pith. sign in

arxiv: 2605.17629 · v1 · pith:AOI5K2A3new · submitted 2026-05-17 · 💻 cs.IT · math.IT

DL-Driven Optimization for ISAC System Equipped With Pinching and Movable Antennas

Nemanja Stefan Perovi\'c , Keshav Singh , Chih-Peng Li This is my paper

Pith reviewed 2026-05-19 22:16 UTC · model grok-4.3

classification 💻 cs.IT math.IT
keywords ISACpinching antennasmovable antennasdeep learning optimizationsum-rate maximizationsensing SINR thresholdantenna positioningunsupervised training
0
0 comments X

The pith

Deep learning tunes positions of pinching and movable antennas in ISAC systems to raise sum rate over fixed-antenna baselines, with larger gains at higher transmit power.

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

The paper sets out to maximize the communication sum-rate in an integrated sensing and communication system while meeting a minimum sensing quality target. It derives a closed-form expression for the best sensing receive combiner and then trains an unsupervised neural network to jointly set the locations of the pinching antennas at the transmitter, the locations of the movable antennas at the users, the communication precoders, and the sensing transmit beamformer. The training loss directly rewards the sum-rate objective and adds penalty terms that keep the antenna positions inside allowed regions. Simulations indicate that the resulting sum-rate exceeds what is achievable with fixed antennas alone and that the gap widens as the allowed transmit power increases.

Core claim

An unsupervised deep-learning network, trained on a loss that combines the sum-rate objective with soft penalties on antenna-position constraints, converges to positions for the transmit pinching antennas and user movable antennas together with communication precoding matrices and a sensing transmit beamformer that jointly deliver higher sum-rate than fixed-antenna ISAC systems; the advantage grows with maximum transmit power, while communication performance is modestly more sensitive to the sensing SINR threshold than sensing performance itself.

What carries the argument

Unsupervised deep-learning network whose loss is the sum-rate objective plus penalty terms on pinching-antenna and movable-antenna position constraints; it jointly optimizes those positions, the communication precoders, and the sensing transmit beamformer while a closed-form solution supplies the sensing receive combiner.

If this is right

  • The sum-rate advantage of pinching and movable antennas grows as maximum transmit power is increased.
  • Communication sum-rate is slightly more sensitive to the sensing SINR threshold than the sensing performance itself.
  • The closed-form sensing receive combiner is fully determined once the other variables are fixed.
  • The same unsupervised training procedure can be reused for different numbers of antennas or users provided the loss function is adjusted accordingly.

Where Pith is reading between the lines

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

  • The approach may scale to scenarios where users move continuously if the network is updated on a slower time scale than the channel coherence time.
  • Hardware imperfections such as phase noise in pinching elements could shrink the reported gains and would need separate modeling.
  • The same position-optimization idea could be applied to multi-cell ISAC deployments where inter-cell interference becomes an additional constraint.

Load-bearing premise

The unsupervised network converges to near-optimal antenna positions and beamformers when trained only with the sum-rate objective and position-penalty terms.

What would settle it

A simulation run in which the sum-rate obtained with the learned pinching and movable antenna positions remains no higher than the sum-rate of an otherwise identical fixed-antenna ISAC system at high transmit power would falsify the claimed performance advantage.

Figures

Figures reproduced from arXiv: 2605.17629 by Chih-Peng Li, Keshav Singh, Nemanja Stefan Perovi\'c.

Figure 1
Figure 1. Figure 1: The proposed ISAC system with the transmit PAs and users’ [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Diagram of the considered DL network. expressed as RC,k = log2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Convergence of the proposed neural network. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Sum-rate versus the maximum transmit power. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
read the original abstract

