Abstract
Mechanical deformation remains a central obstacle to practical multimode fiber (MMF) imaging because bending and torsion alter modal interference, change the effective transmission state, and degrade image reconstruction. Most existing strategies address this problem by recalibrating the transmission matrix, compensating the perturbed output field, or improving the reconstruction model, while the role of the input encoding itself has received less attention. Here, we show that radial encoding position is a controllable factor that strongly affects deformation robustness in far-field step-index MMF imaging. Seven radial holographic encodings were defined on a digital micromirror device and evaluated using experimentally acquired far-field speckle patterns under controlled bending, torsion, and sequential bending–torsion perturbations. Under bending, classification accuracy exhibits a clear radial dependence: outer encodings degrade substantially more slowly than inner encodings, and the outermost channel retains more than 60% accuracy at 30
∘
bending, exceeding the innermost encoding by more than 45 percentage points. In contrast, torsion produces comparable degradation across radial encodings, with no systematic inner-to-outer robustness hierarchy. Sequential bending–torsion loading further reveals that the system response is influenced not only by the final deformation parameters but also by the loading trajectory. These results identify radial encoding as a practical design variable for deformation-resilient MMF imaging and clarify the distinct roles of bending, torsion, and deformation history in far-field speckle-based image transmission.