems with those problem few.

Either way, I'm just pointing out that there are hard clips and then there are hard clips, so it's nice to know what's going on just in case.
Old 5th September 2007
  #14
Lives for gear
 
Nordenstam's Avatar
 

Verified Member
Good illustrations, Alistair!


Quote:
Originally Posted by duskb View Post
(BTW the link you included helped answer a few questions but prompted even more.
Did you get the answers you sought? I think they're there in the illustrations above, but please do ask if there's anything unclear. :-)


Andreas
Old 6th September 2007
  #15
Gear Head
 

Hey guys...lots of food for thought. Still trying to digest it.

You know alot of this stuff is review for me, technically. In college I was required to take a digital theory class but back then I was an arrogant bastard and refused to do homework on the grounds that digital was from the "Dark Side". I used to tell my instructor I'd always have an 827 with me so his wanky digital theory wouldnt amount to a hill of beans in my life. (Boy was i wrong....now I own an HD rig. : ( )

Anyways I must've really tuned out the day when we got into converter theory...I had no idea about the +6dBfs thing. From what I recall in digital 0dBfs was the limit, whether if the waveform was clipped, cropped, or limited it would still come out the other side clipped, cropped, or limited to 0dBfs (and probably sound like crap). If I understand you guys correctly you are saying that because the converters try to remanufacture the missing transient over 0dBfs we have signals that could potentially exceed 0dBfs on "playback". Right?

Since I was an analog guy before digital it might help me to reference this to something I AM familliar with, +4dBu. As I recall this equates to 1.228V. Does the voltage where digital and analog ultimately break up meet at the exact same intersection? Are we just using two different scales to express the same break up point (i.e +22dBu for analog gear versus +6dBfs for digital?) Not sure if I'm comparing apples to apples here but if I understand this whole dBfs in terms of voltage it might get swallowed a bit easier. Somehow these scales have to relate but for the life of me I can't figure out how.
Dusk
Old 6th September 2007
  #16
Gear Guru
 
lucey's Avatar
 

Verified Member
The line in the sand as much as about ears and the chain as numbers. Distortion is everywhere in recorded music and playback is unpredictable.
Old 6th September 2007
  #17
Gear Guru
 
UnderTow's Avatar
 

Verified Member
Quote:
Originally Posted by duskb View Post
Anyways I must've really tuned out the day when we got into converter theory...I had no idea about the +6dBfs thing. From what I recall in digital 0dBfs was the limit, whether if the waveform was clipped, cropped, or limited it would still come out the other side clipped, cropped, or limited to 0dBfs (and probably sound like crap). If I understand you guys correctly you are saying that because the converters try to remanufacture the missing transient over 0dBfs we have signals that could potentially exceed 0dBfs on "playback". Right?
I think the missing pieces of the puzzle for you is how square waves are constructed and the reconstruction filter in the DAC.

Square waves can be seen as a combination of sine waves: The fundamental of the square wave frequency and odd harmonics into infinity (3rd, 5th, 7th etc).

Check out the animation on this page: Square wave - Wikipedia, the free encyclopedia

You can see that with just the fundamental frequency, the waveform is just a sine wave. As more and more odd harmonics are added, the more the resultant wave starts looking like a square wave. If an infinite number of odd harmonics are added, you get a perfect square wave. (Of course in reality this never happens. There are no real perfect square waves in nature but you can ignore this for now).

So if you turn this around and you start with a square wave and remove harmonics, the more harmonics are removed, the more the wave starts looking like a sine wave. If all the harmonics are removed, you get a perfect sine wave.

When you clip a digital signal, you get sample values that represent a square wave in the digital domain. When this reaches your converter, a low-pass filter (the reconstruction filter) removes all the harmonics above the nyquist frequency (half the sampling rate).

If the signal being sent to the converter is, let's say, a square wave at 20Khz, all the harmonics will be removed by the reconstruction filter as they are all above the nyquist frequency. So this theoritical square wave has become a perfect sine wave.

Now the important thing to know is that the peak of the fundamental sine wave of a square wave is higher than the square wave itself! You can best see this on this animation: http://mathserv.swarthmore.edu:8080/...tt1/SqWave.jsp

Hit the "Show the convergence" button. The animation will start with the fundamental sine wave and then add harmonics and show the resultant wave form as each harmonic is added.

The important thing to notice is that the fundamental sine wave (in red) peaks higher than the theoretical perfect square wave (in black)! It peaks about 3 dB higher.

So if we go back to our digital "square wave" at 20Khz and assume it is at 0 dB FS, after hitting the reconstruction filter in your DAC, you end up with a sine wave that peaks at the equivalent of +3 dB FS!

