Near-Field Communications with Different Array Geometries: Rayleigh Distance, Channel Estimation, and Transmission Design
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Electrical Engineering. The day in electrical engineering: signals, systems and the hardware that carries them.
Rosa: We're looking at the paper: "Near-Field Communications with Different Array Geometries".
Dev: Here's the gist of the paper. This work establishes a framework of near-field communication under different array geometries for extremely large-scale multiple-input multiple-output (XL-MIMO).
Rosa: The proposed method provides a general channel estimation algorithm that employs the learned prior information to support compressed sensing methods.
Dev: Finally, it proposes optimization algorithms for designing array topologies and hybrid precoding in the downlink.
Rosa: Near-Field Channel Modeling The paper first formulates the near-field spatially non-stationary channel model which is characterized by the distance between the user and each antenna on uniform and modular curved arrays.
Title and authors: Rosa: So we're looking at this paper now called Near-Field Communications with Different Array Geometries: Rayleigh Distance, Channel Estimation, and Transmission Design. It seems like it tackles a really complex area of communication where the physical shape of the antenna array matters a lot.
Dev: Exactly. The authors are looking at how near-field communication works when you change the geometry of the array, specifically things like uniform or modular curved arrays. They also want to see how this affects channel estimation and transmission design for these extremely large-scale multiple-input multiple-output, or XLMIMO, systems.
Taro: What I find interesting is that they are looking at how changing the curvature of the array affects things like the Rayleigh distance, which is basically how far you can effectively communicate in that near-field zone.
Rosa: Right. So they're fixing the total number of antennas but changing how curved they are, and seeing a non-trivial impact on extending that near-field region for cell edges. It suggests that curvature isn't just a minor tweak; it really changes the reach of the signal.
Dev: And then for the estimation part, which is where I get interested in the engineering side, they propose a general channel estimation algorithm that uses learned prior information to support compressed sensing methods. This is significant because it doesn't rely on specific codebooks anymore.
Taro: That sounds like they are trying to solve one of those research gaps mentioned earlier—the need for a general method without being tied down to specific array geometries like ULA or UPA, which is tough when things get complex with hybrid channels and spatial non-stationarity.
Rosa: So the core idea is using a denoising autoencoder, or AE, to learn the low-dimensional channel manifold. This learned prior then acts as a regularizer for the compressed sensing problem they're setting up for estimation.
Dev: They formulate it by minimizing an objective function that looks like it balances clean reconstruction ability and denoising capability through a parameter called alpha alpha ae L s recon plus one minus alpha ae L s denoiser N train, which is how you tune that balance.
Taro: I wonder how robust this AE-aided approximated message passing algorithm, or AE-AMP, actually performs when the world misbehaves and the channel is spatially non-stationary. The paper claims it's more general than conventional deep learning benchmarks.
Rosa: They did show numerical results demonstrating that this AE-AMP algorithm can effectively estimate those spatially non-stationary near-field channels with robustness and generality compared to several other methods. They also showed a comparison between UPA and MCA, suggesting MCA can enhance performance not just at the cell edge but also at the cell center.
Dev: And when they use these estimates for precoding design, they find an obvious gain compared to using simple least squares channel estimation. The algorithm quickly gets a correlation coefficient of one with the true channels, which is pretty strong evidence for its accuracy.
Title and authors: Taro: So, if we think about what this means practically for deploying these massive systems outside the lab—like on a mobile platform or in a real cell environment—this general approach to CSI acquisition without pre-defined structures seems really useful.
Rosa: It definitely opens up the door for designing array topologies and hybrid precoding jointly to maximize the sum user rate. They recast that as minimizing a non-convex objective function, and they use block coordinate descent to solve it.
Dev: The optimization for array geometry is formulated by minimizing the expectation of the normalized multiuser channel matrix H bar under random user distribution within the considered service area, which sounds like a very holistic way to design it.
Taro: I'm looking forward to seeing how this joint optimization translates into practical benefits, like making the near-field region expansion more effective for users at cell edges.
