Levels and gain staging

[wooden]

Hi people I get to the new house at last.

 

And of course I have some questions

 

1. Could you explain me about VU I know a little bit about dBs and the different scales but nothing about VU

 

2. How to set an appropiate gain structure? For example if I plug a keyboard into the mixer i put the fader at 0 dB then i move the key's volume until i get the loudest possible sound, the I move the mixer's fader until it is sending a good level to my sound card... but i have no clue how to set the master fader appropiately and I just mess with it, but also i want to do it right , i mean knowing the basics of gain staging. Like if you connect a pre amp to an input in your interface and how much gain to apply in the pre amp and how much in the interface. Thanks.

[Phil O'Keefe]

Hi Wooden - glad to see you made it over.

 

Yes, gain staging is VERY important... but let's try to use your setup as a real world example... but in order to do that, we're gonna need to know your exact signal paths. What mic, preamp, compressor, EQ, interface are you using? IOW, please give us ome specifics on what you're using...

[Phil O'Keefe]

BTW, in a nutshell, the idea is to start at one end of the chain, and optimize the levels / gain that feed into the next device / stage of the chain so that you're giving it a healthy signal (which helps to give you a better S/N ratio), but not overloading it (causing it to distort) either. Repeat with each subsequent stage...

[wooden]

Well i dont have much equipment but lets say a chain like

 

mic > pre amp > compressor > EQ > ADC

 

Ca you explian a little bit more?

[Phil O'Keefe]

Mic into pre... set the preamp for a good level. The idea is to not set it so high that it's clipping, but not so low that you're not getting a good level to your DAW. I normally track in the -18 / -15 dBFS range. Does your mic preamp have any metering on it? A "clipping" indicator? Just one "gain" knob, or two?

 

Mic pre into compressor. Adjust to taste, and then adjust the output gain control so that you are not getting a level increase or decrease when you switch between bypass and active on the compressor.

 

Compressor into the EQ... watch the boosting - EQ boosting can significantly increase the output levels! If your EQ has a make up gain control or output trim (volume knob), the same rules apply to it as they do to the compressor... adjust the EQ until it sounds the way you want, then toggle the bypass and adjust the output level so that it's the same either way - bypassed or active.

 

You Do know that you can put the mic pre into the EQ and then into the compressor, right? Try it both ways and see which one you prefer the sound of - EQ post compression or pre compression. Personally, I normally prefer compression to be after the EQ, but there's no rule that says you HAVE to do it that way.

 

If you can tell me what specific products we're talking about, I can give you more specific directions / suggestions.

[wooden]

Thanks phil!!

 

Well, like I said I just had a sm 57 running through a little soundcraft mixer to the interface to the daw.

 

Do you record so the level in you daw is at -15 -18 dBFS? wow!!! I have been recording as close of the zero as i can but thanks for telling me, that is another question i was going to ask you. Can you tell me why do you choose this range?

 

Thanks phil, you are very kind for expaining me all this.

[Kiwiburger]

That's assuming you are recording at 24 bits. If you are still using using 16 bits, I would suggest tracking as hot as you can.

 

Personally, I still track as hot as I can. With digital, the resolution of your sound improves right up to 0dB - and then, splat - clipping. But peak meters are accurate - if you aren't getting clipping then why settle for less resolution?

 

I personally don't agree with tracking at around -18 dBFS, but I can think of some reasons why others might:

 

The Protools TDM effects bus - I might be corrected here, but I believe it's still 24 bit fixed. So if your effects need headroom then you have no option but to give them some headroom.

 

I use Cubase SX myself, which has 32 bit floating throughout - so it's actually impossible to clip internally. You can whack all your tracks up to +24dBFS and they won't clip. (Unless you send them to your 24bit bit D/A converters, which will clip).

 

My thinking is that if you are going to use compression on a track - which lowers the resolution even more - why not start off with as much resolution as possible. Adding make up gain is empty bits - it gets it louder again, it but can't restore lost detail.

