PART 1: Gain staging in general
What is gain staging?
You've probably heard of it. Those nit-picking audio geeks and hardcore engineers go on about it again and again, stressing its importance. But what is gain staging actually? Can someone please explain in simple terms?
Gain staging is awareness of audio levels at all crucial points in the signal path and adjusting the gain or attenuation at various points in this path to achieve the desired sound.
This path can be a chain of several analog boxes wired one after the other, or several plugins inserted on a channel one after the other. It's also the path of the source file through all the plugins, sends, the fader, submix buses, all the way to the master stereo bus.
Why is gain staging important?
In the most immediate sense, gain staging avoids nasty sounding clipping on one side and too much background noise on the other. But in reality, gain staging is much more than that. It's the delicate art of navigating all the non-linearities of the system, to take advantage of the useful ones and safely avoid the unpleasant ones.
It means to push the volume into the processors (plugins or hardware) that react pleasantly to a certain volume and leave enough headroom with the ones that are only happy "on the safe side". Tape recorders are a good example of such processors.
With that awareness in the back of your head throughout the recording and/or mixing process, you will achieve a better signal-to-noise ratio and warmth, impact and weight without hard-clipping anywhere.
How to go about it?
Know your levels! It's as easy as that, once you dig up the right information about all the parts of your system. And to make sense of that information, you have to know the basics. You know, the decibels. These are a unit of ratio of the signal, compared to the chosen reference. The chosen reference is, well, referenced by the additional letters, attached to the "dB". Let's take a look at the most common ones in audio:
Digital domain - as long as you're in the box
dBFS reigns in the box. It stands for "decibel (in regards to) full-scale" - a ratio of signal, compared to "1.0" floating point number. If that seems a bit odd, the missing information is probably "where did the 24-bit representation from the AD converter go?". It got converted from a 24/16-bit integer type value to a floating point type, where either full 24 or 16 bits from the AD converter reading represent a floating point number of "1.0".
The internal resolution of most DAWs is usually at least 32 bits (64-bit "double precision" is standard), but since most of the processing plugins have their parameters mapped to -inf to 0dBFS range, same as most of the DAW's meters, it's best practice to stay within these levels.
This way, you can be sure that all of the plugins that you use work within the intended range, which is especially important if they have non-linear transfer characteristics built into them. You have effectively done gain staging between the plugins.
Quick tip: Some of the "analog modeling" plugins have the normal working level quite a bit below 0dBFS. -18dBFS might seem scary at first in terms of signal resolution degradation, but in reality you're only scaling the exponent of the floating point number.
Transitioning between the two worlds
So how does all that translate if you want to use a hardware insert with your setup? Again, it's all about the reference. And the most common is dBu - level, relative to the root mean square voltage which delivers 1 mW (0 dBm) into a 600-ohm resistor. This amounts to ~0.775 Vrms or ~2.2Vpeak-to-peak.
But what's most important and useful info is what's your converter's point of reference in regards to the 0dBFS. This is largely dependent on the internal voltages of the circuitry around the DA converter, but ballpark values are somewhere around 10dBu for small, USB powered interfaces and anywhere between +14dBu to +24dBu for externally powered devices. This means that if your source level in the DAW reads 0dBFS, that's the analog level that the outputs are going to produce. This information can usually be found under "Technical specifications" on the manufacturers website, or in the user's manual.
Going out of the DA
When you're out of the DA at a know volume, then it's all about knowing your boxes! Some EQs have enough headroom to boost at a certain frequency even if the input signal is at +20dBu, and others will start to distort if you aren't careful and serve them no more than +4dBu, as the level inside the unit can reach the sum of the input level and the additional boost.
For example, if an EQ has internal headroom of +15dBu, boosting a +6dBu 1kHz sine wave input for +10dB with a bell filter, centered at 1kHz, will result in distortion even before the signal reaches the output stage, and no amount of lowering the output gain will solve it. The problem happens in the internals of the EQ!
Compressors can be even more dependent on the level, as one of the most important settings is usually the threshold. With compressors in particular it's important to apply good gain staging practices since they are in essence tools for gain management as well as sound shaping. Giving them a good level, but not too much will result in both great tone and utility.
And then there's transformers and tubes. Units with this kind of design can, most of the time, go from pristine transparency (usually driving them at around 0dBu) through warm and gutsy, to a sweet melting overdrive that still sounds nothing like the ugly digital clipping. That's why we love them and the levels are key to get them do exactly what we want.
