本文敘述並探討以 FPGA可程式化系統晶片針對生醫訊號處理之時域及頻域分析需求，以 Pipeline MIPS 架構為基礎設計生醫晶片訊號處理系統晶片指令集，目的在發展一套以
FPGA 為核心的 SoC 生醫訊號處理系統平台，並可整合控制周邊裝置、LCD、USB、Flash Memory、SRAM Memory；以建構可程式化 RISC 之 MIPS 運算系統及即時平行處理輸入訊號快速傅利葉 (FFT) 轉換後之功率頻譜，以可程式化之運算方式用於多樣變化之生理訊號，利用程式執行程序處理並能進一步分析生理訊號達到即時量測分析的效果，其指令集分為兩大類：尋常指令集在生醫晶片上主要功能為搬動資料及判斷比較資料等；特殊指令集為啟動平行運算旗標，並可做即時訊號處理、資訊儲存及資訊顯示等。本文以心率變異度分析為例研發生醫電訊號之應用，將SoC系統晶片與平台用於分析心率變異度特殊目的之處理，並討論即時生醫訊號處理系統雛型之建置與驗證。系統平台主頻率為 50 MHz、MIPS指令執行速度為 10 MHz、系統心電圖取樣率為 303 Hz，分析心率變異度以 5 Hz重新取樣等表現生醫電訊號分析之應用，即時心電訊號擷取與處理運算結果之 FFT頻譜與 MATLAB運算 FFT頻譜結果作驗證比對，其驗證頻譜數值結果線性相關度達 0.9926，此結果顯示，本系統平台與一般現行之監視系統更具有擴充性與實用性，可提升居家看護或 3C產品之開發運用。

The objective of this research is to develop a biomedical signal processing System on Chip (SoC) platform which is based on Field Programmable Gate Array (FPGA). The FPGA-based SoC platform is capable of integrating and controlling the peripheral devices, such as LCD, USB, Flash Memory, and SRAM Memory. It contains a RISC architecture microcontroller unit (MCU) and employs the microprocessor with interlocked pipeline stages (MIPS) technology. With a dedicated
co-processor, the platform was used to extract the power spectra of the acquired bio-signals through the real-time and in parallel Fast Fourier Transform (FFT) processes. The FPGA-based SoC was designed specifically for the frequency- and time-domain analysis of the bio-signals. There were two types of instruction sets in the algorithm control. One was the regular MCU instruction set whose function was to control the operation flow, to move data, and to perform arithmetic and logic operations. The other was the special instruction set used to set up parallel computing flags in order to real-time deal with the signals, to govern the data storage and to display the results. In this study, the processes for real-time analyzing the heart rate variability (HRV) were presented to validate the performance of the designed FPGA-based SoC platform. In the SoC platform, the main frequency and MIPS instruction frequency were selected to be 50 MHz and 10 MHz, respectively. The analog ECG signals were sampled by 303 Hz, and R-to-R intervals were re-sampled by 5 Hz before HRV analysis. Power spectra of the real-time ECG signals using the current platform were compared with those using the built-in algorithm of the MatLab. A correlation coefficient of 0.993 was found between the two kinds of power spectra. This suggests that the SoC platform proposed can be applied to real-time process the time- and frequency-domain analysis of the acquired bio-signals. The present reported SoC platform would be much easier to process bio-signal for the development of 3C consumer product.