| 摘要: | 近代衛星通訊技術快速發展，毫米波頻段因高衰減特性，需採用天線陣列以增強信號強度。在遠距離衛星通訊中，優化波束場型以提升 SINR 和指向解析度至關重要，同時通訊延遲問題也促使預編碼 (Codebook) 技術的應用，以實現快速波束切換。然而，若需適用於混合波束成型(Hybrid beamforming) 的 Codebook 在高精度需求下將佔用大量儲存空間。因此，本研究提出一種減少係數儲存的方法，透過演算法設計與硬體實現，希望大幅降低 Codebook 所需的記憶體資源，同時滿足系統規格。
本研究利用 JADE (Adaptive Differential Evolution Algorithm) 一種改良後的差分進化演算法，優化波束場型，使主波束寬度與旁瓣電平達到最佳化。觀察優化後的係數呈現對稱且遞減特性，因此可透過在係數之間加上硬體友善的數值關係限制，使我們能僅儲存部分係數，並在硬體端還原完整係數，進而降低記憶體需求。
波束指向採用 DFT-based codebook，並調整粗波束在空間的分布，以減少 beam broadening 和指向誤差。DFT-based codebook 之 Codeword 係數補償天線接收相位延遲，可透過硬體計算獲得。對於二維天線陣列，方位角與仰角的係數分別計算後，再透過克羅內克積(Kronecker product) 計算完整波束係數，進一步降低儲存需求。
波束搜尋階段採用階層式方法，透過二元搜尋確定目標方向所屬的粗調波束範圍，再計算細調波束補償量，以快速獲得精確的指向係數。在硬體設計方面，結合 CORDIC 與 Binary Search 演算法，快速計算指向參數並重建波束係數，並透過定點數模擬優化字長，以減少硬體資源與功耗。
本研究有效解決大規模天線陣列與高角度解析度需求下 Codebook 儲存空間過大的問題，以MATLAB模擬並驗證該方法是否符合目標的性能規格，並以硬體實作，實現能快速切換波束方向之功能。;In recent years, satellite communication technology has rapidly advanced. Due to the high attenuation characteristics of the millimeter wave band, antenna arrays are essential for enhancing the strength of transmitted and received signals. In long-distance satellite communication, optimizing the beam pattern is crucial for improving SINR (Signal to Interference plus Noise Ratio) and enhancing the resolution of the pointing angle to improve service quality. However, long-distance transmission leads to higher communication latency; therefore, rapid switching of beams in different directions is desirable through precoding (Codebook) technology.
Existing spatial filtering codebooks for hybrid beamforming architectures require significant storage space for high accuracy.
This paper proposes a method to reduce coefficient storage and, through algorithm design and hardware implementation, decrease codebook size while meeting system specifications.
This study utilizes JADE (JA Differential Evolution Algorithm), an improved differential evolution algorithm, to optimize the beam pattern, achieving optimal main lobe width and side-lobe level. The optimized coefficients exhibit symmetry and a decreasing trend. By imposing hardware-friendly numerical constraints among the coefficients, it is possible to store only a subset of them and reconstruct the complete set in hardware. This approach not only reduces the algorithm′s dimensionality while maintaining optimization performance but also significantly lowers memory requirements.
For beam steering, a DFT-based codebook is employed, with adjustments made to the spatial distribution of coarse beams to mitigate beam broadening and pointing errors. The Codeword coefficients in the DFT-based codebook compensate for phase delays caused by signal arrival at different antenna elements and can be computed in hardware. For a two-dimensional antenna array, the coefficients for azimuth and elevation are separately computed and then combined using the Kronecker product to obtain the final beamforming coefficients, further reducing storage requirements.
The beam search process adopts a hierarchical approach, first using a binary search to determine the coarse beam range in which the target direction resides, followed by the computation of fine beam compensation to rapidly obtain precise beamforming coefficients. In terms of hardware design, this study integrates the CORDIC and Binary Search algorithms to efficiently compute steering parameters and reconstruct beam coefficients. Fixed-point simulations are conducted to optimize word length, thereby reducing hardware resource consumption and power dissipation.
This research effectively addresses the challenge of excessive codebook storage requirements in large-scale antenna arrays and high angular resolution scenarios. The proposed method is validated through MATLAB simulations to ensure compliance with target performance specifications, and a hardware implementation is carried out to enable rapid beam switching functionality. |
