The hook
Back in lesson 0.4, feedback had one answer: pull the fader. That was the right answer for a survivor: instant, safe, undoable. But notice what it costs. The worship leader asked for more vocal in her wedge this week, and you couldn't give it to her, because two clicks more send and the wedge starts to sing that same low hoot it always sings. The fader-pull answer means the loudest her monitor can ever be is "just under the ring."
Here's the thing the survivor's answer hides: the wedge isn't on the edge of feeding back everywhere. It's on the edge at one frequency — that hoot — while every other frequency still has margin to spare. The whole system is being held hostage by its single worst spot. Remove just that spot, a sliver of one octave, and the level everyone actually wants comes back.
That's notching. It's the first EQ skill in this course that will make the band audibly happier by next Sunday.
The lesson
Quick recap of the machine (0.4, now with Level 4 eyes): mic hears speaker, speaker replays mic, and when the loop's round-trip at any frequency comes back as loud as it left, that frequency runs away. The key insight is at any frequency. The room, the wedge and the mic don't treat all frequencies equally; between the speaker's response, the mic's response and the room's reflections, some frequencies come back a few dB hotter than their neighbors. Push the whole system up and the hottest one crosses the line first, alone. That's why feedback is always a note, never a roar.
Step one: learn to name the note. When a ring starts, your first job is a rough address on the 4.1 map, by ear:
- A low, hollow hoot or howl — the classic wedge complaint — lives in the low mids, roughly 200–500 Hz.
- A singing, vowel-like tone, like the system holding a pitch, sits in the middle: 500 Hz–2 kHz.
- A whistle or sharp squeal is presence territory, 2–5 kHz.
- A thin, glassy sizzle at the top: 5 kHz and up.
You already trained for this: it's the sweep drill from 4.1 pointed at a live problem. Engineers who seem psychic about feedback are just fast at this translation.
Step two: the notch. On the ringing channel or, better, the monitor bus EQ (the ring belongs to the path, usually mic-to-wedge; cut it at the bus and it's fixed for every channel feeding that wedge):
- Take a parametric band, narrow Q (6–10).
- Go to the neighborhood your ear named.
- Cut 3 dB and sweep slowly through the zone while the system sits just at the edge of ringing. When you cross the guilty frequency, the ring dies mid-breath. Unmistakable.
- Settle the cut at −3 to −6 dB, as narrow as still works. Done.
A cut that narrow is close to inaudible on music and speech; you removed a hair's width of spectrum. But at the one frequency holding the system hostage, you just bought real, usable level.
Step three: ring out before the service. Don't wait to meet the hoot live. Empty room, band's mic in its real position, wedge at real angle: slowly push the monitor send past its comfort zone until the first ring starts. Notch it. Push again: a different frequency rings next (the second-worst spot; there's always one). Notch that. Maybe a third. Then stop and back the send down to performance level. You've claimed several dB of honest margin, mapped the path's worst spots, and nobody heard a thing.
The stopping matters. Each notch buys less than the last. The first removes a peak that stood clearly above the rest; by the fourth or fifth you're shaving spots basically tied with a dozen others, and spending audible tone on nothing. Three or four notches per monitor path is the working ceiling. If it still can't get loud enough after that, EQ is no longer the problem: it's mic choice, mic position, wedge angle, or the request itself (this is where Level 6's tools pick up).
Go deeperoptional
The loop condition, stated properly. The mic-console-speaker-room path is a feedback loop with a round-trip transfer function. Sustained oscillation at frequency f requires loop gain ≥ 1 (0 dB) and phase accumulation of an integer multiple of 360° at f, both at once (0.4's deep end, now with teeth). The room's frequency response is jagged, tens of dB of peaks and dips packed into every octave (comb filtering from reflections, speaker and mic response ripple), so as you raise system gain, whichever in-phase peak touches 0 dB first oscillates alone. Notching that peak by even 3 dB pushes it back below its neighbors. The next candidate is whichever peak stands next-tallest among the in-phase set, which is why the ring-out proceeds one frequency at a time.
Why you hear ringing before howling. Just below oscillation (loop gain approaching 1 from below), the system is a resonator with an effective decay time that stretches dramatically: energy at the critical frequency recirculates, losing only a fraction per trip. That's the "bloom" from 0.4 (reverb-like hang on one note), now explained: the pole of the loop transfer function slides toward the stability boundary, and decay time diverges as gain approaches unity. Practically, that stretch is your early-warning instrument. The ring-out procedure works because the hang is audible several dB before sustained howl, which gives you a controlled window to hunt in.
The economics of notches. Measurements and long practice agree on the budget: the first two or three notches typically reclaim 3–6 dB of usable gain-before-feedback in total, because they remove peaks that genuinely stood above the crowd. Beyond that, returns collapse: the response's remaining peaks are near-tied, so each new notch buys a fraction of a dB while carving audible holes in the tone. (A dozen deep notches can leave a wedge measurably quieter-sounding at the same feedback margin; you've spent tone to stand still.) Compare the alternatives: getting the mic 6 dB closer to the mouth (inverse square, 0.4) or choosing a tighter polar pattern aimed away from the wedge (2.5) each buy as much as the entire notch budget, for free, forever. EQ is the third-best feedback tool. It's just the one you can use from the booth.
Why notches age, and graphics vs parametrics. The room's response isn't static: temperature shifts move the speed of sound (about 0.6 m/s per °C), which slides every reflection-based peak slightly; a full congregation changes absorption wholesale. A hair-thin notch placed in an empty morning room can sit a few Hz off the peak by the second service. That's one reason working engineers prefer slightly wider-than-minimum notches (Q 6–8 rather than 16) as insurance, and why feedback suppressors that auto-place ultra-narrow filters drift out of date. The venerable 31-band graphic on the monitor bus embodies the same trade: its fixed third-octave bands (roughly Q 4.3) are wider than ideal for a single resonance, but fast, predictable, and wide enough to survive the room's drift. That's why it held the wedge-rig job for decades before parametrics on every bus became free (3.4).
Remember
- Feedback is always one frequency at a time: the system's tallest peak crossing the line alone. Remove the peak, and the whole system gets room to climb.
- Name the note by ear first: hoot = low mids, sing = mids, whistle = presence, sizzle = top. The 4.1 map is the weapon.
- Notch narrow and shallow: Q 6–10, −3 to −6 dB, on the path (monitor bus) rather than the channel when the ring lives in the wedge.
- Ring out before the service: push, notch, repeat, three or four times, then stop. Later notches buy tone-damage, not level.
- EQ is the third-best feedback tool. Mic closer and speakers aimed better still beat every notch you'll ever cut.
Your turn
- Ring out one wedge with a partner: three notches maximum, frequencies logged, send returned to performance level.
- Drill the naming reflex: next time anything rings, say the zone out loud before touching a knob, then verify with the sweep.
- Find where your console puts EQ on the monitor buses (parametric or graphic), so you're cutting the path, not just one channel.
- Revisit your 0.4 habits with this lesson's lens: how much of your feedback margin is being spent on mic distance and wedge angles that a five-minute stage conversation could fix?