Does this clear things up a bit?

Alistair
Old 6th September 2007
  #18
Gear Head
 

Quote:
Originally Posted by lucey View Post
The line in the sand as much as about ears and the chain as numbers. Distortion is everywhere in recorded music and playback is unpredictable.
While I can't understand everything you said one thing I will contend with is your last statement. Sure maybe on some modern rock record where stuff is shoved up against the wall distortion is everywhere but the world I came from the engineer would be considered incompetent if he delivered tracks to the ME with distortion on them. Distortion can be added or removed at will by the engineer and only he has total control over how much he may or may not add.

Again I'm not saying I don't use distortion for creative purposes, who of us doesn't? The point of my initial post was, "Hey, I'm hearing distortion artifacts that I didn't put there in post mastered material (and mind you the ME gain staged appropriately) while the meter doesn't show any overs, how can this be?" I was never even aware of the overs until I got a meter that clearly displayed them. Sure we cut great music on VU meters for half a century but we also used a hammer and nails for a millenia until someone figured out how to use an arc welder. If an intersample meter helps us be more cognizant of distortion then maybe ME should make a habit out of using them, not ignoring them. YMMV.


My main concern, and what led me here to begin with, was I typically like to exhibit a level of competence in my work. By understanding how this crap works I might have a better chance of controlling distortion on masters that first and foremost playback on my system and hopefully, eventually, someone elses system as well.

Dusk
Old 6th September 2007
  #19
Gear Guru
 
lucey's Avatar
 

Verified Member
The line in the sand is the job of mastering these days. It's not absolute. It's client driven. It takes listening and communication with the client, not meters.

Compression makes distortion, clipping makes distortion, AD converters make distortion. Guitar amps make distortion. Its all become a rather grey area. If you want to play it safe don't clip a converter and keep it under 0.3 dbfs.

But the client has the last word on your "competence". Some clients are all about certain distortions, other are not as they want the compromises. When I'm mastering I'm hearing all kinds of noise that I wish wasn't there, but that's life. So it's more about the clients definitions than my own.

Listen to Audioslave/Audioslave and tell me that Vlado Meller is not "competent". He clearly is, but not in the way you might be, or she might want, or he might be.
Old 6th September 2007
  #20
Lives for gear
 
Nordenstam's Avatar
 

Verified Member
Made an illustration of a severely bad case, a 32 bit float file fed directly to the 24 bit fixed(integer/normal) setup. Noise! Not a theoretical worst case, though very close, half a dB or so lower than the worst case 6dB "hidden" peaks. The waveform display shows sample dot values, not reconstructed signal. Real signal goes nearly a dB above the ceiling. Waveform continues like that with similar sized pulses.

Truth in metering?-badcasemetering.png


Andreas Nordenstam

Last edited by Nordenstam; 6th September 2007 at 08:02 PM.. Reason: fixed math error
Old 6th September 2007
  #21
Lives for gear
 

Quote:
Originally Posted by Lupo View Post
IMO, the false/sample-dot metering being used in 99,999% of digital equipment can take most of the blame for the current state of typically unecessarilly bad digital sound. In recent years, there's been a lot of talk about mixing and working at lower levels in the digital domain. This would all had been avoided in the first place if people used true peak metering.
Yup, The "use every bit" mentality , especially fosterd during the transition from tape to 16 bit(the standared at the time) is a seemingly irrepairable curse ! If peak meters had'nt been imlemented, and the VU , RMS type of metering would have prevailed , we'd be better off.



Quote:
Originally Posted by Lupo View Post
I'm aware of it through metering. My playback system does not reveal this as much of a flaw. Good DACs, such as the ones used in mastering, will handle the intersample peaks without any problems.
Well then lets just get everybody some good , top quality converters then!!!!!!! Then those boom box ghetto blasters won't have those extra harmonics though!!!

Quote:
Originally Posted by Lupo View Post
Straining some cheap transistor circuits, like a cheap mixer preamp, can give you an idea of the effect.
my first cd player , bought years ago, sounds better, analog low pass ???
Quote:
Originally Posted by Lupo View Post

I think this explains most of why it is such a popular thing to do these days. By pushing the signal above the digital ceiling(0dBFS), the ME can cram in some extra level where there is not supposed to be any. Problem is of course that this requires quite formidable equipment to play back without unpredictable distortion. The ME does have that formidable equipment, end user does not. This is why I've been calling this the antithesis of translation.
In the future, when mere mortals like me are trying to wrap our heads around digital theory and reconstruction, we would appreciate you not using words like " antithesis " !!!!!!!!!!!!!

