Breaking the Coupling Among N ² Fully‐Connected Linear Polarization Channels with Polarization Diffraction Neural Network

Polarization multiplexing is one of the key technologies for multi‐dimensional information transmission. When constructing an N 2 ‐dimensional linear polarization multiplexing network with N input states of distinct polar angle and corresponding N outputs ( N > 2), there are inherent couplings for all channels due to polarization decomposition and superposition, exhibiting as information crosstalk among non‐orthogonal polarization channels. Hereby employing a polarization diffraction neural network (PDNN) to minimize couplings in a global manner, N 2 fully‐connected independent transmission channels are demonstrated by multiplexing N linear polar angles. Taking a chiral‐assisted metasurface as a physical platform, which enriches physical freedoms by breaking symmetry in the longitudinal direction, a single‐layer PDNN is verified through ultra‐low crosstalk in digital holographic imaging. Then, PDNN is extended to a limited‐aperture metasurface scenario, achieving 36 independent polarization channels multiplexing within an aperture of 18.7λ 0 × 18.7λ 0 . These results exemplify the architecture's capability for complete polarization multiplexing, with direct applicability to optical linear computing, advanced holography, and high‐security optical encryption.