In array signal processing, there is a fundamental tension between mathematical convenience and physical reality.
On one hand, high-resolution Direction-of-Arrival (DoA) estimation and channel prediction algorithms heavily favor the simplest of geometries: the Uniform Linear Array (ULA). When sensors are arranged in a straight line with equal spacing, their spatial response forms a Vandermonde matrix. This structure possesses a property known as translation invariance (or shift invariance)—shifting down the array by one element alters the phase by a constant, predictable multiplier. Translation invariance is the mathematical engine behind search-free, super-resolution algorithms like ESPRIT and Root-MUSIC. Instead of performing computationally ruinous multi-dimensional grid searches over space, engineers can solve closed-form polynomials or extract eigenvalues to pinpoint a signal's origin instantly.
On the other hand, physical applications routinely reject the ULA. A linear array suffers from a 180-degree front-back ambiguity and cannot provide full 360-degree spatial coverage. To see in all directions, we must bend arrays into circles, cylinders, or spirals. Furthermore, when a signal source is close to the array (the "near-field" regime), the incoming wavefronts are spherical rather than planar, meaning the phase changes quadratically rather than linearly across the sensors. In both cases, the clean Vandermonde structure shatters, forcing practitioners back into slow, exhaustive grid searches.
The research direction outlined below resolves this paradox through a shared mechanism: virtual array transformation. Rather than rewriting core signal processing algorithms to accommodate complex physical geometries or curved wavefronts, this paradigm designs mathematical operators that warp the received data. By mapping non-linear physical manifolds into virtual, shift-invariant spaces, these techniques trick classical linear algorithms into working on arbitrary hardware configurations.
Straightening the Circle: Beamspace and Phase-Space Transformations
The most direct application of this paradigm is converting Uniform Circular Arrays (UCAs) or cylindrical arrays into virtual ULAs. In a UCA, sensor phases vary according to a sinusoidal function of the arrival angle, completely lacking translation invariance.
To bridge this gap, researchers employ mathematical transformations that project the physical element space into a virtual "beamspace." As demonstrated by Xiaolong Su et al. in Fast BSC-based algorithm for near-field signal localization via uniform circular array and Ling Wang et al. in Modelling method with missing values based on clustering and support vector regression, Beam Space Conversion (BSC) maps the non-linear circular array manifold into a virtual space that mimics a Vandermonde matrix. This representational shift allows search-free, closed-form parameter estimation while preserving the UCA's 360-degree coverage.
Physical Domain (UCA) Transformation Matrix Virtual Domain (ULA)
[ Non-Linear Phase ] ---> [ Beamspace / Fourier ] ---> [ Vandermonde Structure ]
(360° coverage, no symmetry) (Manifold Mapping) (Shift-Invariant, Fast)
This concept scales directly to multi-dimensional and sparse configurations:
- Enabling Subspace Algorithms: Zhengqin Xu et al. in A Robust Direction of Arrival Estimation Method for Uniform Circular Array and Di Zhao et al. in Low complexity sparse beamspace DOA estimation via single measurement vectors for uniform circular array use these virtual transformations to restore the exact translation invariance required to run the standard ESPRIT algorithm on circular data.
- Cylindrical Arrays: For massive Internet-of-Things (mIoT) networks and millimeter-wave systems, Zhipeng Lin et al. in Nested Hybrid Cylindrical Array Design and DoA Estimation for Massive IoT Networks introduce a phase-space transformation that linearizes the non-linear phase response of cylindrical configurations.
- Shift-Invariance via Fourier Transforms: Similarly, Zhipeng Lin 0001 et al. in Joint Estimation of Multipath Angles and Delays for Millimeter-Wave Cylindrical Arrays with Hybrid Front-ends apply a quasi-discrete Fourier transform to cylindrical data, mapping the complex geometry into a low-dimensional space characterized by shift-invariant properties.
Conformal Mapping for Complex Geometries
When array geometries become even more irregular—such as the multi-arm Archimedes spiral—simple beamspace projections are insufficient. Spirals are highly desirable because they suppress grating lobes and offer exceptional bandwidth, but they possess no inherent physical symmetry.
To unlock fast subspace processing for these shapes, Yuxin Bai et al. in MAASA: Multi-Arm Archimedes Spiral Array for Arrival Angle Estimation introduce a conformal coordinate transformation. This mathematical mapping warps the physical coordinates of the spiral array into a synthetic, virtual domain where the elements appear to have perfect translation invariance. By creating this virtual uniform grid, the authors allow the computationally efficient ESPRIT algorithm to operate on a physical structure that completely lacks physical translation symmetry.
