Vector Sensor Antennas Overview

• Vector sensor antennas are multi-component electromagnetic sensors that measure both electric (via dipoles) and magnetic (via loops) fields at a common phase centre.
• They enable enhanced direction-of-arrival, polarimetric, and imaging capabilities by leveraging six collocated channels to capture full electromagnetic field signatures.
• Recent studies demonstrate that optimized vector sensing boosts Fisher information, improves self-calibration, and supports advanced applications from RF interferometry to nanoscale optical metrology.

Vector sensor antennas are electromagnetic sensors and sensor arrays that measure multiple field components of an incident wave rather than a single scalar observable. In the radio-astronomy configuration considered for low-frequency space interferometry, a vector sensor consists of three orthogonal dipole antennas and three orthogonal loop antennas with a common phase centre, so that electric- and magnetic-field information is acquired co-locally (Kononov et al., 25 Aug 2025). In related direction-of-arrival work, the same concept is described as six collocated antennas that measure all electromagnetic field components of incident waves (Yu et al., 2020). Across the literature, this vectorial measurement capability is exploited for direction finding, polarimetry, self-calibration, transient pulse reconstruction, and, at optical frequencies, for reconstructing not only the electric field $\mathbf{E}(\mathbf{r})$ but also the associated magnetic field $\mathbf{H}(\mathbf{r})$ and current density $\mathbf{J}(\mathbf{r})$ of infrared optical antennas (Olmon et al., 2010).

1. Canonical architectures and measured quantities

The canonical radio-frequency vector sensor architecture comprises three orthogonal dipoles, which measure electric-field components, and three orthogonal loops, which measure magnetic-field components; all six elements are arranged at right angles and share a common phase centre (Kononov et al., 25 Aug 2025). Dipoles and loops are aligned with the coordinate axes, yielding a collocated six-channel measurement of the incident electromagnetic field. In the direction-of-arrival formulation, this six-element arrangement provides rich spatial and polarization information for estimating the angle of arrival and polarization of incident sources (Yu et al., 2020).

A simpler comparator architecture is the tripole, which has only three orthogonal dipoles. The distinction is operationally significant because a pair of vector sensors forms $36$ cross-correlation products, whereas a pair of tripoles forms $9$ (Kononov et al., 25 Aug 2025). In interferometric settings, that measurement multiplicity is directly tied to Fisher information and spherical-harmonic mode coverage.

Vector sensing also appears in broader forms. In the Pierre Auger Observatory literature, the relevant problem is not a six-element collocated sensor but calibrated recovery of the three-dimensional electric field of transient cosmic-ray pulses using at least two perpendicularly oriented antennas per station together with a vector effective length characterization (Abreu et al., 2012). In optical near-field metrology, vector sensitivity is realized by s-SNOM probe tips engineered for selective sensitivity to in-plane and out-of-plane electric-field components, enabling independent measurements of $\mathbf{E}_y$ and $\mathbf{E}_z$ above infrared optical antennas (Olmon et al., 2010). This suggests that “vector sensor antenna” is best understood as a measurement paradigm centered on multi-component electromagnetic observables, not only as a single fixed hardware topology.

2. Response formalisms and field reconstruction

For space-based vector sensor interferometry, the directional and polarization response is represented by a Jones matrix. The response matrix given for the vector sensor is

$$\bm{A}_{\text{vs}}(\theta, \phi) = \begin{bmatrix} -\cos\theta\cos\phi & \sin\phi \ -\cos\theta\sin\phi & -\cos\phi \ \sin\theta & 0 \ -\sin\phi & -\cos\theta\cos\phi \ \cos\phi & -\cos\theta\sin\phi \ 0 & \sin\theta \end{bmatrix},$$

where the first three rows correspond to dipoles and the next three to loops (Kononov et al., 25 Aug 2025). The sensor response to a plane wave is then written as

$$\bm{v}_{\text{vs}}(\theta, \phi, \alpha, \delta) = \bm{A}_{\text{vs}}(\theta, \phi)\bm{e}(\alpha, \delta).$$

This formalism makes explicit that loop and dipole outputs are distinct linear combinations of the incident polarization state.

