Joint Source And Channel Coding Based On Unequal Error Protection

Shannon's theory that source coding and channel coding can be optimized separately is based on the following two assumptions. That is the source codes and channel codes could be infinite length and the coding process could be unlimited complexity. These two assumptions are impractical. In fact, the "Separate Theory" restricted the system to get optimal result. Thus, people began to lay their eyes on joint source and channel coding (JSCC) to design better systems. The basic concept of JSCC is considering the characteristics of source and channel synthetically to obtain an optimal method which could provide better error protection for information communication and more efficient bandwidth usage. Unequal error protection (UEP) is one branch of JSCC. People notice that the different parts of coded source stream are of unequal importance. The observation leads to the idea of UEP which means providing scaled protection for different parts of the source signal. For instance, video communication is an error tolerance service. The QOS of different parts of coded video streams are different since the importance distribution of video streams is not uniform. According to the interior characteristics of coded video data, we could design joint source and channel coding system to improve the system's performance.In this dissertation, we first integrated the works of joint source channel coding based on UEP in recent years and give a catalog of the work of JSCC. Then we analyze the method of UEP based on GOP in video communication and conclude some problems need to be solved: (1) How to make judgments on the importance of the data encoded by video compression system? (2) How to make a decision on the code rate of channel codes? To solve these problems, we analyze the error concealment method utilized by H.264 and construct a channel distortion estimation method. That is the pixel interpolation prediction method used by intra-frame coded macro-blocks and"copy macro-blocks"method used by inter-frame coded macro-blocks. Furthermore, the approach takes into account the error propagation problem of video communication and considers the influence of the non-linear in-loop filter used by H.264. After that, we construct a channel rate distortion model composed by two parts which are caused by channel error and error propagation. The experiments'results demonstrate that our model is of high efficiency.On the basis of previous analysis, we construct a channel rate allocation model for the purpose of obtaining the minimum channel distortion. Since our problem doesn't satisfy the constraints of the traditional methods which are used to solve optimization problems, we choose Genetic Algorithm to get the optimal channel code rate allocation result. Combined with the property of Genetic Algorithm, we analyze the convergence property of our problem and set value of some important parameters. In our experiment, we utilized H.264 standard. From the result of channel code rate allocation results, we could notice that our model combined with error concealment property, the moving property of original sequence. So the experiments'results tell us that the frames which are in the tail position might be protected by more robust channel code. Besides that, we analyze the channel code rate allocation results of sequences of different kind and conclude that the smoother the channel distortion curve is the less mutation the channel code rate allocation curves are. Experiments show that, compared with existed approach and equal error protection method, our rate allocation method effectively improves the system performance. Especially in the situation of awful channel condition, our system could gain 3-4dB compared with EEP.