隨著第五代行動通訊（5G）對高速率與寬頻資源的需求日益增加，毫米波（mmWave）頻段因擁有豐富且未充分利用的頻譜資源，被視為擴充容量與提升傳輸速率的潛力解方。然而，毫米波訊號在傳輸過程中易受阻擋，且伴隨嚴重的路徑損耗與自由空間衰減，導致其涵蓋範圍與穩定性受限。為改善傳輸品質，現行系統普遍採用大規模多輸入多輸出（massive MIMO）與波束成形技術，以透過天線陣列增益對抗路徑損耗。然在實作層面，若採用全數位架構，需為每根天線配置獨立的射頻鏈（RF chain），將導致系統硬體成本與功耗大幅上升，難以應用於大型天線陣列。因此，本研究採用混合波束成形架構，藉由結合類比與數位處理，於維持方向性控制能力的同時，顯著降低系統複雜度與功耗。本文所提出之架構中，類比波束成形係以預設 DFT codebook 為基礎，並透過 trace-inverse 原則進行初始 beam selection；數位波束成形則採用區塊對角化（Block Diagonalization, BD）演算法，以抑制多使用者間干擾。在總發射功率受限的前提下，研究目標為最大化下行鏈路的總資料傳輸率（sum rate）。為達成此目標，本文進一步將原先僅適用於 MISO 系統的兩階段 beam selection 流程擴展至多使用者 MIMO-OFDM 架構，並引入交替最佳化（Alternating Optimization, AO）策略，同時針對發射端與接收端類比波束進行迭代更新。此外，本研究亦整合資料流層級的功率分配（Water-Filling）機制，使 beam selection 與功率分配得以聯合優化。最後，透過實際信噪比（SNR）對應之調變與編碼指標（MCS index）進行速率估算，替代理想化的 Shannon 容量模型，以更貼近真實系統行為。模擬結果顯示，所提出方法在各種系統參數組合下皆可維持穩定且顯著之傳輸增益，並優於平均功率分配的基準方案。整體結果驗證本方法於寬頻毫米波 massive MIMO-OFDM 系統中，具有良好的效能表現與實務應用潛力;To address the escalating demand for high data rates and wideband access in fifth-generation (5G) networks, the millimeter-wave (mmWave) spectrum has gained prominence due to its abundant and underutilized frequency resources. However, mmWave signals are highly susceptible to severe path loss and blockage, significantly limiting their transmission reliability and coverage. To overcome these limitations, massive multiple-input multiple-output (MIMO) systems employing directional beamforming techniques have emerged as a promising solution for enhancing link quality via antenna array gain. Despite these benefits, fully digital architecture remains impractical in large-scale MIMO deployments, as dedicating a radio frequency (RF) chain to each antenna incurs prohibitive hardware costs and power consumption. This thesis proposes a hybrid beamforming architecture that reduces implementation complexity by combining analog and digital processing. Analog beamforming is performed using a predefined DFT codebook and initialized based on a trace-inverse metric, while digital beamforming is conducted using block diagonalization (BD) to suppress inter-user interference. The core objective of this study is to maximize the downlink sum rate under a total transmit power constraint. To achieve this, we propose a two-stage beam selection algorithm tailored for the MIMO-OFDM setting, which integrates and generalizes key ideas from trace-inverse-based MISO designs to support multiple users and spatial streams per user. An alternating optimization (AO) strategy is employed to iteratively refine both transmit- and receive-side analog beamformers. Furthermore, we integrate a stream-level power allocation strategy based on the water-filling principle, enabling joint optimization of beam directions and power distribution. Instead of relying on idealized Shannon capacity expressions, this work evaluates system throughput using realistic modulation and coding scheme (MCS) mappings. SNR values are mapped to MCS indices according to 3GPP CQI standards with a 10% BLER requirement, and the index table is extended beyond CQI-15 to reflect high-SNR conditions. This allows for practical and accurate estimation of the achievable data rate. Simulation results validate the effectiveness of the proposed design across various configurations. Compared to baseline methods such as random selection or equal power allocation, our approach consistently achieves higher sum rate while maintaining low algorithmic complexity. These findings demonstrate that the proposed framework offers strong potential for deployment in wideband mmWave multi-user MIMO-OFDM systems.
