negative feedback

Negative feedback is the trick that turns a wild, imprecise amplifier into a calm, trustworthy one. The idea sounds almost too simple: take a fraction of the output, flip its sign, and feed it back to fight the input. Picture steering a car on the highway — you don't compute the exact wheel angle in advance; you watch the lane, and the moment you drift, you nudge back. The correction is driven by the error itself, so the system constantly hunts toward where you want it. An amplifier wrapped this way stops obeying its own messy, temperature-dependent, part-to-part-varying raw gain and starts obeying the feedback network you built from precise resistors instead.

The deal you're making is to spend gain to buy everything else. A bare op-amp might have an open-loop gain A of 100,000 — enormous, but sloppy: it sags with temperature, drifts between chips, and falls off with frequency. Feed back a fraction β of the output and the closed-loop gain becomes A/(1 + Aβ). The product Aβ is the loop gain T, the amount of correction available, and when T is large the whole expression collapses to roughly 1/β — set purely by your resistor ratio, not by the unreliable A. That same large loop gain is what divides down distortion, flattens the response, widens the usable bandwidth, and stabilizes the input and output impedances. You threw away most of your raw gain on purpose, and got precision, linearity, and predictability back in exchange.

This is the quiet principle that makes analog reliable from imperfect parts. You can't manufacture a transistor whose gain is exact, but you can manufacture two matched resistors, and negative feedback lets the trustworthy ratio override the untrustworthy device. The catch — and the reason phase margin, dominant-pole, and Miller compensation exist — is that the fed-back signal arrives delayed. If the loop gain is still above one when the phase has slipped a full half-turn, your subtraction quietly becomes addition and the amplifier oscillates instead of settling. So feedback is never free: you design for enough loop gain to be accurate, and enough phase margin to stay stable while you do it.