Our work “A neural network assisted 171Yb+ quantum magnetometer” has been published in npj Quantum Information. We demonstrate the power of combining machine learning with quantum sensing by integrating neural networks into the data processing stage of a trapped-ion magnetometer, significantly extending its operational regime.
The Challenge
A versatile quantum magnetometer must deliver a readable response across a wide range of target field parameters. Standard atomic magnetometers operate in a regime leading to harmonic sensor responses, where RF target parameters are encoded in simple sinusoidal patterns. However, when the RF driving induces responses beyond this standard harmonic behavior — for instance, with large shot noise or in continuous single-shot measurement scenarios — the simple relationship between sensor responses and target parameters breaks down, making accurate field characterization extremely challenging.
Our Approach
We experimentally demonstrated that combining 171Yb+ atomic sensors with adequately trained neural networks enables us to investigate target fields in distinct challenging scenarios. Specifically, we characterized radio frequency (RF) fields in the presence of large shot noise, including the limit case of continuous data acquisition via single-shot measurements. The neural network learns the complex mapping between measured sensor responses and the underlying field parameters, effectively deciphering the information contained in non-trivial sensor responses.
Key Results
In the scenario of reduced measurement averaging, our neural network achieved an average estimation accuracy of 98.76% with only 100 measurement shots per data point. Even with as few as 30 shots, the accuracy remained at 98.31%. For continuous single-shot measurements, the regression between NN outputs and targets showed correlation coefficients exceeding R = 0.999. These results demonstrate that the hybridization of machine learning tools with quantum sensing techniques is a strategy that can be readily extended to other quantum platforms, such as nitrogen-vacancy centers in diamond.
Fig. 1. Energy levels of the 171Yb+ atomic sensor (a, b), the needle trap configuration (c), and the neural network architecture (d). The sensor response from the photomultiplier tube (PMT) is processed through a well-trained neural network to extract RF field parameters from complex sensor responses.
This work was a collaboration with researchers from the University of the Basque Country (UPV/EHU) in Spain.
Find the article here: https://doi.org/10.1038/s41534-022-00669-2