Pith. sign in

REVIEW 4 major objections 6 minor 34 references

RIS Codebook Index Assignment under Imperfect Control Links Using TSP-Inspired Optimization

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The shortest Hamiltonian path through a RIS codebook, labeled with Gray codes, is the ordering that minimizes average SNR loss under single-bit feedback errors.

desk verdict The TSP-reduction claim is mathematically wrong, but the three-phase heuristic and benchmark study are real and the paper as an engineering artifact could be salvaged with a corrected objective. read the letter →

arxiv 2507.18727 v1 pith:PT7MLTJ4 submitted 2025-07-24 cs.IT eess.SPmath.IT

Liangshun Wu , Wen Chen , Qingqing Wu , Xudong Bai , Kunlun Wang This is my paper · ORCID
classification cs.ITeess.SPmath.IT MSC 90C2705C45
keywords reconfigurableintelligentsurfacecodebookindexassignmenttravelingsalesmanproblemfeedbackerrorresilienceGraycodeSNRlossminimizationcombinatorialoptimizationbinarysymmetricchannel
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Reconfigurable intelligent surfaces (RIS) are tuned by sending a short binary index over a wireless control link, and a single flipped bit can make the surface apply the wrong phase pattern and degrade the received signal. This paper claims that the damage from such one-bit feedback errors can be minimized by reordering the codebook: treat each codeword as a city, use the relative SNR loss caused by swapping two codewords as the distance, and find the shortest Hamiltonian path through all codewords. Assigning Gray-code labels along that path makes the most probable error land on a neighboring, minimally damaging configuration. The paper also contributes a three-phase heuristic that finds near-optimal orderings in polynomial time, scaling to codebooks of 1024 codewords, so the protection is achievable without changing the RIS hardware.

What carries the argument

The carrying object is a complete graph on the $K$ RIS codewords, with edge weight $d(c_i,c_j)=|1-\mathrm{SNR}_j/\mathrm{SNR}_i|$ measuring the relative SNR loss from applying $c_j$ in place of $c_i$. The optimization target is the Hamiltonian path $\pi^*$ minimizing $L(\pi)$, and a Gray code — a binary labeling in which consecutive labels differ in exactly one bit — is the indexing rule that turns path adjacency into one-bit index adjacency. The solver that makes the approach practical has three phases: provision (builds layered nearest-neighbor candidate sets from the loss distribution), shotgun (samples $3K^2$ random routes with selection probability $1/(1+(d/(\mu\bar d))^2)$ and keeps the best $3K$), and fuzzy concatenation (reinforces frequently used edges while adaptively decaying greediness). The worst-case complexity is $O(K^4)$.

What would settle it

On a fixed RIS codebook and channel realization (e.g., $K=256$, $N=256$, $b=8$), compute the expected SNR loss in eq. (7) under the TSP-path Gray-code assignment and under an assignment that directly minimizes the all-Hamming-neighbor sum in eq. (8); if the direct minimizer achieves lower loss, the claim that the Hamiltonian path is the optimal single-bit-error index assignment is falsified.

Watch

Extended reading notes

Core claim

The central claim is that RIS codebook index assignment under a binary symmetric feedback channel reduces to a Traveling Salesman Problem: with equal priors and single-bit errors dominating, the expected SNR loss is proportional to $\sum_i \sum_{j:\mathrm{Ham}(i,j)=1} d(c_i,c_j)$, and the paper argues that minimizing the Hamiltonian path cost $L(\pi)=\sum_{k=0}^{K-2} d(c_{\pi(k)}, c_{\pi(k+1)})$ and labeling the path with consecutive Gray codes achieves this robustness. The optimal permutation $\pi^*$ is the one minimizing $L(\pi)$, found by a three-phase solver whose worst-case complexity is $O(K^4)$. In simulations the solver matches the SNR loss of strong exact and heuristic TSP solvers while running substantially faster on large codebooks, and it retains its advantage on clustered and exploded edge-weight distributions.

Load-bearing premise

The load-bearing premise is that a Gray-code labeling of a Hamiltonian path makes every single-bit index error fall on a path-adjacent codeword, even though each binary index has several one-bit neighbors and only two of them sit next to it on the path.

