在本論文中,我們提出一顆可將資料做雙基底對數運算的數位信號處理器,利用其專屬的指令集將語音辨識演算法實現於數位信號處理器。使用硬體電路搭配軟體演算法整合設計的方式,來設計出最佳的電路架構以及專屬的特殊指令集。在電路架構設計方面,處理器具有雙基底對數運算的設計。若處理器中無對數運算的指令,在演算法中需使用對數運算,則必須利用其他方法來實現對數運算,因而費時且精準度不符要求,為此我們特地設計出一個具有對數運算能力的數位信號處理器,來快速處理演算法中的對數運算並且能得到較精確的結果。最後此晶片應用於TSMC 0.13um製程合成,其合成完成後的Gate count約為55萬,工作頻率最高約為50MHz。
In this study, we proposed a digital signal processor which supports dual logarithm based and use its specific instruction set to implement the speech recognition algorithm. We use hardware-software co-design methodology to optimize the processor architecture and instruction set. In terms of architecture, the processor supports dual logarithm based operator. If the processor doesn’t support logarithm operator, we must use other method instead logarithm operator when using it. But it cost a lot of time and its resolution is not enough precise to we need. So we design a processor which supports logarithm operator to evaluate logarithm quickly and get more precise resolution. Finally, the total gate count of this processor is about 550,000 synthesized and estimated with TSMC 0.13 um standard library. The maximum clock frequency of this processor is about 50MHz.