Critical Wavenumbers in the Classification of Fractal Radiation Patterns

Motivation: Fractal models apply to radiation patterns from the so-called fractal antennas and to the description of wave propagation through a complicated environment. In antenna design the prototype antenna array factor has been provided by Weierstraß functions and their band- limited approximations. In antenna characterization one may want to assign a given (synthesized or measured) radiation pattern to a fractal class. Fractal Pattern Synthesis: Array antenna factors are synthesized by means of Weierstraß functions [1]. Two- dimensional fractal radiation patterns can be synthesized by separation of variables. Classification of Synthesized Fractal Patterns: The spectrum enhancement (ση) algorithm, introduced a few years ago [2 and references quoted therein] operates on the Fourier transform of a function of two variables. It evaluates derivatives of integer [3] or of fractional order [4], followed by some non-linear operations. By applying ση to a synthesized fractal pattern one obtains a vector of morphological descriptors, which are submitted to a trainable classifier. A typical classification result is displayed by Figure 1 where the center panel shows the fractal pattern class centroids on the plane of the first two principal components z1 and z2 . Each of the five classes (labelled 1, 3, 5, 6, 9) is composed of patterns such that b = 14 and 16. Moreover, in class 1, D = 1.1; in class 3, D = 1.3 (sample pattern on the right panel); in class 5, D = 1.5; in class 6, D = 1.6 and in class 9, D = 1.9 (sample on the left panel). Problem to be Addressed: There are critical values of the wavenumber [5] at which the nu- merical estimation of dimB [graph[f ]] is affected by relevant errors. The goal of this investigation is to determine whether or not the same wavenumber values are critical for the ση algorithm as well and, in case they are, provide an explanation.