Rosa: They found that using modular cylindrical arrays can expand the near-field region so that favorable spherical-wave propagation conditions are enabled for both the cell center and the cell edge. That's a key result regarding geometry design.
Dev: But there is a caveat here, as they note, they still have research gaps in enhancing reliability and quality for near-field communication. For example, existing methods often rely on specific array geometries like ULA or UPA to exploit the mathematical structure of steering vectors and design specific codebooks ten–twelve <ref:2605.17690#pg2,to exploit the mathematical structure of steering vectors and design specific codebooks>.
Taro: That's true. As soon as the shape and topology of arrays change, you have to redesign the near-field codebook, which isn't practical when you’re dealing with extremely large antenna counts in XLMIMO systems.
Rosa: The paper is really pushing for a general method without relying on those specific assumptions, especially when dealing with hybrid-field channels and spatial non-stationarity. They are trying to make the estimation process less dependent on knowing the exact geometry beforehand.
Dev: And when it comes to transmission design, they propose optimization algorithms that can reformulate digital and analog precoding using alternating optimization algorithms that yield closed form optimal solutions with low complexity for the RF chain variables vtilde mRF.
Taro: Low complexity is important because in real-time loops, latency is everything. If the solution has to be solved through a complex iterative process, it might not work fast enough for high-speed systems.
Rosa: The authors also showed that their proposed AE-AMP algorithm achieves the best performance except for the smallest signal to noise ratio, which shows clear gains compared to conventional codebook-based algorithms and even some end to end learning methods.
Dev: That small SNR limitation is a practical constraint we have to keep in mind for any real deployment, especially when we think about real-time loop performance under noisy conditions.
Title and authors: Taro: So, what's the bigger picture here? It seems they are providing a flexible framework that can handle the complexity of non-stationary channels and arbitrary array geometries without needing specific pre-defined structures.
Rosa: Exactly. They establish this framework for XLMIMO communication under different array geometries, which is a big step toward building more adaptable systems.
Dev: We've seen how they tackle the channel estimation problem using the AE-AMP algorithm, and now they show how to use that accurate CSI to design the array geometry and hybrid precoding jointly for maximum sum user rate.
Taro: The joint optimization of array geometry parameters like curvature angle, subarray number, and subarray spacing to minimize the average channel condition number leads to more stable Zero Forcing or Minimum Mean Square Error precoding.
Rosa: That better spatial separation of users is a direct benefit for performance in multiuser systems. It means better data rate even when users are close together.
Dev: So, looking at the overall findings of this paper, it’s about providing a general channel estimation algorithm that uses learned prior information to support compressed sensing methods, and then using those estimates to design array topologies and hybrid precoding for downlink optimization.
Taro: It's a lot of work to get that level of robustness across different geometries. I wonder if this framework scales well when we move from these controlled simulations out into a messy real-world environment.
Rosa: That’s the next big question, I think. But for now, the numerical results show superiority over several benchmarks, and it demonstrates a capability to learn channel manifolds even with more complicated geometries.
Dev: Before we wrap up this discussion on Near-Field Communications with Different Array Geometries: Rayleigh Distance, Channel Estimation, and Transmission Design, we need to summarize these points.
Taro: I just want to say that the paper shows a general approach that can handle the complexity of non-stationary channels and arbitrary array geometries without needing specific pre-defined structures.
Rosa: And it's definitely a step forward because they demonstrate how to jointly design array geometry and hybrid precoding for downlink optimization.
Dev: The main contribution is the AE-AMP algorithm, which provides robust channel estimation that supports compressed sensing methods through a learned regularizer, and the optimization algorithms for designing the array topology.
Taro: I just want to emphasize that while it's very general, we have to keep in mind that they still have gaps in enhancing reliability and quality for near-field communication.
Rosa: That’s fair, but what they did show is a strong performance gain compared to methods that rely on specific codebooks or simpler channel estimation techniques.
Dev: It's a solid contribution to the field because it provides a framework that addresses the need for general solutions in XLMIMO systems where things get complicated.
