大規模多使用者多輸入多輸出分時雙工的通道追蹤估測技術研究

Abstract

隨著科技的進步、半導體製程的技術演進，硬體可以越做越小，手機已成為現今每個人主要行動上連網的裝置，然而在移動上通訊的需求越來越大，未來5G更是比4G快上至少10倍以上，能產生如此龐大的速度，其中一個被大多數人認定的重要技術就是大規模多輸入多輸出(Massive MIMO)。 大規模多輸入多輸出是指在基地台(Base Station)端相較於行動(Mobile Station)端放置大量的天線數，基地台的天線數可能是100，128或更大…，而行動端可能數個天線，然而使用大量的天線傳送可以大幅增加資料傳輸率，且利用空間領域在單位時間頻率資源上有更高的效率，但在通道估測上難度正比於傳送天線數，然而採用分時雙工(Time-Division Duplexing)形式，可以利用互惠特性(reciprocal property)來估計一行通道得知通道資訊。 傳統上在估計多輸入多輸出及大規模多輸入多輸出通道使用特徵值分解(Eigenvalue Decomposition)、奇異值分解(Singular Value Decomposition)方法和線性處理方法像是zero forcing、MF，然而特徵值分解及奇異值分解法在實際上實現非常複雜，為了減少複雜度，我們使用了子空間追蹤(subspace tracking)演算法，利用近似正交的特性去找出特徵向量改善表現，以及每次遞迴的去更新上一次的子空間藉此來降低複雜度的問題，另外使用少許領航訊號，改善特徵向量角度模糊的問題。最後，我們也討論了zero forcing法以及另外四種不同方法，針對不同的環境，提出一些看法與建議。

As the advancement of technology and the semiconductor process technology evolution, the hardware can be smaller and smaller. Mobile phone has become main device which the most people connect to the Internet in the outdoors. However, the requirement of the mobile communication substantial increases. The next generation 5G is more than ten times faster than 4G. It can produce the huge data rates because of the massive MIMO that is one of the most important technology identified by many people in 5G. The massive MIMO is equipped large number of the antennas at base station than mobile station several antennas. The base station antennas may be one hundred or one hundred and twenty-eight or bigger. Large scale antennas at base station can greatly raise data rates and have more efficiency exploiting spatial domain at time and frequency resources, but the channel estimation complexity is proportional to transmission antennas. Nevertheless, we can adopt TDD technique which has reciprocal property to only estimate uplink or downlink channel informations. Traditionally, estimating the MIMO and massive MIMO channel uses Eigenvalue Decomposition, Singular Value Decomposition, and linear signal processing methods such as MF and zero forcing, but EVD and SVD are very complexity in practice. To reduce complexity, we utilize subspace tracking algorithm. It exploits the approximation characteristic to find eigenvector improving performance and updates the last subspace per iteration to decrease complexity. Besides, using a few pilots improves phase ambiguity of the eigenvectors. Finally, we discuss zero forcing and four different methods to propose some suggestions.