For direction-of-arrival estimation with a single vector sensor, the received data model is

$X_{6 \times N} = AS + n,$

with $\mathbf{H}(\mathbf{r})$0 formed from electromagnetic steering vectors for $\mathbf{H}(\mathbf{r})$1 sources (Yu et al., 2020). The steering vector uses elevation $\mathbf{H}(\mathbf{r})$2, azimuth $\mathbf{H}(\mathbf{r})$3, and polarization parameters $\mathbf{H}(\mathbf{r})$4, so the array manifold encodes angular and polarization structure jointly rather than through phase progression alone.

In transient wideband sensing, the central object is the vector effective length (VEL). The antenna response is written in the time domain as

$\mathbf{H}(\mathbf{r})$5

and in the frequency domain as

$\mathbf{H}(\mathbf{r})$6

where $\mathbf{H}(\mathbf{r})$7 is direction-, frequency-, and polarization-dependent (Abreu et al., 2012). The VEL therefore serves as a compact response operator linking incident vector fields to measurable voltages, including system effects through the realized VEL.

In optical antenna metrology, interferometric homodyne s-SNOM measures the tip-scattered near-field in a collinear backscattering geometry, with detected signal

$\mathbf{H}(\mathbf{r})$8

By scanning the probe tip over the sample and varying the tip-sample distance, the three-dimensional electric near field can be mapped with phase and amplitude information (Olmon et al., 2010). There, vector sensing is not the reception of a propagating far-field wave at a conventional antenna array, but a local near-field measurement that can subsequently be converted into $\mathbf{H}(\mathbf{r})$9 and $\mathbf{J}(\mathbf{r})$0.

3. Direction finding, polarimetry, and interferometric imaging

A recurrent use of vector sensor antennas is the joint estimation of propagation direction and polarization. Because both electric and magnetic field vectors are measured, the architecture supports unambiguous determination of signal direction and wave polarization for incident electromagnetic waves and directly enables full-Stokes polarimetric analysis (Kononov et al., 25 Aug 2025). The associated visibility matrix for a sensor pair is

$\mathbf{J}(\mathbf{r})$1

where $\mathbf{J}(\mathbf{r})$2 is related to the Stokes parameters $\mathbf{J}(\mathbf{r})$3 (Kononov et al., 25 Aug 2025).

For all-sky imaging, the same work replaces flat-sky Fourier imaging with spherical harmonic imaging because a spaceborne interferometer array is non-coplanar and has full $\mathbf{J}(\mathbf{r})$4 steradian field-of-view. The visibility equation is

$\mathbf{J}(\mathbf{r})$5

which is expanded in spherical harmonics to obtain

$\mathbf{J}(\mathbf{r})$6

followed by the maximum-likelihood solution

$\mathbf{J}(\mathbf{r})$7

In practice, Tikhonov regularization is used to stabilize the inversion (Kononov et al., 25 Aug 2025). The Fisher matrix

$\mathbf{J}(\mathbf{r})$8

governs the information content of the recoverable sky modes, with total Fisher information defined as $\mathbf{J}(\mathbf{r})$9.

A common misconception is that loop measurements are redundant because $36$0 and $36$1 are related for a plane wave. In the astronomical imaging setting, that conclusion does not hold: signals arrive from all directions, and the loop measurements are not simply proportional to the dipole measurements but provide distinct linear combinations through the response matrix (Kononov et al., 25 Aug 2025). This non-redundancy underlies the reported result that vector sensor arrays provide four times more total Fisher information than tripole arrays for a given number of nodes, because each vector-sensor baseline provides $36$2 visibilities compared with $36$3 for tripoles (Kononov et al., 25 Aug 2025).

The same gain appears in modal coverage. A single vector sensor can measure the $36$4, $36$5, and $36$6 spherical harmonic modes, while a single tripole is blind to the $36$7 modes and has higher uncertainty in the $36$8 modes (Kononov et al., 25 Aug 2025). This directly links vector sensing to low-order global sky reconstruction in sparse or resource-constrained arrays.

4. Calibration, self-calibration, and learning-based inference

Vector sensor arrays complicate calibration because each channel can have a distinct complex gain and phase error, but they also improve identifiability because the directional response is encoded in both magnitude and phase. In the self-calibration framework, the array data model is

$36$9

with $9$0, $9$1 (Ramamohan et al., 2021). For scalar sensor arrays, unique calibration typically requires at least one reference sensor and one calibrator source. For vector sensor arrays, specifically AVS arrays in that work, a calibrator source is not required; it is sufficient to designate one AVS channel as a reference (Ramamohan et al., 2021). The paper attributes this to the modulation of both magnitude and phase in the AVS response to DOA.

