Electromagnetic Signal and Information Theory: A Continuous-Aperture Array Perspective

arXiv:2605.12910 · cs.IT, math.IT · Submitted 2026-05-13 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Computer Science. The day in computer science: software, systems and theory on arXiv.

Tom: I'm Tom.

Jane: And I'm Jane, here with Lu, Meng and our in-house large language model, Lalam.

Tom: We're looking at the paper: "Electromagnetic Signal and Information Theory".

Title and authors: Tom: Moving on, the title of this paper is "Electromagnetic Signal and Information Theory: A Continuous-Aperture Array Perspective." It tells you right away that they are framing everything through the lens of EM signal and information theory.

Jane: That framing is important because it signals that this isn't just a simple hardware design tweak; it’s a deeper theoretical shift in how we should analyze these systems.

Lu: They are saying that conventional spatially discrete arrays, or SPDA, which are finite sets of separated antenna elements, need to be replaced by something more like a continuous EM aperture.

Meng: So what does that actually mean for the hardware team? Are we ditching our current array designs for something fluid?

Tom: Not ditching them entirely. It means acknowledging that the physical reality is continuous, which motivates studying things through Maxwell’s equations rather than just finite vector math.

Jane: Exactly. It forces us to look at the underlying channels and signals not as static vectors, but as continuous fields and operators governed by those fundamental electromagnetic laws.

The paper's summary: Lu: The authors summarize this by saying that they are moving away from discrete array models toward a continuous-aperture arrays, or CAPAs.

Tom: They explain that a CAPA can be interpreted as the limiting form of a spatially discrete array with increasingly dense spatial sampling and increasingly fine aperture controllability.

Jane: That means conceptually, as you sample more and more densely, you are approaching this continuous limit in terms of how well you control the field over the aperture.

Lu: From an EM perspective, they argue that the physically available spatial degrees of freedom are determined by the aperture size and propagation environment rather than just how dense your sampling is alone.

The paper's improvements: Meng: What specific improvements are they proposing beyond just modeling it as a continuous field? What’s the tangible benefit we can actually build with this concept?

Tom: The key improvement is that they introduce wavenumber domain optimization, which lets us convert those continuous functional optimization problems directly into finite-dimensional matrix problems.

Jane: That’s huge because it means we can use standard matrix tools for power allocation and precoding, enabling AI systems to determine optimal continuous transmit current profiles by solving a discretized version of the problem formulated in equation one hundred forty.

Lu: Another big move is calculus of variations, which provides an exact structure for the optimal current distribution by operating directly on the continuous problem without intermediate model approximations like wavenumber truncation.

Meng: So you’re saying this allows AI to find a more fundamentally structured answer than just throwing a truncated approximation at it?

Tom: Yeah, because calculus of variations lets us reveal that exact structure of the optimal current distribution through integral equations like equation one hundred seventy.

Conclusion: Jane: So to wrap up, the main implication is that we’re using EM theory as a way to build more sophisticated processing tools for CAPA systems.

Tom: They showed how to maintain the continuous space structure while developing modeling and analysis tools that give us these tractable finite-dimensional descriptions.

Lu: The paper's contribution is providing the framework to rigorously maintain that continuous space while building tools that yield these finite-dimensional descriptions for analysis and optimization.

Meng: For production, this means we can finally start thinking about designing higher gain transmission using these continuous beamformers, which are the space analog of the classical MMSE beamformer.

Tom: It’s about being able to transmit with more flexibility than traditional discrete array methods allow by leveraging that continuous nature.

Lu: The future work seems to be focused on taking these continuous structures and applying them to even more complex scenarios where things get even messier.

