Negative feedback changes how references, disturbances and model errors reach the output.

For loop transfer function

$$L(s)=C(s)G(s),$$

define the sensitivity function

$$S(s)=\frac{1}{1+L(s)}$$

and the complementary sensitivity function

$$T(s)=\frac{L(s)}{1+L(s)}.$$

They satisfy

$$S(s)+T(s)=1.$$

A large loop gain $|L|$ makes $|S|$ small at frequencies where the feedback remains well behaved. This suppresses many plant-side disturbances and reduces the effect of moderate plant-model errors on the closed-loop response.

The same feedback cannot make both $S$ and $T$ arbitrarily small at the same frequency. Measurement noise often reaches the output through a path involving $T$, so aggressive high-frequency loop gain can amplify noise.

Feedback design is therefore a trade-off: strong correction where disturbances and uncertainty matter, without creating excessive noise amplification or loss of stability.