A related geometric strategy appears in tomographic imaging. In Motion Detection in Diffraction Tomography by Common Circle Methods, Michael Quellmalz et al. utilize a stereographic projection to map curved, hemispherical data structures in Fourier space onto flat, planar surfaces. This transformation acts as a mathematical bridge, enabling linear tomographic reconstruction algorithms to solve non-linear wave scattering problems. Even in small-scale microphone arrays, Wuyang Jiang et al. in Two-stage Localisation Scheme Using a Small-scale Linear Microphone Array for Indoor Environments show that a coordinate rotation can artificially shift a signal source onto an optimal geometric plane, simplifying subsequent distance estimation.
Flattening the Wavefront: Near-Field to Far-Field Mapping
The breakdown of the Vandermonde structure does not only happen because an array is physically curved; it also happens when a source is close enough to the array that its wavefronts are spherical. In the near-field, the phase of the arriving signal curves across the array face, requiring a simultaneous search over both angle and range.
Spherical Wavefront (Near-Field) Focusing / Transformation Matrix Planar Wavefront (Virtual Far-Field)
) ) ) ||||| ---> [ Phase Compensation ] ---> | | | |||||
(Curved phase geometry) (Wavefront Flattening) (Linear phase progression)
To bypass this multi-dimensional search, researchers use "focusing" or wavefront transformation matrices to flatten the phase geometry:
- Phase Curvature Compensation: Yide Wang et al. in Near-Field Source Localization by Using Focusing Technique apply a specialized focusing matrix directly to the signal covariance data. This operation mathematically compensates for the phase curvature inherent in spherical waves, mapping near-field observations into a far-field-equivalent model where fast plane-wave estimation algorithms can be deployed.
- Extremely Large Antenna Arrays (ELAAs): This problem is amplified in 6G systems using ELAAs, where the near-field region extends significantly further from the antenna array. In Wavefront Transformation-based Near-field Channel Prediction for Extremely Large Antenna Array with Mobility, Weidong Li et al. introduce a wavefront transformation matrix that maps complex spherical near-field wavefronts into virtual far-field plane waves. By flattening the phase fluctuations mathematically, the authors allow standard far-field signal processing tools to predict channels for high-speed mobile users seamlessly.
Taming Irregular and Non-Standard Hardware
Ultimately, virtual array transformation decouples the signal processing software from the physical hardware constraints. This allows engineers to build highly irregular, sparse, or hardware-optimized arrays without losing the ability to perform high-resolution subspace estimation.
For example, lens antenna arrays are widely used in millimeter-wave systems to focus signal energy physically, but their mathematical response follows a sinc-function rather than an exponential progression. To resolve this, Zhengwei Ni et al. in DoA Estimation for Lens Antenna Array via Root-MUSIC, Outlier Detection, and Clustering employ a trigonometric identity to transform the array's sinc-function response into a polynomial structure. This re-expression maps the lens response into a format compatible with Root-MUSIC, achieving high precision through algebra rather than hardware redesign.
When the hardware layout is completely arbitrary or non-uniform, Yue Wang 0019 et al. in Super-Resolution Channel Estimation for Arbitrary Arrays in Hybrid Millimeter-Wave Massive MIMO Systems employ the Jacobi-Anger expansion. This mathematical tool expands the received signals into a series of cylindrical waves, effectively decoupling the physical antenna coordinates from the signal arrival angles. The result is a mathematical mapping of an arbitrary physical array into a virtual uniform domain, enabling high-precision channel estimation independent of physical layout constraints.
Finally, this paradigm applies to sparse, non-uniform lattice geometries like coprime or hexagonal arrays. As shown by Conghui Li et al. in Coprime Sensing via Chinese Remaindering Over Quadratic Fields - Part II: Generalizations and Applications, coordinate mapping techniques can transform non-uniform coarrays onto a virtual rectangular grid. This allows sparse physical arrays—which use fewer sensors to achieve high degrees of freedom—to utilize standard subspace-based algorithms and spatial smoothing techniques that traditionally demand uniform, dense rectangular geometries.
Key Foundational Work
Modelling method with missing values based on clustering and support vector regression (2010) — contributed 2 concept elements to this direction.
Near-Field Source Localization by Using Focusing Technique (2008) — contributed 2 concept elements to this direction.