Levels in Digital Audio

2014 Center for Lydteknik, written by Holger Lagerfeldt. No part of this text may be reproduced without permission.

www.centerforlydteknik.dk/links.html
About this Article
Unless otherwise stated, this article deals with floating point DAWs. This includes
sequencers such as Logic Pro, Cubase/Nuendo, Ableton Live, Pro Tools LE, etc.

Word Explanations
dBFS = decibels relative to full scale digital audio
0 dBFS = maximum possible level in digital audio before clipping occurs
A/D = analog to digital conversion (going into the sound card)
Bits = dynamic range in digital audio
Clipping = exceeding the available headroom and flat-topping the waveform
DAW = Digital Audio Workstation
D/A = digital to analog conversion (output from the sound card)
Floating point = calculations that allow for high internal dynamic range and lossless scaling
Headroom = amount of dB left before clipping
ITB (In-The-Box) = working in a computer based system
SNR = Signal-to-noise-ratio: level of a signal compared to background noise
Summing = adding individual tracks in a bus or master output
Unity = a default level setting where the signal is passed through unchanged

Bit Depth when Recording

Record in 24 bits instead of 16 bits for the highest dynamic range.

LEFT - 16 bit dynamic range (CD quality equivalent) is a theoretical 96 dB

RIGHT - 24 bit dynamic range is a theoretical 144 dB in comparison
NOT PICTURED - 32 bit floating point internal range (for overflow) exceeds 1,500 dB

The extra bits when working in 24 bits mean better resolution (or SNR), which is
especially audible with low level signals. But it also enables you to record at lower peak
levels with more headroom before clipping the input - without sacrificing sound quality.
Even though your target bit depth may be 16 bits (CD) or MP3 there is a lot to be gained
from using more bits during the recording, mixing and mastering process.

A Word about Meter Displays

Some DAWs cap the displayed overload value at e.g. +6 dB but the actual overload could
be higher. Also, some DAWs do not always display a - (minus sign) in front of dBFS values
due to space restrictions, but it is implied. Some DAWs disagree on how many
consecutive clipped samples constitute an over but this is of little practical relevance.

Levels when Recording

Record with a maximum peak of -3 dBFS to get the best dynamics and signal-to-noise
ratio during A/D conversion in your sound card or converter.
You can also record at lower levels, e.g. with a peak of -6 dBFS or -12 dBFS without any
perceptible loss in signal quality. Record too low though, and the quality begins to suffer.

A The peak level is too low - you are not taking advantage of the dynamic range
B Peaking at around -12 dBFS is recommended
C Peaking at around -6 dBFS is recommended, too
D Peaking at -3 dBFS is the recommended maximum peak level when recording
E The peak level is too high - you are likely to get irreparable artifacts or distortion

Recording very hot levels near the clipping point of 0 dBFS can result in the A/D
producing inferior results - even before it officially overloads the input. Notice that the
channel meter in a DAW will not necessarily show if you overloaded during recording. It
may show a 0 dBFS reading instead, no matter the amount of overload. So you need to
watch your levels before it is too late and always leave some headroom.

Debunking some Myths

Here are some examples of when overloading inside a DAW does not degrade or even
change the result - as long as you pay attention. This does not mean that overloading is
encouraged since other stages of the audio chain can be susceptible to overloads.
Myth 1: Overloading the output of a channel will cause irreparable damage to the audio!
Fact: There is no damage or even change to the audio if you lower the master fader until
the overload goes away on the master bus.

Overloading a channel by 6 dB and lowering the master fader by the same amount
makes the overload go away even though the channel output is still overloading. This is
possible due to the lossless scaling used in the DAW summing engine.

LEFT Channel fader overloads by 6 dB but the output fader recovers perfectly
RIGHT Channel fader down by 6 dB and the output fader scales the signal up perfectly
Both results are 100% identical in sound, despite the left example having an overloading
channel, and the right example is using an upscaled signal.
Myth 2: Using high levels on many tracks can cause problems in the summing engine!
Fact: While there are some definite workflow benefits to using lower levels during mixing
theres nothing inherently better sounding about using lower levels as long as you pay
attention to not overloading plug-ins or overloading the D/A.

