Deep Unfolded Optical Decoder for LDPC Codes with System Noise Mitigation

We propose an all-optical low-density parity-check (LDPC) decoder using deep unfolding with learnable damping parameters to mitigate system noise. Our approach enhances the performance of tensor-computable belief propagation (Tensor-BP) algorithm in noisy optical implementation while preserving its matrix-based parallel processing capabilities. Simulation results show that our method significantly outperforms conventional BP decoder under various system noise conditions, approaching ideal noiseless BP decoder performance in high signal-to-noise ratio (SNR) region. This work advances practical optical implementations of LDPC decoders for next-generation high-speed communication systems.