 

Obviously if all your tracks are peaking at around -6dBFS, you are going to have to attenuate some so the master bus has some headroom left for the mastering guy.

 

As long as you don't clip, I can't see it is possible to track 'too hot' with digital. You can always throw resolution away, but you can never retrieve it.

 

Things might change if integrating with external hardware which has been calibrated to allow for the sweet spot of each unit.

[The Audio Cave]

 

Originally posted by Kiwiburger With digital, the resolution of your sound improves right up to 0dB - and then, splat - clipping. But peak meters are accurate - if you aren't getting clipping then why settle for less resolution?

 

This has been beaten to death but most daw meters are not accurate and cannot by design represent the true peak value of the actual analog waveform they're representing. Knowing that, leaving a few db of headroom and setting the target a little lower is good practice. I shoot for -6 when tracking and mixing. The following paper explains why.

 

The Consequences of Traditional Digital Peak Meters

 

Also.. the "resolution" of the sound does not improve as you get closer to 0. It just gets louder. At 24-bits a-10 (for example) mix will be sonically identical to a -1 (non-clipping) mix, just not as loud. Hitting 24-bit digital really hot for more "sound resolution" is a myth. If it sounds "better" it's because it's louder maybe?

 

Lawrence

[Badside]

 

Originally posted by The Audio Cave Also.. the "resolution" of the sound does not improve as you get closer to 0. It just gets louder. At 24-bits a-10 (for example) mix will be sonically identical to a -1 (non-clipping) mix, just not as loud. Hitting 24-bit digital really hot for more "sound resolution" is a myth. If it sounds "better" it's because it's louder maybe?

 

 

Technically, yes the -1 waveform will have more resolution than the -10. But realistically, once you're in 24-bit, this is not a difference big enough for anyone to care. Remember that an audio CD is 16-bits and 24-bits is 256 times more resolution! Since you're gonna be butchering that waveform through further processing and mixing, you need as much resolution as possible (which is why I would never record in 16-bit for multi-tracking), but if you end up using only half the available headroom, you still have 128 times more resolution than a CD.

 

Problem is, multi-tracking in 16-bits was so bad (once you stack 24 tracks and run them through compression/reverb/delay/chorus/etc, the small discrepancies add up and your recording sounds like 8-bits!) that people would do anything possible to maximize resolution. They simply continued doing the same thing when technology improved! And since recording "hot" was the thing with 2" tape as well (depending on the desired result), this thinking is anchored in our minds.

 

Reality is that, digital clipping is the most annoying form of distortion, so screw that 0.1% difference in resolution and make sure you have all the headroom available at the ADC.

[Kiwiburger]

 

This has been beaten to death but most daw meters are not accurate and cannot by design represent the true peak value of the actual analog waveform they're representing. Knowing that, leaving a few db of headroom and setting the target a little lower is good practice. I shoot for -6 when tracking and mixing. The following paper explains why...

 

The Audio Cave - I both agree and disagree with your comments. I too shoot for -6 when tracking. Phil was saying he shoots for -15, and I was trying to explain why I personally aim higher.

 

Most DAW peak meters are highly accurate. The only possible area of debate is whether you decide that 3 consequative full digital blacks constitute an over, or whether you register a clip at only 1. Also there are two camps when you come to average rms scales. But a DAW will definately tell you if you have clipped - and if you havent clipped then you have the highest possible resolution undamaged waveform.

 

You referenced a document that refers to CDs and DVDs. These simply stream data to converters with little or no DSP. I fully understand that the peak meters on these things may well be inaccurate - but i'm talking about DAWs such as Cubase, Protools etc. I am sure the makers of these DAWs would disagree with your idea that their meters are inaccurate. With DSP, there is simply no reason for them not to be accurate.

 

You have caused me to think of an important issue though: the analog side of consumer CD's and DVD's. It's probably fair comment that the last 6dB of these devices is probably very dodgy in the analog circuity. Probably a lot of distortion there, not due to the digital data but due to the crap analog chips in the converters.