The paper's summary: Rosa: So we're looking at this paper now called Near-Field Communications with Different Array Geometries: Rayleigh Distance, Channel Estimation, and Transmission Design. It tackles how to handle extremely large antenna systems when the physical shape of the array changes—like curved or modular setups—and it focuses on getting a good channel estimate and designing the transmission for maximum speed.
Dev: Right. The core idea is they build this whole framework so you can get accurate channel information even when you're dealing with these complicated, non-stationary near-field environments without needing a fixed set of assumptions about the array shape beforehand.
Taro: What I find really interesting is how they use a deep learning method to learn what those complex channel conditions look like in a low dimension, which acts like this learned prior to help the estimation process.
Rosa: Exactly. They propose an algorithm called AE-AMP, which uses this autoencoder to learn the shape of the channel space and then uses that knowledge as a regularizer for their compressed sensing technique. It's designed to be general enough that it doesn't care too much about whether you have a uniform array or something more complex.
Dev: From my side, I'm looking at the real-time aspect here. They also propose optimization algorithms that let you design both the array shape and the hybrid precoding at the same time to maximize how much data users can send.
Taro: That joint optimization part is key because they formulate it by minimizing something related to how spread out those user channels are, which helps them design a better physical layout for the antennas.
Rosa: And when you look at the results, they show that this method beats several existing benchmarks in terms of estimation accuracy and actually improves performance even when you have users close together or near the edge of the cell coverage area.
Dev: They found that their channel estimates are so good, they get a correlation coefficient of one with the true channels pretty quickly, which is a strong signal for reliability in real systems.
Taro: But they did admit there's still some room to improve reliability and quality, especially when you push the limits of near-field communication because those specific geometric assumptions still matter a lot.
Rosa: That’s true. The paper shows it’s a powerful tool for handling the complexity of non-stationary channels without being tied down to just one specific array setup.
Dev: So, it gives you a robust way to get accurate CSI and then use that information to tailor the physical antenna layout and signal transmission for better user performance in XLMIMO systems.
Taro: It’s about making the system adaptable enough that it can handle real-world variations in shape and channel behavior gracefully.
Rosa: It really moves things forward by providing a general method for both channel estimation and the subsequent optimization of the physical hardware design itself.
The paper's improvements: Taro: So, we're looking at how they suggest improving things in that framework—specifically for making the system more robust when you're dealing with those complex near-field conditions and geometry changes.
Rosa: Basically, they are suggesting a way to make the estimation process even more resilient to those weird spatial variations that happen in real environments, like when the antenna is moved or things shift around.
Dev: They focus on making sure that the optimization process for the array shape doesn't get stuck in bad local minimums during those iterative steps they use.
Taro: They seem to be pushing for a way to design those hybrid precoding systems using alternating optimization algorithms that give you closed form solutions with low complexity, which is something engineers really want for real-time operation.
Rosa: That sounds like it’s about making the actual signal processing part faster and less computationally heavy when designing the hardware setup.
Dev: Yeah, they're aiming for those closed forms so the latency doesn't blow up when you have hundreds of antennas involved in that optimization loop.
Taro: And they are also refining how they use that learned prior information to ensure the final channel estimates aren't just good in simulation but actually hold up when you test it outside the lab.
Rosa: So, it’s about moving from a good simulation result to something that performs well when the world misbehaves, which is what we need for field robotics and mobile communication.
Dev: It’s about building a safety case where we know that even if the channel changes unexpectedly, our precoding design won't completely fail or degrade severely.
Taro: They are also looking at how to use that improved array geometry design to better handle interference, specifically by minimizing the average channel condition number, which keeps the users spatially separated.
Rosa: So they’re not just getting a better estimate; they’re using that estimate to actively shape the physical antennas in a way that fights interference from other users.
Dev: That links right back to how well you can actually separate those multiple users when you've got massive antenna arrays operating in close proximity.
Taro: It seems like their next step is showing this framework working on more complex, real-world scenarios where the channel non-stationarity is even more extreme than what they simulated.
Rosa: That’s the real test—how long this setup can stay stable when it's constantly changing, and if it still performs well after a few hours of field trials.
