Channel coding and source coding are two important problems in communications. Although both channel coding and source coding (especially, the distributed source coding (DSC)) can achieve their ultimate performance by knowing the perfect knowledge of channel noise and source correlation, respectively, such information may not be always available at the decoder side. The reasons might be because of the time−varying characteristic of some communication systems and sources themselves, respectively. In this dissertation, I mainly focus on the study of online channel noise estimation and correlation estimation by using both stochastic and deterministic approximation inferences on factor graphs.

In channel coding, belief propagation (BP) is a powerful algorithm to decode low−density parity check (LDPC) codes over additive white Gaussian noise (AWGN) channels. However, the traditional BP algorithm cannot adapt efficiently to the statistical change of SNR in an AWGN channel. To solve the problem, two common workarounds in approximate inference are stochastic methods (e.g. particle filtering (PF)) and deterministic methods (e.g. expectation approximation (EP)). Generally, deterministic methods are much faster than stochastic methods. In contrast, stochastic methods are more flexible and suitable for any distribution. In this dissertation, I proposed two adaptive LDPC decoding schemes, which are able to perform online estimation of time−varying channel state information (especially signal to noise ratio (SNR)) at the bit−level by incorporating PF and EP algorithms. Through experimental results, I compare the performance between the proposed PF based and EP based approaches, which shows that the EP based approach obtains the comparable estimation accuracy with less computational complexity than the PF based method for both stationary and time−varying SNR, and enhances the BP decoding performance simultaneously. Moreover, the EP estimator shows a very fast convergence speed, and the additional computational overhead of the proposed decoder is less than 10% of the standard BP decoder.

Moreover, since the close relationship between source coding and channel coding, the proposed ideas are extended to source correlation estimation. First, I study the correlation estimation problem in lossless DSC setup, where I consider both asymmetric and non−asymmetric SW coding of two binary correlated sources. The aforementioned PF and EP based approaches are extended to handle the correlation between two binary sources, where the relationship is modeled as a virtual binary symmetric channel (BSC) with a time−varying crossover probability. Besides, to handle the correlation estimation problem of Wyner−Ziv (WZ) coding, a lossy DSC setup, I design a joint bit−plane model, by which the PF based approach can be applied to tracking the correlation between non−binary sources. Through experimental results, the proposed correlation estimation approaches significantly improve the compression performance of DSC.

Finally, due to the property of ultra−low encoding complexity, DSC is a promising technique for many tasks, in which the encoder has only limited computing and communication power, e.g. the space imaging systems. In this dissertation, I consider a real−world application of the proposed correlation estimation scheme on the onboard low−complexity compression of solar stereo images, since such solutions are essential to reduce onboard storage, processing, and communication resources. In this dissertation, I propose an adaptive distributed compression solution using PF that tracks the correlation, as well as performs disparity estimation, at the decoder side. The proposed algorithm is

tested on the stereo solar images captured by the twin satellites system

of NASA’s STEREO project. The experimental results show the significant PSNR improvement over traditional separate bit−plane decoding without dynamic correlation and disparity estimation.