應用於聯網自駕車之可見光通訊及相對定位系統

Abstract

聯網自駕車車隊被視為能夠大提升道路使用率、能源效率、行車安全的解決辦法之一。在車隊中，車間保持5到10公尺的車距，以相同的速度行駛。為實現車隊的運作，車子必須透過車間通訊即時的傳送當下的行車數據（如速度、位置、油門煞車位置及方向盤轉向等）給車隊中的其他車輛，使所有車輛能達到行駛狀態的一致性。因此，車間低延遲通訊及精準定位為實現車隊的兩大技術元素。另一方面，現今的車輛廣泛使用發光二極體在頭燈、尾燈等燈光系統中。發光二極體即是一個很適合的通訊元件，可用以發展可見光通訊。因光的直線前進特性，使得傳輸過程的干擾較小，在通訊及定位上都具有極大的優勢。如能使車燈具有通訊及定位的功能，將可大幅降低成本，以利技術在市場上的推廣。 在本篇論文中，我們基於可見光發展了兩項通訊技術與一項車輛相對定技術，用以實現車隊運作。首先，我們開發一套用以傳輸行車數據的可見光通訊系統，以降低車隊中採用無線射頻傳輸資料的比例，減少訊號壅塞的可能性。此系統基於用數位微型反射鏡裝置、感光元件與發光二極體，為一套可根據傳輸端位置動態縮小與變換可視角範圍的可見光通訊系統。該系統可過濾環境光以及來自其他傳輸端的干擾，藉此提升訊噪比，增加資料傳輸率。在高速公路上實測的結果顯示，系統在35公尺的距離下具有超過90%的偵測率，資料傳輸量為607.9 kbps。針對第二項可見光通訊系統，我們的目標為設計出一款具有高空間解析度之可見光通訊，支援多個傳輸端同時傳輸資訊，可作為旁路系統實作資安協定，確保無線射頻通訊安全。我們利用動態視覺感測器作為可見光通訊的接收端。動態視覺感測器可偵測畫面中亮度有變化的地方做輸出，直接過濾環境光造成的干擾。我們首先測試感測器的特性，觀察其對於不同通訊波形的輸出表現。我們選擇以具有曼徹斯特編碼支開關鍵控（OOK with Manchester coding）作為訊號調變設計封包，並訓練一個基於長短記憶模型（LSTM）的分類器來對封包中的每一個位元進行解碼。在高速公路上實測的結果顯示，此系統的資料傳輸量為2.96kbps在30公尺下可達到低於15%的位元錯誤率，此傳輸量可用交換密鑰，確保車隊裡的通訊安全。最後，我們利用投影機模擬LED陣列頭燈開發一套車輛相對位置定位系統，接收端為自行設計之顏色感測器陣列。我們利用差分相移鍵控（Differential phase shift keying）設計投影圖案，使接收端僅需一次觀察即能解碼觀察到的空間訊號，計算出感測器與投影像素的關係。接著，透過分析接收到的頻率與解碼後的相位進行定位。戶外行車實驗結果顯示，此系統達到公分等級的精準度，在20公尺時，定位誤差為30公分，相對行進方位誤差為9度。

The platoon formation of connected autonomous vehicles (CAVs) is crucial in the next-generation intelligent transportation system (ITS), improving road utilization, fuel efficiency, and safety. In a platoon, vehicles continuously share their status information (e.g., the speed, positioning information, throttle, and brake application) via vehicle-to-vehicle (V2V) communications to maintain a uniform driving strategy. As a result, V2V communication and positioning are regarded as two essential components in realizing platooning. Meanwhile, modern vehicles are equipped with LED lighting systems for headlights, taillights, and other lighting. LED is a suitable transmitter (sometimes even receiver) for developing visible light communication (VLC). Leveraging LED technology for communication offers several advantages, including line-of-sight propagation, The line-of-sight propagation of the light, which reduces interference caused by multipath effects and enhances communication and positioning performance. Transforming automotive lighting into a triple-functionality device - communication, positioning, and illumination - can lead to significant cost savings and efficient utilization of resources. This dissertation focuses on enabling communication and positioning capabilities using visible light, presenting two visible light communication systems and one visible light positioning system. First, we present RayTrack, aiming at offloading the data transmission within platoons to limit the network congestion. RayTrack is an interference-free outdoor mobile VLC system composed of a digital micromirror device, photodiodes, and an LED. It dynamically narrows and adjusts its field-of-view according to the transmitter location, effectively mitigating interference from the environment and other transmitters, thus boosting the system throughput. Real-world driving experiments show that RayTrack can achieve a data rate of 607.9 kbps with over 90% detection accuracy at 35 m at driving speeds of 70 - 100 km/hr. The second VLC system targets high spatial resolution for supporting simultaneous data reception from multiple transmitters. We adopt a new type of CMOS sensor called dynamic vision sensor (DVS). DVS monitors the brightness change in each pixel and only generates outputs when it detects a significant change. To investigate the potential of a DVS for communication, we examine the event detection performance of DVS with various modulation schemes.ㄒAccording to the preliminary results, we use OOK with Manchester coding for modulation and train an LSTM-based classifier to decode the OOK symbols. Real-world driving experiments show that the DVS-based VLC system achieves a data rate of 2.96 kbps with a bit error rate (BER) lower than 15% at a distance of 30 m. The proposed DVS-based VLC system can serve as a supplementary channel for enhancing the security of vehicle-to-vehicle (V2V) communications in platoons. In the last part of this dissertation, we introduce MatrixLoc, a vehicle localization system based on a projector serving as the headlight with a customized color sensor array. We adopt differential phase shift keying (DPSK) to create a fringe pattern, enabling single-shot positioning. Accurate positioning is achieved by analyzing the received frequency and demodulated phase information. Real-world driving results show that MatrixLoc can achieve centimeter-level positioning with a positioning error of 30 cm with a bearing error of 9 degrees at a detection distance of 20 m.