pith. sign in

arxiv: 2605.22067 · v1 · pith:DXYADQL5new · submitted 2026-05-21 · 📡 eess.SP · cs.IT· math.IT

Learning Energy-Efficient Modular Arrays under Hardware Non-linearities

\"Ozlem Tu\u{g}fe Demir , Alva Kosasih This is my paper

Pith reviewed 2026-05-22 04:20 UTC · model grok-4.3

classification 📡 eess.SP cs.ITmath.IT
keywords energy efficiencysparse arrayspower amplifier non-linearitiesdeep neural networkBussgang decompositionmodular arraysMIMOantenna activation
0
0 comments X

The pith

Unsupervised DNN jointly optimizes power allocation and modular activation to raise energy efficiency in sparse arrays under amplifier non-linearities.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that an unsupervised deep neural network can learn the mapping from channel features directly to the operating point that maximizes energy efficiency in sparse extremely large aperture arrays when power amplifiers introduce non-linear distortions. A reader would care because the joint choice of which sub-arrays to activate and how much power to allocate is a combinatorial, non-convex problem that grows intractable with array size, while conventional sparse-array designs lose efficiency once hardware non-linearities are taken into account. The authors first obtain a closed-form spectral-efficiency expression via Bussgang decomposition that folds the distortion into the optimization, then train the network to predict distortion-aware power levels, total power scaling, and sub-array activation patterns from singular-value and geometric channel features. If the learned mapping is accurate, the method supplies a practical, low-complexity alternative to repeated numerical optimization for real-time energy-efficient operation.

Core claim

The unsupervised DNN learns the non-linear mapping between singular-value and geometric channel features and the optimal energy-efficiency operating point, jointly outputting distortion-aware power allocation, total transmit power scaling, and modular sub-array activation; numerical results show that the resulting arrays achieve significant energy-efficiency gains over conventional sparse arrays.

What carries the argument

Unsupervised deep neural network that maps singular-value and geometric channel features to the joint choice of power allocation, total power scaling, and modular sub-array activation while maximizing energy efficiency under non-linear distortions.

If this is right

  • The closed-form spectral-efficiency expression derived from Bussgang decomposition makes non-linear distortion effects tractable inside the energy-efficiency objective.
  • The DNN replaces repeated solution of a combinatorial non-convex problem with a single forward pass once trained.
  • Modular sub-array activation combined with per-antenna power scaling improves energy efficiency compared with fixed sparse layouts.
  • The approach scales to array sizes where exhaustive search or convex relaxations become impractical.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same unsupervised learning structure could be applied to joint optimization problems involving other hardware impairments such as phase noise or quantization.
  • Adding more channel statistics or array geometry parameters as input features might tighten the learned mapping without increasing inference cost.
  • The framework points toward online retraining or transfer learning when array size or amplifier characteristics change after deployment.

Load-bearing premise

The Bussgang decomposition supplies an accurate analytical expression for achievable spectral efficiency under power amplifier non-linearities, and singular-value plus geometric channel features are sufficient for the unsupervised DNN to learn the optimal joint operating point.

What would settle it

Running exhaustive or high-accuracy numerical search for the true energy-efficiency optimum on a held-out set of channel realizations and comparing those values to the energy efficiency produced by the DNN outputs would reveal whether the network reaches or falls short of the optimal operating points.

Figures

Figures reproduced from arXiv: 2605.22067 by Alva Kosasih, \"Ozlem Tu\u{g}fe Demir.

