基於EMD運算之呼吸訊號處理系統設計與實作

Abstract

在此篇論文中，我們發展一個基於FPGA平臺的EMD呼吸訊號處理系統，這個系統包括資料擷取系統、基於FPGA的 EMD 加速器、與呼吸訊號經由EMD運算完成後顯示IMF結果的圖表系統，並透過高速網路將這三個系統串連起來與完成資料傳遞，可以達到一個低延遲、即時的呼吸訊號處理系統。此篇論文基於FPGA設計4級管線EMD加速器並採用FIFO結構來降低Cubic Spline的計算延遲；對於精準度方面，本篇採用38位元的浮點運算來近似雙精度正確性；為了處理大量資料EMD計算而採用了外部記憶體方式來設計此應用。我們設計的EMD加速器與使用210MHz ARM11 MPcore平臺的效能比較上，當處理10K資料量的情況下，所提出的系統運算速度比使用一顆核心運算的ARM CPU快29.21倍，比使用2顆核心同時運算的ARM CPU快20.20倍，比使用4顆核心同時運算的ARM CPU快15.47倍。

In this thesis, an EMD-based breathing signal processing system using FPGA is developed. The system consists of 3 sub-systems including data acquisition, FPGA based EMD accelerator and display system which can shows EMD processing results. A high speed internet links these 3 sub-systems such that low latency, low delay and real time data transportation is achieved. In this work, the four-pipelined EMD accelerator is implemented and FIFO structure is used to reduce the computation delay of Cubic Spline. In terms of accuracy, 38-bit floating point data format is adopted to approach the double-precision floating-point accuracy. An external SSRAM architecture is adopted to process a large amount of data. Comparing the EMD accelerator and 210MHz ARM11 MPcore platform, for 10K data size, the EMD accelerator can performs faster by 29.21 times than ARM11 processor running single core, 20.20 times than ARM11 processor running 2 cores, 15.47 times than ARM11 processor running 4 cores, respectively.