Editorial extensions

If this is right

  • Existing RIS codebooks can be reordered in place, with no new phase hardware or feedback-channel changes, to reduce SNR loss from the dominant single-bit error events.
  • The $O(K^4)$ heuristic completes a 1024-codeword reordering in about two minutes, making large-codebook robustness practical.
  • The solver keeps its advantage on clustered and exploded edge-weight distributions, where generic fast heuristics lose 20% or more in SNR compared with the best orderings.
  • If feedback error rates are not tiny, multi-bit errors become non-negligible and the paper's own limitation note says the TSP simplification should give way to the full quadratic assignment formulation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The path objective in eq. (9) controls only the two path neighbors of each index, while the true single-bit objective in eq. (8) sums over all $\log_2 K$ Hamming neighbors; an ordering that directly minimizes eq. (8) would be a strictly stronger defense and is a natural follow-up the paper leaves open.
  • The same TSP-plus-Gray-code machinery should transfer to other finite codebooks whose entries have a quantifiable confusion cost, such as beamforming or quantization codebooks, not just RIS phase configurations.
  • Under non-uniform codeword priors, the equal-prior simplification in eq. (7) breaks, and the optimal ordering would need to weight pair losses by the prior probabilities of the intended codeword.
  • A direct comparison between the proposed solver and generic TSP solvers on 'exploded' edge-weight distributions suggests that neural solvers trained on uniform Euclidean TSP instances may be systematically worse on quantized RIS codebooks, a hypothesis testable with the paper's published setup.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper studies index assignment for RIS codebooks under imperfect feedback links. The authors propose to assign binary indices to RIS codewords so that single-bit index errors cause minimal SNR degradation. They formulate the problem as a shortest Hamiltonian path (TSP) over codewords, with edge weights given by relative SNR loss, and then label the codewords sequentially along the path using a Gray code. A three-phase heuristic (provision, shotgun, fuzzy concatenation) is proposed to solve the resulting TSP, and simulations compare the achieved average SNR loss against natural indexing, random indexing, and several TSP solvers.

Significance. The practical problem is relevant: feedback errors on RIS control links are rarely modeled, and a principled index assignment could improve robustness. The paper also contains a substantial amount of benchmarking against standard TSP solvers (LKH3, Concorde, GLOP, POMO, ELG, 2-opt, etc.), with runtime and scaling results. However, the central theoretical contribution—the equivalence between the single-bit-error objective and a shortest Hamiltonian path—is mathematically incorrect. Because this equivalence is the foundation for the problem formulation, the proposed optimality claim, and the interpretation of all simulation results, the paper's main advertised result is not supported. The heuristic may still be useful as an engineering heuristic, but the manuscript does not provide a valid basis for claiming near-optimal robustness under single-bit errors.