Zhaolin Wang, Chongjun Ouyang, Kuranage Roche Rayan Ranasinghe, Shuai S. A. Yuan, Giuseppe Thadeu Freitas de Abreu de Abreu, Emil Bjornson Fellow IEEE and Yuanwei Liu Fellow IEEE

University of Hong Kong · Queen Mary University of London · Constructor University · Aalto University · KTH Royal Institute of Technology

cs.IT, math.IT

Submitted: 2026-05-13

Updated: 2026-11-18

Comments: Submitted to an IEEE journal for possible publication

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

The gist: This tutorial provides a comprehensive overview of continuous-aperture array (CAPA) systems through the lens of electromagnetic signal and information theory (ESIT)…

Key concepts

EM Foundations
This involves using Maxwell's equations to describe how electromagnetic waves radiate from an antenna array. It distinguishes between near-field and far-field regions and uses Green's functions to calculate power flow, helping engineers understand the physical propagation characteristics of signals across the aperture.
Continuous Beamforming
Instead of using discrete matrix math, this technique optimizes continuous current distributions over a wide antenna aperture. It allows for beamforming that is a true analog signal manipulation, enabling higher gain and more flexible transmission patterns than traditional finite array methods.
Channel Estimation via Dictionary Learning
Since directly measuring an infinite channel response is impossible, the system approximates it using a dictionary of physically motivated basis functions (like far-field models). By solving a compressed sensing problem, the system estimates the coefficients of this dictionary to reconstruct the continuous channel with high accuracy.
Spatial Degrees of Freedom (DoF)
This measures how many independent spatial modes or directions are effectively present in a continuous array. For Line-of-Sight (LoS) channels, it scales quadratically with aperture size, while for multipath channels, it is limited by the number of effective modes that can be resolved by the system.

Terminology

Summary

This tutorial provides a comprehensive overview of continuous-aperture array (CAPA) systems through the lens of electromagnetic signal and information theory (ESIT), focusing on bridging the gap between discrete array models and physics-consistent continuous-aperture formulations. p. 0 The review systematically covers the electromagnetic foundations, practical hardware implementations, channel modeling (including line-of-sight and multipath scenarios), continuous beamforming techniques, channel estimation methodologies, fundamental degrees of freedom (DoF), and capacity limits inherent to CAPA systems. p. 0 A key methodological focus is on clarifying analytical tools such as wavenumber domain methods and functional analysis used to transform infinite-dimensional problems into tractable finite-dimensional ones while rigorously maintaining the underlying physical structure. p. 0

The EM foundations are detailed, examining wave radiation governed by Maxwell's equations, classifying spatial regions (reactive near-field, radiative near-field, radiative far-field) based on dyadic Green's function scaling, and analyzing power using both flux and source methods to distinguish between radiated and dissipated power. Mutual coupling is addressed via a dyadic mutual-impedance kernel derived from the Green's function. The continuous signal model relates a continuous source-current distribution over the transmit aperture to an induced field function over the receive aperture, contrasting this with conventional finite-dimensional matrix models. p. 8 Practical implementations approximate this continuity by projecting the field onto finite observation modes, resulting in a unified circuit-to-field description involving an effective loaded admittance matrix Y. p. 8

Channel modeling encompasses LoS channels using the dyadic Green's function, physics-based multipath models describing scatterers and propagation paths, and correlation-based models characterizing channels via angular power spectra. Time-domain and doubly-dispersive channel models extend this by incorporating delay spread and Doppler spread through time-varying channel impulse responses derived from the time-domain Green's function. Beamforming optimization is achieved by introducing finite representations: wavenumber-domain optimization converts functional problems into MIMO matrix problems, while calculus of variations operates directly on the continuous problem to determine optimal current distributions. p. 0 Channel estimation acknowledges that direct pointwise estimation of the infinite-dimensional kernel is impractical, instead showing how optimal beamforming currents are found by solving a finite-dimensional linear system derived from the user-channel correlation matrix R, leading to continuous analogs of classical beamformers like MMSE and ZF. p. 0

The end-to-end model connects digital data streams, circuit domain hardware response, and continuous EM fields: c hat = S hat r r hat S hat t H(r)h(r, s) w tilde(s) c tilde ds dr + z. p. 9 The optimal beamformer is derived as w tilde(s) = h*(s) I K + 1/sigma squared mu R-1 P hat, where P hat is related to the channel structure. p. 1

