移動感測系統的省電技術研究

Abstract

現今移動感測裝置普遍配置全球定位系統(GPS)，提供準確的地點資訊，但因GPS的耗電量高，所以必須在省電與地點準確度上取得平衡。根據不同的需求，我們提出了相對應的省電機制。在山難搜救方面，目標是收集越多個定位準確的資訊越好，從收集到的登山路徑資料，我們觀察到，GPS收訊的情況與時間跟空間有關，因此我們提出一個可調式工作週期(Adaptive Duty Cycle, ADC)機制來解決這個問題，並使用真實的登山客行走軌跡來驗證我們的ADC機制，並得到帕累托最優解(Pareto optimum)。 在軌跡記錄方面，需求是讓移動感測裝置能記錄完整且均勻的軌跡資料，從GPS模組的使用特性，我們觀察到，GPS定位失敗的耗電量遠大於GPS成功的耗電量，因此我們提出了預算基底自調式工作週期(Budget-based Duty Cycle, BDC)機制，這個機制讓使用者輸入需要記錄的時間長度與限定耗費的電量，並採編列電量預算的方式，保留電力以度過沒有GPS訊號的區域，BDC機制將自動調整GPS的開啟週期。 在環境感測方面，我們提出了自調式返回感測(Adaptive Return-to-Home Sensing, ARS)機制，當無人機在開放空間中進行環境感測時，我們的目標是讓無人機能在預定的路線上感測環境資料後，並能順利返回補充機體電力與上傳環境資訊，我們同時也提出了動態調整ARS機制參數的演算法，結合了單純貝氏分類器(Naive Bayes Classification, NBC)與二元搜索法(Binary Search, BS)，不旦提高了無人機的返回成功率也使得感測的區間分佈的更均勻，最後ADC、BDC與ARS，不僅實作簡單，也針對不同清況下，完成任務並有效的提高能源的使用效率。

Mobile location sensing applications (MLSAs) represent an emerging genre of applications that exploit Global Positioning System (GPS) technology and facilitate location-based services. The design of MLSAs must incorporate a trade-off between information accuracy and energy efficiency because GPS technology is energy expensive and unaffordable for most MLSA platforms, which are battery-powered and therefore resource-constrained. Each scenario has different requirements and presents unique challenges. For example, the hiker tracking scenario requires timely and accurate location information, and as many location coordinates as possible must be collected. Based on our observation that the reception of GPS signals is spatially and temporally correlated, we propose an algorithm called the Adaptive Duty Cycle (ADC) scheme to exploit the spatio-temporal localities in the design of GPS scheduling algorithms. Using a comprehensive set of evaluations, as well as realistic hiker mobility traces, we evaluate the ADC scheme in terms of data granularity and power consumption. The results demonstrate that the scheme can achieve the Pareto optimum in all test cases. For the trace logging scenario, using mobile devices to continuously log a trace over a period of time presents many opportunities for emerging applications. Most such applications are related to recreational activities that use mobile devices instead of paper maps to determine precise locations. GPS is preferred over GSM or Wi-Fi based position systems because of its accuracy. Duty-cycling GPS provides a trade-off between positioning accuracy and lower energy consumption. However, a non-uniform trace will make interpretation of the logging trace more challenging. To address these issues, we present Budget-based Duty Cycle (BDC) scheduling for time-bounded tracking. The method enables a mobile device to effectively log a complete trace over a period of time, while consuming a given amount of the device’s energy. More importantly, BDC uses a series of techniques that preserve power to ensure that the trace is completed and its sampling interval is uniform. BDC was motivated by our observation that GPS locks are not always successful during the GPS duty-cycle, and the power cost of a failed lock is greater than that of a successful lock. The method uses budget power to preserve power for failed GPS locks and automatically calculates the time interval of a lock from the remaining energy. Budget power employs the BDC-Hybrid function, which is a combination of two functions, namely, the BDC-Linear and BDC-Step functions. The former is a naive method that is used when GPS locks succeed, while the latter is more analytical and is used when GPS locks fail. Budget power is concerned with power preservation as well as the uniformity of the sampling of a trace. For the environmental sensing scenario, we propose an algorithm called Adaptive Return-to-Home Sensing (ARS) for a drone sensing system deployed in an open area to conduct periodic environmental sensing. The ARS scheme can perform as many rounds of environmental sensing as necessary without drastic oscillations between consecutive sensing attempts and still conserve sufficient energy for the drone to return home. We also present a parameter-tuning algorithm that combines Naive Bayes Classification (NBC) and Binary Search (BS) to adapt the ARS scheme parameters effectively on the fly. Finally, we evaluate the ARS scheme under a variety of environmental difficulties. The results demonstrate that the scheme is effective in mitigating oscillations of spatial distance between consecutive sensing attempts. The NBC enhanced ARS scheme is better able to guarantee the Return-To-Home (RTH) feature, and it is more cost-effective in terms of parameter tuning than other machine learning based approaches. Moreover, the ADC, BDC and ARS schemes are simple, effective, and generalizable to other mobile location sensing applications in different scenarios.