Dev: We need to see those failure modes clearly laid out so we know exactly what kind of world-misbehavior we're talking about and how the system should recover gracefully.
Conclusion: Rosa: So to wrap up, this paper, "Near-Field Communications with Different Array Geometries: Rayleigh Distance, Channel Estimation, and Transmission Design," shows that we can build a general framework for XLMIMO communication regardless of the antenna array shape.
Dev: Right. The main implication is that you no longer have to be tied down to one specific array geometry—like just a uniform array—to get high-quality channel information and good transmission design.
Taro: I think the real takeaway is that this method gives us a flexible tool for designing systems that can adapt to whatever physical constraints we face in the field.
Rosa: It's about having a method for estimating those near-field channels robustly, even when the environment is constantly shifting, which is crucial for any field robotics deployment.
Dev: From an engineering standpoint, it means we can design systems where the hardware geometry and the signal processing are optimized together to maximize data rate without needing constant manual adjustments.
Taro: I just think this flexibility helps us push the boundaries on what’s possible with massive antenna arrays in unpredictable settings.
Rosa: Exactly. The AE-AMP algorithm is strong, but we still have that caveat about achieving the highest accuracy at the very lowest signal to noise ratio, which means we need more power to really see those gains.
Dev: That low signal level limitation is a practical thing; it tells us that for real deployment, we’ll still be limited by what hardware can reliably transmit at a certain distance.
Taro: We’ll have to figure out how to bridge that gap, maybe by integrating better physical layer security techniques or improving the denoising capability further in the next iteration.
Rosa: That sounds like a good path forward; focusing on making that learned regularizer even smarter for handling those extreme low SNR conditions.
Dev: So, looking at the whole paper on "Near-Field Communications with Different Array Geometries: Rayleigh Distance, Channel Estimation, and Transmission Design," we have a solid foundation for adaptive XLMIMO systems.
Taro: It sets a clear direction for future research on making these AI-aided estimation methods even more resilient to those real-world distortions.
Kangda Zhi, Yi Song, Tianyu Yang, Tuo Wu, Tengjiao Wang, Songyan Xue, Fangzhou Wu, § Giuseppe Caire
Technische Universitat Berlin
eess.SP
Submitted: 2026-05-17
Updated: 2026-11-07
Comments: 13 pages
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 87/100
The gist: This work establishes a framework of near-field communication under different array geometries for extremely large-scale multiple-input multiple-output (XL-MIMO).
Key concepts
- Near-Field Channel Modeling
- This describes how the communication channel is modeled when users are very close to the antennas on curved arrays. The model accounts for both direct line-of-sight (LoS) paths and non-line-of-sight (NLoS) paths, using a visibility region set to define these interactions.
- AE-AMP Algorithm
- This is a deep learning approach used for estimating the channel when array geometries are arbitrary. It uses a denoising autoencoder (AE) to learn the low-dimensional shape of the channel and acts as a learned regularizer for compressed sensing, providing robust estimates.
- Array Geometry Analysis
- This section investigates how changing the curvature of an antenna array affects near-field distances. It shows that increasing curvature can reduce Rayleigh distance for normal incidence but increase it for side incidence, impacting coverage at cell edges.
- Joint Transmission Design Optimization
- This involves designing both the physical layout (array geometry) and the signal processing (hybrid precoding) simultaneously to maximize the total data rate. This is achieved by minimizing an objective function using WMMSE methods and block coordinate descent.
