本論文運用全息希爾伯特頻譜分析（Holo-Hilbert Spectral Analysis， HHSA）探討工作記憶精準度的腦波振盪基礎。HHSA 為一種適應性的雙層分解方法，能在不預設固定頻帶、亦不假設訊號呈正弦波形的前提下，量化非線性的跨頻率交互作用。33 名年輕成人執行連續回報的顏色作業，記憶負荷分為四種水準（1、2、4、6），同時記錄 64 通道腦電波圖（EEG）。回憶精準度以三成分混合模型所估計的集中度參數 κ 作為指標。 試次平均的 HHSA 分析（第五章）顯示，後頂枕區 alpha/beta 呈現隨記憶負荷遞增的抑制。額葉中線 theta 則未顯示顯著的負載量效果。在個別差異層次上，較高的額葉 theta 與較低的後側 alpha/beta 分別與較高的 κ 共變。回溯提示出現後，後側 alpha/beta 包絡的第二層振幅調變（amplitude modulation， AM）訊號亦與 κ 共變。額葉 theta 對後側 alpha/beta 的相位－振幅耦合隨記憶負荷增加而增強，且以額葉 theta 相位領先，但未能預測 κ。後續的源定位分析（第七章）將第五章的各項神經相關指標定位至大腦皮質層次。額葉 theta 定位於內側前額葉皮質／背側前扣帶皮質（mPFC/dACC）及背內側前額葉皮質（dmPFC）等分區，後側 alpha/beta 則對應至視覺與頂葉皮質。在全腦層次上，僅 IMF4（beta）於後頂枕皮質呈現顯著的 κ 負向精準度叢集，IMF5（alpha）則未達顯著。 單一試次分析（第六章）進一步檢驗上述特徵。受提示半視野的空間注意力解碼器於 +1.37 秒達到峰值（AUC = 0.569，p < .0005），其特徵重要性落於對側後側 alpha。精準度解碼器在第一層包絡上未達顯著；僅於後側 alpha 與 beta 接近趨勢水準的振幅變化達到顯著（峰值 r = 0.067，位於 +1.83 秒；叢集校正 p = .0065）。一旦納入後側叢集，額葉特徵即不再攜帶獨立的精準度訊號，顯示額葉的貢獻僅限於協調功能。顏色內容編碼則僅於反應階段出現，在記憶維持階段並未發現。 上述發現支持一個三成分的解釋架構。後側 alpha/beta 包絡作為感覺內容的增益調節器，其中 beta 追蹤回憶保真度，alpha 追蹤記憶負荷；額葉中線 theta 為執行控制的組織節律；跨頻率耦合則為一條協調通道，其強度隨負載量增加，但並不決定試次層次的精準度。本論文提出一套源定位、以 HHSA 為基礎的時間延遲嵌入式隱馬可夫模型（time-delay embedded hidden Markov model， TDE-HMM），作為未來工作及分析的共同架構。;This thesis examined the oscillatory basis of working memory (WM) precision using Holo-Hilbert Spectral Analysis (HHSA)， an adaptive two-layer decomposition that quantifies nonlinear cross-frequency interactions without fixed bands or sinusoidal assumptions. Thirty-three young adults performed a continuous-report color task at four set sizes (1， 2， 4， 6) with a target probe while 64-channel EEG was recorded. Recall precision was indexed by κ from a three-component mixture model. Trial-averaged HHSA (Chapter 5) showed load-graded parieto-occipital alpha/beta suppression. Frontal midline theta showed no significant load effect. Across individuals， higher frontal theta and lower posterior alpha/beta each covaried with higher κ. A post-retrocue second-layer amplitude-modulation (AM) signal in the posterior alpha/beta envelope also covaried with κ. Frontal-theta-to-posterior-alpha/beta phase-amplitude coupling scaled with set size， with frontal theta phase leading， but did not predict κ. Subsequent source analysis (Chapter 7) localized the Chapter 5 correlates to cortex. Frontal theta was localized to mPFC/dACC and dmPFC parcels， and posterior alpha/beta to visual and parietal cortices. At the whole-brain level， only IMF4 (beta)， not IMF5 (alpha)， carried a significant κ-negative precision cluster over parieto-occipital cortex. Single-trial decoding (Chapter 6) tested these features. A cued-hemifield spatial-attention decoder peaked at +1.37 s (AUC = 0.569， p < .0005)， with importance in contralateral posterior alpha. The precision decoder was null for the first-layer envelope. Significant was only found for the near-trend envelope component of posterior alpha and beta (peak r = 0.067 at +1.83 s; cluster-corrected p = .0065). Frontal features carried no independent precision signal once posterior clusters were included， confining the frontal contribution to coordination. A color-identity code emerged only at the response stage. These findings support a three-component account. Posterior alpha/beta envelopes act as a gain modulator on sensory content， beta tracking fidelity and alpha tracking load. Frontal midline theta is the executive organizing rhythm. Cross-frequency coupling is a coordination channel that scaled with load but did not determine trial-level precision. A HHSA-informed time-delay embedded hidden Markov model (TDE-HMM) is proposed as the main analysis framework for future works.
