Design and Implementation of a Distributed Video Decoder

在傳統視訊編碼中有許多標準化的視訊編碼格式，像是動態影像專家組(Moving Picture Experts Group)提出的MPEG格式、國際電信聯盟電信標準化部門提出的H.26X格式等等，這些傳統視訊編碼的特點是盡量讓編碼器探索冗餘的時間和空間資訊，也就是使用影格內編碼或影隔間編碼來壓縮視訊，在這樣架構下的編碼器負擔很大。但是傳統的視訊編碼不適用於只擁有低運算複雜度的編碼器或考量到低功耗的編碼器，像是無線視訊感測器或監視系統，分散式視訊編碼技術因此被提出，分散式視訊編碼的根本以Slepian-Wolf理論和Wyner-Ziv理論為基礎。 在此論文中，我們設計並實現一個分散式視訊解碼器，提出的架構包含累積低密度奇偶檢查碼解碼器、相關雜訊模型、軟輸入計算、邊資訊產生，實作採用TSMC 90nm GUTM製程完成分散式視訊解碼器，針對QCIF大小的測試序列可解碼達到每秒30張的影像，此晶片工作頻率可達到100MHz，使用的邏輯閘數目為690K，晶片大小為4.67 mm2。In conventional video coding, lots of coding techniques were proposed to exploit redundancy of the spatial and temporal signals in the encoder. These coding techniques are focused on compressing video data as high as possible, such as MEPG (Moving Picture Experts Group) and H.26X proposed by International Telecommunication Union. These conventional techniques tried to extremely reduce spatial and temporal redundancies, so all the computing efforts were majorly spent on the encoder. Conventional video coding is not suitable for the situations where little encoders and big decoders are used, like wireless video sensors that have a low computing capability and need a low power consumption. To meet the requirement of a low-complexity and low-power encoder, a new video coding paradigm, distributed video coding, based on Slepian-Wolf Theorem and Wyner-Ziv Theorem, was proposed. In this Thesis, we design and implement a distributed video decoder which is majorly composed of LDPCA, correlation noise modeling, soft input computation, and side information creation. Our proposed DVC decoder, implemented in TSMC 90nm GUTM process technology, can meet the requirement of decoding a QCIF video with a speed of 30fps. The maximum operation frequency is 100MHz, the chip area is 4.67 mm2, and gate count is 690K