快速獨立成份分析於多重氣體感測器應用之低功耗演算法暨架構共同設計

Abstract

隨著科技發展，多重感測器的大小與功耗之需求逐漸升高，要將其嵌入可攜式裝置也變得相對困難。然而，等待未來3DIC技術發展成熟後，感測器系統將可以與電路整合並置入手機內。一個感測器系統是訊號由類比前端接收 (AFE)，類比數位轉換 (ADC)，數位訊號處理，最後無線系統來傳輸至其他裝置。 本論文設計數位訊號處理的部分，嘗試優化並改善獨立成分分析系統 (ICA)。此系統可將多重氣體感測器接收的混雜訊號分離成原始的氣體訊號。我們的目的是要利用改良演算法以及基礎架構的方式設計出一個低功耗的快速獨立成分分析系統 (FICA) 且能在場式可程式閘陣列 (FPGA) 上實現。我們完成了功耗、計算時間和分離訊號準確度的分析與比較，並找出最適合的設計方式。在基本的獨立成分分析系統上，我們還附加了一些創新的功能來優化計算時間並達到系統的要求。動態系統可讓氣體訊號尚未完全擷取前即開始運作，以加速訊號處理。此外，穩定度預測元件可以偵測氣體訊號穩定度。待所有氣體都穩定後，會告知系統提早進入空閒模式 (IDLE mode)，再利用時鐘閘控 (Clock Gating) 來節省能量。 我們利用這些演算法與架構的低功耗方法所設計完成的快速獨立成分分析系統，耗能和電路面積都大大的降低，分離出來的訊號準確度也符合氣體感測器之標準。最後，我非常期待攜帶式多重氣體感測器的產品─電子鼻能在不久的將來在市面上出現。

Since the size and energy consumption of multi-sensors developed in recent years are large, it is complex to make them portable. However, with the help of the technique of future 3DIC, the sensor system can be integrated into mobile applications. The sensor system is composed of AFE receiving, ADC, digital processing, and wireless transmission. We designed the digital processing part, trying to optimize and improve Independent Component Analysis system, which recovers the mixed signals received by Multi-Gas Sensor into original source signals. The aim of this paper is to co-design a low power Fast ICA system and implement into FPGA by optimizing algorithm and basic architecture. The analysis and comparison of different sizes system are accomplished to trade-off between power, delay, and accuracy of extracted signals. Several novelties are attached in the design to meet the requirements and optimize the gas signal extraction. The system starts before all signals approach to speed up the signal processing. In addition, stability check helps the system enters termination mode by clock gating when the gas signal is stable. Finally, the power as well as area can be reduced so the portable multi-gas sensor concept, even electronic nose can be fulfilled.