Figure 1
Figure 1. Figure 1: SE versus power amplifier compression parameter [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: EE versus power amplifier compression parameter [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Example of the modular array structure with [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: EE comparison between the proposed DNN-based [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: SE comparison between the proposed DNN-based [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
read the original abstract

This paper investigates the joint optimization of power allocation and antenna activation in sparse extremely large aperture array systems operating under power amplifier non-linearities. We first derive an analytical expression for the achievable spectral efficiency (SE) of point-to-point MIMO channels affected by non-linear distortions using the Bussgang decomposition. To address the combinatorial and non-convex nature of the energy-efficiency (EE) maximization problem, we employ an unsupervised deep neural network (DNN) that learns the non-linear mapping between the channel state information and the optimal EE operating point. The DNN jointly predicts distortion-aware power allocation, total transmit power scaling, and modular sub-array activation based on singular-value and geometric channel features. Numerical results demonstrate that the proposed DNN-based arrays achieve significant EE gains over the conventional sparse arrays.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 3 minor

Summary. The paper investigates joint optimization of power allocation and modular sub-array activation for energy efficiency (EE) in sparse extremely large aperture array (ELAA) MIMO systems impaired by power amplifier non-linearities. It first derives a closed-form achievable spectral efficiency (SE) expression via Bussgang decomposition for point-to-point MIMO channels, then trains an unsupervised DNN that maps singular-value and geometric channel features to distortion-aware power allocation, total power scaling, and binary activation decisions. Numerical results claim that the resulting DNN-based designs achieve significant EE gains relative to conventional sparse arrays.

Significance. If the Bussgang-derived SE remains an accurate surrogate once activations and scalings are jointly optimized, the unsupervised DNN approach offers a scalable alternative to combinatorial EE maximization in hardware-impaired massive MIMO. The explicit use of an analytical objective inside the unsupervised loss is a methodological strength that could enable reproducible, low-complexity inference at deployment. However, the overall significance hinges on whether the reported EE gains survive verification against the true (non-Bussgang) rate; without that check the numerical claims rest on an unconfirmed modeling assumption.

major comments (1)
  1. [SE derivation and numerical results] The headline numerical claim (significant EE gains) is obtained by training the DNN to maximize the Bussgang-based SE expression. Bussgang requires the effective transmit vector to be circularly symmetric complex Gaussian with distortion uncorrelated to the input; both conditions become approximate when the optimizer is free to activate or deactivate entire modules and to apply per-module power scaling. The manuscript should therefore include, in the numerical-results section, a direct comparison of the analytical SE against Monte-Carlo-estimated mutual information for the final optimized activation patterns; absence of this check leaves the central EE improvement claim only weakly supported.
minor comments (3)
  1. [Abstract and numerical results] The abstract and numerical section report 'significant EE gains' without stating the quantitative improvement (e.g., percentage or dB) or the precise baseline configurations (e.g., which conventional sparse arrays and power-allocation schemes are used).
  2. [DNN design and training] No description is given of the DNN architecture (layer sizes, activation functions), the precise unsupervised loss implementation, training hyperparameters, or the size and diversity of the channel realizations used for training and testing.
  3. [Numerical results] The paper does not report error bars, number of Monte-Carlo trials, or statistical significance tests on the EE curves, which is needed to substantiate the reliability of the plotted gains.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback and for identifying a key point that can strengthen the validation of our numerical results. We address the major comment below.

read point-by-point responses

  1. Referee: [SE derivation and numerical results] The headline numerical claim (significant EE gains) is obtained by training the DNN to maximize the Bussgang-based SE expression. Bussgang requires the effective transmit vector to be circularly symmetric complex Gaussian with distortion uncorrelated to the input; both conditions become approximate when the optimizer is free to activate or deactivate entire modules and to apply per-module power scaling. The manuscript should therefore include, in the numerical-results section, a direct comparison of the analytical SE against Monte-Carlo-estimated mutual information for the final optimized activation patterns; absence of this check leaves the central EE improvement claim only weakly supported.