Quote:
Originally Posted by Lupo View Post
Using a meter to watch how far above zero the signal goes tells a lot about how the processing is faring with the signal. Some limiters and processing is more prone to create intersample peaks than others. Most of the records laden with such peaks only have a wee bit of intersample peaks and these have a bigger chance of surviving end user equipment, MP3'ing, broadcast and so forth. I've seen a consistency in the output of some mastering houses that tells me that these either don't know about intersample peaks or don't care at all. Some have more than 3dB of intersample peaks and that's quite a feat to create! Often, such processes are accompanied by aliasing distortion. Some effects, like pure digital clipping, creates excess amounts of high frequencies. These ultrasonics have no where to go in the bandwidth limited digital signal, and they rear their ugly heads as aliasing distiortion in addition to the intersample peaks.

USE EVERY DAM BIT.
We want all of them put to work , dammit!!!!!



Quote:
Originally Posted by Lupo View Post
This sounds quite unpleasant and it always puzzles me why the ME doesn't use plain old fashioned harmonic distortion instead to achieve the same loudness effect.
Compress those harmonics, not the entire envelope!!




Lupo, your post ROCK!!



heh
hehheh
hehhehheh
Old 6th September 2007
  #22
Gear Guru
 
lucey's Avatar
 

Verified Member
It's silly to assert that MEs dont hear distortion because of better converters. I use the Grado phones and hear a single clip with ease. Lesser converters are so distorted in general they mask a lot of distortions audible in a studio.

The only issue, the practical one, is still about taste and the clients wishes.

I agree that RMS metering and ears are all you really need.
Old 7th September 2007
  #23
Gear Guru
 
UnderTow's Avatar
 

Verified Member
Quote:
Originally Posted by lucey View Post
I use the Grado phones and hear a single clip with ease.
LOL. Aren't you being just a tiny bit arrogant here? So you can hear a single sample clip even if it is a distorted electronic snare sample causing the single sample clip? (just an example). I doubt it.

If it is a violin or a piano or something like that, sure, but it many cases you can clip some transients inaudibly.

Alistair
Old 7th September 2007
  #24
Gear Guru
 
lucey's Avatar
 

Verified Member
Quote:
Originally Posted by UnderTow View Post
LOL. Aren't you being just a tiny bit arrogant here? So you can hear a single sample clip even if it is a distorted electronic snare sample causing the single sample clip? (just an example). I doubt it.

If it is a violin or a piano or something like that, sure, but it many cases you can clip some transients inaudibly.

Alistair
Of course you can. Yet if it's audible to anyone, I can hear it in the cans. Clear? I'm not talking about any sample, but any audible distortion of either the analog or digital variety. Anyone mastering at a high level can hear the 'worst' case. That's part of the job.

Now some clients don't care about some sounds, others are hyper sensitive to certain frequency ranges, some just want it loud and don't care about the compromises. So it's a client's call what kind and degree of distortions are too much. My point was that a mastering converter hiding distortions that would be more audible to a consumer is backwards logic. Consumer systems are more distorted, and thus mask more.
Old 7th September 2007
  #25
Gear Guru
 
UnderTow's Avatar
 

Verified Member
Quote:
Originally Posted by lucey View Post
My point was that a mastering converter hiding distortions that would be more audible to a consumer is backwards logic. Consumer systems are more distorted, and thus mask more.
The idea is that better converters have more headroom in the DACs. Especially in the analogue stages. These converters can handle inter-sample peaks better than cheaper converters with less or no headroom.

I don't ever listen to iPods, getto blasters, car CD players or anything like that so I don't know how valid this idea is...

On a different note, there was an interesting discussion on the Pro Audio mailing list about the effects of room acoustics making certain kind of distortions more audible. The idea being that distortion that would be just below the hearing threshold on cans or in an anechoic chamber could become audible in more typical room acoustics. So listening on cans might not always be the most revealing playback systems for certain types and/or levels of distortion.

Here are some quotes:

Bruno Putzeys talking about listening to stuff in an anechoic chamber:

Quote:
Originally Posted by Bruno Putzeys
Another surprising bonus is that the kind
of electronic artefact (amplifier distortion etc) that you'd usually pick up
easily in even the crudest of dbt's suddenly don't seem to matter anymore.
It reminds me of Floyd Toole's notion that reflections off side walls
*increase* listeners' ability to discern certain small changes in the
signal.
Paul Frindle's response:

Quote:
Originally Posted by Paul Frindle
This is a hugely significant point! I have talked about this at length
before and will not reitterate it all, but one can glean lots from the
way we perceive sounds in the natural environment - and how in unnatural
environments we can be more easily fooled by errors. And we must
remember that real sounds are virtually never experienced under anechoic
conditions.
In a later post Paul broadens this to listening to headphones.