 

Which raises an important question: why the hell is the record industry cramming all their new music into the last 6dB of the CD media?

 

That's a rhetorical question btw ... we all know they are greedy and stupid.

[Phil O'Keefe]

The only possible area of debate is whether you decide that 3 consequative full digital blacks constitute an over, or whether you register a clip at only 1.

 

Digital black is zero signal. I think you probably meant to say "three samples".

 

Personally, I avoid overs like the plague.

[Kiwiburger]

Thanks Phil - temporary brain fade. I didn't mean digital black at all - i guess I meant digital white.

 

What's the word for when you've maxed out all your bits? Digital zero I guess. Most peak counters consider three of these in a row must be a clip.

 

You're right - we shouldn't be getting anywhere near there. I'm just saying I shoot for -6dB max. If there are advantages in going lower, I'd like to know about them, but i'm not seeing them at my end.

[Phil O'Keefe]

No problem Kiwi. I figured that's what you meant.

 

The old Sony 1630's had an over standard of three samples at 0 dBFS (dB Full Scale).

 

As far as levels, we need to make sure we're all on the same page here. I shoot for a -15 dBFS "average" level (RMS), but I will almost certainly have peaks that shoot higher than that. If you're talking about tracking with your peaks at -6 dB, then we're on the same general page... but if you're printing at -6 dBFS with your average levels, IMO, you're probably printing too hot. But hey, if it works for you, who am I to argue?

[Kiwiburger]

Ah .. I think that explains our apparant differences ...i'm definately talking about peaking at no more than -6dB. I'm talking Peak and you are talking RMS.

 

I admit i'm a bit hazy on the correct terms here. Most software DAW's I know, by default, have peak level meters. They can hold the highest peak, so it's easy to tell.

 

I guess the original poster did ask specifically about VU, which is a specific RMS scale. But then went on to mention soundcard and DAW and I presumed the discussion was about the peak level meters in the DAW. 0dbFS means Full Scale, refering to digital.

 

When talking RMS power in a DAW, it gets really messy. Apparantly different DAW makers use different scales. I understand that Cubase use the mathematically correct scale, where a sine wave peaking at 0dB gives a -3dB RMS reading. But I believe some audio pro's, including Bob Katz with his K system, use a different convention where a sine wave peaking at 0dB gives a 0dB RMS reading. Correct me if I'm wrong - but i've seen some debates on this by people much more clever than me, and it's very confusing.

 

Long story short - if your comparing RMS between Protools, Cubase, Sonar, Logic etc - I have NO IDEA how they compare, and whether you have to add or subtract 3dB or not.

 

Which is why I just use Peak - I can understand peak ...

[gsHarmony]

 

Technically, yes the -1 waveform will have more resolution than the -10.

 

 

Ok, now I'm confused. I thought the resolution was constant over the entire dynamic range. In fact, I thought that resolution was defined as decibels per bit. I had always thought that maximizing levels was to improve signal to noise ratio. Apparently, I am missing something. Could someone explain where I am going wrong?

 

The only thing I can think of is that the resolution changes to match the logarithmic decibel scale, but I have no idea why it would be done this way.

[wooden]

 

Originally posted by gsHarmony Ok, now I'm confused. I thought the resolution was constant over the entire dynamic range. In fact, I thought that resolution was defined as decibels per bit. I had always thought that maximizing levels was to improve signal to noise ratio. Apparently, I am missing something. Could someone explain where I am going wrong? The only thing I can think of is that the resolution changes to match the logarithmic decibel scale, but I have no idea why it would be done this way.

 

 

I think (and i am only guessing) that they say a hotter signal has more resolution because hotter signals needs more bits to be represented thus using almost all of the bits available. Lets suppose that your siganl need 12 bits to be represented, and then you reaise the level so now it needs18. In the first place you had 12 bit representing nothing, like saying the first 12 bits of your sample would be zeros, then you just have 6, so you are using "more resolution" but all of them are samples of 24 bits.

