音響定位之研究

Abstract

一個定位系統是由多個固定端對於待測物的時間差與距離量距，來估測它的位置。主要應用包括機器人與UWB(Ultra-Wideband)無線感測網路。 [針對室內定位，我們希望利用聲音當作傳輸的信號。因為音響相對於射頻訊號，具有繞射，較低聲速與低硬體成本的優勢。 定位之前的距離估測主要的困難包括。電腦的隨機延遲所造成的同步問題，因環境而異的聲速，因阻礙物造成的估計誤差，以及如何有效追蹤運動中的機器人等議題。 首先我們提出了用參考麥克風來解決同步及聲速校準的問題，不同於傳統射頻信號同步(收發機成本高)，及飛行時間差(時間估測可靠度低) 的方法，本方法具有高準確及節省成本的好處，且由自製機器人實驗中，得到初步不錯的定位效果。其他議題包含喇叭與麥克風的替換，以及利用泰勒展開的最小方差法完成的多聲道定位。接著專注降低非直視線的誤差以及利用訊號能量來提升測距準確度的研究。最後作動態定位研究，利用過去估測的軌跡及速度，來修正時間記憶參數與泰勒展開參考值，以提升即時定位的準確度。

Localization aims to estimate the geographical position of an object based on range measurements of its surrounding sensors. Main applications include robot localization, UWB wireless sensor networks, and so on. We focus on acoustic localization of an indoor robot, where the sound is chosen as the transmitted signal because of its advantages of diffraction, low propagation velocity, and low cost over radio-frequency equipments. The range measurements usually suffer from some problems, such as synchronization problem cased by PC’s latency, sound velocity calibration, NLOS (Non–line–of-sight) error, which in turn degrade the accuracy of localization. We will propose a microphone-aided scheme to overcome difficulties of the synchronization and sound velocity calibration. Conventional methods include the auxiliary high-cost RF transmitters / receivers and the less accurate Time- difference-Of-Arrival approach. By comparison, our proposed method has both the advantages of high accuracy and low cost. The preliminary experiments of our robot setup show promising good localization performance. We will also investigate the issue of the position exchange of the microphone and loudspeaker, and multi-channel localization using the Taylor-series-based linearized least squares approach. In the issue of range estimation, a path-loss model is proposed to enhance performance of the conventional time-of-flight model. Improved NLOS mitigation algorithms and NLOS identification method will be proposed as well. Finally we focus on the adaptive localization of a moving robot. Its trajectory and velocity can be used to determine the time-variant forgetting factor and the reference point of the Taylor-Series-based least squares solution to facilitate fast adaptive position tracking. Computer simulations and real-time robot implementation of acoustic localization will be performed to demonstrate the effectiveness of the proposed algorithms.