Terminology
Summary
This work establishes a framework of near-field communication under different array geometries for extremely large-scale multiple-input multiple-output (XL-MIMO). p. 0 The proposed method provides a general channel estimation algorithm that employs the learned prior information to support compressed sensing methods. p. 1 Finally, it proposes optimization algorithms for designing array topologies and hybrid precoding in the downlink. p. 0 p. 0
Near-Field Channel Modeling
The paper first formulates the near-field spatially non-stationary channel model which is characterized by the distance between the user and each antenna on uniform and modular curved arrays. The channel is expressed as a general form involving a line-of-sight (LoS) path and Lk non-LoS (NLoS) paths, characterized by a visibility region (VR) set Φk,l. p. 1 The m-th element of hk,0 in the channel expression denotes the near-field LoS channel between user k and the m-th antenna of the BS. p. 2
Channel Estimation Algorithms
To estimate the channel under arbitrary array geometries and arbitrary fields, a deep learning-based approximate message passing (AMP) algorithm is proposed which uses a denoising autoencoder (AE) to learn the low-dimensional channel manifold and serves as a learned regularizer for the compressed sensing problem. This AE-AMP algorithm is evaluated by establishing its state evolution (SE) and replica bound. p. 1 The estimation algorithm is formulated to solve the antenna-domain channel estimation problem hˆk = arg min hk∥yk − Vulhk∥2+ λ∥hk − RΘ hT∥2 for arbitrary array geometries and arbitrary fields. p. 6
Array Geometry Analysis
The paper investigates the impact of array curvature on near-field distances by fixing the total number of antennas while varying the degree of curvature, uncovering a non-trivial impact on extending the near-field region for cell edges. For a uniform curved array, Proposition 1 shows that as array curvature angle θ increases, the Rayleigh distance for normal incident direction reduces while the Rayleigh distance for side incident direction increases. p. 3 Similarly, Proposition 2 indicates that for a uniform cylindrical array with fixed horizontal arc length Lh and fixed vertical length (Mv − 1)d, as array curvature angle θ increases, the Rayleigh distance for normal incident direction reduces while the Rayleigh distance for side incident direction increases. p. 4
Transmission Design Optimization
The paper proposes an optimization algorithm to maximize the sum user rate for sub-connected XL-MIMO systems by jointly designing the array geometry and hybrid precoding in the downlink. The problem is recast using the WMMSE method [39] to minimize a non-convex objective function, leading to a tractable problem solvable by block coordinate descent. p. 9 Design of array geometry is formulated as minimizing the expectation of the normalized multiuser channel matrix H̄ under random user distribution within the considered service area. p. 9
Simulation and Results
Numerical results demonstrate that the proposed AE-AMP algorithm can effectively estimate the spatially non-stationary near-field channels with robustness and generality compared to several conventional and deep learning-based benchmarks. The comparison between UPA and MCA shows that MCA can enhance the performance not only at the cell edge, but also at the cell center. p. 3 With highly accurate CSI, precoding design based on channels estimated by AE-AMP algorithms achieves an obvious gain compared to that with LS channel estimation. p. 0 The proposed algorithm quickly achieves a correlation coefficient of 1 with true channels. p. 0
Conclusion
This paper investigated the challenging problem of XLMIMO communication with array geometry optimization and CSI acquisition. The proposed method provides a general deep learning-based channel estimation algorithm that employs the learned prior information to support compressed sensing methods. p. 1 Finally, it proposes optimization algorithms for designing array topologies and hybrid precoding in the downlink. The numerical results demonstrate its superiority over several benchmarks. p. 0
