pith. sign in

arxiv: 2604.09261 · v1 · submitted 2026-04-10 · 📡 eess.SP

Joint Device Pairing and Bandwidth Allocation Optimisation for Semantic Feature Multiple Access Networks

Jiaxiang Wang , Zhaohui Yang , Mingzhe Chen , Mohammad Shikh-Bahaei This is my paper

Pith reviewed 2026-05-10 16:55 UTC · model grok-4.3

classification 📡 eess.SP
keywords semantic communicationmultiple accessuser pairingbandwidth allocationcross-user attentionsemantic distortionSwinJSCCminimum-weight perfect matching
0
0 comments X

The pith

SFMA superimposes semantic features for paired users and jointly optimizes pairing with bandwidth allocation to reduce overall distortion under latency and energy limits.

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

This paper introduces a Semantic Feature Multiple Access framework that extends single-user semantic codecs to allow two users to share the same time-frequency resources through superimposed transmissions. A Cross-User Attention module lets paired devices exchange semantic features by exploiting image similarity while controlling interference. The authors decompose the resulting mixed-integer resource problem into a matching step for pairing and a convex feasibility check for bandwidth, solved via a polynomial-time algorithm. A sympathetic reader would care because the approach promises more efficient use of scarce wireless spectrum for semantic tasks such as image reconstruction without separate resource blocks for each user.

Core claim

By extending SwinJSCC to a two-user superimposition paradigm and adding a Cross-User Attention module, SFMA enables simultaneous semantic transmission to multiple users over shared resources; the joint pairing and allocation problem is decomposed into a Minimum-Weight Perfect Matching subproblem and a convex bandwidth-allocation feasibility check whose semi-closed-form bounds come from a strictly concave rate expression, yielding a Blossom-matching plus bisection-search algorithm that reduces global semantic distortion while meeting bandwidth, latency, and energy constraints.

What carries the argument

Minimum-Weight Perfect Matching for user pairing combined with bisection search over bandwidth bounds derived from a strictly concave rate expression, enabled by the Cross-User Attention module that performs controlled feature exchange between paired users.

If this is right

  • The polynomial-time algorithm solves the originally intractable joint problem while satisfying all physical-layer constraints.
  • Reconstruction quality improves across multiple pairing modes compared with baselines that do not share resources.
  • Overall semantic distortion decreases because paired users exchange relevant features through the attention module.
  • The framework remains feasible for any pairing that admits a feasible bandwidth allocation under the concave rate bounds.

Where Pith is reading between the lines

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

  • The same superimposition-plus-matching pattern could be tested on video or point-cloud semantic streams where inter-sample similarity is also present.
  • If the attention module generalizes beyond two users, spectrum efficiency gains might compound in dense multi-user deployments.
  • The derived semi-closed-form bandwidth bounds may serve as building blocks for latency-constrained semantic scheduling in other wireless settings.

Load-bearing premise

The Cross-User Attention module can leverage inter-image similarity to exchange features and mitigate interference without introducing unmodeled performance losses, and the decomposition into matching and convex checks preserves near-optimality for the original mixed-integer problem.

What would settle it

A direct comparison on ImageNet-100 in which the proposed joint optimization produces higher or equal semantic distortion than non-paired, separate-resource transmission under identical bandwidth, latency, and energy constraints would falsify the central performance claim.

Figures

Figures reproduced from arXiv: 2604.09261 by Jiaxiang Wang, Mingzhe Chen, Mohammad Shikh-Bahaei, Zhaohui Yang.

Figure 1
Figure 1. Figure 1: System architecture of the proposed SFMA framework. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: PSNR versus SNR over AWGN channel. is optimally allocated using the closed-form solution derived from the KKT conditions in Lemma (2) (Random + KKT) [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Average MSE over various Bmax. allocate resources only to feasible and distortion-sensitive pairs. While Channel-Balanced+EqualBW and Random+KKT show moderate gains, they are still inferior due to the lack of joint pairing and resource optimisation. The superiority of our approach arises from the MWPM-based pairing, which minimizes semantic distortion while satisfying latency and energy constraints. V. CON… view at source ↗
read the original abstract

This paper presents a Semantic Feature Multiple Access (SFMA) framework for multi-user semantic communication in downlink wireless systems. By extending SwinJSCC to a two-user superimposition paradigm, SFMA enables simultaneous semantic transmission to multiple users over shared time-frequency resources. A key innovation is the Cross-User Attention (CUA) module, which facilitates controlled semantic feature exchange between paired users by leveraging inter-image similarity while mitigating interference. We formulate a joint user pairing and resource allocation problem to minimize global semantic distortion under constraints on bandwidth, end-to-end latency, and energy. This mixed-integer non-convex problem is decomposed into a Minimum-Weight Perfect Matching (MWPM) sub-problem and a convex bandwidth allocation feasibility check, with semi-closed-form bandwidth bounds derived from a strictly concave rate expression. A polynomial-time algorithm based on Blossom matching and bisection search is proposed. Extensive simulations on ImageNet-100 show that SFMA significantly improves reconstruction quality across pairing modes, and the proposed optimization effectively reduces overall distortion while satisfying physical-layer constraints.

