Filter Construction

Filter models are represented as block diagrams consisting of interconnected filter elements. Filter elements may be either built-in numeric processing elements, or references to other filter models. Thus, filter models are hierarchical, in that they may be composed of other filter models. However, all filter models must be composed at the lowest level of the predefined filters described in Chapter 11 “Data Analysis” on page 422.

Filters operate on vectors. Vectors are discrete and ordered sets of numeric data, organized as entries. Each entry consists of an abscissa and an ordinate value. These are double-precision floating point numbers. A filter model may operate on one or more vectors and combine them to form its output, which must consist of just one vector. The vectors that are fed into the filter are called input vectors; the result of the filter’s processing is called the filter’s output vector.

A filter model block diagram must abide by several requirements concerning its structure. First, the graph which it consists of must be acyclic, so that none of a filter element’s input vectors depend on the calculation of its output vector. Second, each filter model must have exactly one unused output vector among all of its filter elements. A filter element’s output is used if it serves as the input of at least one other filter element. The unused filter element output is the output of the overall filter model.

Filter models are executed using a data-flow approach. A filter element can compute its output vector only if all of its input vectors are available. Initially, only the input vectors supplied by the user in the Results Browser are considered to be available. These will allow certain filter elements to be executed. Once these elements produce their output vectors, other elements will in turn be executed, until finally the last filter element is executed and the result of the overall filter model is produced.

There are two types of filter model inputs that the user may supply when executing a filter model from the Results Browser. These inputs are vectors (as described previously), and parameters. The input vectors of the filter model are needed to satisfy the filter element inputs that are not attached to the outputs of other filter elements. These are called unused inputs.

Certain filter elements have parameters which configure them to perform a precise operation. For example, the “windowed average” filter element has a parameter controlling its window size. The user may choose to specify parameters of filter elements when constructing the filter, or to leave these unresolved and to specify them at the time when the filter model is executed. This approach allows the construction of a library of filter models that are easily reusable.