Selection and performance of probabilistic tables used in non-model based signal prediction

Conditional probabilistic tables are used as a means to estimate the future value of a signal when little if any knowledge on the system that gives rise to the signal is available. The size of the probabilistic table, namely its dimension and the quantization employed, determines its ability to distinguish between very similar actual conditions on the signal, and therefore, when correctly selected, enables accurate predictions to take place. In the paper, the maximum number of point intersections within an appropriate time window that the signal induces on a zero slope line forms the basis of an heuristic rule for the selection of the dimension of the probabilistic table. The performance of tables with smaller and larger than necessary dimensions are compared against each other and also against two other prediction schemes which involve linear observers and a "zero-order hold".