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

2 major / 2 minor

Summary. The manuscript proposes the Semantic Feature Multiple Access (SFMA) framework extending SwinJSCC to a two-user superimposition paradigm for simultaneous semantic transmission over shared resources in downlink systems. It introduces a Cross-User Attention (CUA) module for controlled feature exchange between paired users leveraging inter-image similarity. The central contribution is a decomposition of the joint user-pairing and bandwidth-allocation problem (minimizing global semantic distortion subject to bandwidth, latency, and energy constraints) into a Minimum-Weight Perfect Matching (MWPM) subproblem solved by the Blossom algorithm and a per-pair convex feasibility check solved by bisection on semi-closed-form bandwidth bounds derived from a strictly concave rate expression. Polynomial-time implementation and ImageNet-100 simulations are reported to show reduced reconstruction distortion across pairing modes while satisfying physical-layer constraints.

Significance. If the MWPM decomposition with precomputed weights is shown to preserve near-optimality for the global objective and the CUA module is validated to mitigate interference without unmodeled losses, the work would offer a practical polynomial-time method for multi-user semantic communications that improves resource efficiency. The use of strictly concave rate properties for semi-closed-form bounds and the explicit handling of superimposition are positive technical elements, but the absence of suboptimality analysis or small-instance exhaustive-search validation limits the strength of the claimed performance gains.

major comments (2)
  1. [optimization formulation and algorithm description (abstract and §4)] The decomposition into MWPM (using a weight matrix) followed by per-pair bisection on semi-closed-form bounds assumes that pair-specific semantic distortion can be captured by weights computed independently of the final bandwidth allocation. However, because CUA-enabled feature exchange and residual interference after superimposition make each pair's achievable distortion a non-separable function of both the pairing choice and the exact bandwidth split (subject to latency/energy constraints), an MWPM weight that ignores the post-allocation rate-distortion curve can select pairings that are suboptimal once bisection is executed. No suboptimality bound or exhaustive-search validation on small instances is provided to support that the separated procedure solves the original mixed-integer non-convex problem.
  2. [rate expression and bandwidth bounds (abstract and §3)] The claim that the proposed optimization 'effectively reduces overall semantic distortion' rests on the MWPM + bisection procedure, yet the abstract and description provide no derivation details, error analysis, or verification that the pairing subproblem and feasibility check together solve the original problem. The strictly concave rate expression is invoked for bounds, but without showing how the CUA module and superimposition interference are incorporated into the rate-distortion mapping, the support for the central claim remains incomplete.
minor comments (2)
  1. [simulation results] The abstract states that the CUA module 'facilitates controlled semantic feature exchange ... while mitigating interference,' but no ablation study isolating the contribution of CUA versus standard attention is mentioned; adding such a study would strengthen the empirical section.
  2. [problem formulation] Notation for the global minimization objective and any weighting parameters should be shown to be independent of the simulation data or prior fitted models to address potential circularity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments on our manuscript. We address each major comment point by point below, providing clarifications on the decomposition and derivations while remaining faithful to the current manuscript content. We will incorporate additional details and discussion in the revised version where this strengthens the presentation without altering the core claims.

read point-by-point responses

  1. Referee: [optimization formulation and algorithm description (abstract and §4)] The decomposition into MWPM (using a weight matrix) followed by per-pair bisection on semi-closed-form bounds assumes that pair-specific semantic distortion can be captured by weights computed independently of the final bandwidth allocation. However, because CUA-enabled feature exchange and residual interference after superimposition make each pair's achievable distortion a non-separable function of both the pairing choice and the exact bandwidth split (subject to latency/energy constraints), an MWPM weight that ignores the post-allocation rate-distortion curve can select pairings that are suboptimal once bisection is executed. No suboptimality bound or exhaustive-search validation on small instances is provided to support that the separated procedure solves the original mixed-integer non-convex problem.

