針對建立於藍芽4.0之動作辨識應用之電源管理方法

Power Management for Motion Applications Based on Bluetooth 4.0 Low Energy Technology

黃凱伶

指導教授 ： 周百祥

國立清華大學/電機資訊學院/資訊工程學系/碩士(2013年)

https://doi.org/10.6843/NTHU.2013.00713

摘要

利用可穿戴式的感測器做動作辨識的應用中，功率消耗一直是一個很大的挑戰。常為了節省功率的消耗，而造成系統的敏感度(sensitivity)、特異性(specificity)降低和延遲時間(latency)增加。因此，在此論文中，我們針對動作辨識這一類型的應用，提出了一個全新的動作辨識方法，此動作辨識方法解決了以往會因不同使用者而辨識率降低的問題，並針對此動作辨識方法，我們提出了三個電源管理的方法。第一個方法是階層式滯後性閾值(tiered hysteretic threshold)，此方法能讓系統在未使用時進入低階省電模式，但當系統利用硬體偵測到動作時，又會調整到高階模式，以維持一定的敏感度和特異性。但有鑒於第一個方法的閾值都是事先經由開發者先定義好，無法適用於每個不同的使用者，因此我們提出第二個電源管理方法名為自適應式閾值(adaptive threshold)。最後，由於不同的無線傳輸協定會影響功率的消耗，因此在第三個電源管理中，我們提出使用藍芽4.0低功耗，並找出對此類型應用最佳的幾個藍芽4.0低功耗的參數。經由實驗驗證，此動作辨識方法精準度相當高，並且這三個電源管理方法確實為我們系統省下高達70%的耗電量。

關鍵字

動作辨識 ； 低功耗 ； 可穿戴式裝置 ； 電源管理

並列摘要

Power consumption is one of the most challenging aspects in designing wearable sensors for motion tracking, as power must be saved without sacrificing sensitivity, specificity, and latency. In this thesis, we propose a vertical suite of techniques for building power-efficient wearable motion sensing systems. First, we propose a new motion recognition method based on edit distance, which is able to recognize each action over a wide range of motion amplitudes while lending itself to power management techniques at other levels. Of the three power management methods we propose for prolonging the battery life, the first is tiered hysteretic threshold, which increases sensitivity at the lower tier to
wake up the next higher tier only when it deems necessary. Thus, the second power management method is adaptive thresholds, which adjusts the threshold levels automatically over time to adapt to the specific users. The third power management method exploits specific features of the wireless
protocol, Bluetooth 4.0 Low Energy Technology (BLE), in integrating our tiered hysteretic and adaptive thresholding techniques to make our system energy-efficient while being directly compatible with smartmobiles without the high Rx power during idle listening. Experimental results validate
the ability to recognize motions with high accuracy. These three power management methods help motion application reduce power consumption, thereby enhancing the wearability of our body sensor systems.

參考文獻

[13] BOURKE, A., VAN DE VEN, P., GAMBLE, M., O’CONNOR, R., MURPHY, K., BOGAN,

[9] AGRAWAL, S., CONSTANDACHE, I., GAONKAR, S., ROY CHOUDHURY, R., CAVES, K., AND

[2] Bluetooth Low Energy. http://www.bluetooth.com/Pages/low-energy.aspx.

[4] Python Programming Language a˛V Official Website. http://www.python.org.

[5] Silicon Laboratories. http://www.ember.com.