Cramér-Rao Bound Optimization for Near-Field Sensing With Continuous-Aperture Arrays

Hao Jiang, Zhaolin Wang, Yuanwei Liu, Arumugam Nallanathan

TL;DR

This paper introduces a framework for optimizing Continuous-Aperture Arrays (CAPAs)—the "holographic" evolution of antennas—to achieve unprecedented accuracy in near-field sensing. By treating the antenna as a continuous electromagnetic surface rather than a collection of discrete points, the authors demonstrate a tenfold improvement in sensing precision and provide a computationally efficient algorithm (SMGD) to design the optimal "current distribution" across the array.

What Problem Does This Paper Address?

As we move toward 6G, wireless systems are shifting to higher frequencies (mmWave and THz) and larger apertures. This pushes many sensing targets into the near-field region, where the curvature of the incoming wavefront becomes significant.

Traditional systems use Spatially Discrete Arrays (SPDAs)—antennas made of individual elements spaced half a wavelength apart. However, SPDAs have two major drawbacks:

  1. Limited Degrees of Freedom (DoFs): Sensing resolution is physically capped by the discrete placement of elements.
  2. Complexity: Managing thousands of individual antenna elements in "gigantic MIMO" setups leads to massive computational overhead.

The authors propose using CAPAs, which transmit signals over a truly continuous surface. While CAPAs theoretically offer "infinite" spatial DoFs, they are mathematically difficult to optimize because you are no longer tuning a vector of discrete values; you are tuning a continuous function (the source current) across a 2D plane.

Methods

The researchers tackle this functional optimization problem through several key steps:

  • EM-Based Modeling: Instead of simple geometric rays, they use electromagnetic theory and Green’s functions to model how a continuous current distribution creates an electric field that bounces off targets and returns to the receiver.
  • CRB Derivation: They derive the Cramér-Rao Bound (CRB) for this continuous system. The CRB is a fundamental lower limit on how accurately a system can estimate a target's position.
  • Subspace Manifold Gradient Descent (SMGD): Minimizing the CRB is a "non-convex, integral-based functional optimization problem"—essentially a math nightmare. The authors prove that the optimal source current must reside in a specific subspace spanned by the array responses of the targets. This allows them to transform a complex functional problem into a more manageable manifold optimization problem, which they solve using their proposed SMGD algorithm.
  • Maximum Likelihood Estimation (MLE): They develop a continuous-form MLE method to actually estimate target locations (distance and angle) from the received "echo" signals.

Datasets

The paper does not use external real-world datasets. Instead, it relies on numerical simulations based on standard electromagnetic parameters (e.g., 28 GHz carrier frequency, 1-meter aperture). The performance is evaluated by comparing the CAPA model against a benchmark SPDA model with standard half-wavelength spacing.

Key Insights and Ideas

  • The "Infinite" Advantage: Because CAPAs aren't restricted by discrete element placement, they can "tailor" the electromagnetic field with much finer precision. This allows them to exploit the full spatial DoFs of the environment.
  • Structure of the Optimal Current: A major "aha moment" is Lemma 1, where the authors prove that the optimal continuous current isn't some arbitrary messy function; it's a linear combination of the transmit array responses. This "optimal structure" is what makes the optimization computationally feasible.
  • 2D Positioning in the Near-Field: Unlike far-field sensing (which mostly sees angles), the near-field curvature allows CAPAs to perform joint distance and angle estimation with high accuracy using a single monostatic setup.
  • Complexity Reduction: By using the subspace method, the complexity of the optimization depends on the number of targets, not the number of antenna elements (which is effectively infinite in a CAPA).

Key Results

  • 10x Performance Boost: Simulations show that CAPAs achieve a tenfold improvement in sensing accuracy (lower CRB) compared to traditional discrete arrays (SPDAs) of the same size.
  • Sharper MLE Peaks: The Maximum Likelihood Estimation spectrum for CAPAs shows much sharper "spikes" and lower side-lobes than discrete arrays, meaning the system is much less likely to be confused by noise or ghost targets.
  • Efficient Convergence: The proposed SMGD algorithm converges to the optimal solution significantly faster than standard "wavenumber-domain sampling" methods, which require tens of thousands of samples for a 1-meter aperture at 28 GHz.

Limitations and Future Work

  • Narrowband Assumption: The current model assumes a narrowband signal. Real-world 6G sensing will likely require wideband signals to achieve high range resolution.
  • Prior Knowledge: The CRB optimization requires an initial estimate of the target's position (though the authors show the system is robust to small initial errors).
  • Hardware Realization: While the paper treats the aperture as "truly continuous," practical implementations (like Metasurfaces or Holographic MIMO) are still quasi-continuous. Bridging the gap between the theoretical continuous model and sub-wavelength discrete hardware remains an open challenge.
  • Multi-static Scenarios: This work focuses on a monostatic (transmitter and receiver at the same spot) setup; future work could extend this to distributed sensing networks.

Summary updated

Abstract

A Cramér-Rao bound (CRB) optimization framework for near-field sensing (NISE) with continuous-aperture arrays (CAPAs) is proposed. In contrast to conventional spatially discrete arrays (SPDAs), CAPAs emit electromagnetic (EM) probing signals through continuous source currents for target sensing, thereby exploiting the full spatial degrees of freedom (DoFs). The maximum likelihood estimation (MLE) method for estimating target locations in the near-field region is developed. To evaluate the NISE performance with CAPAs, the CRB for estimating target locations is derived based on continuous transmit and receive array responses of CAPAs. Subsequently, a CRB minimization problem is formulated to optimize the continuous source current of CAPAs. This results in a non-convex, integral-based functional optimization problem. To address this challenge, the optimal structure of the source current is derived and proven to be spanned by a series of basis functions determined by the system geometry. To solve the CRB minimization problem, a low-complexity subspace manifold gradient descent (SMGD) method is proposed, leveraging the derived optimal structure of the source current. Our simulation results validate the effectiveness of the proposed SMGD method and further demonstrate that i) the proposed SMGD method can effectively solve the CRB minimization problem with reduced computational complexity, and ii) CAPA achieves a tenfold improvement in sensing performance compared to its SPDA counterpart, due to full exploitation of spatial DoFs.
BibTeX
@article{jiang2026cramer,
  author = {Hao Jiang and Zhaolin Wang and Yuanwei Liu and Arumugam Nallanathan},
  title = {{Cramér-Rao Bound Optimization for Near-Field Sensing With Continuous-Aperture Arrays}},
  journal = {IEEE Trans. Wireless Communications},
  year = {2026},
  doi = {10.1109/twc.2025.3628549},
}

Loading related content...

Figure