用於基於腦機介面的物連網遍佈式雲霧計算架構

Abstract

近年來由於感測器效能的進步與穿戴式設備的興起，基於腦電圖的腦機介面研究在現實生活下的應用開始變得可能，但是碰到了一些挑戰，最為重要的問題是如何在動態的生活環境中，可靠且實時分析預測使用者的腦狀態，並及時反饋，為了得到可行的結果，產生了三個需求：(1) 持續傳輸並收集資料；(2) 實時的運算與反饋；(3) 匯流不同感測器資料。為此需求，我們採用雲霧計算的概念，在前端的感測器與後端的雲端伺服器間，加入霧伺服器作為近端，藉由靠近使用者(移動設備)與感測器的霧伺服器，來卸載它們的計算負擔，從而延長其電池壽命，並提高其數據處理性能。 為了達到持續傳輸、匯流、收集感測器資料的需求，本論文在霧端實作FogConsole，做為一個包裝了必要的功能模組的跨平台霧端軟體，方便使用者簡單、快速將手邊設備做為一個霧伺服器加入三階層式架構，並提出一個事件驅動的軟體架構和友善的介面，方便開發者開發FogConsole上的服務。利用此框架實作資料錄製的服務，解決收集資料的需求、並作為一個框架範例。在雲端上，本論文採用相同的框架設計，提出一個雲端服務框架，並以此實作繪製錄下資料的圖表之服務。在霧端與雲端服務的背後，本論文採用物連網中的MQTT做為感測器通訊協議串連設備、傳輸資料流，並以REST為服務風格，設計雲霧兩端的服務介面，以期達到高互動性。最後在測試中顯示，在資料傳輸面跟服務控制面上，此平台能有足夠的能力去達到腦機介面在可靠且實時應用上之需求。

In recent years, due to advances in miniature sensors and wearable devices, EEG-based Brain-computer interfaces (BCI) becomes possible in real-world applications, but faces some challenges in order to become truly useful in real-world environment. The main difficulty is to make reliable real-time analysis and even prediction of users’ cognitive states in dynamic real-life situations, then feedback to user on time. For finding reliable way to determine users’ brain state, it produces three requirements: (1) real-time and long term data streaming and archive, (2) real-time computing and feedback, (3) converge not only EEG signal but also other signals such as motion. With these requirements, we adopt pervasive Fog and Cloud Computing infrastructure for Internet-of-Things to overcome the challenges. It adds Fog Servers into Cloud computing architecture serving as the near-end computing proxies between the front-end devices and the far-end servers, and it can offload their computing burden so as to prolong their battery life and enhance their data processing performance. For requirement of long term data streaming, data converging, and data archiving, in Fog side, we implement a all-in-one and cross-platform software – FogConsole for users to easily add their nearby devices to be a Fog Server, then join this device to three-tier infrastructure. And we propose a event-driven framework and well-defined software interface for developers easily adding their service on it. Using this framework, we implement a streaming-archive service for users and researchers to easily collect the sensors data. In Cloud, we use same framework design and implement a data plotting service for users to plot a simple overview of recorded data. Finally, preliminary tests demonstrated its effectiveness to meet the requirement of real-time and reliable application.