Our work was recently published in Optics Express and was highlighted as the Editor’s Pick. In this work, we present a method to fully characterize the transmission matrices of complex media using neural networks.
While similar methods existed, few of them could measure the relative phases between rows of the transmission matrix. Relative phases are necessary for coherent control of light after it propagates through given complex media, allowing their applications in optical networks, biomedical imaging, and quantum information processing.
Doing simple modifications to our setup and performing randomised measurements allows full recovery of transmission matrix using (what we call) multi-plane neural networks (MPNN). We show that our technique performs a much more accurate measurement as compared to the standard existing method of measurements on the same physical setup. Moreover, our technique is extremely robust to noise, retrieving a high-quality transmission matrix even when the measured data is majorly just noise (upto SNR =0.8)!
We also demonstrate the scalability of this method, to characterize multiple complex media simultaneously in a highly non-trivial and non-convex system.
You can read more about this work which is open access at doi.org/10.1364/OE.500529 . We have included all the codes, experimental and simulational datasets with the paper which can be found on Zenodo.
BBQ Lab 