major comments (4)
  1. [Section II.B, Eqs. (8)-(9)] The reduction from the true single-bit-error objective to a shortest Hamiltonian path is invalid. In Eq. (8), the objective sums over all index pairs with Hamming distance 1; with log2(K)-bit indices, each index has log2(K) Hamming neighbors, so the sum contains K log2(K) directed terms. In contrast, the Hamiltonian path objective Eq. (9) sums only K-1 path edges. Under a sequential Gray-code labeling, only the two path-adjacent labels (one at each endpoint) differ by one bit from a given interior label. For K=256, Eq. (9) controls at most 2/log2(256)=25% of the single-bit-error events, and for K=16 it controls at most 2/4=50%. The paper's claim that 'any single-bit index error leads to jumping to an adjacent codeword on this path' is therefore false.
  2. [Section II.B, Table II example] The K=16 example in Table II concretely demonstrates the failure of the claimed equivalence. In the listed Gray sequence, the codeword labeled 0000 (position 0) has Hamming neighbors 0001 (position 1), 0010 (position 3), 0100 (position 7), and 1000 (position 15). Only position 1 is adjacent on the TSP path; the other three Hamming-neighbor pairs are not constrained by the path cost in Eq. (9). Thus the path objective does not control the majority of single-bit confusions even in the paper's own illustrative example.
  3. [Section IV, simulation methodology] The simulations do not close the gap between the optimized objective and the evaluated metric. Step 5 of the simulation procedure evaluates average SNR loss over all codeword pairs whose indices differ by one bit, i.e., the objective in Eq. (8). The proposed solver, however, minimizes only the path cost in Eq. (9). Consequently, Figs. 4-10 and Tables VI-VII report the heuristic's performance on a metric that the algorithm does not directly optimize, and the comparisons against other TSP solvers compare solutions to an objective that is not equivalent to the stated problem. The reported 'near-optimal' behavior therefore does not validate the advertised TSP equivalence.
  4. [Appendix B] Appendix B repeats the unsupported assertion that 'Hamming-distance-1 neighbors correspond exactly to adjacent nodes on the TSP path.' This is the same incorrect statement as in Section II.B. A Gray-code sequence of length K has many non-adjacent pairs at Hamming distance 1; only the path-neighbor relation is controlled by Eq. (9). Since this appendix is used to justify the entire simulation model, the invalid assertion is load-bearing and not a minor presentation issue.
minor comments (6)
  1. [Title] The title contains a typo: 'Control Link s' should read 'Control Links'.
  2. [Section II.B] The text says the codewords can be 'labeled sequentially or via Gray code.' Sequential binary labeling does not guarantee that adjacent path positions differ by one bit; only a Gray-code labeling does. This ambiguity should be clarified.
  3. [Section IV] There are two experiments labeled 'Experiment III' (the ablation study and the comparison with other TSP solvers). The second should be renumbered.
  4. [Section IV.D] The sentence 'fix the number of codebooks at K = 256' should read 'fix the number of codewords at K = 256'.
  5. [Section II.A / Eq. (6)] The loss d(ci,cj) defined in Eq. (6) is not symmetric, yet the algorithm accumulates unordered pair counts (Algorithm 1, step 22) and treats the path as undirected. The manuscript should state explicitly whether the edge weights are symmetrized for the TSP formulation.
  6. [Table III] The many heuristic parameters (l1, l2, l3, nshot, kshot, ncate, kcate, µ0, σ, µmin, z, kmin, T) are given without derivation or sensitivity analysis. This does not affect the central mathematical error, but it limits reproducibility and makes the claimed scalability results less informative.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the SNR-loss objective and benchmarks are independent; the TSP reduction is mathematically questionable but not circular.

full rationale

The paper's central derivation is not circular. The SNR-loss metric d(c_i,c_j) in Eq. (6) is defined from the physical channel model, and the true objective Eq. (8) is derived from BSC single-bit error statistics, independent of the solver or its outputs. The proposed path objective Eq. (9) is a separate quantity; although the paper asserts equivalence ('Hamming-distance-1 neighbors correspond exactly to adjacent nodes on the TSP path', Appendix B), this is a correctness/validity gap rather than a self-referential reduction: for K=16 a 4-bit Gray code has non-adjacent Hamming-neighbor pairs (e.g., 0000 and 0010), so Eq. (8) contains terms not controlled by Eq. (9). The three-phase heuristic hyperparameters (Table III) are hand-set and not fitted to the reported SNR losses, and the simulations are benchmarked against external solvers (LKH3, Concorde, GLOP, POMO, ELG), so the claimed gains are not fitted inputs renamed as predictions. No load-bearing self-citation chain is used; the TSP mapping cites prior external work [27]. Consequently, the paper has no significant circularity.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical entities. Its free parameters are the many heuristic hyperparameters in Table III, plus the informal Type I/II distribution classification. The most consequential axiom is the asserted equivalence between the true objective (8) and the path objective (9), which is invalid and constitutes the main correctness risk. The asymmetry of the loss metric is another unflagged modeling assumption.