[Kiwiburger]

A similar analogy is to consider digital graphics. Take a photo with a 8 Mega Pixel digital camera. If you printed that picture on a poster, it should look ok. But if you scaled that down to a small 100kb jpg - it would look great on your PC screen, but when you print it out on a poster, it would look like Picasso during his cubic period. Extreme pixelation.

 

So the moral with digital pictures is very simple: never throw away resolution. You can always make a small picture bigger - but if the quality has gone, it's gone forever.

 

It's not that different with digital audio.

 

Imagine a pure sinewave generator putting out an analog signal with 1V peaks. With analog gear, you can multiply that analog wave form by any factor, or multipy by any factor, and still essentially get a sinewave. Maybe in a tube preamp that sinewave would be represented by 300V peaks. Maybe in an IC circuit that sinewave would be represent by 0.5V peaks.

 

With analog, you can make a signal smaller, and then amplify it again without losing significant detail.

 

Once that signal has been digitized, everytime you make it smaller in the digital domain, you lose bits.

 

It's like turning your picture of a sine wave into a connect-the-dots picture. Then when you make it smaller, you throw away a lot of the dots. Then - when you make it bigger, you don't get the dot's back that you threw away. So yes, you can make it bigger, but it will get squarer too.

 

If you record at 24 bits to start with, you can actually afford to throw away a lot of bits without noticing any drop in audio quality. But in the course of a mix, these tracks could have multiple gain changes. Every time you lower the gain, you throw away resolution. Every time you raise it back again, you aren't actually adding resolution - you are just adding empty bits.

 

So - in my opinion - with digital audio you should try to avoid gain changes as much as possible, especially multiple lowering and raising the gain.

 

Compression is an example of lowering the gain. Usually a compressor also adds makeup gain, because often the purpose of compression is to ultimately maximise the volume. So unnecessary digital compression can be lossy. In my view, automating the channel fader can achieve better control of dynamics with less loss of resolution.

[wooden]

So I guess i was right.

 

The thing is, if you record your signal using a level that requires 8 bits and then in the mix you raise it to a level that requires 20 bits, those extra bit will give the pixelation or degradation of the signal because they werent there at the first place.

[Kiwiburger]

With digital audio, binary zeros and ones, each bit represents a doubling in volume, which is 6dB.

 

If you record you audio using only 8 bits of the available 16 bits, then the damage is already done. The waveform will be 'squarer' than if you had recorded using, let's say 15 bits (leaving that last 6dB of headroom).

 

Gain is a multiplying calculation - so if you make your 8 bit grainy sound louder, you are just multiplying the numbers, and introducing rounding errors too. There is nothing in the calculation that can figure out what the original waveform might have looked like.

 

To use an extreme example - you might record a sine wave with say 4 bits and it ends up looking like a square wave. If you boost or normalise that wave, it will just end up being a big square wave.

 

Consider that 16 bits fullscale = 65535 and 24 bits fullscale = 16777215.

 

If you use the "join-the-dot" analogy, 24 bits gives you 256 times as many dots to describe your waveform. For only a 50% increase in data size. It's a no-brainer to record at 24 bits wherever possible.

 

 

 

 

 

 

[gsHarmony]

 

Lets suppose that your siganl need 12 bits to be represented, and then you reaise the level so now it needs18. In the first place you had 12 bit representing nothing, like saying the first 12 bits of your sample would be zeros, then you just have 6, so you are using "more resolution" but all of them are samples of 24 bits.

 

 

I'm not sure if this is right. I think you are always using all the bits, some of the them just happen to be zero. The only reason it seems that they are not being used is because of the encoding scheme (0000, 0001, 0010, 0011, etc.). There is absolutely no reason why you couldn't use a different encoding scheme (for example, gray code) and get the same exact same end result, in which case the lower levels would not have zero's in the high order bits.

 

I will try to post a picture from a DSP book sometime tommorrow to illustrate what I am thinking.