Other references for dB
There's a couple more of those that you may come across, especially if you deal with older equipment. I will only list a few of them to get you acquainted and if you find that you enjoy their company, here's a link to most of them.
dBm stands for "decibel in regards to milliwatt", most often into a 600-ohm load. In that case, they equal dBu readings.
dBTP stands for "decibel true peak". This is used in digital systems when you're trying to approximate what's going to happen when the signal gets reproduced by the DA converter. In rare cases, the actual reproduced voltage could spike above the converter's reference and, in the case of a "too tight" design, cause unwanted distortion.
dBV stands for "decibels in regards to one volt". Finally, this one is pretty self-explanatory. No complication.
PART 2: Best practices - myths and common knowledge
The magic -18dB
There's quite a few proponents out there of keeping the levels inside you DAW at around -18dB and after reading a bit of theory above, it makes quite a bit of sense, especially if you're running a hybrid digital-analog setup. If staying at around that level, you can rest easy inserting all kind of analog equipment into the processing chain and you can be pretty sure that the level is going to be appropriate. It also makes using EQs a more intuitive process, when you don't have to fiddle around with input/output gain just to avoid clipping when boosting a narrow peak filter to "hunt" for an offending frequency.
Cutting is better than boosting
This is somehow related to the first one. If you end up with levels way above the -18dB in you DAW, cutting parts of the spectrum with an EQ will almost always* lower the volume of the track and end up closer to safer levels in general, whereas boosting will probably only bring you closer to 0dBFS.
* In rare cases, the cutting filter's transfer function will cause some of the transients to actually peak above the original level. You don't have to worry about it if you're in the "safe" range, but if you're doing filtering on the master bus where levels are usually pretty close to 0dBFS, it might be a good idea to keep an eye on the true peak value.
Plugins and DAWs sound better when not pushed
That's a geeky one, so proceed at your own risk.
This is not a question of how the saturating plugins react to the incoming signal, but a more subtle notion that some of them start to eat away at the detail in the sound when the levels get high.
This was a well-based fear in the early days of computer audio, when some of the software was using fixed-point arithmetic. In that case, boosting and attenuating the signal for large amounts in the same signal chain could result in the last few digits being lost in the process and causing distortion.
With practically all modern systems now using double precision floating point arithmetic to do the math, this phenomenon is not a problem any more. Even boosting or for ridiculous amounts like 100dB or more and attenuating again will not result in any loss of data. If you're really nit-picky, it is a fact that all real numbers can't be binary represented perfectly (1/3 for instance), but the errors that could emerge from that are magnitudes smaller than the thermal noise of any real world electric circuit, including even the most bad-ass audio converters available.
Nerd info: Even if floats are represented as 32-bit data, they in fact don't simply add another 8 bits of resolution on top of the 24 that a most common AD converter would output. They represent the numbers as 9 bits for a signed exponent of two and 23 bits for the multiplier, or mantissa. A more complete explanation can be found here.
Soft and hard clipping
So what's the deal with that clipping? Is it bad or not?!? It depends on the nature of it.
Very broadly speaking, clipping is a process of "compressing" the signal peaks with a fairly high ratio that at one point approaches infinity, without timing parameters. So with no attack and release parameters, the only property of the process that affects the sound is the transfer curve. Its symmetry, positive and negative polarity knee shapes and thresholds directly influence the type and amplitude of the generated higher harmonics when the signal exceeds the linear region.
Those characteristics come in all different shapes, sizes and flavors, but the general idea is somewhat like this: Symmetric transfer curves and softer knees sound more pleasant and asymmetric transfer curves with little or no knee sound harsh and "clicky".
As DAWs don't implement other than linear transfer curves for their channels and buses, the only clipping that can occur there is hard clipping. You can get away with going "into the red" a little bit when hitting the summing bus, but that's when the clipping of decimals takes place, so it's not exactly ideal.
But if you exceed the 24-bit value when going to the DA converter... that's hard clipping. And it will sound ugly. Gain staging here means to organize the levels of audio before that point happens.
Credit where it's due
A substantial amount of knowledge on the digital audio systems and their math came from a gentleman named Sâmm White, who kindly offered a good explanation on much of the key points about floating point math in the DAWs, after reading the original article. Sâmm, thank you!