References
[1] K. Zhi et al., “A general channel estimation method for spatial nonstationary mixed near- and far-field XL-MIMO channels,” 2026
[2] H. Lu, Y. Zeng, C. You, Y. Han, J. Zhang, Z. Wang et al. p. 0, “A tutorial on near-field XL-MIMO communications toward 6G,” IEEE Commun. p. 0 Surveys Tutorials
[3] T. Mao, H. Li, S. Tan, P. Wang, G. Liu et al., “NEFT: A unified transformer framework for efficient near-field CSI feedback in XLMIMO systems,” arXiv preprint arXiv:2509. p. 112748
[4] S. Yang, C. Xie, W. Lyu, B. Ning, Z. Zhang, and C. Yuen, “Near-field channel estimation for extremely large-scale reconfigurable intelligent surface (XL-RIS)-aided wideband mmwave systems,” IEEE J. p. 0 p. 0 p. 0 Sel. Areas Commun., vol 42
[5] K. Zhi et al., “Performance analysis and low-complexity design for XL-MIMO with near-field spatial non-stationarities,” IEEE J. p. 0 Sel. Areas Commun., vol 42
[6] Y. Chen, X. Guo, G. Zhou, S. Jin, D. W. K. Ng, and Z. Wang, “Unified far-field and near-field in holographic MIMO: A wavenumber-domain perspective,” IEEE Commun Mag. p. 0, vol 63
[7] P. Tang, H. Xu, J. Zhang, X. Liu et al., “A tutorial on 3GPP rel-19 channel modeling for 6G FR3 (7-24 GHz): From standard specification to simulation implementation,” arXiv preprint arXiv:2602. p. 007623
[8] M. Liu, C. Pan, K. Zhi, H. Ren et al., “Low-complexity iterative precoding design for near-field multiuser systems with spatial nonstationarity,” IEEE Trans Signal Process. p. 1, vol 74
[9] Y. Zhang, C. You, R. Zhang, B. Zheng, H. C. So, D. Niyato, and T. Q. Quek, “Rotatable antenna enabled multi-cell mixed near-field and farfield communications,” IEEE Trans Wireless Commun. p. 0, 2026
[10] M. Cui and L. Dai, “Channel estimation for extremely large-scale MIMO: Far-field or near-field? p. 0” IEEE Trans Commun. p. 0, vol 70
[11] Z. Wu and L. Dai, “Multiple access for near-field communications: Sdma or ldma? p. 0” IEEE J. Sel. Areas Commun., vol 41
[12] Z. Wu, M. Cui, and L. Dai, “Enabling more users to benefit from near-field communications: From linear to circular array,” IEEE Trans Wireless Commun. p. 0, 2023
[13] K. Zhi, T. Yang, S. Xue, and G. p. 0 Caire, “Near-field 3D localization and mimo channel estimation with sub-connected planar arrays,” arXiv preprint arXiv:2510. p. 020274
[14] C. Wu, C. You et al., “Two-stage hierarchical beam training for nearfield communications,” IEEE Trans Veh. p. 0 Tech., 2023
[15] X. Zhang, H. Zhang, J. Zhang, C. Li, Y. Huang, and L. Yang, “Codebook design for extremely large-scale MIMO systems: Near-field and farfield,” IEEE Trans Commun. p. 0, 2023
[16] W. Liu, H. Ren, C. Pan, and J. Wang, “Deep learning based beam training for extremely large-scale massive MIMO in near-field domain,” IEEE Commun Lett. p. 0, vol 27
[17] P. Wang, X. Lei, M. Wu, and L. Fan, “Robust near-field XL-MIMO channel estimation via dual-stage optimization and residual refinement,” IEEE Trans Commun. p. 0, 2026
[18] A. Tang, J.-B. Wang, Y. Pan et al., “Joint visibility region and channel estimation for extremely large-scale MIMO systems,” IEEE Trans Commun. p. 0, vol 72
[19] W. Xu, A. Liu, M.-j. Zhao, and G. Caire, “Joint visibility region detection and channel estimation for XL-MIMO systems via alternating MAP,” IEEE Trans Signal Process. p. 0, vol 72
[20] H. Lu and Y. Zeng, “Communicating with extremely large-scale array/surface: Unified modelling and performance analysis,” IEEE Trans Wireless Commun. p. 0, vol 21
[21] P. Zhang, V. Petrov, and E. Bjornson, “Near-field boundary distance in mmwave and THz communications with misaligned antenna arrays,” IEEE Trans Wireless Commun. p. 0, vol 25
[22] A. Kosasih and E. Bjornson, “Finite beam depth analysis for large arrays,” IEEE Trans Wireless Commun. p. 0, vol 23
[23] C. Ouyang, Y. Liu, X. Zhang, and L. Hanzo, “Near-field communications: A degree-of-freedom perspective,” arXiv preprint arXiv:2308. p. 000362
[24] Z. Zhang, Y. Liu, Z. Wang, X. Mu, and J. Chen, “Physical layer security in near-field communications,” IEEE Trans Veh. p. 0 Tech., vol 73
