Pink noise signal showing unexpected level in DAW

[johnlink]

"Mastering Audio" by Bob Katz arrived today and I can see that I am going to learn a lot by reading that book. Bob, thank you in advance.

I downloaded the pink noise file from http://www.digido.com/User/Assets/Ac...cor_st_441.WAV and loaded it into my DAW. I was surprised to see that the level meter shows -11.3 db instead of -20 db. Why would that be?

John Link

[bob katz]

Originally Posted by johnlink

"Mastering Audio" by Bob Katz arrived today and I can see that I am going to learn a lot by reading that book. Bob, thank you in advance.

I downloaded the pink noise file from http://www.digido.com/User/Assets/Ac...cor_st_441.WAV and loaded it into my DAW. I was surprised to see that the level meter shows -11.3 db instead of -20 db. Why would that be?

John Link

My pleasure, John. Hope you like the book. The reason is your DAW is reading the instantaneous PEAK level of the pink noise instead of its RMS level. By the way, if you have a metering system that purports to be RMS but it does not read EXACTLY -20 dBFS on that test signal I have provided, then you can blame the meter. There are several reasons why metering manufacturers have made some classic mistakes so you can depend on this pink noise signal to separate the accurate from the inaccurate, or the simply incorrect from the AES-17-compliant.

BK

[johnlink]

Thank you, Bob. You answer makes perfect sense to me. The meter in my DAW is obviously a peak meter.

I like your book very much, and I'm sure that I'll have more questions as I work through it and apply your ideas.

John Link

[UnderTow]

Originally Posted by bob katz

By the way, if you have a metering system that purports to be RMS but it does not read EXACTLY -20 dBFS on that test signal I have provided, then you can blame the meter. There are several reasons why metering manufacturers have made some classic mistakes so you can depend on this pink noise signal to separate the accurate from the inaccurate, or the simply incorrect from the AES-17-compliant.

BK

Hi Bob,

What are the AES-17 specs? Every single meter and/or application I have shows the RMS level to be -22.94 dB. Only in Audition when I change the RMS settings from "0 dB = FS Square Wave" (default) to "0 dB = FS Sine Wave" does it show the level as -19.93 dB. Is this the correct setting?

And what window size should be used? (Deafult is 50 ms).

Thanks,

Alistair

[bob katz]

Originally Posted by UnderTow

Hi Bob,

What are the AES-17 specs? Every single meter and/or application I have shows the RMS level to be -22.94 dB. Only in Audition when I change the RMS settings from "0 dB = FS Square Wave" (default) to "0 dB = FS Sine Wave" does it show the level as -19.93 dB. Is this the correct setting?

And what window size should be used? (Deafult is 50 ms).

Thanks,

Alistair

Dear Alistair. I don't know what context they are using for "window". Do they mean for the RMS calculation or for attack or release time? 50 ms is probably a little too low a window for good low frequency response for RMS program material, I would suggest 100 ms.

Anyway, the latter is the correct setting. Sine waves should read the same dB level on both peak and RMS on all meters that have been created since time immemorial, at least for 60 or more years.

It's not just AES-17, actually, it's common sense. It's just that many of the guys doing software-based metering have lived in their ivory towers and never lived in the analog world. For the past 60 years, until DAWs came along, this has been the rule:

On Sine wave whose peak level is 0 dBFS, the RMS level of that
sine wave is DEFINED to be 0 dBFS. This is also true of the AVERAGE level.

So if you have a square wave peaking at 0 dBFS, the RMS level should read OVER 0 dBFS! But this freaked out designers who justified their choice also by the right triangle rule and their knowing that the RMS value of a sine wave is 0.707 of its peak level. But the mistake they made is that they then felt this implied that the RMS level must be 3 dB down. But decibels are ALWAYS relative and must ALWAYS have a reference.

