用於生醫訊號處理之高效率十六通道獨立成分分析RISC-V處理器

Abstract

獨立成分分析(Independent Component Analysis)能夠分離多通道生理訊號中的雜訊以提升訊號品質，以提供各種訊號處理應用，像是癲癇偵測。然而，使用硬體進行獨立成分分析運算會有兩個瓶頸，分別是面積限制與硬體可利用性。獨立成分分析會有大量矩陣運算，複雜度會是O(n2~3)，因為面積的緣故，通道數為四到十六通道，難以超過此數量。使用硬體進行即時的獨立成分分析對於生醫訊號處理可以提升訊號品質並且降低通道數量。然而，在電極片與位置固定的狀況下，在獨立成分分析的特徵值分析所得到的特徵值在每日的動物實驗中前後變化不大，只需要一次性的特徵值分析，因此本論文提出使用RISC-V處理器與韌體去取代硬體實現特徵值分析，比起傳統的純硬體加速器而言，可以節省大量面積，並且提高通道數量，而且可以整合其他相關功能，變成系統的核心。

To improve the performance of epileptic seizure detection, independent component analysis (ICA) is applied to multi-channel signals to separate artifacts and signals of interest. FastICA is an efficient algorithm to compute ICA. However, the number of channels will limit by chip area. ICA has much matrix operations. The complexity of matrix is O (n2~3). It is hard to extend channel out of four to eight channels. In the case of ECoG signal, the electrode patches are fixed after installation. Some of the FastICA algorithm can be reduced because of fixed electrode patches. We only need one-time singular value decomposition for eigenvalue at first time. We can implement singular value decomposition by firmware in stand of hardware to reduce area. We can have more area to get more channels. The microcontroller can implement other protocol and signal process making this chip more applicable. The performance of the chip was verified by human dataset.