Integrated sensing and communication (ISAC) is considered to be a promising technology for future wireless systems due to its ability to provide communication and sensing services using shared hardware and spectrum resources. Moreover, the introduction of recently developed pinching antennas (PAs) and movable antennas (MAs) has the potential to further improve the performance gains of ISAC. Therefore, our goal is to study the optimization of the sum-rate for an ISAC system equipped with PAs and MAs, capable of satisfying minimal sensing requirements. To achieve it, we derive a closed-form solution for the optimal sensing receive combiner, and show that it is determined by other optimization variables. For these other variables (i.e., the positions of the transmit PAs, the positions of the users' MAs, the communication precoding matrices, and the sensing transmit beamformer), we propose a deep learning (DL) network that finds their optimal values. To train the network in an unsupervised manner, we formulate a loss function consisting of the objective function, as well as the penalty terms related to the constraints for the PAs and MAs positions. Simulation results show that using PAs and MAs in ISAC systems provides a larger sum-rate compared to ISAC systems with only fixed antennas, and that this performance advantage is increased with the maximum transmit power. Furthermore, we demonstrate that the communication performance of the considered system is a bit more affected by the sensing signal-to-interference-plus-noise ratio (SINR) threshold compared to the sensing performance.

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 / 3 minor

Summary. The paper considers an ISAC system equipped with pinching antennas (PAs) at the transmitter and movable antennas (MAs) at the users. It derives a closed-form expression for the optimal sensing receive combiner (shown to depend on the remaining variables) and proposes an unsupervised deep learning network to jointly optimize the transmit PA positions, user MA positions, communication precoding matrices, and sensing transmit beamformer in order to maximize the communication sum-rate subject to a minimum sensing SINR constraint. The network is trained by minimizing a composite loss that directly encodes the sum-rate objective together with penalty terms enforcing the position constraints. Simulation results are reported to show that the PA/MA configuration yields higher sum-rate than a fixed-antenna baseline, with the gap widening as maximum transmit power increases, and that communication performance is somewhat more sensitive to the sensing SINR threshold than sensing performance itself.

Significance. If the learned solutions are shown to be competitive with conventional optimization methods, the work would demonstrate a practical route to exploiting the additional spatial degrees of freedom offered by PAs and MAs inside ISAC systems. The closed-form combiner derivation supplies an analytical reduction that simplifies the remaining joint optimization, and the unsupervised loss construction (objective plus explicit penalties) is a clean way to embed constraints without requiring labeled data. These elements, together with the reported power-dependent gains, could inform the design of flexible-antenna ISAC transceivers for 6G.

major comments (2)
  1. [DL network and loss function formulation] The central performance claims rest on the unsupervised DL network producing near-optimal values for the joint PA/MA positioning and beamforming problem. However, the non-convex nature of the optimization means gradient descent on the composite loss has no global-optimality guarantee and may converge to feasible but suboptimal points. The manuscript supplies no small-scale verification (exhaustive search on discretized positions, SDR relaxation, or alternating optimization benchmark) that would quantify the optimality gap; without such evidence the reported sum-rate advantage over fixed antennas, while possibly real, cannot be confidently attributed to the full potential of the PA/MA architecture.
  2. [Simulation results] Simulation results (presumably §V) state that the performance advantage of PAs/MAs increases with maximum transmit power and that communication is “a bit more affected” by the sensing SINR threshold than sensing performance. These qualitative statements are load-bearing for the paper’s conclusions yet lack accompanying quantitative metrics, confidence intervals, or explicit figure/table references that would allow the reader to assess effect sizes and statistical reliability.
minor comments (3)
  1. [Closed-form combiner derivation] The abstract and method description mention a “closed-form solution for the optimal sensing receive combiner” but provide no derivation steps or explicit expression; including the key algebraic steps (even if relegated to an appendix) would improve verifiability.
  2. [Deep learning implementation] Network architecture details (number of layers, activation functions, input/output dimensions) and training hyperparameters (learning rate schedule, batch size, number of epochs) are not summarized; a concise table or paragraph would aid reproducibility.
  3. [System model] Notation for the position vectors of the PAs and MAs should be introduced once and used consistently; occasional re-use of symbols for different quantities risks confusion.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We respond to each major comment below and indicate the revisions planned for the manuscript.

read point-by-point responses

  1. Referee: [DL network and loss function formulation] The central performance claims rest on the unsupervised DL network producing near-optimal values for the joint PA/MA positioning and beamforming problem. However, the non-convex nature of the optimization means gradient descent on the composite loss has no global-optimality guarantee and may converge to feasible but suboptimal points. The manuscript supplies no small-scale verification (exhaustive search on discretized positions, SDR relaxation, or alternating optimization benchmark) that would quantify the optimality gap; without such evidence the reported sum-rate advantage over fixed antennas, while possibly real, cannot be confidently attributed to the full potential of the PA/MA architecture.

