The assumption of equal signal and pump attenuation coefficients is no longer required, while the iteration step-size can be significantly enlarged even under saturated amplification.

Obviously, you'd wonder why the iPhone wouldn't measure signal in a more linear fashion, where the decibel value for each bar is equal.

This determines the adaptation time of the analysis network ("forgetting factor"). The setting of this parameter should be approximately equal to the reverb tail length of the signal.