PART 3: mix:analog equipment
Mix:analog I/O unit
To accommodate for what we think is the most useful range of levels, the AD/DA converters in the mix:analog system, the lovely BURL Mothership B80, outputs +18dBu at 0dBFS. So, if you were ever wondering - that is your output level if you hit 0 on the upper meter of the I/O unit.
At the moment, Rack1 at mix:analog consists of a pair of Pultec programme EQs, a Fairchild 670 vari-mu compressor/limiter, a GML/Sontec inspired fully parametric EQ and a MOS/J-FET Limiter.
Pultec Passive EQs
The Pultecs have an internal headroom of ~22dBu, which means that boosting a moderate amount after outputting very hot out of the DA can cause internal saturation that, unfortunately, is not on the most pleasant side. Its gain make-up stage after the passive filters is a transformer-coupled tube amplifier with another transformer for the output. It will start to saturate in a pretty subtle way when driven a bit hotter, but break pretty abruptly when it reaches the ~22dBu headroom level.
Gain staging for the Pultecs means watching your D/A output levels and keeping in mind that at full blast, there is only about 4dB of actual headroom left before saturation kicks in.
Fairchild 670 Compressor
The Fairchild 670 compressor limiter was designed way back in the days of vinyl-cutting, which means it was originally intended to work at input levels from around 0dBu to +6dBu. The input level control is actually a passive attenuator for the 12dB fixed gain amplifier that follows, so to keep it at unity gain, the approximate setting is about "2 o'clock".
Putting all that together, driving this unit at around +6dBu with the input level control at the default setting will result in the threshold control operating in the most useful range little above the half-way setting. Pushing the Fairchild hard will provide some really subtle thickening, but since it runs at insane voltages, hitting the headroom and achieving serious distortion with this box is nearly impossible.
Gain staging with the Fairchild is not only about the input gain level but also about the threshold and ratio because it does not have a direct output level control - the combination input gain, threshold and ratio command what sort of level will be coming out of it and into the processor after it.
Precision Series Mastering EQ
The Precision Series Mastering EQ has an internal headroom of +24dBu. If you exit the Fairchild pretty hot, be sure not to do too much boosting EQ, as it's capable of 10dB of boost on each band. If you still want to do some work and keep the high levels, try achieving the effect by cutting what's "too much" instead of boosting.
Precision Series Brickwall Limiter
Last is the MOS/J-FET Precision Limiter. Its internal headroom is around +22dBu so it's best practice to achieve the peak reduction with lowering the ceiling (clockwise), rather than cranking the input gain. If you want to exit at high volume to drive the AD converter nearly perfectly, use the output gain control, as it provides plenty of gain - enough to clip the converter inputs if you would like to get that extra 1dB of loudness!
Gain staging the precision limiter revolves around managing its internal headroom with the ceiling control and the input gain, while optimizing the level for the next processor with the output controls.
The most important information here is what level the "0" on the threshold dial corresponds to. As you might have expected, it's 0dBu! But since this is a soft-knee compressor, it will start to slowly grab at your transients well below this level at lower ratios, but very close to it at the highest, 1:10 ratio. The internal headroom is ~22dBu, but the ratio setting range is probably going to be pushing you lower the DA output level to somewhere around -10dBFS/+8dBu.
With the settings maxed out and the input level as hot as the DA can produce, this box will show some of the well-known English solid state console character. It's most often described as tightening the lows/low-mids and gluing the mix together even with minimal compression - "just touching the needle".
Our beloved tape machine, the Telefunkem Magnetophon 15, is calibrated to output +6dBu at "zero VU", the reference level on the special calibration tape. This means that driving it from -18dB to -12dB, with unity gain on the analog volume control unit, it will perform "to specification".
Clean, undistorted, with decent signal-to-noise ratio. But push it harder, and magic starts to happen in the low end. Kick drums gain a couple kilograms and start "molding" around the bass guitar or low synths, and snare drums get a certain snap that no other processing delivers.
If you utilize the analog gain unit after exiting the DA converter already pretty hot, you can exceed +20dBu and really push the tape. But don't expect it to break down just like that, that's proper German engineering right there, with loads of headroom and very pleasant saturation well before things would get ugly.
1176 rev. A
The trick with this box is pretty simple - drive it as hot as possible until you get too much compression, and then back off a little bit. This early revision does not have the "LN" - low noise circuit modification implemented yet, so it can get a bit noisy if driven at too low a level. It also helps to "strip silence" or expand/gate the source track before hitting this box, as it will, especially in "all-buttons-in" mode, pull out even near-inaudible noise, present in the source file.