⚡ electrical engineering

Fundamental Analysis of Scalable Fluid Antenna Systems: Identifiability Limits, Information Theory, and Joint Processing

This paper establishes an observation entropy framework for scalable fluid antenna systems to unify identifiability limits and system design, revealing that joint processing outperforms sequential methods by overcoming information bottlenecks and proposing a joint MUSIC algorithm to approach theoretical capacity bounds.

Original authors: Tuo Wu, Kai-Kit Wong, Jie Tang, Ye Tian, Baiyang Liu, Maged Elkashlan, Kin-Fai Tong, Hing Cheung So, Matthew C. Valenti, Fumiyuki Adachi, Kwai-Man Luk

Published 2026-04-02

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Tuo Wu, Kai-Kit Wong, Jie Tang, Ye Tian, Baiyang Liu, Maged Elkashlan, Kin-Fai Tong, Hing Cheung So, Matthew C. Valenti, Fumiyuki Adachi, Kwai-Man Luk

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 you are trying to listen to a crowded room full of people talking at once. Your goal is to figure out exactly where each person is standing and what they are saying. This is essentially what wireless networks do when they try to locate devices or track signals.

For decades, engineers have used a "fixed microphone array"—a rigid row of microphones stuck in one place. This paper introduces a revolutionary new idea: a Fluid Antenna System (S-FAS). Think of this not as a row of microphones, but as a swarm of intelligent, shape-shifting drones that can instantly rearrange themselves into different formations to solve the problem.

Here is the breakdown of the paper's breakthroughs using simple analogies:

1. The Problem: The "Rigid Room" vs. The "Swarm"

Old Way (Fixed Arrays): Imagine a rigid row of 32 microphones. If 33 people start talking, the system gets confused and can't tell them apart. It hits a hard wall. Also, because the microphones are packed tight, they interfere with each other (like people standing too close in a hallway), making the signal muddy.
New Way (Fluid Antennas): The S-FAS is like a swarm of drones. It can instantly change its shape. It can squeeze into a tight, dense formation or stretch out into a long, wide line, depending on what the situation needs.

2. The Two "Modes" of the Swarm

The paper shows that this swarm has two superpowers, but it can't use both at full strength at the same time without a special trick.

Mode A: The "Compact Crowd" (Compressed Configuration)
- What it does: The drones pack very tightly together.
- The Benefit: Because they are so close, they can hear everyone without any "echoes" or confusion (no "grating lobes"). It's like a tight huddle where everyone hears the speaker clearly.
- The Catch: Because they are packed so tight, they interfere with each other (mutual coupling). To fix this, the system has to "mute" the drones on the very edges. It's like a choir where the singers on the ends are told to stay quiet to stop the noise. This reduces the number of active microphones, limiting how many people it can count.
Mode B: The "Wide Stretch" (Extended Configuration)
- What it does: The drones spread out far apart.
- The Benefit: They don't interfere with each other, and because they are spread out, they have a huge "listening range" (aperture). This gives them incredible precision to pinpoint exactly where a sound is coming from.
- The Catch: If the sound source is very close (like a person whispering right next to the mic), the math gets complicated because the sound waves aren't flat anymore.

3. The Big Discovery: The "Entropy Budget"

The authors realized that counting antennas isn't enough. They introduced a concept called "Observation Entropy."

The Analogy: Think of "Entropy" as the information budget or the "clarity allowance" the system has.
- Every time you add a source (a person talking), you spend part of your budget.
- You need to keep a little bit of budget left over just to hear the "background silence" (noise). If you spend all your budget on the people talking, you lose the ability to distinguish them from the silence, and the system crashes.
- The Rule: You can only identify $N-1$ people if you have $N$ microphones, because you need one "spare" microphone's worth of budget to listen to the silence.

4. The Trap: The "Two-Stage" Bottleneck

Previously, engineers thought: "Let's use the Compact Crowd first to get a rough idea of where people are, then switch to the Wide Stretch to get the exact details."

The Paper's Warning: This is a trap!
The Analogy: Imagine you take a blurry, low-resolution photo of a crowd (Stage 1) and then try to use that blurry photo to guide a high-resolution camera (Stage 2). No matter how good the second camera is, it can only see what the first blurry photo told it to look for. If the first photo missed a person, the second camera will never find them.
The Result: The whole system is limited by the "weak" first stage. You lose the potential of the powerful second stage.

