Reservoir computing using multimode fiber and spatial light modulator for mode decomposition - 10m MMF

Tao, Houchun

Reservoir computing using multimode fiber and spatial light modulator for mode decomposition - 10m MMF

Contributing person datacite.contributor.DataCollector	Houchun Tao
Description of the data datacite.resourceType	The data was collected on a 10m MMF
Type of the data datacite.resourceTypeGeneral	Collection
Total size of the dataset datacite.size	54167848067
Author dc.contributor.author	Tao, Houchun
Upload date dc.date.accessioned	2026-10-02T14:00:24Z
Publication date dc.date.available	2026-10-02T14:00:24Z
Publication date dc.date.issued	2026-10-02
Abstract of the dataset dc.description.abstract	Multimode fibers (MMFs) are attractive for many applications, such as endoscopic imaging and internet data trans mission. However, modal dispersion leads to crosstalk effects. Robust mode decomposition is therefore crucial—whether to harness the inherent mode mixing for physical layer security (PLS) or to correct it in spatial division multiplexing (SDM). While traditional holographic methods can untangle this crosstalk, they require a stable reference beam, which is impractical for longhaul transmission due to environmental sensitivity. Furthermore, although recent deep-learning approaches enable reference-less decomposition, they typically suffer from high computational latency and massive training costs for dynamic channels. To overcome these limitations simultaneously, we propose a reference free, low-latency method for mode decomposition and channel reconstruction that leverages the linear complex scattering of the MMF as a passive high-dimensional spatial feature map, whose nonlinearity is provided by square-law detection at the camera. Following a reduction of the sampling rate in the frequency domain to decrease feature dimensionality, a computationally efficient linear readout layer inverts the scattering process to reconstruct the transmitted complex fields. Although the final channel reconstruction still requires digital computation steps, this readout process is limited to a single, computationally efficient linear matrix multiplication. This reduces both latency and computational overhead compared to conventional electronic neural networks, which typically exhibit high energy consumption and high latency. The physical limitations and trade-offs of this framework when scaling to higher modal complexities are also discussed. We experimentally demonstrate high-precision, reference-less mode decomposition for multiplexed classical optical fields.
Public reference to this page dc.identifier.uri	https://opara.zih.tu-dresden.de/handle/123456789/2974
Public reference to this page dc.identifier.uri	https://doi.org/10.25532/OPARA-1582
Publisher dc.publisher	Technische Universität Dresden
Licence dc.rights	CC0 1.0 Universal	en
URI of the licence text dc.rights.uri	http://creativecommons.org/publicdomain/zero/1.0/
Specification of the discipline(s) dc.subject.classification	3::32::308
Title of the dataset dc.title	Reservoir computing using multimode fiber and spatial light modulator for mode decomposition - 10m MMF
Funding Acknowledgement opara.project.fundingAcknowledgement	We would like to express our sincere gratitude to the German Research Foundation for funding the Reinhart Koselleck project for highly innovative research (CZ 55/61-1, project number: 560574412). This research was also funded by the German Federal Ministry of Education and Research within the framework of the 6G-life (funding code: 16KISK001K) and QUIET (project code: 16KISQ092) projects.