Semiempirical optimization of frequency-diverse holograms for localization

We present a new design technique for frequency-diverse holograms used in millimeter- and submillimeter-wave imaging. A neural-network backed imaging method utilizes quasirandom hologram structure, which disperses the field in the region of interest across the WR-3.4 (220-330 GHz) band. The global optimum for the hologram structure is generally not known. A new, semiempirical optimization method utilizes the experimental data used in localization tasks. The data is a collection of the of the local field at the object plane measured with a probe ($S _{\mathbf{21}}$), and the corresponding localization performance derived from the reflection spectra of a corner cube ($S _{\mathbf{11}}$). The frequency diversity of the local field at the object is correlated with the localization error. The proposed design technique uses the obtained empirical correlation data to inform a new hologram design.