Ultra-wide broadband and low-loss optical transformers for structured light

Optical transformations are widely used for image processing, optical communications,
quantum information and sensing. However, the design processing for these approaches
usually results in transformations that are only suitable for a few select wavelengths and are
implemented using inefficient diffractive optical approaches. The aim of the project is to
pioneer the development of a novel AI assisted approach to design cascaded optical transforms
made of multiple conformal mappings. This design approach will enable the creation of
wideband and low loss mode dependent optical couplers for integrated photonics and will
serve as the interface between free-space or fibre for a range of applications demonstrated in
both Quantum or Classical communication and sensing.

Multiple Plane Light Conversion (MPLC) technology has been demonstrated for a range of
applications including space division multiplexing, environmental simulations, and digital
holography [Fontaine]. MPLCs are computationally generated using wavefront matching
techniques, which are powerful approaches for the iterative generation of a unique set of masks
to map a set of modes at the input plane to a set of modes at the output plane. Fontaine et al.
have demonstrated the demultiplexing of 210 spatial modes using 7 phase planes with optima
performance at 1550nm with approximately 4dB higher transmission losses at the extremes of
the c-band [Fontaine]. The number of required planes has been hotly debated, where for general
transformations it is commonly considered that one needs at least N mask for the
transformation of N modes. However, in particular cases, the number of masks can be greatly
reduced if the input and output sets are chosen correctly. A key question in the research area is
what the optimal number of screens are. A further core issue with the wavefront matching
techniques, is they are inherently wavelength selective as the phase profile of each optical
element is iteratively varied and a loss function is minimised to produce the optimised set of
phase masks. However, this approach commonly leads to non-physical solutions that are not
valid for experimental realisation as these methods are based on Fourier propagation modelling
[Fontaine].

The project will develop a new approach in the design of MPLC optical systems, which will
utilise AI to determine the minimum number of conformal mappings to transform from the input
to the output. Hossack et al., outlined a proof that any single plane can perform a conformal
mapping in the far field of that plane [Hossack]. Cascaded conformal mapping could
subsequently transform any arbitrary input to any other output and in fact would determine the
theoretical minimum number of transformations required. When such mappings are well
defined, as previously demonstrated by [Berkhout, Lavery], two planes can be used to sort over
50 spatial modes. However, the determination of the correct conformal mapping is nontrivial for
arbitrary transformations. Utilising a physical propagation trained neural networks, which could
be designed to recognise possible combinations of transformations to achieve the desired
reshaping from a known input to a required output.

Once designed, these surfaces will be fabricated using diamond turning from an external
provider and utilised for hybrid communication and sensing applications in both fibre and in
free space.