模組最佳化設計運用在離散時間積分三角類比數位轉換器

Abstract

傳統積分三角調變器的設計上，需要不斷的嘗試各種不同的規格，是一個相當耗時的工作。本論文是第一個提出使用模組化設計的方法去設計積分三角調變器。在時間上，使用模組化的設計方法將會比使用行為模擬的方法快上萬倍。由於之前非理想雜訊及失真模型不完備，模組化設計的方法一直無法真正的去實現。現在，因為積分器充放電雜訊模型以及放大器的非線性直流增益諧波失真模型的推出，使得模組化設計方法得以實現。另外使用行為模擬的方法在最佳化的過程中，必須要搭配演算法讓設計時間縮短，例如基因演算法或模擬退火法。最後我們還比較模組化設計和行為模擬搭配模擬退火法之間的優缺點。

Conventional high-level ΣΔM synthesis is mainly based on behavior simulation which is very time-consuming. The model-based approach can be 50000 times faster than simulation-based approaches, but it is never realized before due to the incompleteness of non-ideality models. The Recent establishment of a settling noise model and an OTA distortion model facilitates fast model-based ΣΔM designs which involve no time-consuming behavior simulations. Nonetheless, new problems associated with model-based approaches arise, notably the correlation between noises and the overload on quantizers. In this thesis, noise correlation models are derived, and a ΣΔM input magnitude estimator is proposed to avoid quantizer overload. Quantitative comparisons between model-based and simulation-based approaches are made. It is demonstrated that, compared with simulated annealing by the simulation-based method, direct search by the model-based method is much faster and reliable, and produces better results.