| 摘要: | 在無線通訊中位置定位是個很重要的課題，但非直視性(non-line-of-sight, NLOS) 誤差一直是影響目標物追蹤準確性相當重要的因素之一，為了提高無線定位在直視性 (line-of-sight, LOS) 與非直視性混合環境下的準確性和可靠性，我們考慮使用到達時間(Time of Arrival, TOA) 測量技術來對目標物做追蹤，在本篇論文中，NLOS 的測量被模擬成擾亂雜訊所呈現，標準的濾波器如擴展卡爾曼濾波器 (Extended Kalman Filter, EKF) ，在非高斯雜訊情況底下會產生很大的誤差，於是利用 Masreiez 濾波器 (MF) 克服這點，因此我們同時並行兩個濾波器在交互性多模型 (Interacting Multiple Model, IMM) 的框架中，在 LOS 環境時，EKF 可以有很高的精確度，而 MF 可以加強當 NLOS 傳輸時的狀態估計，但目標物行走在街道上不太可能只有等速度一種行為而已，所以我們加入了等加速與座標轉彎模型來改善追蹤性能，之後再次使用交互性多模型分別在 LOS 與 NLOS 的情況下對三種運動模型作相互間的狀態估計，這就是我們所提出的雙層交互性多模型 (Nested Interacting Multiple Model) 架構。在數值分析中，我們提出的演算法性能會優於只使用 IMMEKF 或者 IMMMF，並且我們提出的演算法的均方誤差 (Root Mean Square Error,RMSE) 會比較靠近 Cramer-Rao Lower Bound(CRLB)。
Localization of mobile nodes is an important issue in wireless communications. The non-Gaussian noise resulted from the non-line-of-sight (NLOS) measurement greatly affects the tracking accuracy. In order to enhance the tracking reliability in mixing line-of-sight (LOS)and
NLOS environments, we develop a new algorithm for mobile node tracking based on time of arrival (TOA) measurements. In this thesis, two kinds of different non-Gaussian noises modeled by the mixture of Gaussian and Laplacian noises are considered as the NLOS noise. In non-Gaussian noise
environments, the extended Kalman filter (EKF) loses the optimality for state estimation. Here, we employ the Masreliez filter (MF) to deal with the non-Gaussian noise. Besides the concern of the noise model, three possible dynamic models, i.e., constant velocity motion (CV), constant acceleration motion (CA), and coordinated turn
(CT), are usually adopted for maneuvering target tracking. Since the conventional EKF has higher tracking precision in the LOS environment while the MF provides robust state estimation in the NLOS environment, a new interacting multiple model (IMM) algorithm possessing two conventional IMM operation in parallel, called Layered IMM algorithm, is
used to simultaneously accommodate two different measurement models and three different dynamic models. Numerical results show that the Layered IMM algorithm outperforms the conventional IMM-EKF algorithm and the IMM-MF algorithm. The root mean squared error (RMSE)analysis also indicates that the tracking error performance of the proposed algorithm is quite close to the posterior Cramer-Rao lower bound (CRLB)in steady state. |