    Authors: The concern about non-separability is valid in general for such joint problems. In the manuscript, the MWPM weights are explicitly computed as the minimal per-pair semantic distortion obtained by first solving the convex feasibility check (bisection on the semi-closed-form bandwidth bounds) for that specific pair, incorporating the CUA module's feature exchange (via inter-image similarity) and the residual interference model from superimposition. This ensures the weight reflects the best achievable distortion under the full set of constraints for that pairing. The global MWPM then selects the matching minimizing the sum of these values, after which bisection confirms feasibility for the chosen pairs. We acknowledge that this yields a high-quality but not necessarily globally optimal solution to the original mixed-integer non-convex problem, and no theoretical suboptimality bound is derived in the current version. Simulations on ImageNet-100 demonstrate consistent gains over alternative pairings. We will add a dedicated discussion of the decomposition's approximation properties and include exhaustive-search validation for small user counts (e.g., 4-6 users) as supplementary material in the revision. revision: partial

  2. Referee: [rate expression and bandwidth bounds (abstract and §3)] The claim that the proposed optimization 'effectively reduces overall semantic distortion' rests on the MWPM + bisection procedure, yet the abstract and description provide no derivation details, error analysis, or verification that the pairing subproblem and feasibility check together solve the original problem. The strictly concave rate expression is invoked for bounds, but without showing how the CUA module and superimposition interference are incorporated into the rate-distortion mapping, the support for the central claim remains incomplete.

    Authors: The abstract and §3 summarize the approach due to space constraints, but the full rate expression and bounds derivation appear in §3 and the supplementary material. The strictly concave rate is derived from the effective SNR after superimposition interference (modeled as additive noise scaled by power allocation), combined with the semantic rate-distortion function of the extended SwinJSCC. The CUA module is incorporated by modulating the effective feature quality (and thus the distortion-rate parameters) according to the cross-user similarity metric, which reduces the required rate for a target distortion when similarity is high. The semi-closed-form bandwidth bounds follow from inverting the concave rate function subject to latency and energy constraints, enabling the bisection feasibility check. We agree that expanded derivation details, explicit incorporation steps for CUA/interference, and bisection error bounds would improve clarity. These will be added to §3 in the revision, along with a verification that the combined procedure satisfies the original constraints. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation uses external rate properties and standard algorithms

full rationale

The paper formulates a joint pairing-allocation problem and decomposes it into MWPM (using Blossom) plus per-pair convex feasibility via bisection on semi-closed-form bounds derived from a strictly concave rate expression. These rate bounds and the matching algorithm are independent of the semantic distortion objective and are not fitted to the target data. The CUA module and SwinJSCC extension are presented as modeling choices validated by simulation on ImageNet-100, without the objective or weights reducing to the simulation outputs by construction. No self-citation chain, self-definitional loop, or fitted-input-renamed-as-prediction is exhibited in the derivation steps.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

Only the abstract is available, so the ledger is necessarily incomplete; the paper appears to introduce the CUA module and SFMA framework as new constructs while relying on standard properties of wireless rate functions.

axioms (1)
  • domain assumption The rate expression is strictly concave
    Invoked to obtain semi-closed-form bandwidth bounds in the feasibility check sub-problem.
invented entities (2)
  • Cross-User Attention (CUA) module no independent evidence
    purpose: To enable controlled semantic feature exchange between paired users by leveraging inter-image similarity while mitigating interference
    Presented as the key innovation extending SwinJSCC to the two-user superimposition paradigm.
  • Semantic Feature Multiple Access (SFMA) framework no independent evidence
    purpose: To support simultaneous semantic transmission to multiple users over shared time-frequency resources
    The overarching proposed system for multi-user semantic communication.

pith-pipeline@v0.9.0 · 5490 in / 1545 out tokens · 86387 ms · 2026-05-10T16:55:34.773813+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

14 extracted references · 14 canonical work pages

  1. [1]

    A vision of 6g wireless sy stems: Applications, trends, technologies, and open research pro blems,

    W. Saad, M. Bennis, and M. Chen, “A vision of 6g wireless sy stems: Applications, trends, technologies, and open research pro blems,” IEEE network, vol. 34, no. 3, pp. 134–142, 2019

    work page 2019

  2. [2]

    Performance optimization for semantic communications: A n attention- based reinforcement learning approach,

    Y . Wang, M. Chen, T. Luo, W. Saad, D. Niyato, H. V . Poor, and S. Cui, “Performance optimization for semantic communications: A n attention- based reinforcement learning approach,” IEEE Journal on Selected Areas in Communications , vol. 40, no. 9, pp. 2598–2613, 2022

    work page 2022

  3. [3]