[25] Z. Wang, X. Mu, and Y. Liu, “Near-field integrated sensing and communications,” IEEE Commun Lett. p. 0, vol 27
[26] S. Zhang, F. Wang, Y. Mao, A.-L. Jin, and T. Q. Quek, “Rate-splitting multiple access for near-field communications with imperfect CSIT and SIC,” IEEE Trans Commun. p. 0, 2025
[27] Y. Chen, M. Chen, H. Xu, Z. Yang, K.-K. Wong, and Z. Zhang, “Joint beamforming and antenna design for near-field fluid antenna system,” IEEE Wireless Commun Lett. p. 0, vol 14
[28] L. Zhu, W. Ma, Z. Xiao, and R. Zhang, “Movable antenna enabled near-field communications: Channel modeling and performance optimization,” IEEE Trans Commun. p. 0, vol 73
[29] S. Liu, X. Yu, J. Xu, and R. Zhang, “Near-field communication with massive movable antennas: A functional perspective,” IEEE Trans Wireless Commun. p. 0, vol 25
[30] Y. Guo, Y. Zhang, L. Pang, Y. Ren, Y. Chen, and Y. Liu, “Design of non-uniform 3D arrays for near-field XL-MIMO communications,” IEEE Wireless Commun Lett. p. 0, 2025
[31] H. Shen, Y. Chen, C. Han, and J. Yuan, “Hybrid beamforming with widely-spaced-array for multi-user cross-near-and-farfield communications,” IEEE Trans Commun. p. 0, 2025
[32] M. Wu, C. You, and X. Lei, “Beamfocusing analysis for modular XLarrays: Effect of time synchronization errors,” IEEE Wireless Commun Lett. p. 0, 2026
[33] Y. Han, S. Jin, C.-K. Wen, and X. Ma, “Channel estimation for extremely large-scale massive MIMO systems,” IEEE Wireless Commun Lett. p. 0, vol 9
[34] A. G. Davoodi and S. A. Jafar, “Aligned image sets under channel uncertainty: Settling conjectures on the collapse of degrees of freedom under finite precision CSIT,” IEEE Trans Inf Theory, vol 62
[35] 3GPP, “NR; physical channels and modulation,” no. 38.211, 2026
[36] K. Zhang, Y. Li, W. Zuo, L. Zhang, L. Van Gool, and R. Timofte, “Plug-and-play image restoration with deep denoiser prior,” IEEE Trans Pattern Anal Mach Intell. p. 7, vol 44
[37] C. A. Metzler, A. Maleki, and R. G. Baraniuk, “From denoising to compressed sensing,” IEEE Trans Inf Theory, vol 62
[38] D. Guo and S. Verdu, “Randomly spread CDMA: Asymptotics via statistical physics,” IEEE Trans Inf Theo., vol 51
[39] Q. Shi et al., “An iteratively weighted MMSE approach to distributed sum-utility maximization for a MIMO interfering broadcast channel,” IEEE Trans Signal Process., vol 59
[40] K. Zhi et al., “Is RIS-aided massive MIMO promising with ZF detectors and imperfect CSI? p. 0” IEEE J. Sel. Areas Commun., vol 40
Where the gains come from
The improvement of data rate by using modular curved arrays with the estimated channel is also validated. The sum user rate maximization problem with hybrid precoding constraints and array-geometry constraints can be formulated as a non-convex problem, which is solved by introducing auxiliary variables uk and vk ≥ 0 and applying the WMMSE method [39]. p. 8 The optimization of array geometry is formulated as minimizing the expectation of the normalized multiuser channel matrix H̄ under random user distribution within the considered service area. p. 9 The modular cylindrical array can expand the near-field region so that favorable spherical-wave propagation condition is enabled for both the cell center and the cell edge. p. 9 The resulting performance shows that MCA can enhance the performance not only at the cell edge, but also at the cell center. p. 3 The proposed AE-AMP algorithm achieves the best performance except for the smallest SNR, showing obvious gains compared to conventional codebook-based algorithms [11] and end-to-end learning methods. p. 0 The similar estimation precision achieved between UPA and MCA also shows the capability of the algorithm to learn the channel manifold even with more complicated geometries. p. 1 The main reason is that MCA has a better beamfocusing capability to mitigate multi-user interference even when users are located with closed angles. p. 3 The proposed AE-AMP algorithm only uses a network to learn the low-dimensional channel manifold and assists the compressed sensing algorithm. p. 1 The overall training objective is Θ∗ = arg min Θ N Xtrain s=1 αaeL(s) recon + (1−αae)L(s) denoiser Ntrain, where αae balances the clean reconstruction ability and the denoising capability. p. 6 The proposed AE-AMP algorithm is general, effective, and robust. p. 0 The performance gain is demonstrated compared to that with LS channel estimation. p. 1