Also, the cumulation of distortion in lesser systems might bring something above the hearing threshold that isn't audible on a high-end system.

I think it is a bit dangerous to assume that a lesser system will never reveal something that is inaudible on a high-end system or that listening on headphones will always be more revealing than listening to stuff on speakers.

Alistair
Old 7th September 2007
  #26
Lives for gear
 
Nordenstam's Avatar
 

Verified Member
Quote:
Originally Posted by lucey View Post
It's silly to assert that MEs dont hear distortion because of better converters.
Agreed! The distortion caused by altering the waveshape, and the aliasing commonly caused by this, should be easier to hear on good equipment.

There's three typical distortion sources. The waveform is bent and altered, making a different signal. This in turn often gives rise to aliasing distortion. The third source is the thing I'm at, the analogue parts. Both in the DAC itself and following parts that may crap out with signals that are hotter than 0dBFS sine wave.

Quote:
Originally Posted by lucey View Post
Lesser converters are so distorted in general they mask a lot of distortions audible in a studio.
Perhaps. A lot of distortion going on everywhere, indeed. The analogue part going wack effect can be quite dramatic though. A fictitious situation to illustrate this; battery powered device where the output circuit is built with a swing going close to the power rails. The engineer will typically design this to handle a sine wave at full scale. That is a seemingly reasonable spec, but it will not tolerate such signals as illustrated earlier in the thread. There's no headroom left at all. Say a one volt peak to peak device with a 1.2 volt supply trying to do ~1.4 volts (3dB intersample peak). Splat! Solid state function-before-art engineered electronics are generally not favoured for distortion.

Perhaps most important - it makes the signal a vulnerable source for further digital processing. Like resampling, equalizers, bass boosts and lossy coding. A lot of consumer listens to it this way through broadcast and personal players. Such procesing creates a new set of sample dots, often revealing the intersample peak'ed nature of the real signal, making a jump in the level of the sample dots. Since there is no space for these new dots to be stored above 0dBFS, this will program further distortion into the signal. Lossy coders can distort both at in- and output in the actual math crunching too. Add a typical miniature MP3 player and the low voltage scenario mentioned above and there's a couple of layers of distortion there, that is in effect programmed by the ME, totally out of control.

The problem may be transparent at 0.1dB overshot and a horror at +3dB. Without the oversampled meter - there's no accurate way to know just how far off limit it goes.


One of the big points is to use this as a way to gauge what the processing does to the signal. There's no reason, as confirmed by listening to a lot of equally freaking loud modern masters with relatively good sound, to go far above zero. There's other and better ways to do it. A lot of guys create at least as hot masters without resorting to pushing stuff far above 0dBFS.


Andreas N
Old 8th September 2007
  #27
Lives for gear
 
Nordenstam's Avatar
 

Verified Member
Quote:
Originally Posted by duskb View Post
Since I was an analog guy before digital it might help me to reference this to something I AM familliar with, +4dBu. As I recall this equates to 1.228V. Does the voltage where digital and analog ultimately break up meet at the exact same intersection?
The output level of the reconstruction is dictated by circuit design. The digital level is free to be converted to any analogue level desired. The deciBel scale is used in both domains for convenience sake, but does not have any fixed relationship between dBFS and anything else.

Quote:
Originally Posted by duskb View Post
From what I recall in digital 0dBfs was the limit, whether if the waveform was clipped, cropped, or limited it would still come out the other side clipped, cropped, or limited to 0dBfs (and probably sound like crap). If I understand you guys correctly you are saying that because the converters try to remanufacture the missing transient over 0dBfs we have signals that could potentially exceed 0dBfs on "playback". Right?
Right. The sample points program the DAC to recreate a waveform that exceeds zero.

Each sample dot in the digital audio is an impulse. The computer believes this is an infinitely short event in time, but it's far from the final truth. This infinite event requires infinite frequencies and that's hard to practically realize. The signal is filtered to a continuous audio bandwidth signal on reconstruction. The impulse response of the filter resembles a sinx/x function - the ideal low pass filter.

Note that the peaks and through lines neatly up. In the digital audio system, these leads and tails on the pulse lines up at the sample frequency. The output signal is the sum of a long series of such positive and negative pulses. Each pulse alone only tell the value at that instant, the sum of the pulses AND all the wiggles around the pulse train fills in the space between.

Dan Lavry have a nice sampling theory paper (.PDF) with a lot better explanation of this if you want to dig in. The point is that the sample dots only represents a very small part of the final picture.