    On p rivacy, security, and trustworthiness in distributed wireless lar ge ai models,

    Z. Y ang, W. Xu, L. Liang, Y . Cui, Z. Qin, and M. Debbah, “On p rivacy, security, and trustworthiness in distributed wireless lar ge ai models,” Science China Information Sciences , vol. 68, no. 7, p. 170301, 2025

    work page 2025

  4. [4]

    Energy efficient semantic communication over wireless networks with rate splitting,

    Z. Y ang, M. Chen, Z. Zhang, and C. Huang, “Energy efficient semantic communication over wireless networks with rate splitting,” IEEE Journal on Selected Areas in Communications , vol. 41, no. 5, pp. 1484–1495, 2023

    work page 2023

  5. [5]

    Optimizing model splitting and device task assignment for deceptive si gnal assisted private multi-hop split learning,

    D. Wei, X. Xu, Y . Liu, H. V . Poor, and M. Chen, “Optimizing model splitting and device task assignment for deceptive si gnal assisted private multi-hop split learning,” IEEE journal on selected areas in communications, 2025

    work page 2025

  6. [6]

    Trans former based collaborative reinforcement learning for fluid anten na system (fas)-enabled 3d uav positioning,

    X. Xu, H. Xu, D. Wei, W. Saad, M. Bennis, and M. Chen, “Trans former based collaborative reinforcement learning for fluid anten na system (fas)-enabled 3d uav positioning,” IEEE Journal on Selected Areas in Communications, 2025

    work page 2025

  7. [7]

    Deep joi nt source- channel coding for wireless image transmission,

    E. Bourtsoulatze, D. B. Kurka, and D. G¨ und¨ uz, “Deep joi nt source- channel coding for wireless image transmission,” IEEE Transactions on Cognitive Communications and Networking , vol. 5, no. 3, pp. 567–579, 2019

    work page 2019

  8. [8]

    Non-o rthogonal multiple access enhanced multi-user semantic communicati on,

    W. Li, H. Liang, C. Dong, X. Xu, P . Zhang, and K. Liu, “Non-o rthogonal multiple access enhanced multi-user semantic communicati on,” IEEE Transactions on Cognitive Communications and Networking , vol. 9, no. 6, pp. 1438–1453, 2023

    work page 2023

  9. [9]

    Compression ratio allocation for probabilis tic semantic communication with rsma,

    Z. Zhao, Z. Y ang, Y . Hu, C. Zhu, M. Shikh-Bahaei, W. Xu, Z. Z hang, and K. Huang, “Compression ratio allocation for probabilis tic semantic communication with rsma,” IEEE Transactions on Communications , 2025

    work page 2025

  10. [10]

    Semantic feature multiple access (sfma) over wireless net works,

    J. Wang, Z. Y ang, C. Huang, Z. Zhang, M. Shikh-Bahaei, an d M. Chen, “Semantic feature multiple access (sfma) over wireless net works,” in IEEE INFOCOM 2025-IEEE Conference on Computer Communicati ons W orkshops (INFOCOM WKSHPS), pp. 1–6, IEEE, 2025

    work page 2025

  11. [11]

    Deep learning based superpos ition coded modulation for hierarchical semantic communications over broadcast channels,

    Y . Bo, S. Shao, and M. Tao, “Deep learning based superpos ition coded modulation for hierarchical semantic communications over broadcast channels,” IEEE Transactions on Communications , 2024

    work page 2024

  12. [12]

    Generative ai empowered semantic featur e multiple access (sfma) over wireless networks,

    J. Wang, Y . Y ang, Z. Y ang, C. Huang, M. Chen, Z. Zhang, and M. Shikh-Bahaei, “Generative ai empowered semantic featur e multiple access (sfma) over wireless networks,” IEEE Transactions on Cognitive Communications and Networking , 2025

    work page 2025

  13. [13]

    Swin transformer: Hierarchical vision transforme r using shifted windows,

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, Z. Zhang, S. Lin, and B. Guo, “Swin transformer: Hierarchical vision transforme r using shifted windows,” in Proceedings of the IEEE/CVF international conference on computer vision , pp. 10012–10022, 2021

    work page 2021

  14. [14]

    Blossom v: a new implementation of a min imum cost perfect matching algorithm,

    V . Kolmogorov, “Blossom v: a new implementation of a min imum cost perfect matching algorithm,” Mathematical Programming Computation , vol. 1, no. 1, pp. 43–67, 2009

    work page 2009