    Authors: We agree that confirming the accuracy of the Bussgang-based SE expression for the jointly optimized activation patterns and power scalings is essential to support the reported EE gains. Although the derivation applies the Bussgang decomposition to the effective (post-activation and scaled) transmit vector and the unsupervised loss directly maximizes this closed-form expression, the circularly symmetric Gaussian assumption becomes approximate under binary sub-array decisions. In the revised manuscript we will add, in the numerical-results section, a direct comparison of the analytical SE against Monte-Carlo estimates of mutual information evaluated on the final optimized activation patterns and power allocations. This verification will quantify any gap and thereby strengthen the central claims. revision: yes

Circularity Check

0 steps flagged

No significant circularity; SE derivation and DNN optimization remain independent

full rationale

The paper first derives a closed-form SE expression via Bussgang decomposition applied to the point-to-point MIMO channel under PA non-linearities. This analytical expression is then employed as the objective function inside the unsupervised DNN loss, which maps singular-value and geometric channel features to joint power allocation, total power scaling, and sub-array activation decisions. No equation or step reduces the final EE result to a fitted parameter by construction, nor does any load-bearing claim rest on a self-citation chain whose validity is internal to the present work. The Bussgang-based SE is a standard external tool whose assumptions are stated separately from the DNN training procedure, making the overall derivation self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the accuracy of the Bussgang model for non-linear distortions and the sufficiency of the chosen channel features for the DNN to learn near-optimal EE points.

axioms (1)
  • domain assumption Bussgang decomposition accurately models the effect of power amplifier non-linearities on the transmitted signal in point-to-point MIMO channels.
    Invoked to derive the analytical spectral efficiency expression.

pith-pipeline@v0.9.0 · 5662 in / 1259 out tokens · 71426 ms · 2026-05-22T04:20:30.338040+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

18 extracted references · 18 canonical work pages

  1. [1]

    Near-field MIMO communications for 6G: Fundamentals, challenges, potenti als, and future directions,

    M. Cui, Z. Wu, Y . Lu, X. Wei, and L. Dai, “Near-field MIMO communications for 6G: Fundamentals, challenges, potenti als, and future directions,” IEEE Commun. Mag. , vol. 61, no. 1, pp. 40–46, 2023

    work page 2023

  2. [2]

    Ex ploiting the depth and angular domains for massive near-field spatial mul tiplexing,

    P . Ramezani, A. Kosasih, A. Irshad, and E. Bj¨ ornson, “Ex ploiting the depth and angular domains for massive near-field spatial mul tiplexing,” IEEE BITS the Inf. Theory Mag. , pp. 1–12, Oct. 2023

    work page 2023

  3. [3]

    Extremely large aperture array (ELAA) communic ations: Foundations, research advances and challenges,

    S. Y e, M. Xiao, M.-W. Kwan, Z. Ma, Y . Huang, G. Karagiannid is, and P . Fan, “Extremely large aperture array (ELAA) communic ations: Foundations, research advances and challenges,” IEEE Open J. of the Commun. Soc. , vol. 5, pp. 7075–7120, 2024

    work page 2024

  4. [4]

    Sparse arra y enabled near-field communications: Beam pattern analysis a nd hybrid beamforming design,

    C. Zhou, C. Y ou, H. Zhang, L. Chen, and S. Shi, “Sparse arra y enabled near-field communications: Beam pattern analysis a nd hybrid beamforming design,” IEEE Trans. on Wireless Commun. , Early Access 2025

    work page 2025

  5. [5]

    Can sparse arrays outperform colloc ated arrays for future wireless communications?

    H. Wang and Y . Zeng, “Can sparse arrays outperform colloc ated arrays for future wireless communications?” in Proc. IEEE Globecom W orkshops, Malaysia, Dec. 2023, pp. 667–672

    work page 2023

  6. [6]

    Near-field multiu ser communications based on sparse arrays,

    K. Chen, C. Qi, G. Y . Li, and O. A. Dobre, “Near-field multiu ser communications based on sparse arrays,” IEEE J. Sel. Top. Signal Process., vol. 18, no. 4, pp. 619–632, May 2024

    work page 2024

  7. [7]

    Near-field beamfocusing, localization, and channel estimation with modular linear a rrays,

    A. Kosasih, ¨O. T. Demir, and E. Bj¨ ornson, “Near-field beamfocusing, localization, and channel estimation with modular linear a rrays,” arXiv preprint arXiv:2505.07991, 2025

    work page arXiv 2025

  8. [8]

