Inverse design of a spatial demultiplexer for free-space optical communications: direct optimization over turbulence statistics
Nicolas Barré · 2026
Read Inverse design of a spatial demultiplexer for free-space optical communications: direct optimization over turbulence statistics by Nicolas Barré (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-18760.txt.
Introduction
Barré redesigns receiver optics for ground links from satellites. Air turbulence blurs incoming wavefronts and hurts coupling into single-mode fibers. A spatial demultiplexer spreads the field over a fiber bundle for coherent recombination. Barré first compares modal bases and finds mode spatial support matters far more than basis choice. A compact two-plane refractive system is then tuned over many turbulence draws. Gradient descent drives the design with no modal constraint imposed. The result beats direct fiber coupling and nears ideal modal projection across turbulence strengths.
Catalog record
| CREATOR | Nicolas Barré |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31514 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.18760
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Inverse design of a spatial demultiplexer for free-space optical communications: direct optimization over turbulence statistics Nicolas Barré Abstract Atmospheric turbulence severely limits the coupling of received optical wavefronts into single-mode fibers in satellite-to-ground free-space optical links. Spatial demultiplexing receivers address this challenge by distributing the incoming field across a bundle of single-mode fibers whose outputs are recombined coherently, relaxing the requirements on wavefront correction. In this work, we investigate the design of such receivers from two complementary angles. We first compare the power coupling statistics achieved by several modal bases and show that the spatial support of the modes matters far more than the specific choice of basis, questioning the relevance of mode-selective approaches for this application. We then present the inverse design of a compact two-plane refractive system optimized directly over an ensemble of turbulence realizations using stochastic gradient descent, with no constraint imposed on the input modal decomposition. The optimized system significantly improves over direct coupling into the fiber bundle, approaches the performance of an ideal modal projection, and remains competitive across a broad range of turbulence conditions. 1 Introduction Free-space optical (FSO) communications offer a promising alternative to radio-frequency links for high-throughput ground-to-satellite and inter-building connectivity [ 12 ] . A central challenge in FSO systems is the coupling of the received optical wavefront into single-mode fibers, which is severely degraded by atmospheric turbulence. Turbulence introduces wavefront distortions that reduce the spatial coherence of the received field, leading to coupling losses and signal fading that directly limit the link budget. Two main strategies have been developed to mitigate this effect. The first relies on adaptive optics, where a wavefront sensor and a deformable mirror correct the incoming wavefront in real time to restore single-mode fiber coupling [ 19 , 13 ] . The second strategy, which is the focus of this work, relies on spatial demultiplexing: the turbulent wavefront is distributed across a set of single-mode fibers, and the outputs are recombined coherently to recover the transmitted signal [ 5 , 6 , 9 , 7 ] . Two main device architectures have been explored to implement this spatial demultiplexing. Photonic lanterns, originally developed for astronomical instrumentation [ 18 ] , interface a bundle of single-mode fibers with the incoming multimode beam through an adiabatic waveguide taper [ 8 ] . Ultrafast laser inscription has since extended this concept to complex three-dimensional geometries [ 11 ] , improving robustness and design flexibility. Multi-plane light conversion (MPLC), on the other hand, realizes a fully controlled unitary transformation between a prescribed set of orthogonal spatial modes through successive free-space phase modulations [ 17 , 22 , 1 ] . A recent experimental comparison over a simulated GEO-to-ground FSO link has shown that a seven-port photonic lantern achieves on average 1\, dB lower coupling losses than a commercial 15-mode MPLC module, at the cost of higher signal fading [ 7 ] . The design of multi-element diffractive optical systems, in which a wavefront undergoes successive phase modulations separated by free-space propagation, has a long history predating the optical neural network literature. As early as 1993, Fienup established gradient-based inverse design of cascaded