free parameters (7)
  • Provision phase layer sizes (l1, l2, l3) = sqrt(K), 2 sqrt(K), K/3
    Hand-set in Table III to control candidate neighborhoods.
  • Exhaustive tail threshold f = 4
    Hand-set in Table III.
  • Shotgun phase counts (nshot, kshot) = 3K^2, 3K
    Hand-set in Table III.
  • Fuzzy concatenation counts (ncate, initial kcate, kmin) = 200K, 4K, 200
    Hand-set in Table III.
  • Perturbation coefficient and decay (mu0, sigma, mu_min) = 0.5, 0.01, 0.15
    Hand-set in Table III; controls greediness of sampling.
  • kcate decay rate z and max iterations T = K/20, sqrt(K)
    Hand-set in Table III.
  • Type I/II distribution classification = unimodal vs multimodal, no quantitative test
    The decision to use layered candidate sets depends on a subjective classification of the pairwise-loss distribution; no statistical test is specified.
assumptions (4)
  • domain assumption Feedback channel is a binary symmetric channel with small bit error rate q, so only single-bit errors matter.
    Used in Section II.B to simplify the objective; valid only for very small q, and the paper acknowledges multi-bit errors are neglected.
  • domain assumption All codewords are equally likely to be selected (uniform prior on i).
    Stated in Section II.B before eq. (7); simplifies Pi = 1/K.
  • ad hoc to paper The pairwise loss d(ci,cj) can be treated as a symmetric edge weight in the TSP.
    Eq. (6) defines |1 - SNR_j/SNR_i|, which is asymmetric; the TSP formulation and eq. (9) assume symmetry without comment.
  • ad hoc to paper Equation (8) is equivalent to minimizing the Hamiltonian path length in eq. (9) when labels are assigned by Gray code.
    This equivalence is asserted in Section II.B but is false because a binary index has log2(K) Hamming neighbors, not just the two path-adjacent positions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RIS Codebook Index Assignment under Imperfect Control Links Using TSP-Inspired Optimization." pith.science (2026). https://pith.science/paper/PT7MLTJ4

@misc{pith2026250718727,
  author       = {Pith},
  title        = {Pith review of: RIS Codebook Index Assignment under Imperfect Control Links Using TSP-Inspired Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PT7MLTJ4}},
  note         = {Machine review of arXiv:2507.18727}
}
read the original abstract

Reconfigurable Intelligent Surfaces (RIS) promise transformative gains in wireless communications by enabling programmable control of the propagation environment through discrete phase configurations. In practical deployments, the control of RIS phase states is typically managed using finite codebooks, with configuration indices transmitted over low latency, yet imperfect, wireless feedback channels. Even rare feedback bit errors can lead to significant mismatches between intended and applied RIS states, degrading system performance. This paper addresses the challenge of robust RIS codebook index assignment by formulating it as a combinatorial optimization problem, equivalent to the Traveling Salesman Problem (TSP), where codewords are "cities" and edge weights reflect SNR degradation under codeword confusion. A novel three-phase heuristic algorithm is proposed to solve this, consisting of a provision phase, a shotgun phase, and a fuzzy concatenation phase. Simulation results show that the method outperforms conventional indexing strategies and achieves near-optimal robustness to index errors, while also being scalable and hardwareagnostic for real time deployment. Future work includes multiple bits error correction and online adaptive mapping for time varying channels.

Figures

Figures reproduced from arXiv: 2507.18727 by the authors.

Figure 1
Figure 1. RIS-assisted downlink MISO communication scenario [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. RIS codebook quantization and IA to TSP formulation. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The proposed TSP solver. III. THE PROPOSED ALGORITHM To efficiently find a near-optimal index permutation for the RIS codebook, we propose a novel TSP solver [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Average relative SNR loss versus BSC SNR for three [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Relative SNR loss under varying system parameters. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 7
Figure 7. Figure 7: Average SNR loss of different solvers as a function of [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Average SNR loss of different solvers as a function of [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Comparison of relative SNR loss for different solver [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Execution time of different solvers as a function of [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: Visualization of codeword-pair (edge weight) dist [PITH_FULL_IMAGE:figures/full_fig_p011_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 32 canonical work pages

  1. [27]

    Codebook index ass ignment by an approximate solution of the traveling salesman problem,

    A. Spira, R. Mayrench, and D. Malah, “Codebook index ass ignment by an approximate solution of the traveling salesman problem, ” in 2000 10th European Signal Processing Conference . IEEE, 2000, pp. 1–4

    work page 2000

  2. [1]