If you buy a Dorrough meter and feed it a sine wave, the VU and peak will read the same. If you buy an old analog Dolby quasi-peak meter, it will read the same decibels when fed a sine wave. If you buy a BBC or IEC peak-reading meter, they've all been calibrated to read 0 dB when fed 0.775 volts RMS.

By having a consistent reference based on a sine wave of a particular level being called 0 dB, the whole system works, or "worked", until digital audio designers came along, some of whom had never used a meter in the analog world.

For the past 60 or more years, ALL analog peak and averaging and RMS meters read the same decibel level for a sine wave. In other words, by definition, 0.775 volts RMS is defined as 0 dBu, 0 dBm, and 0 dB on both peak and RMS-responding voltmeters.

The same goes for VU meters. External VU meters are calibrated to a sine wave of a specific dBFS level even if measured on a peak instrument.

So if you encounter a "VU" or an "Averaging" or an "RMS" meter whose level reads 3 dB (or 2.94) lower than its peak level with a sine wave, please send a letter to the designer and tell him to read AES-17.

Hope this helps,

BK

[UnderTow]

Bob, waow. Thanks for the extensive explanation. That all makes alot of sense!

Alistair

[Lagerfeldt]

@Bob, great explanation

[rufus13]

Originally Posted by bob katz

<snip> Sine waves should read the same dB level on both peak and RMS on all meters that have been created since time immemorial, at least for 60 or more years.<snip>
BK

I'm not sure if the manufacturers of audio equipment are doing anyone a favor by displaying peak, rms, and average as the same value.

For a sine wave, peak value of 1, the RMS value is .707, and the Average rectified value is .636. If you have a meter that reads the same when you read in peak mode or RMS mode/Average mode with a sine wave input, the meter has something wrong with it (or it has 3 different reference levels, one for each mode).

Typical calibration of (electro-mechanical) meters is "average responding, R.M.S. indicating" at a specified frequency or range of frequencies, BUT this is true for sine wave only. The displayed value of a complex alternating waveform (compared to equivalent DC heating) is less accurate than the sine wave the meter is calibrated with (almost always, check the mfg spec's).

Thanks for the heads up on DAW meters. One more thing to re-check with every update.

[bob katz]

Originally Posted by rufus13

I'm not sure if the manufacturers of audio equipment are doing anyone a favor by displaying peak, rms, and average as the same value.

For a sine wave, peak value of 1, the RMS value is .707, and the Average rectified value is .636. If you have a meter that reads the same when you read in peak mode or RMS mode/Average mode with a sine wave input, the meter has something wrong with it (or it has 3 different reference levels, one for each mode).

Not exactly, not if you think in dB. RME has an excellent explanation of it in the manual on the Totalmix metering. In one direction (what you advocate) you are technically correct, but it is far more convenient for all decibel-based metering to be based on the same reference. And besides, it's a standard... AES-17.

[minister]

Originally Posted by bob katz

For the past 60 years, until DAWs came along, this has been the rule:

On Sine wave whose peak level is 0 dBFS, the RMS level of that
sine wave is DEFINED to be 0 dBFS. This is also true of the AVERAGE level.

i didn't know the dB FULL SCALE meant anything in 1946!

Originally Posted by bob katz

So if you encounter a "VU" or an "Averaging" or an "RMS" meter whose level reads 3 dB (or 2.94) lower than its peak level with a sine wave, please send a letter to the designer and tell him to read AES-17.

are you sure the problem isn't pan law? if i put a -20 sine wave generator on a mono track in, say Pro Tools, i get a -22.5 signal showing left and right. this is pan law, not failure to follow AES-17.

...or, what am i missing in what you are saying?

[bob katz]

Originally Posted by minister

i didn't know the dB FULL SCALE meant anything in 1946!

Actually, it did. Around that time, or by the early 50's the very first analog peak reading meter was developed. It was calibrated so that sine wave reading 0 dBm would also read 0 dB on the peak meter. I first saw a quasi peak-responding meter in 1973, but I forget who the manufacturer was. Probably Klein and Hummel.