    Authors: We concur that gradient-based training of the unsupervised network offers no global optimality guarantee for this non-convex joint optimization. The primary goal of the work is to demonstrate that the additional spatial degrees of freedom from PAs and MAs, when optimized via the proposed DL framework, yield measurable sum-rate gains over fixed-antenna ISAC baselines under identical sensing constraints. Because antenna positions are continuous, exhaustive search over a discretized grid is computationally intractable at the problem scale considered. In the revised manuscript we will add a limited-scale comparison against an alternating-optimization benchmark (with positions discretized on a coarse grid) to quantify the observed optimality gap and thereby strengthen the attribution of performance gains to the PA/MA architecture. revision: partial

  2. Referee: [Simulation results] Simulation results (presumably §V) state that the performance advantage of PAs/MAs increases with maximum transmit power and that communication is “a bit more affected” by the sensing SINR threshold than sensing performance. These qualitative statements are load-bearing for the paper’s conclusions yet lack accompanying quantitative metrics, confidence intervals, or explicit figure/table references that would allow the reader to assess effect sizes and statistical reliability.

    Authors: We thank the referee for highlighting the need for greater quantitative precision. In the revised simulation section we will replace the qualitative descriptions with explicit numerical results (e.g., sum-rate gains in bits/s/Hz or percentage improvements at selected transmit-power values), add direct references to the corresponding figures and tables, and report variability across multiple independent training runs where feasible. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation or optimization chain

full rationale

The paper first derives a closed-form expression for the optimal sensing receive combiner that depends explicitly on the other variables. It then trains an unsupervised neural network whose loss is defined directly as the sum-rate objective plus explicit penalty terms for the position constraints on PAs and MAs. This loss is not constructed from the network outputs themselves, nor does any step rename a fitted quantity as a prediction or rely on a self-citation chain for a uniqueness claim. The reported sum-rate gains are obtained by comparing the DL-optimized configuration against fixed-antenna baselines in simulation; the comparison is external to the training procedure and does not reduce to a tautology. The approach is a standard heuristic for non-convex optimization and remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard convex optimization assumptions for the combiner derivation and on the ability of gradient-based training to solve the remaining non-convex joint optimization problem. No new physical entities are postulated.

free parameters (1)
  • Deep learning network weights and biases
    Trainable parameters of the neural network that map system parameters to antenna positions, precoding matrices, and beamformer.
axioms (1)
  • domain assumption A closed-form optimal sensing receive combiner exists and is fully determined by the remaining optimization variables.
    Invoked to reduce the search space before applying the DL network.

pith-pipeline@v0.9.0 · 5815 in / 1362 out tokens · 46565 ms · 2026-05-19T22:16:25.513434+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

19 extracted references · 19 canonical work pages

  1. [1]

    Rate region of

    Ouyang, Chongjun and Wang, Zhaolin and Liu, Yuanwei and Ding, Zhiguo , journal=. Rate region of. 2026 , publisher=

    work page 2026

  2. [2]

    IEEE Trans

    Pinching Antenna Systems for Integrated Sensing and Communications , author=. IEEE Trans. Wireless Commun. , volume=. 2026 , publisher=

    work page 2026

  3. [3]

    Integrated sensing and communications for pinching-antenna systems

    Zhang, Zheng and Wang, Zhaolin and Mu, Xidong and He, Bingtao and Chen, Jian and Liu, Yuanwei , volume=. Integrated sensing and communications for pinching-antenna systems. IEEE Commun. Lett. , year=

    work page

  4. [4]

    Multi-waveguide pinching antennas for

    Mao, Weihao and Lu, Yang and Xu, Yanqing and Ai, Bo and Dobre, Octavia A and Niyato, Dusit , journal=. Multi-waveguide pinching antennas for. 2025 , publisher=

    work page 2025

  5. [5]