    6G Workshop 2025 - Tdoc 6GWS-250004,

    Nokia, “6G Workshop 2025 - Tdoc 6GWS-250004,” 3GPP Techn ical Document, Mar. 2025. [Online]. Available: https://www.3g pp.org/ftp/ workshop/2025-03-10 3GPP 6G WS/Docs/6GWS-250004.zip

    work page 2025

  9. [9]

    6G Workshop 2025 - Tdoc 6GWS-250083,

    Ericsson, “6G Workshop 2025 - Tdoc 6GWS-250083,” 3GPP Te chnical Document, Mar. 2025. [Online]. Available: https://www.3g pp.org/ftp/ workshop/2025-03-10 3GPP 6G WS/Docs/6GWS-250083.zip

    work page 2025

  10. [10]

    Impact of backward crosstalk in 2 × 2 MIMO transmitters on NMSE and spec tral efficiency,

    P . H¨ andel, ¨O. T. Demir, E. Bj¨ ornson, and D. R¨ onnow, “Impact of backward crosstalk in 2 × 2 MIMO transmitters on NMSE and spec tral efficiency,” IEEE Trans. on Commun. , vol. 68, no. 7, pp. 4277–4292, 2020

    work page 2020

  11. [11]

    Hardware distortion corre- lation has negligible impact on ul massive MIMO spectral effi ciency,

    E. Bj¨ ornson, L. Sanguinetti, and J. Hoydis, “Hardware distortion corre- lation has negligible impact on ul massive MIMO spectral effi ciency,” IEEE Trans. on Commun. , vol. 67, no. 2, pp. 1085–1098, 2019

    work page 2019

  12. [12]

    F luid antenna systems,

    K.-K. Wong, A. Shojaeifard, K.-F. Tong, and Y . Zhang, “F luid antenna systems,” IEEE Trans. Wireless Commun., vol. 20, no. 3, pp. 1950–1962, 2021

    work page 1950

  13. [13]

    A tutorial on movable antennas for wireless networks,

    L. Zhu, W. Ma et al. , “A tutorial on movable antennas for wireless networks,” IEEE Commun. Surveys Tuts. , 2025, to appear

    work page 2025

  14. [14]

    Antenna sel ection in measured massive mimo channels using convex optimization,

    X. Gao, O. Edfors, J. Liu, and F. Tufvesson, “Antenna sel ection in measured massive mimo channels using convex optimization, ” in Proc. IEEE Globecom W orkshops, 2013, p. 129–134

    work page 2013

  15. [15]

    Minimizing energy consumption in MU-MIMO via antenna muti ng by neural networks with asymmetric loss,

    N. Rajapaksha, J. Mohammadi, S. Wesemann, T. Wild, and N . Rajatheva, “Minimizing energy consumption in MU-MIMO via antenna muti ng by neural networks with asymmetric loss,” IEEE Trans. on V ehic. Technol., vol. 73, no. 5, pp. 6600–6613, May 2024

    work page 2024

  16. [16]

    The Bussgang decomposition of non-linear systems: Basic theory and MIMO extensions,

    ¨O. T. Demir and E. Bj¨ ornson, “The Bussgang decomposition of non-linear systems: Basic theory and MIMO extensions,” CoRR, 2020. [Online]. Available: http://arxiv.org/abs/2005.01597

    work page arXiv 2020

  17. [17]

    How much training is need ed in multiple-antenna wireless links?

    B. Hassibi and B. M. Hochwald, “How much training is need ed in multiple-antenna wireless links?” IEEE Trans. on Inf. Theory , vol. 49, no. 4, pp. 951–963, 2003

    work page 2003

  18. [18]

    Fundamentals of energy-efficient wireless links: Optimal ratios and scalin g behaviors,

    A. Enqvist, ¨O. T. Demir, C. Cavdar, and E. Bj¨ ornson, “Fundamentals of energy-efficient wireless links: Optimal ratios and scalin g behaviors,” in 2024 IEEE 99th V ehicular Technology Conference (VTC2024-S pring). IEEE, 2024, pp. 1–6

    work page 2024