5. The Solution: "Joint Processing" (The Super-Stack)

The paper proposes a brilliant solution: Don't choose one mode; use both at the same time.

The Analogy: Instead of taking a photo and then another, imagine you have two cameras taking a picture of the same crowd simultaneously—one zoomed in and one zoomed out. You then stack the images on top of each other to create one giant, super-detailed 3D map.
The Magic: By combining the "Compact Crowd" data and the "Wide Stretch" data into one giant dataset, the system effectively doubles its "information budget."
The Result: The system can now identify almost twice as many sources as before. If a single array could handle 31 people, this new "Joint" method can handle 57!

6. The New Algorithm: J-MUSIC

To make this work, the authors invented a new math tool called J-MUSIC (Joint Multiple Signal Classification).

Think of it as a super-solver that looks at the combined data from both the tight and wide formations simultaneously. It doesn't just guess; it mathematically proves where every single source is, even if there are more sources than physical antennas.

Summary

This paper proves that by treating a fluid antenna system not just as a set of moving parts, but as a dynamic system with a flexible "information budget," we can break the old limits of wireless technology.

Old Way: Rigid, limited, and prone to bottlenecks.
New Way: Flexible, combines the best of two worlds simultaneously, and can track nearly double the number of devices or targets.

It's like realizing that instead of having one giant telescope, you can have a swarm of drones that can instantly become a wide-angle lens and a zoom lens at the same time, giving you a view of the universe that was previously impossible.

1. Problem Statement

The paper addresses the fundamental limitations of Fixed-Position Arrays (FPAs) in Direction-of-Arrival (DOA) estimation and source identifiability. FPAs suffer from:

Mutual Coupling: Degrades performance, especially in dense arrays.
Static Degrees of Freedom (DoFs): The number of resolvable sources is strictly limited by the fixed number of physical elements ( $M$ ), typically $K_{max} < M$ .
Rigid Architecture: Cannot adapt to varying field regimes (near-field vs. far-field) or optimize aperture dynamically.

While Fluid Antenna Systems (FAS) offer dynamic reconfigurability, the Scalable FAS (S-FAS)—which switches between a "compressed" (sub-wavelength) and "extended" (half-wavelength) configuration—lacks a rigorous theoretical framework to define its identifiability limits. Classical algebraic methods (rank counting) are insufficient because they fail to account for:

Dynamic observation entropy scaling.
Information flow bottlenecks in sequential processing.
The distinction between fundamental DoF exhaustion and algorithmic suboptimality.

2. Methodology: The Observation Entropy Framework

The authors propose a novel Observation Entropy Framework to unify the analysis of S-FAS. Instead of relying solely on algebraic rank analysis, they utilize differential entropy and mutual information to define the "entropy budget" of an observation space.

Core Concepts:

Entropy Budget: For a configuration $\alpha$ with effective dimension $M_{eff,\alpha}$ , the observation entropy is bounded by $H(Y_\alpha) \leq M_{eff,\alpha} \log(2\pi e \sigma_y^2)$ .
Identifiability Condition: Reliable parameter estimation requires the noise subspace to retain at least one dimension to serve as a statistical reference. This leads to the universal bound:
$K_{max} = M_{eff} - 1$
Dual Validation: The paper validates theoretical bounds using two methods:
1. Algebraic: Measuring the dimension of the noise subspace ( $dim(\mathbf{U}_n)$ ).
2. Information-Theoretic: Analyzing the noise entropy ratio ( $\rho_n$ ), which quantifies the fraction of entropy allocated to the noise baseline.

System Model:

Compressed Configuration ( $\alpha_c$ ): Sub-wavelength spacing ( $d_c < \lambda/2$ ) eliminates grating lobes but introduces severe mutual coupling. Mitigated by central subarray selection (removing $p$ edge elements), resulting in effective DoFs $M_c = M - 2p$ .
Extended Configuration ( $\alpha_e$ ): Half-wavelength spacing ( $d_e = \lambda/2$ ) with negligible coupling, utilizing all $M$ elements ( $M_e = M$ ). Supports both far-field and mixed-field (angle-range) estimation.

