• 作者： 郭鍾榮; 徐裕峰
• 作者服務機構： 中正大學電機工程學系; 清華大學電機工程學系
• 中文摘要： 本文在評估統計相關／無關／獨立信號的純量與向量量化效率。倘若我們對原始統計相關信號做KL轉換，那麼即可得到統計無關信號。如果我們先利用循環全通瀘波器把原始信號轉換成具有聯合高斯分佈的信號，再做KL轉換，那麼就可得到統計獨立信號。而藉著把信號由統計相關轉換成統計獨立，我們不僅可提高純量量化的性能更能使其與向量量化相抗衡。我們並發現統計獨立信號可以輕易地得到，而且純量量化也比向量量化更適用於非靜態信號。因此循環全通濾波器，轉換和純量量化的結合可在實際應用上為向量量化的代替品。
• 英文摘要： The efficiency of both scalar and vector quantization for a dependent/uncorrelated/independent signal isevaluated here. The uncorrelated signal is obtained by the Karhunen-Loeve transform of the original dependentsignal. To obtain the independent signal, the original signal is first converted into a joint Gaussian signal by acyclic all-pass filter. Then, the Karhunen-Loeve transform is used to obtain the independent signal. By convertingthe signal from uncorrelated into independent, the performance of the scalar quantization improves and is comparablewith that of vector quantization. It is shown that the independent signal can be easily obtained and that scalarquantization is more suitable for a nonstationary signal than is vector quantization. Therefore, the combinationof a cyclic all-pass filter, transformation and scalar quantization provides a substitute for vector quantization inpractical applications.
• 中文關鍵字： scalar/vector quantization; dependent/uncorrelated/independent random variables
• 英文關鍵字： --