It has nothing to do with pan law. It has to do with establishing a 0 decibel reference that is consistent and always referenced to a sine wave for every meter, regardless of its averaging characteristics.

Oh, I see you said "dB full scale", you're right---that term probably only refers to digital recording. But the principle that "0 dB" is 0 dB on EVERY meter with a sine wave has existed since time immemorial. This was only broken by some latter-day DAW-makers (like Wavelab, which has since mended its ways) who wanted to be technically correct about the 0.707 x peak = RMS with sine wave and they then equated that with the -3 dB point.

BK

[minister]

oh, right, sorry, i kinda mis-read your post :

Originally Posted by bob katz :
So if you encounter a "VU" or an "Averaging" or an "RMS" meter whose level reads 3 dB (or 2.94) lower than its peak level with a sine wave, please send a letter to the designer and tell him to read AES-17.

RMS vs PEAK. sorry... never mind me!

[screwtop]

Originally Posted by bob katz

On Sine wave whose peak level is 0 dBFS, the RMS level of that
sine wave is DEFINED to be 0 dBFS. This is also true of the AVERAGE level.

So if you have a square wave peaking at 0 dBFS, the RMS level should read OVER 0 dBFS!

Sorry for digging up an old thread, but I've only just become aware of this in the past few days, and it's really quite important, and different from what I thought I understood. Thanks again to Bob for the very clear explanation.

I admit I find it counterintuitive that the convention would equate 0 dB peak to 0 dB RMS for a sine wave, since that equivalence (while potentially useful) would only hold for signals with a 3 dB crest factor. Most audio signals are not isolated sine waves, so I'm not sure what the reasoning behind this definition is.

Is it correct to say that "dB FS" as defined in AES-17 is by definition an RMS measurement? That is, if someone says "-20 dB FS", you know that they mean "an RMS level of 20 dB below the RMS level of a full scale sine wave"? Or do you still need to specify peak/RMS? I'm sure I've seen dB FS used to indicate peak levels (perhaps sloppily).

Is there a correct way to signify RMS level in dB with respect to digital full scale, such that a full scale sine wave would actually read -3 dB? I think I may have seen "dB FS digital" or "dB re FS" used for this.

Lastly, what implications does this have for measuring and specifying crest factor? Does a -20 dB FS (AES-17 RMS) signal that peaks at -3 dB FS have a crest factor of 17 dB or 20 dB? I'd say it's the latter, even though it seems numerically strange.

I admit to being a little confused here.

[bob katz]

Originally Posted by screwtop

Sorry for digging up an old thread, but I've only just become aware of this in the past few days, and it's really quite important, and different from what I thought I understood. Thanks again to Bob for the very clear explanation.

I admit I find it counterintuitive that the convention would equate 0 dB peak to 0 dB RMS for a sine wave,

Those of us who come from the analog days of VU meters find the whole thing to be so intuitive and natural that we can't understand the reverse! When lining up a VU meter with a digital system we have been accustomed to using sine waves, set up 0 VU and align it with the dBFS level of choice. Automatically this is the same as the AES and IEC standards, which state simply: (and I paraphrase) "No matter what the averaging or detection method, the decibel level shall be the same with a sine wave signal."

I guess the sine wave was chosen as the one with no harmonics, most commonly available. It seems to work well as the standard. Obviously with pink noise, square waves, triangles, etc. your mileage will vary and it will be different. That would have been true no matter what the standardized test signal that was chosen. But the precedent is there, 60-70-80 years ago, sine wave equivalence for decibels was chosen, the standard is not new at all.

since that equivalence (while potentially useful) would only hold for signals with a 3 dB crest factor. Most audio signals are not isolated sine waves, so I'm not sure what the reasoning behind this definition is.