Summing 60 tracks using low levels and then pulling up the master fader will result in
exactly the same result as summing the same tracks with high or overloading outputs and
then pulling down the master fader. Again, this is possible due to floating point math and
works in exactly the same way as described in myth 1. Even if you overload all 60 tracks
each by 12 dB you would not be able to exceed even half the headroom of the summing
engine.

Myth 3: Using low levels means better channel fader resolution!

Fact: Since a channel fader has most of its resolution at the top of the scale, the myth
stipulates that using high levels will force you to set your faders at the bottom of the scale
where resolution is worse. However, this argument for using low levels has little impact
since you can overload the individual channel outputs or the master output - and then
lower the master fader without any quality loss. Still, there are some workflow benefits to
using lower levels because you avoid having to use gain or trim plug-ins on the busses.

Levels Matter with some Plug-ins

Once you start using dynamic plug-ins without input attenuation the rules change a bit.
This is why gain structure still matters when working in all kinds of DAWs.

So overloading during mixing or processing ITB can happen in two ways:

A Overloading the channel output by pulling up the channel fader
B Overloading the plug-in chain internally due to an increase in level inside a plug-in
C Shows overloading in the plug-in chain but it is hidden due to a lowered channel fader
In the case of A, were not overloading the gain structure of the plug-in chain. This is due
to the fact that the channel fader is located after any processing occurs, and that the
overload occurs at this later point - something which can be rectified by lowering the
master fader.
In the case of B, the overload occurs inside the processing chain before the channel fader.
So any output overload in a plug-in will affect the input of the the next plug-in in the chain.
This can cause distortion or unpredictable behavior in some plug-ins, especially dynamic
plug-ins such as compressors and limiters. You can see the amount of actual overload by
setting the channel fader at unity, but you need to bypass any overloading plug-ins first.
Internal plug-in chain overload is not audible until you insert a plug-in that actually reacts
to the overload, usually a dynamic plug-in. And as example C shows, it is not necessarily
visible on the channel output meter either, i.e. if you lowered the channel fader until the

output signal went below the clipping point. So you need to pay attention to your output
level in each plug-in in your plug-in chain which is why lower levels from the start is smart.

Playback a sine test tone peaking at 0 dBFS on a channel. Insert a gain or trim plugin (or an equalizer) and boost the output level by 6 dB using the plug-in. This overloads the
channel by 6 dB internally, which will also show on the output meter if the channel fader is
set to unity level. Insert a Waves L1 or a similar dynamic plug-in that lacks input
attenuation. Notice how the plug-in will do 6 dB of gain reduction without the possibility of
changing this amount in the plug-in, effectively making the threshold unusable. This
experiment is repeatable to various degrees with several other plug-ins including other 3rd
party and bundled sequencer plug-ins.

Uncontrollable overload of a dynamic plug-in that lacks an input attenuation option

How to Avoid Overloading a Plug-in

The traditional solution is to use the output gain knob in the previous plug-in to lower the
signal being passed on to the next point plug-in. This is similar to the workflow of analog
mixing. There is no quality loss with this method.

A similar option is to attenuate the input of the overloading plug-in. However, not all
plug-ins have an input gain knob but most have an output knob, which is why the above
solution is usually better. Another option is to insert a dedicated gain or trim plug-in before
the overloading plug-in. Use the gain plug-in to lower the level of the signal before it
reaches the next plug-in. There is no quality loss with this method either.

How to Avoid Overloading a Bus or Master

If you are not using any plug-ins on the bus or master channel you can simply lower the
relevant bus or master fader to remove any overloads without problems.
If you are using plug-ins then the best option is to use a gain or trim plug-in as the first
insert on the bus, and use that plug-in to lower the level until it is safe again.

Levels when Processing with Plug-ins

Aim for a peak of around -6 dBFS on individual channels to be on the safe side, though
you are likely not to experience any problems using higher values. This value is not
necessarily the peak value shown on the output meter of the channel since that value can
be offset and hidden by a lowered channel fader, as explained earlier. Rather it is the
internal level of the signal chain which you can see by setting the channel fader at unity or
using a pre-fader metering option in your DAW.

The above suggestions make sense if you care about fidelity. Many professional
mixing engineers will routinely ignore this and clip on purpose using dedicated plug-ins, or
even intentionally overloading the master channel before the mastering process.
Overloading the master channel during the mixing process is not recommended since it
limits what the mastering engineer can do.

