arXiv: Multi-Plane Neural Networks

In our new preprint titled Referenceless characterisation of complex media using physics-informed neural networks, we show how multi-plane neural networks (MPNN) can be used to recover the complex transmission matrix of a commercial multi-mode fibre in a noise-robust manner, without using a reference field! We also show how the MPNN technique can be used to characterise a series of independent complex media, as shown in the figure above. This work will have many applications ranging from classical optical networks, biomedical imaging, to quantum information processing. As just one example, the MPNN technique forms a central part of our previous work on programming high-dimensional quantum gates inside a multi-mode fibre using inverse-design.