The reasoning is, first of all, long and very old precedent (more than 60 years... I can show you a European-made peak-reading test meter marked in dBm. Put in a sine wave whose RMS level is 0.775 volts and it displays 0 dBm. The other reasoning is "might as well pick sine wave"... the human ear doesn't respond to peak level after all. Regardless, complex program material doesn't resemble any standard artificial test signal, so might as well use the sine wave that we were accustomed to. Average responding meters came first, then true-RMS meters, then in the 1940's, quasi-peak meters. It was natural to calibrate the quasi-peak meter to read in dBm with the same sine wave that had already been defined as so many dBm.

Is it correct to say that "dB FS" as defined in AES-17 is by definition an RMS measurement? That is, if someone says "-20 dB FS", you know that they mean "an RMS level of 20 dB below the RMS level of a full scale sine wave"? Or do you still need to specify peak/RMS? I'm sure I've seen dB FS used to indicate peak levels (perhaps sloppily).

Hmmm.... it's predefined first as that any method of measurement should read the same in decibels as any other for a sine wave. This means that, for example, a sine wave whose peak level is full scale digital is called 0 dBFS (peak reading) and that same sine wave is called 0 dBFS RMS reading by the definition.

Is there a correct way to signify RMS level in dB with respect to digital full scale, such that a full scale sine wave would actually read -3 dB? I think I may have seen "dB FS digital" or "dB re FS" used for this.

Some people have tried to name this by various names, but they all are tongue twisters and potentially ambiguous. I've seen "RMS +3 dB", all kinds of stuff. It's a wild wild west out there.

Lastly, what implications does this have for measuring and specifying crest factor?

Crest factor is always defined using the standard for each metering method. So if you align your meters to the standard, then you measure the crest factor as the ratio between the peak and, for example, the average at a certain moment in time with given program material.

20 dB FS (AES-17 RMS) signal that peaks at -3 dB FS have a crest factor of 17 dB or 20 dB? I'd say it's the latter, even though it seems numerically strange.

It has a 17 dB peak to average ratio, the difference between the peak level and the RMS level, measured using the recommended methods. Again, in years past, that's the way we used to measure and set up VU meters in digital systems. If the VU was adjusted, it was adjusted with a sine wave whose VU level was set to 0 VU and which read -20 dBFS (for example) on the digital peak meter.

[screwtop]

Originally Posted by bob katz

Those of us who come from the analog days of VU meters find the whole thing to be so intuitive and natural that we can't understand the reverse! When lining up a VU meter with a digital system we have been accustomed to using sine waves, set up 0 VU and align it with the dBFS level of choice.

Ah - in the context of calibrating a system using a sine wave as the alignment tone, this makes a lot more sense to me now.

I'm so used to digital peak-reading meters that averaging level meters seem like the new and unfamiliar thing. I've never had the chance to use a VU meter in recording - it would probably be very instructive!

Originally Posted by bob katz

Hmmm.... it's predefined first as that any method of measurement should read the same in decibels as any other for a sine wave. This means that, for example, a sine wave whose peak level is full scale digital is called 0 dBFS (peak reading) and that same sine wave is called 0 dBFS RMS reading by the definition.

I see. So, for measurements of any signal that isn't a sine wave, one needs to specify whether it's peak or RMS. I like your usage "peak-reading" - plain "peak" could be a bit ambiguous.

Oh, I found Rane Note 169: "No Such Thing as Peak Volts dBu" interesting reading on this topic.

I think my confusion stems from the difference between "amplitude" and "level". In digital audio, it's easy and seems natural to measure the amplitude, since that's what every sample is: a measure of amplitude. I had previously thought that dB FS was a measure of (instantaneous) amplitude relative to the largest sample value supported by the system. I'd never have guessed that dB FS for RMS measures would be anything other than the root mean square of the amplitudes within a particular window.