[Kiwiburger]

I think the correct term is "effective bits". In a DAW like Cubase, your internal audio is always 32 bits. But you may have seriously under-recorded your audio and maybe only 'effectively' using 8 bits.

 

Cubase even has feature for measuring the effective bit depth of a wave file. For example, people could take 16 bit samples and convert them to 24 bits and try to sell them as 24 bit samples. You can see if they really are 24 bit samples, or if they are just converted 16 bit samples ...

[gsHarmony]

In terms of voltages, resolution is defined as the voltage range divided by the total number of quantization levels (2^24 in the case of 24 bit recording). This resolution is not dependent on the measured voltage and therefore is constant for any measured voltage. Replace voltages with decibels and you have the resolution of audio at a given bit depth.

 

Therefore, regarding the digital picture analogy, a lower resolution picture would be the equivalent of recording at a lower bit depth (not just using a lower level). A lower level in the spatial domain would correspond to a different color or brightness depending on what is being sampled, not a lower resolution image. Saying that a lower level effects the number of bits used is like saying that oranges pixels take less bits than green pixels in a bitmap. All colors require the same number of bits, just like all decibels levels require the same number of bits in audio.

 

Of course, this does not mean that you will always get the same waveform for every possible level you record at, due to quantization error. The quantization error ranges from +/- 1/2 of the resolution. Depending on the level, the actual amount of quantization error will vary. However, there is no way of optimizing this by changing levels. The reason that 24 bit recording is so much better than 16-bit is that the quantization error is much smaller. This last part is what Kiwi was saying, just in different terms.

[gsHarmony]

 

For example, people could take 16 bit samples and convert them to 24 bits and try to sell them as 24 bit samples. You can see if they really are 24 bit samples, or if they are just converted 16 bit samples ...

 

 

This is true. The reason this wouldn't work is that the 16 bit samples would already form a waveform based on the quantization error from a bit depth of 16. When converted to 24-bits, the waveform would be identical to the 16-bit waveform, it would just use more bits for the same waveform.

[Badside]

 

Originally posted by gsHarmony Therefore, regarding the digital picture analogy, a lower resolution picture would be the equivalent of recording at a lower bit depth (not just using a lower level). A lower level in the spatial domain would correspond to a different color or brightness depending on what is being sampled, not a lower

 

 

You're mixing bit depth and resolution. In the colorful world of pictures, the 24-bit refers to the color of the pixel and resolution to the number of pixel (which would be like the bit rate). THe thing is that those 24-bit actually contain 3 different information: the level for 3 different colors (I'm guessing it's 3 groups of 8 bits, one for red, one for green, one for blue... I'm just guessing).

 

So each of those 24-bit words contains information about both level and color mixing. But the analogy still holds: if you record a darker picture, you still have less "resolution" of the color. Proof being: try to brighten that pic and it will look like dog s...

 

In the sound world, those 24-bit are purely there to represent the level of the signal. If the ADC converter receives the maximum voltage it is designed to received, all 24 bits will be equal to 1. If it receives 0 voltage they will be equal to 0.

 

So yes the 24-bits are always used. But it doesn't mean they are effectively used.

 

Let's imagine a 3-bit resolution, which gives us 8 different levels. Run a sine wave at half the available input level and you only get the 4 first levels (well the 4 middle actually), thus your sine wave is not as defined. The 3 bits are there, but only 2 are effectively used.

[wooden]

Originally posted by Kiwiburger I think the correct term is "effective bits". In a DAW like Cubase, your internal audio is always 32 bits. But you may have seriously under-recorded your audio and maybe only 'effectively' using 8 bits. Cubase even has feature for measuring the effective bit depth of a wave file. For example, people could take 16 bit samples and convert them to 24 bits and try to sell them as 24 bit samples. You can see if they really are 24 bit samples, or if they are just converted 16 bit samples ...

 

This is what i meant!

 

I know all the samples will be represented by the same amount of bits, but not all of them are effective. And that could (if you record underbited ) make the problem the other people say.