| 摘要: | 隨著現今無線通訊的進步，無論是存取點(access point, AP)還是站點(station, STA)都導致了天線需求量的急速增加。因此在多使用者的多輸入多輸出(MIMO)系統中，採取整體效能的優化是必要方法。通道估計便是協助效能優化的前提，它能夠有效協助我們得知通道內重要的參數，像是角度、增益以及延遲，而進一步去調整通道的能量消耗以及整體系統的涵蓋範圍。
智慧反射面(Intelligent reflecting surface, IRS)在無線通訊系統中扮演一個重要的腳色，因為它能夠提升整體通訊系統的涵蓋範圍、資料的吞吐量以及功率消耗。由於它是被動裝置，相較發送端及接收端，其內部的元件數量也較為龐大。
本論文描述一個針對多使用者的多輸入多輸出，且有IRS輔助的上行系統，採取針對該系統中重要參數的通道估計；同時對IRS內部反射元件進行分組(grouping)的處理，來降低原先較為龐大的運算量。我們創建的訓練資料會在發送端(STA)做傳輸，經過IRS反射後抵達接收端(AP)，該筆資料會以三維形式來表達。我們利用張量(tensor)運算以及訊號的旋轉不變性(rotational invariance)，來完成通道參數的估計。不同使用者會利用指定的訓練子載波(subcarrier)來做估計，以此來區分通道參數對應的使用者。在我們的模擬結果，在反射元件數量為64的情形中，我們採取兩種的分組方法：分成4組及分成16組，兩者的正歸化均方誤差(normalized mean square error, NMSE)分別為 3.2×〖10〗^(-3) 和 2.6×〖10〗^(-3)，前者在乘法器的使用數量則是分別降低了24.2% 和19.5%。
;Following with the advanced wireless communication, both the base station and user equipment cause the requirement of antennas increase extremely. Therefore, it is an essential step to improve the overall performance in a multiuser MIMO system. Channel estimation can help on leverage some significant parameters on channel, i.e., angle, path gain and delay, which have high relationship with performance.
Intelligent reflecting surface (IRS) has become a significant paradigm to create a favorable environment, which include improve the wireless communication coverage, throughput and energy efficiency. IRS is a kind of passive device, which means it doesn’t employee any transmit/receive radio frequency (RF) chain. It becomes a practically challenging task due to its massive number of passive elements.
In this thesis, we focus on an uplink channel estimation for an IRS-aided multiuser multiple-input multiple-output (MIMO) systems. We model the training signal from the user equipment to the base station via IRS as a third-order canonical polyadic tensor with a maximal tensor rank equal to the number of elements. We model different kinds of grouping schemes on elements. We extract the cascaded channel parameters by leveraging the characteristic of tensor computation and signal rotational invariance. In simulation part, we define the number of elements is 64 and there has two kinds of grouping method: 4 Groups and 16 Groups. Their performances on NMSE are 3.2×〖10〗^(-3) and 2.6×〖10〗^(-3), respectively. Focusing on the number of multiplications, the former case alleviates about 24.2% and 19.5% than the letter case, respectively. |