Levels when Outputting from a D/A Converter

When audio is output from the DAW to the D/A converter, the digital sample points are
reconstructed as an analog waveform in the D/A. The sample points do not appear exactly
on the peaks all the time which means that the actual analog waveform reproduced in the
D/A can be higher in amplitude, and potentially overload the output.

Extra D/A headroom is needed for some waveforms once reconstructed (lower row)
If you want to be on the safe side you should output a signal to your outboard gear that
peaks somewhere between -12 dBFS and -6 dBFS. This also has the benefit of matching
the input requirements of most analog gear, though some units will work better with even

lower levels. It also means that you can safely sample rate convert any exported file
without worrying about inter-sample peak overloading.

A sine wave can on rare occasions require up to 3 dB of extra D/A headroom to be

reproduced without distortion. However, artificially created square waves can require 6 dB
or even more D/A headroom due to high level inter-sample peaks. Clipping audio in the
mix can create something similar to square waves. The actual amount of audible distortion
during D/A conversion depends on the quality of the D/A. You can also use an intersample peak meter such as the free SSL X-ISM plug-in to watch for these problems,
though its rarely necessary as long as you follow the advice in this article.

Mix Output Levels

The D/A output level issue is also relevant to the final mix level. This has been covered in
more detail in the PDF document entitled Mixdown for Mastering, available for download
here: http://www.onlinemastering.dk/mastering-faq.html

The mastered mix is subject to the same potential D/A overload problems, in fact
even more so due to the often digitally limited or intentionally clipped audio in combination
with high RMS (average) levels. This kind of processing provides a very loud master but
the trade-off is potential and unpredictable distortion during playback on various D/As,
including sound cards, CD players and iPods. When bouncing or exporting a WAV file to
MP3 conversion you should leave at least 0.3 dB of headroom. This will avoid at some of
the potential distortion during the MP3 conversion process.

Normalizing
Normalizing can never improve the existing sound quality, and except for a very few
situations (usually when dealing with automated batch processing of files) normalizing is
an unnecessary procedure that should be avoided.

Some DAWs have a normalize option in the bounce window. If this function is
enabled your final mix will have its peak level normalized to 0 dBFS. So if you had 3 dB of
headroom in your mix before clipping the master bus, normalizing would raise the total
level by 3 dB. However, if your mix was overloading by 3 dB, floating point normalizing
would bring your mix back to a peak level of 0 dBFS without clipping, and in the process
making your mix 3 dB lower. Logic Pro also offers an option called Overload Protection
Only, which works like floating point normalizing except it will never raise the level, only
lower it if necessary. However, none of these functions are recommended since you
should watch your mix levels and leave some extra headroom for mastering.

Dithering
Explaining dithering is outside the scope of this article. It is always a good idea to apply
dithering when going from a higher bit depth to a lower bit depth. This is especially
important when reducing to 16 bits, where quantizing artifacts in the lowest bits can
sometimes be heard as low level distortion. If you are going to re-process (e.g. with EQ) a
bounced file with dither then choose a flat type of dither such as TPDF. Noise shaped
dithers should be reserved for the final master bounce.

Related Articles
Mixdown for Mastering - written by Holger Lagerfeldt
http://www.onlinemastering.dk/mastering-faq.html
Loudness when Producing and Mixing - written by Holger Lagerfeldt
http://www.gearslutz.com/board/music-computers/468170-loudness-when-producingmixing-tips.html
Inter-sample Peaks - written by Andreas Nordenstam, who also provided some pictures for
this PDF (thanks!)
http://www.gearslutz.com/board/tips-techniques/334385-intersample-peaks.html
-Overload in Signal Conversion - written by Sren H. Nielsen and Thomas Lund
-0 dBFS+ Levels in Digital Mastering - written by Sren H. Nielsen and Thomas Lund
-Level Control in Digital Mastering - written by Sren H. Nielsen and Thomas Lund
-Stop Counting Samples - written by Thomas Lund
http://www.tcelectronic.com/loudness/literature-glossary/
Level Practices - written by Bob Katz
http://www.digido.com/media/articles-and-demos.html
Dithering Guide - by Izotope
http://www.izotope.com/products/audio/ozone/OzoneDitheringGuide.pdf