Also, the Rane Note seems to confirm my impression that the AES definition of dB FS is an RMS measure. It suggests "peaks x dB above y dBu" to describe instantaneous amplitude, which I think would translate to "peaks x dB above y dB FS" in the case of digital. This is one of the tongue-twisters you mentioned above, Bob, right? So, for example, a full-scale sine wave would peak 3 dB above 0 dB FS, and would have a level of 0 dB FS on an AES-17 compliant meter, correct? And a full-scale square wave would peak 3 dB above 0 dB FS, and would read +3 dB FS. I can see why I found this confusing! At least the crest factors make sense with this scheme. Someone could write another article titled "No Such Thing as Peak Amplitude dB FS", for the benefit of digi-heads like me.

Just to confirm if I understand correctly, 0 dB above 0 dB FS is 1/sqrt(2) x full scale amplitude. Thus, a peak-reading digital meter should have a scale that goes up to 3 dB above 0 dB FS. Since 0 dB FS is 3 dB below full scale, 3 dB above 0 dB FS is itself full scale. Right, I think I get it now (but can't imagine it will ever be intuitive!).

From now on, I will try to think of "dB FS" as "dB RMS re Full-scale Sinusoid" and avoid using plain dB FS for specifying peak levels (except maybe for sine waves).

I wish there were a Gearslutz wiki/FAQ that could be a home to definitive explanations for some of these questions. There's such a wealth of knowledge and expertise here, and plenty of issues like this that are traps for young players (and that keep coming up in discussion).

[bob katz]

Originally Posted by screwtop

Ah - in the context of calibrating a system using a sine wave as the alignment tone, this makes a lot more sense to me now.

Oh, I found Rane Note 169: "No Such Thing as Peak Volts dBu" interesting reading on this topic.

That Rane paper is hard to follow. I can't tell for sure which position it is arguing for! Anyway, those of us who started in the analog days with VUs have always aligned things with the sine wave and then set the digital meter to a reference point and thus nothing has puzzled us about the ITU standard.

Just to confirm if I understand correctly, 0 dB above 0 dB FS is 1/sqrt(2) x full scale amplitude. Thus, a peak-reading digital meter should have a scale that goes up to 3 dB above 0 dB FS. Since 0 dB FS is 3 dB below full scale, 3 dB above 0 dB FS is itself full scale. Right, I think I get it now (but can't imagine it will ever be intuitive!).

That's essentially true. It would take an artificial test signal to ring past 0 dBFS peak reading but you are correct. And to add further confusion, there are now interpolating upsampling meters which can calculate what is being called the "true peak" reading of any signal that would be observed if seen on the output of a D/A converter, for example.

[screwtop]

Sorry, I didn't say what I meant here:

Originally Posted by screwtop

Lastly, what implications does this have for measuring and specifying crest factor? Does a -20 dB FS (AES-17 RMS) signal that peaks at -3 dB FS have a crest factor of 17 dB or 20 dB? I'd say it's the latter, even though it seems numerically strange.

I shouldn't have said "-3 dB FS" (tough habit to break!) - I actually meant "3 dB below full scale", i.e. "0 dB above 0 dB FS" or "~70.7% of full scale". I think (hope!) in this case we would agree that its crest factor is 20 dB.

Thanks again, Mr Katz.

[screwtop]

Originally Posted by bob katz

That Rane paper is hard to follow. I can't tell for sure which position it is arguing for!

I know what you mean! It's like "common knowledge says this" but "the AES says this". In that sense it wasn't helpful, but the explanation of the AES definition did help me.

And to add further confusion, there are now interpolating upsampling meters which can calculate what is being called the "true peak" reading of any signal that would be observed if seen on the output of a D/A converter, for example.

Ah yes, that seems like a nice idea. But wouldn't it need to take into account the analog headroom after the D/A converter (and other unknowns)? I think if we follow your recommendations on safe level practices, it shouldn't be an issue, at least not for music. Maybe for analog radio broadcast, though...