What is delay feedback?

Delay feedback in plain English

A delay holds on to what you feed it and plays it back a set time later. On its own that gives you one echo. Feedback takes some of that echo and sends it back into the delay's input, where it gets delayed again and comes out as a second echo. Some of the second echo goes back in to make a third, and so on.

The feedback knob decides how much goes back each time. Set it to 30% and each echo is 30% of the level of the one before it, so the trail fades fast. At 80% each echo keeps most of its level and the trail goes on for a long time. That one number sets both how many repeats you hear and how long the delay keeps sounding after the dry sound has stopped.

If you're new to delay, start with the delay lesson, which covers delay time, mix and the other controls. This page stays on feedback: the maths behind it, how long a tail lasts, what happens when you turn it all the way up, and the odd things it does at very short delay times.

Reading the lab

The screen to watch is the echo timeline at the top. Each bar is one repeat, placed at the moment it arrives and as tall as it is loud, from 0 dB at the top of a lane to −60 dB at the bottom. With feedback above zero the bars make a staircase going down to the right, and the slope of that staircase is the feedback: a steep one means each pass loses a lot of level, a shallow one means it loses very little.

The Right now panel counts the repeats louder than −40 dB and gives the time when the echoes fall below −60 dB, measured against the original sound. Both numbers include the mix knob, which sets the level of every echo, while the feedback sets how fast they fade. Hover over any bar on the timeline to read its exact level.

Ducking and ping-pong are locked on this page so the steps only change the controls that affect the feedback loop. The steps start on the vocal, move to the lead at 120 BPM when the timing needs a beat to measure against, and end on the drums.

How feedback sets the number of repeats

Feedback is a multiplier. Every time the sound goes round the loop, its level is multiplied by the feedback amount, so at 50% the first repeat is at full level, the second is half that, the third a quarter and the fourth an eighth. In decibels, halving the level is a drop of 6 dB, so at 50% every repeat is 6 dB quieter than the one before.

The drop per repeat in decibels is 20 × log₁₀ of the feedback as a fraction. You don't need to work it out by hand, since the table below has the common settings. What it shows is that the knob doesn't behave evenly. Going from 10% to 30% feedback changes the drop from 20 dB to 10.5 dB, which doesn't add many repeats, but going from 80% to 90% roughly halves the drop again, from 1.9 dB to 0.9 dB, and more than doubles the length of the tail. Most of the change in tail length happens in the top quarter of the knob.

You won't notice every repeat in that count. Once the drums and bass are playing, an echo 30 or 40 dB below the original is hard to pick out, so at 50% you'll clearly hear four or five repeats, then a few more if you listen for them in a gap. That's why the lab counts repeats above −40 dB separately from the time it takes to reach −60 dB.

The numbers on the knob don't travel well between plug-ins. Nothing obliges a maker to map the feedback control the same way, and some put more of the travel at the top end where the tail changes fastest. A setting of 50% on one delay may not sound like 50% on another, so count the repeats and listen to the tail.

How long does a delay tail last?

The tail is how long the echoes keep going after the dry sound stops. Engineers measure a reverb's tail as its RT60, the time it takes to fall 60 dB, and the same idea works for a delay. Take the delay time, multiply by 60 and divide by the decibels lost per repeat, and you get how long the echoes take to fall 60 dB below the first one. A 1/4 note at 120 BPM is 500 ms, and at 50% feedback each repeat drops 6 dB, so 500 × 60 ÷ 6 = 5,000 ms, a five-second tail.

The delay time is in that sum too, which is easy to forget. With the feedback left at 50%, a 1/8 note tail lasts 2.5 seconds and a 1/16 tail lasts about a second and a quarter, because the repeats arrive faster and each one loses the same 6 dB. A slower tempo does the opposite: at 80 BPM the same 1/4 note at 50% rings for 7.5 seconds.

So if you shorten the delay time and want the tail to stay the same length, raise the feedback. In decibels the rule is short: halve the delay time and halve the drop per repeat. A 1/4 note at 50% loses 6 dB a repeat, so a 1/8 note needs about 71% to lose 3 dB, and a 1/16 needs about 84% to lose 1.5 dB.

What happens at 100% feedback?

At exactly 100%, every repeat comes back at the same level as the one before it, so nothing ever fades. The echoes of one phrase keep cycling while new sound is added on top, and the level in the loop climbs as each new phrase stacks onto the old ones. Above 100% each repeat is louder than the last, so even a single click grows on every pass.

That growth has to stop somewhere, and what stops it shapes the sound. On a tape echo, the tape and the electronics saturate as the level rises, rounding off the peaks and holding the loop at the point where the loss from saturation matches the gain from the feedback. The result is the howling, self-oscillating sound dub engineers played like an instrument, riding the Intensity knob on a Roland Space Echo up and down. A clean digital delay with nothing in the loop to hold it back would keep getting louder until it clipped your output, which is why some plug-ins cap the feedback below 100% or put a limiter or saturation stage in the loop.

The lab stops at 90% so the repeats always die away, and it has a gentle saturation stage in the loop that only acts when the echoes pile up past full scale. At 85% you can already hear the build-up: a vocal phrase is still echoing when the next one arrives, and the echoes blur into a wash.

Runaway feedback can get very loud very fast. Turn your monitors down before you push a delay past 90%, and put a limiter at the end of the return channel while you're experimenting. The lesson on protecting your hearing explains why a few seconds of a howling loop at full volume is worth avoiding.

Some delays have a freeze or hold button that does something related. Ableton Live's Delay calls it Freeze: it stops new sound going into the loop and keeps what's already there repeating at full level. That's 100% feedback with the input shut off, so the level can't build.