3. Key Contributions

A. Derivation of Identifiability Bounds

The paper establishes a complete capacity hierarchy for S-FAS:

Compressed Configuration: $K_{max,c} = M - 2p - 1$ $K_{ma x, c} = M - 2 p - 1$ .
- Insight: Sacrificing edge elements reduces the entropy budget but ensures grating-lobe-free initialization.
Extended Configuration: $K_{max,e} = M - 1$ $K_{ma x, e} = M - 1$ .
- Insight: This bound holds for both far-field (1 parameter/source) and mixed-field (2 parameters/source) scenarios. The limit is determined by the observation dimension $M$ , not the parameter count per source.
Sequential Processing (Two-Stage): $K_{seq} = M - 2p - 1$ $K_{se q} = M - 2 p - 1$ .
- Insight: Using the compressed stage to initialize the extended stage creates an information bottleneck. By the Data Processing Inequality, the mutual information available to the second stage cannot exceed the entropy of the first stage ( $I(\theta; \hat{\theta}_c) \leq H(Y_c)$ ). Thus, the extended array's extra capacity is wasted.
Joint Processing: $K_{joint} = (M_c + M_e) - 1 = 2M - 2p - 1$ $K_{j o in t} = (M_{c} + M_{e}) - 1 = 2 M - 2 p - 1$ .
- Insight: By spatially stacking observations from both configurations, the entropy budget expands, effectively doubling the resolvable sources compared to sequential processing.

B. Theoretical Insights

Bottleneck Diagnosis: The framework explains why sequential processing fails (information loss) rather than just stating it fails.
Fundamental vs. Algorithmic Limits: The noise entropy ratio allows designers to distinguish between hitting the fundamental DoF limit (entropy collapse) and poor algorithmic performance (residual entropy).
Mixed-Field Equivalence: Proved that the identifiability limit for mixed-field estimation is identical to far-field estimation ( $M-1$ ), contrary to naive parameter-counting assumptions.

C. Proposed Algorithm: J-MUSIC

The authors developed Joint-MUSIC (J-MUSIC), which exploits augmented steering vectors from both configurations.

Mechanism: Stacks compressed and extended observation vectors to form an augmented manifold of size $(M_c + M_e) \times K$ .
Complexity: $O((M_c + M_e)^3)$ , dominated by the Eigenvalue Decomposition (EVD) of the augmented covariance matrix.

4. Key Results

Capacity Gains:
- For a baseline $M=32, p=3$ $M = 32, p = 3$ :
  - Compressed limit: 25 sources.
  - Extended limit: 31 sources.
  - Joint limit: 57 sources (an 84% increase over extended-only and 128% increase over compressed-only).
- Sequential processing is capped at 25 sources, confirming the bottleneck theory.
Simulation Validation:
- Algebraic: Noise subspace dimension perfectly matches $M_{eff} - K$ until the theoretical limit, where it collapses to zero.
- Information-Theoretic: The noise entropy ratio $\rho_n$ decays monotonically and collapses to zero exactly at $K_{max}$ .
- Practical Saturation: While the theoretical joint bound is 57, practical simulations show saturation around $K \approx 34-36$ due to manifold conditioning issues (near-linear dependencies in steering vectors) and MDL criterion instability at high source counts. However, this still represents a significant gain over single-array processing.
Performance: J-MUSIC demonstrates:
- 15–30% reduction in RMSE compared to Extended-MUSIC.
- Robustness to high source counts (resolving up to 20 sources where others fail).
- Faster convergence with fewer snapshots.

5. Significance

This paper makes a paradigm shift in array signal processing analysis:

Beyond Algebra: It moves beyond classical rank-counting to an information-theoretic foundation, providing a unified metric (entropy budget) for comparing disparate array geometries.
S-FAS Optimization: It provides the theoretical justification for joint processing over sequential processing, guiding the design of next-generation fluid antenna systems to maximize DoFs.
Diagnostic Tool: The proposed entropy metrics offer a practical way for engineers to diagnose whether system performance limits are fundamental (hardware/physics) or algorithmic.
Scalability: It proves that scalable fluid antennas can break the $K < M$ barrier of traditional arrays, enabling super-resolution and underdetermined source estimation in complex, mixed-field environments.