    Effici ent joint precoding design for wideband intelligent reflecting surfa ce-assisted cell- free network,

    Y . Wang, J. Jiang, X. Du, Z. Lian, Q. Wu, and W. Chen, “Effici ent joint precoding design for wideband intelligent reflecting surfa ce-assisted cell- free network,” IEEE Transactions on Communications , 2025

    work page 2025

  3. [2]

    Spatial multiplexin g oriented channel reconfiguration in multi-irs aided mimo systems,

    Y . Chen, Q. Wu, G. Chen, and W. Chen, “Spatial multiplexin g oriented channel reconfiguration in multi-irs aided mimo systems,” IEEE Trans- actions on V ehicular Technology, 2025

    work page 2025

  4. [3]

    Intel- ligent reflecting surface aided mimo networks: Distributed or centralized architecture?

    G. Chen, Q. Wu, W. Chen, Y . Hou, M. Jian, S. Zhang, and J. Li, “Intel- ligent reflecting surface aided mimo networks: Distributed or centralized architecture?” IEEE Transactions on Wireless Communications , 2024

    work page 2024

  5. [4]

    Irs -aided overloaded multi-antenna systems: Joint user grouping and resource allocation,

    Y . Gao, Q. Wu, W. Chen, Y . Liu, M. Li, and D. B. da Costa, “Irs -aided overloaded multi-antenna systems: Joint user grouping and resource allocation,” IEEE Transactions on Wireless Communications , vol. 23, no. 8, pp. 8297–8313, 2024

    work page 2024

  6. [5]

    Exploiting intelligent reflecting surfaces for interfere nce channels with swipt,

    Y . Gao, Q. Wu, W. Chen, C. Wu, D. W. K. Ng, and N. Al-Dhahir, “Exploiting intelligent reflecting surfaces for interfere nce channels with swipt,” IEEE Transactions on Wireless Communications , vol. 23, no. 5, pp. 4442–4458, 2023

    work page 2023

  7. [6]

    Fairn ess optimization for intelligent reflecting surface aided upli nk rate-splitting multiple access,

    S. Zhang, W. Chen, Q. Wu, Z. Liu, S. Zhang, and J. Li, “Fairn ess optimization for intelligent reflecting surface aided upli nk rate-splitting multiple access,” IEEE Transactions on Communications , 2024

    work page 2024

  8. [7]

    Intelligent reflecting surface emp owered self- interference cancellation in full-duplex systems,

    C. Qiu, Q. Wu, M. Hua, W. Chen, S. Ma, F. Hou, D. W. K. Ng, and A. L. Swindlehurst, “Intelligent reflecting surface emp owered self- interference cancellation in full-duplex systems,” IEEE Transactions on Communications, vol. 72, no. 5, pp. 2945–2958, 2023

    work page 2023

Show all 34 references
  1. [8]

    Multi-functional beamforming design for integrated sens ing, communi- cation, and computation,

    Y . Zhao, Q. Wu, W. Chen, Y . Zeng, R. Liu, W. Mei, F. Hou, and S . Ma, “Multi-functional beamforming design for integrated sens ing, communi- cation, and computation,” IEEE Transactions on Communications, 2024

    2024

  2. [9]

    Intell igent omni surfaces assisted integrated multi-target sensing and mul ti-user mimo communications,

    Z. Zhang, W. Chen, Q. Wu, Z. Li, X. Zhu, and J. Y uan, “Intell igent omni surfaces assisted integrated multi-target sensing and mul ti-user mimo communications,” IEEE Transactions on Communications , 2024

    2024

  3. [10]

    Intellig ent surface empowered integrated sensing and communication: From coex istence to reciprocity,

    K. Meng, Q. Wu, C. Masouros, W. Chen, and D. Li, “Intellig ent surface empowered integrated sensing and communication: From coex istence to reciprocity,” IEEE Wireless Communications , vol. 31, no. 5, pp. 84–91, 2024

    2024

  4. [11]

