Modular Algorithms for Multidimensional Signal Processing

在多维信号处理的模块化算法的标题下，正在研究四个主题。第一个主题是研究可以用级联格形式表示的有理传递函数的子类。其次，采用多通道方法对二维（2-D）信号的所有正交晶格构型进行表征。研究最大熵（ME）方法与模型约束的关系构成了研究的第三个主题。这项工作是基于研究者最近关于递归自回归模型的最优性的结果，即最小化与ME解决方案的一定信息理论距离。最后，将研究二维散射阵列的层剥离算法。本研究将点阵建模方法扩展到多维信号处理。由于其在VLSI和自适应处理方面的优势，这种晶格方法在一维信号（如音频）的情况下非常有影响力。研究人员正在努力解决一些难题，试图将这些优势扩展到二维情况下，以应用于视频和计算机视觉。

Four topics are being studied under the heading of modular algorithms for multidimensional signal processing. The first topic is an investigation of the subclass of rational transfer functions which can be represented in cascade lattice form. Second, a multichannel approach is made to the characterization of all orthogonal lattice configurations for two-dimensional (2-D) signals. A study of maximum entropy (ME) methods in relation to model constraints constitutes the third topic of study. This work is based on the investigators recent results regarding the optimality of recursive autoregressive models in terms minimizing a certain information theoretic distance from the ME solution. Finally, layer-pealing algorithms will be investigated for 2-D scattering arrays. This research concerns the extension of the lattice modeling approach to multidimensional signal processing. This lattice approach has been very influential in the case of one-dimensional signals such as audio because of its advantages for VLSI and also for adaptive processing. The researchers are working on the difficult problems of trying to extend these advantages to the two-dimensional case for application to video and computer vision.