Exploiting multimode fiber speckle for turbulence-resilient OAM classification

Free-space optical communication using orbital angular momentum (OAM) modes offers high channel capacity, but atmospheric turbulence destroys the modal orthogonality required by conventional receivers. We demonstrate that a passive multimode fiber reservoir, operating as an optical extreme learning machine, provides turbulence-resilient OAM classification that outperforms ideal modal decomposition in moderate-to-strong turbulence. Unlike traditional methods that project onto a rigid basis, the reservoir maps distorted wavefronts into a high-dimensional speckle feature space where class identities remain linearly separable. We validate the system using structural (MNIST) and textural (Fashion-MNIST) benchmarks, achieving 97.6% and 87.1% accuracy, respectively. Applied to OAM classification, this passive all-optical approach exceeds the performance of modal decomposition by over 20% in strong turbulence regimes (SR ≤ 0.5). This approach demonstrates that complex wavefront processing can be effectively offloaded to passive optical scattering, enabling robust OAM recovery with simple linear readouts even when modal orthogonality is destroyed.