Funcionamiento, interfaz y formato de los datos en MaxEnt - Gbif.es

AbstractFinite-valued input two-dimensional (2-D) Gaussian channels with memoryrepresent an important class of systems, which appears extensively incommunications and storage. In spite of their widespread use, the workingsof 2-D channels are still very much unknown. In this work we tryto explore their properties from the perspective of estimation theory andinformation theory.At the heart of our approach is a mapping of a 2-D channel to anundirected graphical model, and inferring its a-posteriori probabilitiesusing generalized belief propagation (GBP). The derived probabilitiesare shown to be practically accurate, thus enabling optimal maximuma-posteriori (MAP) estimation of the transmitted symbols. Also, theShannon-theoretic information rates are deduced either via the vector-wiseShannon-McMillan-Breiman theorem, or via the recently derived symbolwiseGuo-Shamai-Verdú theorem. Our approach is also described from theperspective of statistical mechanics, as the graphical model and inferencealgorithm have their analogues in physics.Our experimental study, based on common channel settings taken fromcellular networks and magnetic recording devices, demonstrates that undernon-trivial memory conditions, the performance of this fully tractableGBP estimator is almost identical to the performance of the optimal MAPestimation. It also enables a practically accurate simulation-based estimateof the information rate. Rationalization of this excellent performanceof GBP in the 2-D Gaussian channel setting is addressed.Index Terms: Two-dimensional channels, intersymbol interference, maximuma-posteriori estimation, information rate, generalized belief propagation, clustervariation method, Shannon-McMillan-Breiman theorem, Guo-Shamai-Verdútheorem, magnetic recording channels, multiple-access channels.2