於行動裝置上應用多執行緒實現加速FastICA演算法

Abstract

本篇論文展示將FastICA演算法實作於行動手持裝置中並且利用多執行緒達到加速的目的。藉由JAVA程式語言撰寫與多執行緒的技術，FastICA演算法能夠有效地在四核心的行動手持裝置上處理並分析腦電波的訊號。對於FastICA演算法來說，每一個矩陣的運算能被分割成多個子運算，不同的子運算交由不同的執行緒處理。藉由這樣的方式達到平行化運算的加速效果。根據不同矩陣運算的實驗數據結果，我們採用四個執行緒處理矩陣相乘的運算並在四核心的行動手持裝置中實作FastICA演算法。最後，我們展示在行動手持裝置上利用多執行緒實作FastICA演算法的結果同時驗證此方法的正確性。根據實驗結果數據顯示，此方法與未平行化的程式相比最多可減少43.17%的運算時間。

This thesis presents the implementation of FastICA algorithm acceleration using multi-thread on the mobile device. By using multi-thread technique with JAVA programming language, the FastICA algorithm can be processed efficiently for the EEG signals on the mobile device with quad-core processor. For the FastICA algorithm the matrix operation in the FastICA algorithm can be divided into several parts to utilize the multi-thread technique. Based on the experiment of different matrix operations, we adopt the four-thread technique for the matrix multiplications for the FastICA algorithm on the quad-core processor based mobile device. The implementation of the FastICA algorithm acceleration using multi-thread is validated on the mobile device. According to the experiment result, the proposed implementation of FastICA algorithm acceleration can reduce the execution time up to 43.17% compared to the single-thread implementation.