To handle the unknown bivariate function h(r, s) in equation (182), a projection onto finite transmit and receive codebooks defines projected channel coefficients g m,n (equation 183), transforming the continuous operator into a finite matrix G in C N r times N t. p. 8 A dictionary aligned with dominant propagation mechanisms is then introduced, representing the continuous channel as h(r, s) = theta hat (theta) phi(r, s; theta) d theta (equation 185). Discretizing theta into a codebook yields a finite dictionary phi i(r, s), leading to the sparse approximation h(r, s) about sum from i=1 to N c alpha i phi i(r, s) (equation 186). p. 2 Dictionary functions phi i are chosen based on physics: far-field models use angular-domain kernels phi(r, s; k, kappa) = a r(r; k)a* t(s; kappa) (equation 187), while near-field propagation employs spherical-wave kernels. p. 2 The quality of the dictionary is assessed by the residual h(r, s) - sum from i=1 to N c alpha i phi i(r, s) in L 2(S r times S t) (equation 190). p. 6 Channel training is then formulated as a standard finite-dimensional linear model v = A + z (equation 193), where A is the sensing matrix. p. 0 The estimation of becomes a compressive-sensing problem, minimizing | | 0 or its convex relaxation | | 1, subject to measurement error epsilon. Once estimates hat are obtained, the continuous channel is reconstructed as h(r, s) = sum from i=1 to N c alpha hat i phi i(r, s) (equation 197).

The spatial degrees of freedom (DoF) in CAPA systems are characterized by the number of nonnegligible singular values of the continuous channel operator. p. 0 For LoS channels, Landau’s eigenvalue theorem provides a scaling law: N LoS DoF about A t K r/(2 pi) squared (equation 206), showing quadratic scaling with aperture sizes and inverse square dependence on Tx-Rx distance. For multipath channels, maximum resolvable directional modes are bounded by N LoS DoF at most N eff,t, N eff,r (equation 212), and in rich scattering cases, N LoS DoF about pi (A t, A r)/lambda squared (equation 213). Capacity analysis utilizes continuous covariance kernels, where mutual information simplifies to I(Q) = X m 2 (1 + lambda m) (equation 221), leading to a capacity decomposition C = sum from n=1 to infinity X infinity n=1 2 (1 + sigma squared n P n/N 0) where optimal powers follow a water-filling rule. The Kolmogorov epsilon-capacity offers a deterministic bound: C epsilon about X sqrt P t sigma n> epsilon 2 sqrt P t sigma n /epsilon (equation 232). p. 1 The central message reiterates that effective CAPA design preserves the continuous-space structure imposed by EM theory while developing modeling, optimization, and analysis tools that yield tractable finite-dimensional descriptions. p. 0

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Wavenumber domain optimization allows for direct conversion of continuous functional optimization problems into finite-dimensional matrix optimization problems, enabling standard matrix-based tools for power allocation and precoding. This enables AI systems to determine optimal continuous transmit current profiles by solving a discretized version of the problem formulated in equation 140.

  2. Calculus of variations provides an exact structure for the optimal current distribution by operating directly on the continuous problem without intermediate model approximations such as wavenumber truncation. This allows AI systems to reveal the exact structure of the optimal current distribution and solve complex multi-user beamforming problems through integral equations like equation 170.

  3. The system can implement continuous beamformers, such as the MMSE beamformer described by equation 178, which is the continuous-space analog of the classical minimum mean-squared error (MMSE) beamformer in discrete MIMO systems. This allows for more flexible and higher gain transmission than traditional discrete array methods.

  4. The channel estimation process can be adapted to estimate a continuous-space operator described by the field response h(r, s), which maps transmit current over the aperture to the received field over the aperture. This moves beyond conventional finite-dimensional matrix recovery to address infinite-dimensional uncertainty directly.

  5. Doubly dispersive channel modeling allows AI systems to capture both time dispersion (delay spread) and frequency dispersion (Doppler spread) simultaneously through models like equation 123. This enables the system to design waveforms and receivers that are robust against frequency-selective fading and time-selective fading caused by motion between the transmitter, receiver, and scatterers.

Sources

Related papers