European Conference on Trapped Ions (ECTI)

A versatile magnetometer must deliver a readable response when exposed to targets fields in a wide range of parameters. In this work, we experimentally demonstrate that the combination of a ${}^{171}{\mathrm{Yb}}^{+}$ atomic sensor with adequately trained neural networks enables the characterization of target fields in challenging scenarios. In particular we estimate parameters of radio frequency (RF) driving fields with large shot noise, including the limiting case of continuous data acquisition with single-shot measurements, and in regimes where the magnetometer delivers a response that significantly departs from the standard harmonic behavior.