    Semi-passive intelligent reflecting surface enabled sens ing systems,

    Q. Peng, Q. Wu, W. Chen, S. Ma, M.-M. Zhao, and O. A. Dobre, “Semi-passive intelligent reflecting surface enabled sens ing systems,” IEEE Transactions on Communications , 2024

    2024

  5. [12]

    Rate-splitting multiple access for transmissive reconfig urable intelligent surface transceiver empowered isac systems,

    Z. Liu, W. Chen, Q. Wu, J. Y uan, S. Zhang, Z. Li, and J. Li, “Rate-splitting multiple access for transmissive reconfig urable intelligent surface transceiver empowered isac systems,” IEEE Internet of Things Journal, 2024

    2024

  6. [13]

    Secure intelligent reflecting surface-aided integrated sensing a nd communica- tion,

    M. Hua, Q. Wu, W. Chen, O. A. Dobre, and A. L. Swindlehurst , “Secure intelligent reflecting surface-aided integrated sensing a nd communica- tion,” IEEE Transactions on Wireless Communications , vol. 23, no. 1, pp. 575–591, 2023

    2023

  7. [14]

    Reconfigurable intelligent surface aided vehicular edge c omputing: Joint phase-shift optimization and multi-user power alloc ation,

    K. Qi, Q. Wu, P . Fan, N. Cheng, W. Chen, and K. B. Letaief, “Reconfigurable intelligent surface aided vehicular edge c omputing: Joint phase-shift optimization and multi-user power alloc ation,” IEEE Internet of Things Journal , 2024

    2024

  8. [15]

    Deep-reinforcement-learning-based aoi-aware resource allocation for ris-aided iov networks,

    K. Qi, Q. Wu, P . Fan, N. Cheng, W. Chen, J. Wang, and K. B. Le taief, “Deep-reinforcement-learning-based aoi-aware resource allocation for ris-aided iov networks,” IEEE Transactions on V ehicular Technology , 2024

    2024

  9. [16]

    Sensing-assisted com munication in vehicular networks with intelligent surface,

    K. Meng, Q. Wu, W. Chen, and D. Li, “Sensing-assisted com munication in vehicular networks with intelligent surface,” IEEE Transactions on V ehicular Technology, vol. 73, no. 1, pp. 876–893, 2023

    2023

  10. [17]

    A general framewo rk for ris-aided mmwave communication networks: Channel estimation and mob ile user tracking,

    S. E. Zegrar, L. Afeef, and H. Arslan, “A general framewo rk for ris-aided mmwave communication networks: Channel estimation and mob ile user tracking,” arXiv preprint arXiv:2009.01180 , 2020

    2009 arXiv

  11. [18]

    Channel estimation for ris a ssisted wireless communications—part i: Fundamentals, solutions , and future opportunities,

    X. Wei, D. Shen, and L. Dai, “Channel estimation for ris a ssisted wireless communications—part i: Fundamentals, solutions , and future opportunities,” IEEE communications letters , vol. 25, no. 5, pp. 1398– 1402, 2021

    2021

  12. [19]

    Channel estimatio n for irs- assisted millimeter-wave mimo systems: Sparsity-inspire d approaches,

    T. Lin, X. Y u, Y . Zhu, and R. Schober, “Channel estimatio n for irs- assisted millimeter-wave mimo systems: Sparsity-inspire d approaches,” IEEE Transactions on Communications , vol. 70, no. 6, pp. 4078–4092, 2022

    2022

  13. [20]

    Channel estimation for hybrid ris aided mimo communications via atomic norm minimization,

    R. Schroeder, J. He, and M. Juntti, “Channel estimation for hybrid ris aided mimo communications via atomic norm minimization,” i n 2022 IEEE International Conference on Communications W orkshop s (ICC W orkshops). IEEE, 2022, pp. 1219–1224

    2022

  14. [21]

    Channel estimation for reco nfigurable intelligent surface aided miso communications: From lmmse to deep learning solutions,

    N. K. Kundu and M. R. McKay, “Channel estimation for reco nfigurable intelligent surface aided miso communications: From lmmse to deep learning solutions,” IEEE Open Journal of the Communications Society , vol. 2, pp. 471–487, 2021

