毫米波（mmWave）大規模多輸入多輸出（MIMO）通信是未來蜂巢式網絡中一種有前途的技術。然而，傳統的全數位波束成型每根天線都需要專用的射頻 （RF） 鏈，這將導致龐大的硬體成本和功率消耗。混合波束成型結構在大規模 MIMO 中扮演了重要的角色，因為它可以支持高陣列和復用增益，同時有效降低高昂的硬體成本。本論文在寬頻通道中的上行毫米波多用戶 (MU) 大規模 MIMO 系統提出了三種具動態資料流分配 (DSA) 的新穎混合波束成型設計，旨在最大化整體頻譜效率。正交分頻多工被採用來對抗通道的頻率選擇性。我們提出的演算法由一個兩段式的方法所組成，其中類比和數位波束成型器是分開設計的。考量到性能和複雜度，開發了三種不同的類比波束成型設計，以盡可能地最大化每個子載波上等效基頻通道的通道增益，同時也為每個移動站 (MS) 決定資料流的數量。具體來說，對於第一種設計，類比波束成型器被設計成匹配每個 MS 在整個頻帶上相對應的通道。對於第二種而言，為了顯著降低第一種設計的複雜度，我們可以透過選擇適當的陣列響應向量集合來完全避免矩陣分解。最後一種設計僅對中心子載波的通道執行奇異值分解 (SVD)，以求得類比波束成型器。結合塊對角化 （BD） 和協調發射-接收處理之方法，進一步推導出數位預編碼器和組合器以消除每個子載波上的干擾。 我們也基於現有的演算法開發全數位波束成型和另外兩種混合波束成型設計作為性能基準。透過數據模擬，在為每個 MS 動態分配資料流的數量後，所提出之方案的頻譜效率會顯著提升。接下來，我們研究了在各種參數底下所提出之方案和其他基準方案之間的性能差異。此外，模擬指出所提出的第一和第三種的混合波束成型設計可以達到非常具競爭力的性能，甚至在某些情況下優於全數位的設計。 ;Millimeter wave (mmWave) massive multiple-input-multiple-output (MIMO) communication is a promising technology for the future cellular networks. However, the traditional fully-digital beamforming requires a dedicated radio frequency (RF) chain for each antenna which will lead to huge hardware cost and power consumption. A hybrid beamforming structure plays an important role for massive MIMO because it can support the high array and multiplexing gain while effectively alleviating the high hardware cost. This thesis proposes three novel hybrid beamforming designs with dynamic streams assignment (DSA) in the uplink mmWave multi-user (MU) massive MIMO systems over wideband channels, aiming at maximizing the overall spectral efficiency. The orthogonal frequency-division multiplexing (OFDM) is adopted to combat the frequency selectivity of the channel. Our proposed algorithms consist of a two-stage approach, in which the analog and digital beamformers are designed separately. Considering the performance and complexity, three different analog beamforming designs are developed to maximize the channel gains of the equivalent baseband channel at each subcarrier as much as possible, while the number of data streams are also determined for each mobile station (MS). Specifically, for the first design, the analog beamformers are designed to match the corresponding channel on the entire band for each MS. For the second one, to significantly reduce the complexity of the first design, we can completely avoid the matrix decomposition by selecting the appropriate set of array response vectors. The last design only performs singular value decomposition (SVD) on the channel of the center subcarrier to find the analog beamformers. Combining the block diagonalization (BD) and coordinated transmit-receive processing method, the digital precoders and combiner are further derived to eliminate the interference at each subcarrier. We also develop fully-digital beamforming and another two hybrid beamforming designs based on the existing algorithms as performance benchmarks. Through numerical simulations, the spectral efficiency of the proposed schemes can be significantly improved after dynamically assigning the number of data streams for each MS. Then, we study the performance difference between the proposed and other benchmark schemes under the various parameters. Moreover, simulations indicate the first and third proposed hybrid beamforming designs can achieve a very competitive performance and even outperform the fully-digital design in some situations.