The proposed self-calibration methods are geometry independent convex optimization algorithms derived from both the element-space and covariance data models. In the covariance form,

$9$2

which is vectorized using the co-array manifold $9$3 (Ramamohan et al., 2021). The resulting convex relaxations are reported to outperform Weiss-Friedlander and Paulraj-Kailath in accuracy and robustness in synthetic studies, and an anechoic-chamber experiment with a five-AVS linear array showed that fixing only the first AVS channel as reference was sufficient for accurate DOA recovery (Ramamohan et al., 2021).

Machine learning has also been applied to single-vector-sensor DOA estimation. The neural-network approach in (Yu et al., 2020) uses the covariance matrix

$9$4

as input, retaining only its upper triangle, splitting into real and imaginary parts, and vectorizing the result. The workflow is two-stage: a classifier estimates the number of sources $9$5, and a regression network estimates the DOA pairs $9$6. Architectures and training procedures examined include fully connected feed-forward neural networks, Bayesian Normalization, Levenberg-Marquardt, Scaled Conjugate Gradient, radial basis function networks, early stopping, dropout, and batch normalization (Yu et al., 2020).

The reported performance shows that neural networks can achieve reasonably accurate estimation with up to five sources, which is the theoretical maximum for the six-antenna vector sensor considered. For one source, the best neural-network models achieve RMSE below $9$7 degree after SNR exceeds $9$8 dB; for five sources, the mean absolute error is about $9$9 degrees for both azimuth and elevation at moderate SNR above $\mathbf{E}_y$0 dB (Yu et al., 2020). Performance improves when the field-of-view is limited and degrades near the sphere poles and at azimuth wraparound, where the covariance structure becomes geometrically less discriminative. This suggests that vector sensing does not eliminate angular ambiguities by itself; inference quality still depends on parameterization, training coverage, and symmetry breaking.

5. Transient sensing, vector effective length, and pulse fidelity

For transient radio measurements, especially cosmic-ray induced air showers, vector sensor behavior must be characterized in the time domain as well as the frequency domain. The Pierre Auger Observatory study emphasizes that radio emission pulses are transient, polarized, and arrive from arbitrary directions, so the measured voltage must be related quantitatively to the incident field through a fully characterized antenna sensor (Abreu et al., 2012). The VEL provides that description by encoding amplitude, phase, and polarization response.

The time-domain impulse response of the antenna is the inverse Fourier transform of the realized amplified VEL,

$\mathbf{E}_y$1

and the recorded voltage is the convolution of this impulse response with the incident signal (Abreu et al., 2012). Group delay,

$\mathbf{E}_y$2

determines dispersive broadening. Flatter group delay preserves pulse shape better, whereas group-delay variation broadens the pulse and reduces peak amplitude.

The Auger Engineering Radio Array compared three candidate antennas: Small Black Spider, Salla, and Butterfly. The Butterfly antenna is reported to have the best transient fidelity, lowest internal noise, and robust broadband vector sensing; Salla has good impulse response but lower sensitivity due to the loading resistor; Small Black Spider exhibits significant pulse dispersion, with about $\mathbf{E}_y$3 peak reduction due to group delay variation across the bandwidth (Abreu et al., 2012). On that basis, the Butterfly antenna was selected for the next AERA stage.

Although this literature does not present the six-channel vector-sensor architecture of radio interferometry, it is central to the broader theory of vector sensing because it formalizes how polarization-sensitive antenna systems are calibrated for field reconstruction. The paper states that for dual-polarized vector measurements, the full 3D field can be reconstructed using the measured responses and the known vector response matrices (Abreu et al., 2012). A plausible implication is that VEL-based calibration and six-component vector sensing are complementary rather than competing frameworks: one emphasizes faithful field-to-voltage transduction, the other expands the set of measured electromagnetic observables.

6. Nanoscale optical antennas and vector-resolved near-field metrology

At infrared optical frequencies, vector-sensor concepts appear as experimental reconstruction of the full near-field state of an antenna. The s-SNOM method in “Determination of electric field, magnetic field, and electric current distributions of infrared optical antennas: A nano-optical vector network analyzer” provides nanometer spatial resolution and full phase and amplitude information for the three-dimensional electric near field of a coupled-dipole antenna (Olmon et al., 2010). The enabling hardware is a probe-tip design fabricated by truncating a commercial Si atomic force microscope tip with focused ion beam milling and then depositing an approximately $\mathbf{E}_y$4 nm thick Pt platelet by electron-beam assisted nano-CVD to create a sub-resonant dipole antenna at the tip apex. Simulations and experiments show minimal depolarization of about $\mathbf{E}_y$5 and a spatial resolution of about $\mathbf{E}_y$6 nm (Olmon et al., 2010).