phase elements as a viable computational approach [ 10 ] . Wang and Piestun later applied equivalent functionality using cascaded planar elements optimized through iterative projection algorithms, in the context of volumetric holography and multicolor beam shaping [ 20 ] . Independently, the optical neural network community framed the same physical architecture as an end-to-end differentiable computing substrate, optimized for arbitrary objective functions [ 15 , 16 , 23 , 21 ] . Within this framework, MPLC can be viewed as a particular instantiation that uses phase masks as optical elements and restricts the transformation to a unitary mapping between a fixed set of orthogonal input and output spatial modes, a natural choice for telecommunications where mode orthogonality defines the independent channels that carry information. The approach presented here is fundamentally different: no input modal basis is defined, and no orthogonal mode decomposition is imposed. Instead, the optimization criterion is directly the total power collected into the single-mode fibers, averaged over an ensemble of atmospheric turbulence realizations, which is the physical quantity of interest in free-space optical communications. In this work, we approach the design of FSO spatial demultiplexers from two complementary angles. We first systematically compare the power coupling distributions achieved by several modal bases for turbulent wavefront collection, and show that the choice of detection basis has a limited impact on power collection efficiency, provided the spatial support of the modes covers the collecting pupil efficiently. We then present the inverse design of a compact two-plane refractive system optimized directly over an ensemble of atmospheric turbulence realizations using stochastic gradient descent, without any constraint on the input modal decomposition. The optimized system approaches the performance of an ideal Karhunen-Loève modal projection and significantly improves over direct coupling into the fiber bundle, while remaining competitive across a broad range of turbulence conditions. 2 Modal basis comparison for turbulent wavefront power coupling The fraction of power that can be collected from a turbulent wavefront depends on the choice of modal basis onto which the incoming field is projected. Before designing any physical system, it is therefore instructive to compare the power coupling distributions achieved by different modal bases, independently of any hardware constraint. This section establishes the theoretical upper bound set by the Karhunen-Loève basis, introduces the statistical characterization methodology used throughout this work, and compares the performance of several practically relevant modal bases. 2.1 Turbulence statistics and Karhunen-Loève basis Atmospheric turbulence is modeled using a modified Von Kármán power spectrum as defined in [ 14 ] : \Phi_{\phi}(u)=0.0299\cdot r_{0}^{-5/3}\cdot\frac{\exp\!\left(-u^{2}/u_{o}^{2}\right)}{\left(u^{2}+u_{i}^{2}\right)^{11/6}}, (1) where \Phi_{\phi} is the two-dimensional power spectral density of the wavefront phase, u is the spatial frequency in cycles per meter, r_{0} is the Fried parameter, and u_{i}=1/l_{0} and u_{o}=1/L_{0} are the spatial frequencies associated with the inner scale l_{0} and outer scale L_{0} , respectively. In practice, FSO receivers operate with collecting apertures ranging from a few centimeters to tens of centimeters. Rather than simulating at the physical scale of the receiver, we work in a reduced coordinate system where the pupil diameter is set to D=160\,\mathrm{\mu} m, corresponding to the assumption that a telescope images the collecting aperture onto the input of the demultiplexer. All results depend only on the dimensionless ratio D/r_{0} , which fully characterizes the turbulence strength relative to the aperture size, and are therefore independent of the physical scale of the simulation. Phase screens are generated on a 300\times 300 grid with sampling \Delta x=1\,\mathrm{\mu} m, giving a simulation domain of 300\,\mathrm{\mu} m, slightly larger than the pupil to avoid Fourier artifacts at the pupil boundary. The inner scale is set by the grid sampling, l_{0}=\Delta x=1\,\mathrm{\mu} m. To extend the outer scale and suppress low-frequency artifacts inherent to FFT-based generation, the grid is zero-padded by a factor of 8 prior to spectral filtering, giving an effective outer scale L_{0}=8\times 300\,\mathrm{\mu} m =2400\,\mathrm{\mu} m. A set of N=3000 independent turbulent phase screens is generated at D/r_{0}=8 , corresponding to moderate turbulence conditions representative of a ground-to-satellite FSO link. To identify the optimal modal basis for power coupling, we arrange the N=3000