🔢 mathematics

A Unified Dual Framework for Sparse-Array Near-Field Beam Focusing With Spatial Interference Suppression

This paper establishes a unified dual framework for sparse-array near-field beam focusing that analytically characterizes optimal beamformers as generalized matched filters, provides finite-dimensional convergence guarantees for numerical methods, and derives a closed-form performance bound showing that array order, rather than optimization algorithms, dominates the signal-to-interference ratio.

Original authors: Changhao He, Xiaojuan Zhang, Francois Chin Po Shin

Published 2026-07-14

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Changhao He, Xiaojuan Zhang, Francois Chin Po Shin

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a fleet of low-Earth orbit satellites acting like a giant, floating flashlight. Their job is to shine a super-bright, focused beam of energy onto a specific airplane flying high above the ground. But here's the tricky part: they must do this without accidentally blinding or interfering with the millions of people and devices living on the ground below. It's like trying to shine a laser pointer at a friend in a crowded room without making anyone else squint.

This paper tackles the math behind that "laser pointer" problem for satellite fleets. The researchers found that no matter how clever your computer algorithm is, there is a hard, physical ceiling on how well you can focus the beam while ignoring the ground.

The "Geometry Trap"
The biggest surprise isn't about the software; it's about the shape of the world. Because the satellites are so high up (hundreds of kilometers) and the airplane is relatively low, the target and the ground look almost exactly the same direction from the satellites' point of view. It's like trying to separate two friends standing right next to each other in a photo; if they are too close together, your camera lens can't tell them apart.

The paper proves that because of this "near-collinear" geometry, the signal-to-interference ratio (how clear the signal is compared to the noise) is limited by the number of satellites you have, not by how far apart they are spread out.

The Rule: If you double the number of satellites, you only get about 3 decibels (dB) better performance.
The Proof: The authors ran simulations showing that whether the satellites are spread over 50 km or 1,000 km, the performance stays almost exactly the same. The only thing that matters is adding more satellites to the fleet.

What Doesn't Work (The "Don't Bother" List)
The paper explicitly rules out a few ideas that people might hope would work:

Just making the satellites bigger: Putting a larger antenna on a single satellite doesn't help much. Because the target and the ground look so similar from that high up, a local antenna can't tell the difference between the airplane and the ground. It's like trying to shout at a friend in a noisy room by just shouting louder; you'll just make the room louder too.
Trying to force a "perfect" shape: Some engineers try to force every satellite to use exactly the same amount of power (a "constant-modulus" constraint) because it's cheaper for the hardware. The paper shows that while this is hard to calculate, it actually works almost as well as the "perfect" theoretical solution. You don't lose much by using the cheaper hardware.
The "LCMV" method: There is a classic technique called LCMV that tries to put "nulls" (dead zones) exactly where the ground is. The paper found that if you try to put too many of these nulls, the math breaks down completely. The computer gets confused, the signal to the airplane collapses, and you get terrible results. It's like trying to tie too many knots in a single string until it snaps.

The Solution: A New Way to Think
Instead of just guessing and checking with computers, the authors developed a "unified framework" using a mathematical tool called Lagrangian duality. Think of this as finding a master key that unlocks the structure of the problem.

They proved that the best possible beam is actually a "generalized matched filter." In plain English, it's a filter that matches the shape of the interference it's trying to avoid.
They showed that even though the ground is a continuous surface, the math says the "worst-case" interference only happens at a very small number of specific points (at most the square of the number of satellites). This means computers don't need to check every single inch of the ground; they just need to find these few critical spots.

The Real-World Fix
If you can't fix the problem by spreading the satellites out or making them bigger, what do you do? The paper suggests the answer lies on the other end of the link: the airplane.

Because the satellites can't easily tell the difference between the target and the ground, the airplane needs to help out. If the airplane uses a directional antenna (a "smart" receiver) that points only at the satellites and ignores the ground, it can boost the signal significantly.
The math shows that a small gain on the receiver side is much more powerful than trying to squeeze more performance out of the satellite side.

How Sure Are We?
The authors didn't just guess; they proved these limits mathematically and backed them up with simulations.

They simulated a fleet of 8 satellites and showed that the performance hits a ceiling of about 9 dB.
They tested the "constant-modulus" (cheaper hardware) idea and found it gets within 0.05 dB of the perfect theoretical limit. That's incredibly close.
They also tested what happens if the satellites aren't in their exact perfect spots (due to orbit errors). They found that if the position error is about 2 cm, the signal drops by roughly 2.9 dB. This gives engineers a clear "tolerance budget" for how precise their satellites need to be.

In short, the paper tells us: Stop trying to outsmart the geometry with bigger antennas or wider spacing. The physics of the situation sets a hard limit based on how many satellites you have. To get better performance, you need more satellites or a smarter receiver on the airplane, not a fancier algorithm on the satellite.

Technical Summary: A Unified Dual Framework for Sparse-Array Near-Field Beam Focusing With Spatial Interference Suppression

Problem Statement
This paper addresses the challenge of designing coherent satellite formations (specifically Low-Earth-Orbit, or LEO, constellations) to focus radiated energy on an airborne target while simultaneously suppressing interference over a continuous terrestrial protection region. This scenario arises in regenerative-payload non-terrestrial networks (NTN) where downlink signals to airborne terminals must coexist with ground-based cellular users.

