A different representation of recursive systematic convolutional (RSC) codes is proposed. This representation can be realized by a conventional Tanner graph. The graph becomes a tree by introducing hidden edge. It is shown that the sumproduct algorithm applied to this graph model is equivalent to the BCJR algorithm for turbo decoding with lower computational complexity. The message-passing chain more exact of the BCJR algorithm is shown in the graph. In addition, the proposed representation of RSC codes provides an efficient method to set up the trellis and the conventional Tanner graph for RSC codes provides directly the architecture for decoding.

A new iterative processing for image restoration is proposed based on the turbo iterative principle. For this purpose, two uncorrelated component images issued from an original image are differently filtered and the pieces of information extracted from both filters are mutually exchanged in the iterative process. This scheme has achieved much better performance and reduced computational cost in restoration of image both with additive noise model and multiplicative noise model (such as SAR image).

A concept of statistical multiresolution analysis in amplitude-frequency domain is proposed, which is to employ the wavelet transform on the statistical character of signal in amplitude domain. In terms of the theorem of generalized ergodicity, an algorithm to estimate the transform coefficients based on the amplitude statistical multiresolution analysis (AMA) is presented. The principle of applying the AMA to Synthetic Aperture Radar (SAR) image processing is described, and the good experimental results imply the AMA to be an efficient tool for processing of speckled signals modeled by the multiplicative noise.

提出幅度-频率域统计多分辨率分析的概念，即在幅度域上对信号的统计特征作小波变换。根据广义的各态历经定理，给出幅频统计多分辨率分析的估计算法。描述了该分析方法用于合成孔径雷达图像处理的原理，应用结果表明，幅频统计多分辨率分析为处理具有乘性模型的斑点噪声信号提供了一个有效的新途径。