This is essentially signal normalization. Sensor systems operate on a narrow band of inputs using non-linear components (like e.g. diodes). Adding noise is like

This is essentially signal normalization. Sensor systems operate on a narrow band of inputs using non-linear components (like e.g. diodes). Adding noise is like raising the DC voltage on a signal so that it shifts the signal to within your operating parameters. For example, for a transistor amplifier to work, the base diode needs a 1.4V difference to turn on and amplify a 10mV source signal. What makes white noise interesting is that it has a zero DC component (on average) by randomly adding energy as often as it subtracts, while passing through analog components (like a capacitors where R(s) = 1/(sC) in s-domain). In my view, this is still a form of amplification. Adding Gaussian noise around your source frequency is certainly source amplification because it's equivalent to attenuating the surrounding "noisy" frequencies for the benefit of your sensor equipment. This would increase the signal-to-noise ratio, but you have to know what the base frequency is you are trying to measure. I hope to be corrected by the signal folks, but it seems misleading to me that "the frequencies in the white noise corresponding to the original signal's frequencies will resonate with each other," when resonance does not occur in the signal, but in the resonant body of the sensor, and so this is more about sensor system calibration than boosting an unknown source signal. It seems that the practical implications are that you can have simpler sensors (that work in a narrower band) if you can cheaply control the input noise based on sensor reading.