    2021

  15. [22]

    How often channel estimation is required for adaptive irs beamf orming: A bilevel deep reinforcement learning approach,

    J. Zhang, Z. Wang, J. Li, Q. Wu, W. Chen, F. Shu, and S. Jin, “How often channel estimation is required for adaptive irs beamf orming: A bilevel deep reinforcement learning approach,” IEEE Transactions on Wireless Communications, vol. 23, no. 8, pp. 8744–8759, 2024

    2024

  16. [23]

    Reconfigurable-intelligent-surface-aided space-shift keying with imper- fect csi,

    X. Zhu, W. Chen, Q. Wu, Z. Li, J. Li, S. Zhang, and M. Ding, “Reconfigurable-intelligent-surface-aided space-shift keying with imper- fect csi,” IEEE Internet of Things Journal , vol. 11, no. 7, pp. 11 480– 11 495, 2023

    2023

  17. [24]

    Pseudo-gray coding,

    K. Zeger and A. Gersho, “Pseudo-gray coding,” IEEE Transactions on communications, vol. 38, no. 12, pp. 2147–2158, 1990

    1990

  18. [25]

    A study of vector quantization for noisy channels,

    N. Farvardin, “A study of vector quantization for noisy channels,” IEEE Transactions on Information Theory , vol. 36, no. 4, pp. 799–809, 1990

    1990

  19. [26]

    The hadamard transform- a tool for index assignment,

    P . Knagenhjelm and E. Agrell, “The hadamard transform- a tool for index assignment,” IEEE Transactions on Information Theory , vol. 42, no. 4, pp. 1139–1151, 2002

    2002

  20. [28]

    An extension of the lin-kernighan-helsg aun tsp solver for constrained traveling salesman and vehicle routing proble ms,

    K. Helsgaun, “An extension of the lin-kernighan-helsg aun tsp solver for constrained traveling salesman and vehicle routing proble ms,” Roskilde: Roskilde University, vol. 12, pp. 966–980, 2017

    2017

  21. [29]

    Concor de tsp solver,

    D. Applegate, R. Bixby, V . Chvatal, and W. Cook, “Concor de tsp solver,” 2006

    2006

  22. [30]

    H- tsp: Hierarchically solving the large-scale traveling sal esman problem,

    X. Pan, Y . Jin, Y . Ding, M. Feng, L. Zhao, L. Song, and J. Bi an, “H- tsp: Hierarchically solving the large-scale traveling sal esman problem,” in Proceedings of the AAAI Conference on Artificial Intelligen ce, vol. 37, no. 8, 2023, pp. 9345–9353

    2023

  23. [31]

    Glop: Le arning global partition and local construction for solving large- scale routing problems in real-time,

    H. Y e, J. Wang, H. Liang, Z. Cao, Y . Li, and F. Li, “Glop: Le arning global partition and local construction for solving large- scale routing problems in real-time,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 18, 2024, pp. 20 284–20 292

    2024

  24. [32]

    Pomo: Policy optimization with multiple optima for reinfo rcement learning,

    Y .-D. Kwon, J. Choo, B. Kim, I. Y oon, Y . Gwon, and S. Min, “Pomo: Policy optimization with multiple optima for reinfo rcement learning,” Advances in Neural Information Processing Systems , vol. 33, pp. 21 188–21 198, 2020

    2020

  25. [33]

    Towards gene ralizable neural solvers for vehicle routing problems via ensemble wi th trans- ferrable local policy,

    C. Gao, H. Shang, K. Xue, D. Li, and C. Qian, “Towards gene ralizable neural solvers for vehicle routing problems via ensemble wi th trans- ferrable local policy,” arXiv preprint arXiv:2308.14104 , 2023. APPENDIX A: P ROOF : T HE PROPOSED TSP SOLVER HAS POLYNOMIAL COMPLEXITY ....

    2023 arXiv

  26. [34]

    is the tail probability of the standard normal distribution. In practice, for small q (i.e., in the high BSC SNR regime), the probability of multiple simultaneous bit errors is negl igible, so only single-bit flips dominate the average loss calculati on. Hamming-distance-1 neig...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.