Knowledge-based antenna pattern extrapolation

Michael RobinsonarXiv 2012 · OpenReview

Abstract

We describe a theoretically-motivated algorithm for extrapolation of antenna radiation patterns from a small number of measurements. This algorithm exploits constraints on the antenna's underlying design to avoid ambiguities, but is sufficiently general to address many different antenna types. A theoretical basis for the robustness of this algorithm is developed, and its performance is verified in simulation using a number of popular antenna designs.

Introduction

Characterizing an antenna’s radiation pattern—the spatial map of how it sends or receives signals—traditionally requires exhaustive sampling at many angular locations. This process is often slow and expensive, requiring specialized near-field or far-field ranges. Mathematically, this is an "inverse source problem": attempting to determine internal currents based on external field measurements. Such problems are typically unsolvable due to non-uniqueness, where different internal distributions produce identical external fields.

This paper presents a mathematically rigorous approach to this challenge by blending differential topology with electromagnetic theory. The author argues that because we test antennas with known designs rather than random objects, we can restrict the search to a finite-dimensional "design space," making the problem uniquely solvable and robust to noise.

Methodology

The core of the proposed approach is the application of the Whitney embedding theorem. This theorem suggests that a smooth manifold, such as an antenna's design space, can be embedded into a higher-dimensional measurement space such that the mapping is injective. In practice, this means that with enough samples, each unique antenna design will produce a unique set of measurements.

Parameters are divided into two categories:

  1. Configuration ($\mathcal{C}$): Nonlinear physical layout parameters (e.g., element spacing, horn height).
  2. Excitation ($\mathcal{E}$): Linear parameters representing signal magnitude and phase at the ports.

The algorithm uses a recursive two-step optimization: a nonlinear search through the configuration space combined with a linear constrained least-squares solve for the excitation. Once the best parameters are found, the full pattern is generated using a standard electromagnetic model.

Sampling Strategies and Symmetry

A significant finding of the research is that standard measurement techniques, such as taking only azimuth and elevation cuts, are often suboptimal. Because many antennas possess inherent symmetry, these structured paths can lead to mathematical ambiguities where the algorithm cannot distinguish between distinct designs.

Comparison of sampling strategies: principal plane cuts, contiguous azimuth blocks, and random samples.

Surprisingly, random sampling proves superior to structured scanning. Randomly distributed sample points are far more likely to "break" the symmetries that cause ambiguities, allowing for more efficient and accurate model identification.

Performance and Robustness

The algorithm's performance was validated using simulated datasets for rectangular phased arrays, E-plane horn antennas, and parabolic dishes. For horn antennas, as few as 10 measurements were sufficient to uniquely identify the radiation pattern.

Mean residual error versus number of iterations for various measurement noise levels in rectangular arrays.

The method demonstrated high robustness, successfully converging even with significant Gaussian white noise (up to 1 dB) added to the measurements. Furthermore, the algorithm can identify if the wrong model is being used; if the residual error does not drop near zero, it indicates the physical antenna does not match the assumed design. Notably, the method successfully predicted "off-axis" sidelobes that were never actually measured by correctly inferring the underlying physical structure.

Conclusion

By leveraging design knowledge, this algorithm enables the extrapolation of full antenna patterns from sparse, potentially random measurements. While current simulations use idealized models, the results provide a strong theoretical basis for reducing the time and cost of antenna characterization. Future work will involve testing against physical antenna range data to account for environmental interference and calibration errors.

Summary updated

Abstract

We describe a theoretically-motivated algorithm for extrapolation of antenna radiation patterns from a small number of measurements. This algorithm exploits constraints on the antenna's underlying design to avoid ambiguities, but is sufficiently general to address many different antenna types. A theoretical basis for the robustness of this algorithm is developed, and its performance is verified in simulation using a number of popular antenna designs.
BibTeX
@misc{robinson2012knowledge,
  author = {Michael Robinson},
  title = {{Knowledge-based antenna pattern extrapolation}},
  year = {2012},
  eprint = {1203.2528},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1203.2528},
}
Figure