Gain Staging

Two recordings can be made with the same microphone, in the same room, at the same final loudness — and one hisses while the other is silent between words. The difference isn't how much gain was applied. It's where. Gain staging is the discipline of putting amplification in the right places, and it's the closest thing audio engineering has to pure arithmetic.

One volume, many knobs

Between your vocal cords and a Zoom call, the signal is amplified several times by several different devices. The microphone capsule turns air pressure into a few millivolts. A preamp — the circuit behind the mic's gain dial — lifts that up toward line level. The interface may apply a digital trim after conversion. The operating system has an input slider. The app you're talking into may add gain of its own. Each of these is a gain stage, and the chain of them is your gain structure.

Here's the part that surprises people: the total gain is not the interesting number. If your voice needs 40 dB of amplification to reach a comfortable level, you can get that 40 dB from one cranked preamp, or from a moderate preamp plus digital makeup, or from a dozen small boosts scattered through the chain. The final loudness is identical in every case. The noise floor is not — and that's what gain staging is about.

A discipline from the telephone plant

The mathematics of chained gain stages was worked out long before anyone owned a USB microphone. The Bell System's long-distance lines were a cascade of amplifiers — repeaters spaced along thousands of miles of copper — and engineers needed to predict how much hiss a call would accumulate along the way. In 1944, Bell Labs engineer Harald Friis published the noise-figure formula for cascaded amplifiers that radio and telecom engineers still use: the noise contribution of each stage is divided by the gain of all the stages before it. The consequence is blunt. The first stage in the chain dominates the noise performance of the whole system. Get the first stage right and later stages barely matter; get it wrong and nothing downstream can save you.

Broadcasting turned that insight into an operating tradition. Studios standardized on a reference line level — 0 VU on the meter standardized in 1939 by Bell Labs with CBS and NBC — and every console, tape machine and transmitter link was aligned to pass that level at unity gain: signal out equals signal in. Gain was added once, deliberately, at the preamp, and everything after it simply passed the signal along. When a chain misbehaved, an engineer could inject a line-up tone and find the stage that wasn't at unity. Fifty years later, the advice for a podcaster with a USB mic is the same idea in miniature: amplify once, on purpose, at a known place — don't let three devices each add "a little" without you knowing.

The noise-floor arithmetic

Gain is indiscriminate. A gain stage multiplies everything at its input — your voice, the room's air-conditioning rumble, the hiss of every circuit upstream — by the same factor. That gives gain staging its one governing rule: gain applied after a noise source amplifies that noise; gain applied before it does not.

Run the numbers on a typical desk setup. Say your speech peaks arrive at −40 dBFS and the combined floor of room tone and preamp hiss sits at −80 dBFS: you have 40 dB of signal-to-noise, fixed at the moment of capture. Add 28 dB of gain anywhere downstream and both numbers rise together — peaks at −12, floor at −52. The ratio never improves. Now suppose one stage in the middle of the chain contributes its own hiss. Every decibel of gain you move from before that stage to after it lifts that hiss, decibel for decibel, relative to your voice. Friis's cascade formula is just this observation written out for the whole chain.

There's a second, more practical reason not to solve level problems at one stage: analog circuits are at their worst at the top of their range. A preamp near maximum gain amplifies its own input-stage noise by the largest possible factor, and it drags up hum, power-supply artifacts and radio interference along with it. The quiet, linear middle of the dial is where preamps are designed to live.

Analog gain versus digital gain

The two kinds of gain in a modern chain are physically different things. Analog gain is amplification by transistors before the converter. Some of it is mandatory — a raw microphone signal is far too small to digitize well — and every decibel of it comes with thermal noise attached. Digital gain is multiplication of numbers. In floating-point processing it is essentially perfect: multiplying samples by 4 adds no hiss, no hum, no distortion. Its only limitation is honesty — it amplifies whatever noise was already captured, exactly as much as it amplifies the voice.

The old studio rule — "get your level in analog, digital gain is a last resort" — dates from early converters whose own noise floor punished quiet recordings. A modern 24-bit converter has more than 100 dB of usable dynamic range, and that changes the winning strategy. Record at a moderate, clean analog level with plenty of headroom, then make up the difference digitally. Better still, make it up digitally after noise reduction: once a processor has lowered the captured noise floor, digital makeup gain lifts a cleaner signal than the preamp ever saw. Late digital gain, applied after cleanup, is the one place in the chain where "just turn it up" is mathematically free.

Staging a voice for calls, podcasts and streams

For spoken audio the target is well established: speech peaks around −12 dBFS at delivery — loud enough to sit far above any remaining floor, with 12 dB of headroom below the digital ceiling for laughter and emphasis. The staging that gets you there cleanly: set the mic's physical dial to a moderate position where your voice registers healthily without the preamp working hard; leave every stage you don't control deliberately alone (operating-system boosts and per-app "auto" gains fighting each other are the most common source of mystery noise); and let one known stage at the end bring peaks to target. Then compression and automatic leveling have a clean, predictable signal to work with instead of a noisy one they can only make louder.

In DeskBroadcast

DeskBroadcast splits gain across three deliberate stages. First is the mic's physical dial: the app's guided calibration keeps it in a quiet zone rather than cranked. Second is the device's digital trim. Third is 0–12 dB of clean digital makeup gain applied inside the processing chain itself. The philosophy is exactly the arithmetic above: keep the analog preamp low, because cranked preamps add hiss and hum, and make the level up digitally after noise reduction, where gain is mathematically clean. From there, auto-level holds your speech peaks around the −12 dBFS studio target — one chain, staged once, on purpose.