在非理想環境下使用支持向量機之強健式陣列波束成型技術

Abstract

可適性陣列波束成型技術能透過調整各個天線陣列上的權重係數，可以接收特定方向的信號並且消除其他方向的干擾以及雜訊，在諸多領域已有許多重要應用，傳統的波束成型技術所需要的已知資訊為欲接收信號的入射角方向，此種技術稱為指引式波束成型器，例如LCMV，然而近二十年來發展出了另一種類型的可適性陣列波束成型技術，不需要知道欲接收信號的入射角方向，其運用信號的某些特徵例如週期恆定性或是恆模特性來達到波束成型，此種技術稱做盲目式波束成型器，而本論文即針對「指引式波束成型器」以及「利用信號週期恆定性的盲目式波束成型器」發展出多種強健式技術，以對抗非理想效應的影響。 在論文的第一部分我們嘗試將一種在機器學習領域相當熱門的技術-支持向量機與傳統的強健式波束成型器做結合，並提出一種方法來使得傳統的強健式波束成型器可以符合支持向量機的運算結構，期望能改善傳統指引式波束成型器的收斂速度以及在指引向量誤差下能有更好的強健性；此外我們利用支持向量機並基於傳統的盲目式波束成型演算法：LS-SCORE和CAB，發展出兩種新穎的盲目式波束成型器，同樣期望能提升LS-SCORE和CAB的收斂速度以及效能，由於盲目式波束成型技術所須的已知資訊只有欲接收信號的週期頻率，因此我們所提的演算法在週期頻率誤差下同樣會產生嚴重問題，我們亦針對此部分提出可行的方法來解決，模擬顯示，相較於以往的強健式方法，我們所提出的方法具有許多優點。 在論文的第二部分我們不涉及支持向量機，首先我們利用「自相關矩陣重建法」來改進兩種傳統的強健式波束成型技術：RCB和Cheng’s method，我們將欲接收信號從接收信號當中除去，以期能減輕能量反置的效果並增快收斂速度；此外我們也將此想法應用在一種利用週期恆定特性為限制條件的盲目式波束成型器(Origin CC method)來增快其收斂速度，並與前人的做法相比較具有顯著的優勢；此外我們也分析Origin CC method在週期頻率誤差下的效能衰落並提出兩種新穎的方法來解決週期頻率誤差下的問題，模擬結果顯示我們所提出的方法優於以往的強健式方法且相當接近理想情況下的最優效能。

Adaptive array beamforming which can extract signals of interest from specific angles while suppress interferences and noise by adjusting weights on the array elements has been applied to many areas recently. For conventional beamforming, the a priori information is the direction of the desired signal, and this kind of technique is called steered-beam beamformer, for example LCMV. Over the past two decades, another kind of adaptive array beamforming without the knowledge of the direction of the desired signal has been widely presented. It utilizes some characteristic of the signals to achieve beamforming, for example, signal cyclostationarity or constant modulus and thus it is called “blind” beamformer. The purposed of this thesis is mainly to develop several robust techniques for both steered-beam beamformer and blind beamformer using signal cyclostationarity in order to tackle the performance degradation under non-ideal environments. In the first part of this thesis, we attempt to apply support vector machine (SVM) (which is a popular technique in the area of machine learning) to the robust array beamformers and also present a method to cope with the combination of both techinques. In this way, we expect that the robustness against steering vector errors and convergence rate of conventional steered-beam beamformer can be improved. Besides, we also present two novel blind beamformers utilizing SVM and traditional blind beamforming algorithm－LS-SCORE and CAB and also expect to achieve better convergence rate and performance. Since the a priori information required by performing blind beamforming is only the cycle frequency of the desired signal, the presence of cycle frequency error (CFE) may lead to severe performance degradation. We present an effective method to solve this problem, and the simulation result shows that the proposed method is better than the past robust method. In the second part of this thesis, we won’t use support vector machine. Instead, we use “the reconstruction of the autocorrelation matrix” to improve the performance of two kinds of conventional robust beamforming－RCB and Cheng’s method by removing the desired signal from the receive signal. In this way, we can expect the alleviation of the effect of “power inversion” and the improvement of the convergence rate. On the other hand, we apply this idea to a kind of blind beamformer using a constraint related to signal cyclostationarity (Origin CC method) to improve the convergence rate and we will show that the proposed method is better than the existed method. In the end, we also analyze the performance degradation of Origin CC method due to CFE and present two novel method to cope with this problem. The simulation result shows that the performance of our proposed methods are better than those existed robust method and nearly reach the optimal performance under ideal case.