基於三維馬可夫隨機場模型之迭代式無線視訊解碼系統

Abstract

因為近期的60GHz高頻無線傳輸技術能提供相當大的無線網路頻寬，無線視訊傳輸系統被視為未來家用網路的一大重要特點。然而，在如此高的頻率下進行傳輸，無線訊號往往會受到相當的干擾。換句話說，此類無線視訊系統將會在相當低的訊噪比下進行傳輸。而如何在此種高干擾的無線網路下有效率的利用發送器的能量以及維持系統的穩定性便成了一個相當具有挑戰性的問題。在本論文中，我們提出了一個基於三維馬可夫場模型的迭代式共同視訊-通道解碼系統。提出的三維馬可夫場是一個基於位元層的馬可夫場模型，能成功的描述視訊中空間和時間的冗餘性。對於具有強烈攝影機移動的視訊內容，我們利用了實現在解碼端的運動估計來定位三維馬可夫場中的時間上的冗餘性。透過提出的三維馬可夫場模型，我們更進一步地提出了一個軟進軟出的視訊內容解碼演算法。提出的視訊內容解碼演算法有著比傳統的最大事後機率馬可夫場解碼更低的複雜度以及相當好的效能。最後，基於提出的視訊內容解碼演算法，我們更進一步設計基於此演算法的迭代式共同視訊-通道解碼系統。此系統能在解碼端根據視訊的內容估計三維馬可夫場模型的參數，而不需要在傳送端附加任何的變更，使得我們的解碼系統更能輕易地被整合。此外，在此解碼系統的設計中，我們利用了一個緩衝器來傳遞時間上的冗餘資訊，使得此時間上的冗餘資訊能夠不斷的在傳輸解碼過程中被傳遞並提升解碼效能。我們的模擬結果呈現出提出的模型和系統能夠大幅的增加無線視訊傳輸系統的效能及穩定性。

Due to the emergence of mmWave systems that provide multi-Gbps data rate over 60GHz band, uncompressed video transmission has been considered as a commonly used feature for wireless multimedia transmission over the wireless personal area networks (WPANs). However, mmWaves signals usually have higher attenuation than the conventional low-frequency wireless signals, and therefore supporting transmission in a low SNR condition becomes a challenging problem for such systems. In this thesis, we propose an iterative source-channel decoding method for uncompressed video transmission based on a proposed novel 3D-MRF model. The 3D-MRF model is a bit-plane level MRF model which successfully reveals the spatial and temporal redundancy of video sequences. For those video sequences of intense movement, we utilize the motion estimation at the decoder side to precisely locate the temporal movement under the block-base view, and thus the temporal redundancy can be fully used by the 3D-MRF. Through the 3D-MRF model, we develop the 3D-MRF based soft-in soft-out (SISO) source decoder. The 3D-MRF based SISO source decoder has much lower complexity than conventional iterative MAP-MRF decoder while still significantly improves the decoding performance. We design an iterative source-channel decoding structure which is capable of jointly estimating the MRF parameters at the receiver, and thus the proposed decoder can automatically adjust the amount of the information exchange according to the decoding video sequence. Also, by adding a single buffer into our decoder structure, the temporal extrinsic information is allowed to propagate throughout the video decoding process. We show that the proposed ISCD method can significantly enhance the video quality in terms of peak signal-to-noise ratio (PSNR). Even under very low SNR condition, the video frames are still maintained in good quality.