Simulation Details
The simulation setup involves a BS with M = 48 × 16 = 768 antennas located at [0, 0, 0] where each RF chain connects Ms = 6 antennas, operating at fc = 6.8 GHz to communicate with K=10 users. p. 1 The encoder contains three 1D convolutional layers with a kernel size of 5 and a stride of 2. p. 6 For offline training, we generate 25 × 10 4 channel samples for UPA and MCA where each location of the user is sampled randomly from a 3D area spanning x ∈ [5, 50], y ∈ [-50, 50], and z ∈ [0, 4]. p. 6 The batch size, number of epochs, and initial learning rate are set to 512, 100, and 1e-3 respectively. p. 7 The noisy inputs for AE training are generated with SNR uniformly sampled from 0 dB to 20 dB. p. 6
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
-
Improved Channel Estimation via AE-AMP: The proposed denoising autoencoder aided approximate message passing algorithm can effectively estimate spatially non-stationary near-field channels with robustness and generality compared to several conventional and deep learning based benchmarks. This allows the AI system to obtain accurate channel state information (CSI) under arbitrary array geometries without relying on specific codebooks or dictionaries.
-
Improved Array Geometry Optimization: The optimization framework jointly designs the array geometry and hybrid precoding in the downlink to maximize the sum user rate by minimizing average channel condition number so that they do not need to be adjusted with time. This enables a more effective near-field region expansion and better coverage for users at cell edges.
-
Improved Transmission Design via Joint Optimization: The system can jointly design array geometry parameters like curvature angle, subarray number, and subarray spacing to minimize the average channel condition number, leading to more stable Zero Forcing (ZF) or Minimum Mean Square Error (MMSE) precoding. This results in better spatial separation of users and improved data rate performance.
-
Improved Robustness via Learned Regularization: The use of a learned regularizer RTheta that captures the inherent features of the channel can help the estimation algorithm generate estimates that satisfy certain inherent features of the channel hk, which does not reduce to a simple sparsity assumption modelled by explicit codebooks. This enhances robustness against variations in array geometry and spatial non-stationarity.
-
Improved Downlink Precoding Design: The optimization process for digital and analog precoding can be reformulated using alternating optimization algorithms that yield closed form optimal solutions with low complexity for the RF chain variables vtilde mRF, which is desirable for large scale systems. This facilitates the design of power-constrained RF chains in a structured way.
Sources
- NEFT: A Unified Transformer Framework for Efficient Near-Field CSI Feedback in XL-MIMO Systems
- A Tutorial on 3GPP Rel-19 Channel Modeling for 6G FR3 (7-24 GHz): From Standard Specification to Simulation Implementation
- Near-Field 3D Localization and MIMO Channel Estimation with Sub-Connected Planar Arrays
- Near-Field Communications: A Degree-of-Freedom Perspective
Related papers
- Open-H-RF: A Large-Scale Ultrasound Raw Channel Dataset Initiative
- Scalable Attention for 5G NR Channel Estimation
- Low-Altitude Wireless Networks: A Comprehensive Survey
- Sensing-Assisted OFDM with Index Modulation: A Waveform for Optical Wireless Integrated Communication and Localisation
- Iteration-Free Candidate Pruning and an Index-Domain VLSI Architecture for Sort-Free MIMO Box Decoding up to 1024-QAM
- A Foundation Model for Large-Scale Wireless Network Planning, Operation and Optimization