    IEEE Trans

    Movable antenna enhanced networked integrated sensing and communication system , author=. IEEE Trans. Wireless Commun. , volume=. 2025 , publisher=

    work page 2025

  6. [6]

    IEEE Trans

    Movable antenna enhanced wireless sensing via antenna position optimization , author=. IEEE Trans. Wireless Commun. , volume=. 2024 , publisher=

    work page 2024

  7. [7]

    IEEE Trans

    Movable antenna-assisted integrated sensing and communication systems , author=. IEEE Trans. Wireless Commun. , volume=. 2025 , publisher=

    work page 2025

  8. [8]

    IEEE Trans

    Movable antenna-aided near-field integrated sensing and communication , author=. IEEE Trans. Wireless Commun. , volume=. 2025 , publisher=

    work page 2025

  9. [10]

    2025 , publisher=

    Kang, Jae-Mo and Yun, Sangseok and Kim, Il-Min , journal=. 2025 , publisher=

    work page 2025

  10. [11]

    Ouyang, Z

    C. Ouyang, Z. Wang, Y. Liu, and Z. Ding, ``Rate region of ISAC for pinching-antenna systems,'' IEEE Trans. Commun., vol. 74, pp. 5849--5866, 2026

    work page 2026

  11. [12]

    H. Li, R. Zhong, Z. Pan, C. Dong, J. Lei, and Y. Liu, ``Pinching antenna systems for integrated sensing and communications,'' IEEE Trans. Wireless Commun., vol. 25, pp. 13\,416--13\,429, 2026

    work page 2026

  12. [13]

    Zhang, Z

    Z. Zhang, Z. Wang, X. Mu, B. He, J. Chen, and Y. Liu, ``Integrated sensing and communications for pinching-antenna systems (PASS) ,'' IEEE Commun. Lett., vol. 29, no. 12, 2025

    work page 2025

  13. [14]

    W. Mao, Y. Lu, Y. Xu, B. Ai, O. A. Dobre, and D. Niyato, ``Multi-waveguide pinching antennas for ISAC ,'' IEEE Trans. Wireless Commun., vol. 25, pp. 5846--5858, 2025

    work page 2025

  14. [15]

    W. Ma, L. Zhu, and R. Zhang, ``Movable antenna enhanced wireless sensing via antenna position optimization,'' IEEE Trans. Wireless Commun., vol. 23, no. 11, pp. 16\,575--16\,589, 2024

    work page 2024

  15. [16]

    Jiang, C

    C. Jiang, C. Zhang, C. Huang, J. Ge, D. Niyato, and C. Yuen, ``Movable antenna-assisted integrated sensing and communication systems,'' IEEE Trans. Wireless Commun., vol. 24, no. 8, pp. 6397--6412, 2025

    work page 2025

  16. [17]

    J. Ding, Z. Zhou, X. Shao, B. Jiao, and R. Zhang, ``Movable antenna-aided near-field integrated sensing and communication,'' IEEE Trans. Wireless Commun., vol. 25, pp. 493--508, 2025

    work page 2025

  17. [18]

    Y. Guo, W. Chen, Q. Wu, Y. Liu, Q. Wu, K. Wang, J. Li, and L. Xu, ``Movable antenna enhanced networked integrated sensing and communication system,'' IEEE Trans. Wireless Commun., vol. 25, pp. 5555--5572, 2025

    work page 2025

  18. [19]

    J.-M. Kang, S. Yun, and I.-M. Kim, `` CaMPASS-Net : A deep learning framework on capacity maximization for MIMO pinching antenna systems in IoT ,'' IEEE Internet Things J., vol. 12, no. 21, pp. 45\,917--45\,920, 2025

    work page 2025

  19. [20]

    N. S. Perovi \'c , K. Singh, C.-P. Li, O. A. Dobre, and M. F. Flanagan, ``Optimal beamforming design for multi-user MIMO near-field ISAC systems with movable antennas,'' arXiv preprint arXiv:2512.22620, 2025

    work page arXiv 2025