dc.description.abstract | Multimode fibres (MMF) are miniaturised, flexible and high-capacity information channels
due to characteristics like small cross-section and a large number of spatial modes, promising
to open up new applications in endoscopic imaging and optical communication. However, the
transmission of high-resolution, spatially distributed information through standard MMFs in
real-time is still an open challenge, due to not only the complex mode dispersion and mode
coupling but also the highly variable MMF information channels over time, caused by
environmental changes, such as bending, temperature fluctuations and vibrations. Although
notable progress has been made in this field, such as resorting to optical wavefront shaping,
particularly transmission matrix, to overcome this transmission degradation, this technology
can be extremely complicated in practice and unavoidably deteriorated if the MMF
transmission channel has changed.
In this thesis, an effective solution based on deep learning is implemented to overcome the
long-standing challenge of the high variability and randomness of MMFs when being used as
information channels, leading to significantly improved resilience to MMF information
channel sensitivity, demonstrating highly scalable, high-spatial-density data transmission
through standard MMFs. Specifically, work is carried out on the following three topics: spatial
image transmission through an MMF subject to continuous shape variations using a deep
convolutional neural network, scalable calibration of a dynamically deformed MMF and selfadaptively cross-state focusing through a semi-flexible MMF enabled by a continual generative
adversarial model, and continuous transmission of spatially distributed information through
MMFs using a scalable confidence-based semi-supervised learning model.
The work in this thesis has successfully demonstrated novel improvements to information
transmission through MMFs, and underlined their potential for developing high-resolution and
flexible MMF endoscopes and extremely high capacity and reliable MMF communication
systems. The results of this thesis have enriched research contents on this topic and have
stimulated research interests in further developments in this field. | en_US |