本研究旨在解決全像儲存系統在多工記錄過程中，繞射效率隨記錄頁數增加而呈平方倍顯著衰減的問題。傳統系統多依賴角度或波長多工，但隨容量提升，訊號雜訊比往往因物理限制而大幅上升。本論文採用一種創新的全像儲存架構，透過改良記錄與讀取路徑之光學設計，本架構能有效緩解多頁記錄過程中的能量損耗，並顯著提升系統對於光路高精度對準誤差的容忍度。研究主要聚焦在脈衝訊號的實驗應用，藉由在同一空間區域內進行多組訊號的同步記錄，從物理層面大幅增進系統的資訊儲存密度與資料傳輸量。 在訊號處理與優化方面，本研究結合啟發式演算法與人工智慧技術。針對多個脈衝訊號同步記錄所引起的能量分布不均問題，利用基因演算法計算各訊號點的初始相位值，優化記錄品質。此外，本論文導入深度學習模型，針對重建後的脈衝訊號進行影像預處理與特徵提取。透過神經網路對訊號區域進行自動化分割與分類，取代傳統繁瑣的數值擬合與人工分析，實現對多組脈衝訊號的平行解調。實驗數據顯示，此整合方案不僅大幅縮短了系統的資料讀取與分析時間，更在高雜訊與光學畸變環境下展現出卓越的辨別準確性，為高容量、高傳輸率的全像儲存系統提供了全面的優化方向。;This research aims to address the critical issue of significant diffraction efficiency decay， which follows an inverse-square trend relative to the number of recorded pages during the multiplexing process in holographic data storage systems. Traditional systems primarily rely on angular or wavelength multiplexing; however， as storage density increases， the signal-to-noise ratio typically degrades due to inherent physical limitations. This thesis employs an innovative holographic storage architecture that， through optimized optical design of the recording and retrieval paths， effectively mitigates energy loss during multi-page recording and significantly enhances the system′s tolerance to high-precision optical alignment errors. The focus is centered on the experimental application of pulse signals， achieving a substantial increase in information storage density and data throughput by performing simultaneous recording of multiple signal sets within the same spatial volume. In terms of signal processing and optimization， this study integrates heuristic algorithms with artificial intelligence techniques. To resolve the uneven energy distribution caused by the simultaneous recording of multiple pulse signals， a genetic algorithm is utilized to calculate the initial phase values of each signal point， thereby optimizing recording quality. Furthermore， this thesis introduces a deep learning framework for image pre-processing and feature extraction of the reconstructed pulse signals. By employing neural networks for automated segmentation and classification of signal regions， the proposed method replaces conventional， labor-intensive numerical fitting and manual analysis， enabling parallel demodulation of multiple pulse signals. Experimental results demonstrate that this integrated approach not only drastically reduces data retrieval and analysis time but also exhibits superior recognition accuracy under high-noise and optical distortion environments， providing a comprehensive optimization strategy for high-capacity， high-transfer-rate holographic data storage systems.
