Deep learning-based phase retrieval with embedded data for holographic data storage

This paper proposes a deep learning-based (DL-based) method for phase retrieval with embedded data. Utilizing known embedded data as strong constraints, in the case of multi-level phase encoding, each signal data has a unique solution for the phase difference relative to the embedded data. This innovative coding strategy reduces the arrangements of signal data and its adjacent data from 4⁵ types to just 4, allowing a single labeled sample to contain all possible coding arrangements. The robust constraint imposed by the known embedded data on the diffraction characteristics between adjacent data facilitates precise phase retrieval, even with a smaller recording area of the hologram. Experimental results demonstrate that when the recording area is limited to the Nyquist size, the proposed method outperforms traditional approach by reducing the required number of training samples by a factor of 50 and decreasing the bit error rate (BER) by approximately 9 times.