The reason the sample point value meters are not accurate enough is that they are undersampled. To check the signal with a digital tool you need more information than the relatively scarce sample point values alone can provide. Digital oscilloscopes use at least 10x signal frequency to give a fairly accurate visual representation of the signal. If the signal is to be studied closer in a visual/metering way, more sample dots are needed.

As long as the only intent is to transmit the information there is no need for more samples. Oversampling locally in the peak meter is the way to go. This is the beauty of the Nyquist/Shannon - a lot of information can be crammed into a small information space. Slower sampling rates means better measurement quality/less distortion too.



Flatfinger: Thanks for the comments! Sorry if my English is odd at times. Is it alright if I draw the second language card? BTW; The use of the word Antithesis was not in Hegelian sense. :-)


Cheers,

Andreas N
Old 8th September 2007
  #28
Gear Guru
 
lucey's Avatar
 

Verified Member
I never said that the cans are the be all and end all reference. The cans, the room, and mostly the client tell the tale. I don't bother myself with perfectionism. It's a relative sport, with client driven standards. Musicality is the key.

There's nothing much to debate.... each client, each source, each aim for the music and each clients fear is going to be different and interact in a unique way. Some are afraid of one little hair, some are afraid of not being heard.
Old 11th September 2007
  #29
Lives for gear
 

Quote:
Originally Posted by Lupo View Post
The output level of the reconstruction is dictated by circuit design. The digital level is free to be converted to any analogue level desired. The deciBel scale is used in both domains for convenience sake, but does not have any fixed relationship between dBFS and anything else.



Right. The sample points program the DAC to recreate a waveform that exceeds zero.

Each sample dot in the digital audio is an impulse. The computer believes this is an infinitely short event in time, but it's far from the final truth. This infinite event requires infinite frequencies and that's hard to practically realize. The signal is filtered to a continuous audio bandwidth signal on reconstruction. The impulse response of the filter resembles a sinx/x function - the ideal low pass filter.

Note that the peaks and through lines neatly up. In the digital audio system, these leads and tails on the pulse lines up at the sample frequency. The output signal is the sum of a long series of such positive and negative pulses. Each pulse alone only tell the value at that instant, the sum of the pulses AND all the wiggles around the pulse train fills in the space between.


Andreas N
This explanation is almost correct but there are a few areas that confusion could arise.

Firstly the cause of the higher levels is in fact the reconstruction filter - this is required to decode the PCM - it is not a question of 'joining dots' it's a question of excluding all illegal freqs - and producing a continuous signal from the time sampled codes. And BTW a sin(x)/x filter is far from ideal for this job as it is not steep enough and too sloppy.

Secondly all final reconstruction actually goes on in the analogue domain (it has to as this is the only place it isn't sampled in time) so modern converters tend to use a digital pre-DAC filter that does the exclusion of illegal freqs bit and upsamples to higher rate - so that the analogue filter after the DAC can deal with the much easier to filter high freq components left over from the digital filter. This is much much more accurate than trying to make an analogue filter do it all.

So if so designed - it is possible for the digital filter to overload and clip - or the DAC after the filter to do likewise, completely independently of what may go on in the scaling of the analogue signal.

The reason that this problem may persist (and DAC manus don't simply design for handling +3db signals) is that such an illegal signal could not be generated from an ADC in the first place (so they have some justification that it's your fault) - and designing for max that is higher than strictly legal impacts negatively on their highly competitive spec figures, which people compare when deciding which kit to buy and employ.. For instance if they were to design for the absolute worst case (of 6dB overload tolerance) a product that was formerly specified at 102dB SNR would have to be re-specified at 96dB SNR..

This is bad news for them commercially - and it is arguably not their fault if the professional user routinely sends illegal PCM signals to their final product. Neither is it their fault if makers of DAWs employ metering that prevents you from seeing the problem.

I should add that the reason DAW manus continue with their simple sample value metering is that its too expensive to do otherwise. Each reconstruction meter costs as much processing as a hefty plug-in so peppering them all over the application is simply not an option for them.
Old 12th September 2007
  #30
Lives for gear
 
Nordenstam's Avatar
 

Verified Member
Hi Paul!

Thanks for the elaborations!

It's always great when you share some of your seemingly boundless knowlegde. :-)


Cheers,

Andreas
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
Similar Threads
Thread
Thread Starter / Forum
Replies
Trebor Flow / Mastering forum
279
DPS / Mastering forum
5
soldiaboy / So much gear, so little time
2
David R. / So much gear, so little time
2

Forum Jump
Forum Jump