A deep-learning enhanced computational optical decoding strategy (DECODS) for optical data storage

Abstract Traditional optical data storage (ODS) systems face constraints imposed by the optical diffraction limit. It prevents reliable recognition of closely spaced symbols, forcing heavy reliance on error correction codes (ECC) and capping storage density and efficiency. Here, we introduce a deep-learning enhanced computational decoding strategy (DECODS) that treats optical readout as a data-driven optical read-channel model and performs parallel multi-length (2 T–8 T) learned decisions, extending the effective decoding capability beyond the conventional diffraction-limited threshold at the decision level while retaining the standard Blu-ray optical front end. A high-precision physics-based channel model—including aberrations/defocus/tilt, clock/servo jitter, white noise, intersymbol interference and inter-track crosstalk—generates physically consistent training data. Under the validated experimental conditions of this study, DECODS exhibited 0% raw bit-error rate (BER) on evaluated segments of real readout; this does not represent a universal operating guarantee. DECODS also enabled reliable recognition of symbol spacings approaching one quarter of the conventional diffraction-limited threshold at the decision level. This lowers the raw BER, reduces the redundancy required to achieve a target reliability level, and indicates a potential effective capacity improvement of up to 14.8% while supporting tighter symbol/track spacing. Results show a 5.8 × increase in noise tolerance and ≥ 20% expansion of robust servo margins; combined with ECC-related gains, the theoretical effective-density improvement reaches up to 26.28%. DECODS integrates optical physics, computational modeling, and deep learning to provide a scalable decoding framework with quantified, decision-level strategy to longstanding ODS bottlenecks.