The Use of Statistical Image Classification as a Convergence Criteria in Phaseless Near-Field Antenna Measurements
Authors: Stuart F. Gregson, John McCormick, Clive G. Parini
The utility of a variety of objective, quantitative and robust methods of assessing similarities between antenna measurement data have already been highlighted in the literature. These techniques essentially involved the extraction of interval, ordinal, and categorical features from antenna pattern data sets that can then be effectively compared to establish a measure of their adjacency, i.e. similarity. Hitherto, such techniques have primarily been limited to the purposes of comparing two or more images as a means in itself, e.g. far-field three-dimensional radiation pattern of a given antenna having been characterised using two different facilities. In contrast, this paper discusses the utility of such techniques for the purposes of establishing convergence within an iterative optimisation process, namely the phase retrieval (PR) plane-to-plane algorithm. Within this paper, in addition to the conventional holistic error metrics, an alternative image classification comparison technique is employed as an error metric. The convergence properties, as reported by these various metrics are compared and contrasted using empirical mm-wave measured data taken using a planar near-field scanner and processed using a commonly encountered plane-to-plane PR algorithm.