Free-space optical communications are highly sensitive to distortions induced by atmospheric turbulence. This is particularly relevant when using orbital angular momentum (OAM) to send information. As current machine learning techniques for computer vision allow for accurate classification of general images, we have studied the use of a convolutional neural network for recognition of intensity patterns of OAM states after propagation experiments in a laboratory. The effect of changes in magnification and level of turbulence were explored. An error as low as 2.39% was obtained for a low level of turbulence when the training and testing data came from the same optical setup. Finally, in this article we suggest data augmentation procedures to face the problem of training before the final calibration of a communication system, with no access to data for the actual magnification and level of turbulence of real application conditions.
|Title of host publication||Laser Communication and Propagation through the Atmosphere and Oceans VIII|
|Editors||Jeremy P. Bos, Alexander M. J. van Eijk, Stephen Hammel|
|State||Published - 2019|
|Event||Laser Communication and Propagation through the Atmosphere and Oceans VIII 2019 - San Diego, United States|
Duration: 12 Aug 2019 → 14 Aug 2019
|Name||Proceedings of SPIE - The International Society for Optical Engineering|
|Conference||Laser Communication and Propagation through the Atmosphere and Oceans VIII 2019|
|Period||12/08/19 → 14/08/19|
Bibliographical noteFunding Information:
The authors thankfully acknowledge conversations with F. Bernuy on data augmentation techniques for deep learning. This work was partially funded by the Millennium Institute for Research in Optics (MIRO) and the Advanced Center of Electrical and Electronic Engineering (AC3E) (CONICYT/FB0008).
© 2019 SPIE.
- Laser beam transmission
- digital image processing
- neural networks
- optical vortices
- pattern recognition