The problem is characterized by three principal challenges:

Semi-infinite constraints: The protection region is a continuous spatial continuum, requiring the suppression of field strength over infinitely many points, which typically necessitates numerical cutting-plane methods with uncertain convergence properties.
Geometric degeneracy: In typical LEO scenarios, the target altitude is far smaller than the orbital altitude. This creates a "near-collinear" geometry where the target and the dominant ground interference direction are highly correlated, imposing a fundamental ceiling on the Signal-to-Interference Ratio (SIR) that is independent of the optimization algorithm.
Constant-modulus implementation: Practical satellite payloads utilize saturated power amplifiers, imposing a constant-modulus constraint on the beamforming weights. This transforms the problem into a non-convex manifold optimization, rendering classical closed-form linear solutions (like MVDR) generally infeasible.

Methodology
The authors propose a unified Lagrangian-dual framework to analyze and solve this problem. Rather than treating different beamforming approaches (SOCP, MVDR, LCMV) as distinct designs, the paper derives a common analytical structure for the optimal solution.

Dual Analysis: The authors formulate the beam focusing problem as a semi-infinite Second-Order Cone Program (SOCP). By applying Lagrangian duality, they characterize the optimal beamformer as a generalized matched filter against an effective spatial covariance matrix. This covariance is induced by an optimal dual measure supported on the protection region.
Finite-Support Characterization: Using measure-theoretic arguments (specifically Tchakaloff's theorem), the authors prove that the optimal dual measure has finite support. The number of active "worst-case" interference points is bounded by $M^2$ in general and by $M$ (where $M$ is the number of array elements) for Uniform Linear Arrays (ULA). This provides a theoretical certificate for the convergence of cutting-plane methods.
Asymptotic Analysis: The paper derives a closed-form upper bound on the mean-SIR. Under the near-collinear regime, this bound is shown to scale logarithmically with the array order $M$ (specifically $10 \log_{10} M$ ), rather than with the aperture span.
Algorithmic Specialization: The framework unifies several existing algorithms as special cases of the dual measure formulation. Furthermore, to address the constant-modulus constraint, the authors develop a Riemannian conjugate-gradient algorithm on the unit-torus manifold, initialized by projecting the unconstrained MVDR solution onto the manifold.

Key Contributions

Optimal Beamformer Structure: The paper establishes that every optimal linear beamformer can be represented as a generalized matched filter against a spatial covariance induced by a dual measure. This connects semi-infinite SOCP formulations directly to classical adaptive beamforming (MVDR, LCMV).
Finite-Support Certificate: It is proven that the optimal dual measure has finite support (cardinality $\le M^2$ generally, $\le M$ for ULA). This explains the empirical efficiency of sparse active-set methods and guarantees that cutting-plane algorithms terminate with a finite working set.
Fundamental SIR Limit: A closed-form upper bound on the mean-SIR is derived. The analysis reveals that under near-collinear geometry, the achievable SIR is governed by the array order $M$ , not the aperture span or the specific optimization algorithm.
Unified Algorithmic Interpretation: The framework demonstrates that MVDR, LCMV, and SOCP-based focusing are special cases of a single dual-measure formulation. It also provides a practical manifold optimization algorithm for constant-modulus constraints that approaches the unconstrained performance limit.
Architectural Insights: The analysis identifies three operational invariances:
- Algorithm-invariance: Both unconstrained and constant-modulus beamformers attain near-optimal SIR.
- Topology-invariance: In the near-collinear regime, 1D and 2D constellations with the same $M$ yield essentially identical SIR.
- Transmit-Receive Asymmetry: Increasing transmit aperture (per-platform) yields diminishing returns due to geometric constraints, whereas receiver-side directional antennas provide linear gains in dB, making them the dominant lever for exceeding the SIR ceiling.

Results
Numerical experiments on a representative LEO formation ( $M=8$ satellites, $D=300$ km) validate the theoretical findings:

SIR Scaling: The measured SIR upper bound scales exactly as $10 \log_{10} M$ and is largely independent of the aperture span $D$ , confirming the asymptotic analysis.
Algorithm Performance: The unconstrained MVDR beamformer achieves 9.03 dB, within 0.01 dB of the theoretical bound. The proposed Riemannian manifold algorithm (with MVDR initialization) achieves 9.09 dB, demonstrating that constant-modulus constraints do not significantly degrade performance in this regime.
LCMV Failure: The study quantifies the failure of LCMV under geometric degeneracy; as the number of null constraints increases, the condition number of the constraint matrix explodes, leading to target signal collapse.
Robustness: The system shows sensitivity to position perturbations, with SIR degradation scaling as $1/\sigma_p^2$ (where $\sigma_p$ is the position error standard deviation). However, sub-centimeter orbit determination precision is sufficient to maintain low SIR loss.
Per-Platform Strategies: Simulations show that local processing strategies (boresight, local nulling, or local MVDR) on individual satellite platforms cannot overcome the geometric SIR ceiling. Only a fully joint coherent processing of all physical elements ($MN$ elements) can break this limit, though this requires impractical network-wide synchronization for current systems.

Significance and Claims
The paper claims to provide a theoretical unification of sparse-array near-field beam focusing, bridging the gap between numerical optimization (SOCP) and analytical adaptive beamforming (MVDR/LCMV). Its primary significance lies in shifting the focus from algorithmic improvements to geometric limitations.

The authors assert that for distributed satellite formations operating in near-collinear geometries:

The achievable SIR is fundamentally limited by the array order $M$ and the altitude ratio, not by the sophistication of the optimization algorithm.
Existing hierarchical architectures (where satellites perform local combining before inter-platform coordination) are inherently bounded by this geometric ceiling.
Practical design efforts should prioritize receiver-side directivity or increasing the number of platforms ( $M$ ) rather than attempting to refine transmit-side algorithms or aperture spans, as the latter offers negligible gains in this specific regime.

The work concludes that while constant-modulus constraints are practically necessary, they do not preclude approaching the fundamental performance limits derived from the geometric constraints of the problem.

More like this