Once $\mathbf{E}_y$7 is measured, the magnetic field is obtained from Faraday’s law in the frequency domain,

$\mathbf{E}_y$8

which reduces in the planar geometry used to

$\mathbf{E}_y$9

The current density is then linked to the measured electric field through Hallén’s integral equation,

$\mathbf{E}_z$0

which is solved numerically by the method of moments with pulse-basis/point-matching implemented in MATLAB (Olmon et al., 2010). An alternative current estimate,

$\mathbf{E}_z$1

is described as more sensitive to noise and less preferred because it involves additional derivatives (Olmon et al., 2010).

This optical work is also used to infer mutual impedance and capacitive coupling between antenna elements, which manifest as shifts in current maxima toward the gap region and red-shifts in resonance (Olmon et al., 2010). The method is presented as generalizable beyond the demonstrated infrared dimer antenna: with full 3D vector field data from two orthogonal measurement directions, arbitrary antenna geometries can in principle be characterized, and the same principle can be adapted to THz and microwave regimes with suitable probes (Olmon et al., 2010). This suggests that vector sensing at the nanoscale can function as an experimental analogue of a vector network analyzer, but with local field reconstruction rather than only port-level scattering parameters.

7. Array optimization and current research directions

Recent optimization work extends vector-sensor thinking from measurement and inference to antenna placement and orientation. A wideband signal model based on vector spherical wave functions (VSWFs) expands the electromagnetic field outside a circumscribing sphere and uses modal reception coefficients to map incoming vector fields to port voltages (Lafer et al., 19 Sep 2025). The model explicitly includes frequency, direction, and polarization-dependent characteristics, together with mutual coupling and distortions from surrounding obstacles. For localization, the received signal vector is written as

$\mathbf{E}_z$2

and the Fisher information matrix is

$\mathbf{E}_z$3

with CRLB

$\mathbf{E}_z$4

(Lafer et al., 19 Sep 2025).

Two optimization criteria are formulated with respect to array geometry parameters: A-optimality, which minimizes the trace of $\mathbf{E}_z$5 averaged over the parameter space, and D-optimality, which minimizes the negative determinant of $\mathbf{E}_z$6 integrated over the same domain (Lafer et al., 19 Sep 2025). The demonstration uses a planar three-XETS array, optimizing element positions and rotations. The reported workflow updates the geometry, performs full-wave EM simulation in Ansys HFSS, computes VSWF reception coefficients through modal decomposition, evaluates FIM and CRLB over incident angles and polarizations across bandwidth, and then optimizes using Differential Evolution; up to $\mathbf{E}_z$7 array configurations are simulated (Lafer et al., 19 Sep 2025).

The resulting optimized array exhibits lower average CRLBs for almost all polarization angles except one, and reduced variance across polarizations relative to the initial co-oriented arrangement (Lafer et al., 19 Sep 2025). That outcome is consistent with a broader theme running through the literature: vector sensor antennas are valuable not only because they increase the number of channels, but because they preserve polarization diversity and directional structure in a way that can be exploited by imaging, estimation, calibration, and design algorithms.

Across these subfields, several objective boundaries are also clear. Neural-network DOA estimation has been studied under idealized geometry and does not address multipath or array imperfections (Yu et al., 2020). Convex self-calibration has experimental validation in an acoustic vector sensor array rather than an RF antenna array (Ramamohan et al., 2021). Optical vector-field reconstruction is demonstrated for a linear coupled-dipole antenna in the mid-infrared (Olmon et al., 2010). Spaceborne all-sky imaging with vector sensor interferometry is supported by analytical development and simulation, including a nine-sensor free-space example and spherical CLEAN deconvolution, but not by on-orbit data (Kononov et al., 25 Aug 2025). These limitations do not diminish the central result shared across the cited work: vector sensor antennas provide a physically richer measurement model than scalar or purely electric-field architectures, and that additional structure can be converted into gains in identifiability, polarimetry, imaging fidelity, calibration economy, and electromagnetic interpretability.