Filters inside the feedback loop

Many delays put a high cut and a low cut inside the loop, so the echo goes through the filters on every pass. That makes the feedback different at different frequencies. Take a 3 kHz high cut with a 12 dB per octave slope, like the one in the lab: at 6 kHz it takes off about 12 dB. With 85% feedback the mids lose 1.4 dB per repeat, but at 6 kHz each repeat loses 1.4 + 12.3 ≈ 13.7 dB, so the top end of the echoes has gone after four or five repeats while the mids are still ringing.

The low cut does the same at the bottom. With a 250 Hz low cut in the loop, the bass and low mids drop away on each pass, which stops the echoes of a vocal from building up a muddy hum under the mix. Leave the low cut off at high feedback and the echoes can cloud the whole mix below 300 Hz.

Tape echoes behaved this way without anyone setting a filter. Every pass through the tape lost a little top end and picked up a little noise and saturation, so the repeats seemed to drift into the distance. The tape delay lesson covers that sound in full.

Filtering the loop also makes high feedback easier to use. With both filters in, only the midrange can sustain, so you can set 70 or 80% for a long ambient trail and the echoes stay out of the way of the bass and the cymbals.

Short delay times: feedback turns into a note

Below about 30 ms the repeats come too fast to hear one at a time. Add feedback and they arrive as a steady stream, so many a second that your ear hears the rate as a pitch. The pitch in hertz is 1,000 divided by the delay time in milliseconds: a 10 ms delay rings at 100 Hz, 5 ms at 200 Hz, and 2.27 ms at 440 Hz, the A above middle C.

The higher the feedback, the longer and more clearly it rings, because the stream of repeats takes longer to fade. A delay with feedback at these times is a comb filter: it boosts the frequency set by the delay time and every whole multiple of it, which is why the note sounds metallic. A flanger sweeps that delay up and down, with its feedback acting as resonance; the flanger lesson has it in a lab. Karplus–Strong synthesis, a way of making plucked-string sounds that Kevin Karplus and Alex Strong published in 1983, feeds a burst of noise into a short delay with a filter in the loop.

How to set delay feedback, step by step

This works for most delays on a send, which is how you'll set up a vocal or lead delay most of the time.

1. Start at 0%Set the delay time and the send level first, with one echo, so you know the echo sits in the right place before you add more of them.
2. Raise it in a gapLoop a part of the song with a gap after a phrase. Bring the feedback up until the echoes carry across the gap, which is usually somewhere between 20 and 40%.
3. Stop before the next phraseIf the echoes of one line are still clearly audible when the next line starts, back the feedback off or shorten the delay time.
4. Filter the loopBring the low cut up to 200 or 300 Hz and the high cut down to 3 to 5 kHz. You'll often find you can raise the feedback a little once the filters are in.
5. Check it in the full mixEchoes that sound right on their own are usually too long once the drums and bass come in. Listen to the whole track and trim the feedback there.
6. Automate the long tailsLeave the feedback modest for most of the song and raise it only where you want a long trail, like the last word before a drop.

Feedback settings for common uses

The tail column is worked out at 120 BPM with the formula above. At other tempos, scale it with the delay time.

Automating feedback and throws

You can give one word a long echo by automating the send or the feedback, and the two sound different. Automating the send puts only that word into the delay, at whatever feedback the delay is already set to, so you get a clean trail of that one word. Automating the feedback changes how every echo already in the loop behaves, so raising it holds on to everything that's currently echoing and lets it build.

Use the send for a classic echo throw on the last word of a line. Use the feedback when you want the whole delay to swell, like a dub break where the echoes of the last bar keep cycling and building while the drums drop out. Bring it back down before the tail gets out of hand.

Turning the feedback down to 0% doesn't silence the tail instantly. Whatever is already in the delay still plays once more, so the echoes stop within one delay time of the move. To cut a tail dead, automate the return channel's fader or the delay's mix as well.

Common feedback mistakes

Setting it solo

A long trail of echoes sounds lovely on its own and turns to mush once the rest of the track is playing. Set the feedback with the full mix running, and expect to end up lower than you would have guessed.

Forgetting the delay time

Feedback and delay time set the tail between them. Switching a delay from a 1/8 to a 1/4 note at the same feedback doubles how long it rings, and slowing the project tempo makes every tail longer too. If a tail suddenly feels too long, check the time before you blame the feedback.

Full-range echoes at high feedback

At 70% or more, the low end of every repeat stacks up for many seconds. A low cut of 200 to 300 Hz in the loop, or an EQ on the return, keeps the echoes out of the bass.

Boosts inside the loop

If your delay lets you put an EQ, saturation or another effect inside the feedback path, any boost there adds to the loop gain. A 3 dB bell at 1 kHz with 75% feedback pushes that frequency past 100%, and it will start to howl even though the knob says 75. Cut inside the loop and boost on the return.

Training your ears to hear feedback

The quiz under the lab hides an amount of feedback on a loop at 120 BPM, 15, 45 or 75%, and asks which one you're hearing. The controls and screens are covered and every round is level-matched, so the length of the trail is the only clue. A 15% round has two repeats you can pick out and is over in under a second, 45% gives a handful and about two seconds of tail, and at 75% the repeats run on for around five seconds, into the next note. There are eight rounds, and your best score is saved in your browser.

Away from the quiz, the habit to build is counting. Move the feedback knob without looking at the number, play the vocal and count the repeats you can hear after a phrase. Two or three clear repeats is around 25 to 35%, four to six is about 50%, and a trail you lose count of is 70% or more. Counting works on any delay, whatever its knob says, so it's the one to keep when you move to a plug-in you don't know.

Questions people ask

Keep going