Pith. sign in

REVIEW 3 major objections 6 minor 38 references

A PPO agent that allocates sensor bandwidth from channel state and the Kalman prior cuts LQR control cost below sensing- and rate-oriented rules.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-10 10:15 UTC pith:YWYCFQNT

load-bearing objection Solid SC3 bandwidth-for-LQR engineering paper: PPO beats rate/sensing baselines under a coherent LQG+MI model, but the gains are untested outside that Gaussian rate map. the 3 major comments →

arxiv 2607.08272 v1 pith:YWYCFQNT submitted 2026-07-09 cs.IT cs.SYeess.SYmath.IT

Deep Reinforcement Learning-Empowered Wireless Sensor Networking for 6G Closed-Loop Controls

Chengleyang Lei , Wei Feng , Yunfei Chen , Yongxu Zhu , Ning Ge , Shi Jin This is my paper
classification cs.IT cs.SYeess.SYmath.IT
keywords bandwidth allocationedge information hublinear quadratic regulatorsensing-communication-computing-control closed loopdeep reinforcement learningproximal policy optimizationKalman filterPOMDP
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Remote robots that cannot sense a target alone need a ring of field sensors and a UAV-mounted edge hub to close a sensing–communication–computing–control loop. Communication bandwidth among those sensors is scarce, so how it is shared decides how well the hub can estimate the plant and how costly the resulting control is. This paper models the rate limit as additive Gaussian distortion whose variance falls with mutual information, runs classical Kalman filtering plus LQR control on the distorted measurements, and trains a PPO agent that sees only the channels and the filter’s prior estimate. In simulation the learned continuous allocation yields lower average infinite-horizon LQR cost than both “give more bandwidth to higher-SNR sensors” and “maximize sum rate” baselines across bandwidth, power, time, and noise sweeps. The result matters because it shows that closed-loop cost, not a communication or sensing proxy, can be the direct objective of wireless resource allocation.

Core claim

Under Kalman-filter state estimation and LQR control, a PPO policy that continuously allocates sensor-to-EIH bandwidth from instantaneous channel gains and the filter prior achieves lower average infinite-horizon LQR cost than sensing-oriented and communication-oriented baselines over the tested ranges of total bandwidth, transmit time, power, and sensing noise.

What carries the argument

The rate-to-distortion map σ²_nk,t = σ²_yk,t / (2^(2ρ D_k,t) − 1) that turns allocated bandwidth into the measurement-noise covariance of the Kalman filter, thereby shaping both the LQR cost that becomes the PPO reward and the filter prior that becomes the agent’s observation.

Load-bearing premise

Rate-limited sensing is treated as independent additive Gaussian noise whose variance is set by a fixed fraction of the transmitted mutual information; if real quantization or channel distortion is non-Gaussian or poorly matched by that map, the learned allocations no longer match the physical loop.

What would settle it

Replace the Gaussian rate-distortion model with a realistic quantizer and packet-error channel on the same plant and re-run the trained PPO policy; if its realized LQR cost no longer beats the two baselines, the central claim fails for that setting.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Closed-loop LQR cost can replace sum-rate or sensing-SNR as the design objective for multi-sensor bandwidth allocation in SC3 systems.
  • When plant states are strongly coupled, the learned policy favors better channels; when they are decoupled, it favors weaker channels or less stable open-loop modes.
  • Continuous bandwidth allocation can outperform on/off transmission scheduling under the same total resource budget once the budget is not tiny.
  • The same POMDP framing extends immediately to other continuous resource knobs (power, quantization bits) once the rate-to-distortion map is available.

Where Pith is reading between the lines

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

  • The early-termination rule used to keep training stable already encodes an implicit safety constraint; making that constraint explicit in the reward could yield policies that trade cost against instability risk.
  • Because the downlink is assumed perfect, any joint uplink–downlink allocation would have to re-weight the rate-to-distortion map by the relative packet lengths of sensing data versus control commands.
  • The observed dependence of allocation monotonicity on the plant’s coupling structure suggests a cheap pre-check: estimate the observability Gramian or the open-loop poles before deciding whether to favor strong or weak channels.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper studies sensor-to-EIH bandwidth allocation in a multi-sensor SC3 closed loop for robot control. Rate-limited uplink sensing is mapped to additive Gaussian distortion noise via a mutual-information / rate-distortion relation (Eqs. 3–5), with fixed information utilization ratio ρ. Under a separation-based policy (Kalman filter + infinite-horizon LQR), the allocation problem is cast as a POMDP whose observation is channel state plus the filter prior and error covariance, and a PPO agent is trained to minimize long-horizon LQR cost. Simulations on linear plants report lower average LQR cost than sensing-oriented (SNR-weighted) and sum-rate baselines across bandwidth, transmit time, power, and sensing-noise sweeps, with additional comparisons for K=10 and against a DRL scheduling baseline, plus qualitative analysis of how system coupling shapes allocation.

Significance. If the modeling chain from bandwidth to innovation covariance is accepted, the work is a useful step beyond communication- or sensing-only resource allocation for SC3 loops: it ties continuous multi-sensor bandwidth to closed-loop LQR cost under standard LQG control, adapts PPO with a multi-path feature extractor and instability early-stop, and shows that sum-rate maximization can be suboptimal for control (Fig. 12). The dynamics-dependent allocation findings (Figs. 15–16) are of independent interest. The contribution is primarily methodological and simulation-based rather than a new information-theoretic or control-theoretic theorem; its impact hinges on whether the Gaussian rate map and linear-plant results transfer to more realistic sensing and quantization.

major comments (3)
  1. §II-A, Eqs. (3)–(5) and the simulation setting ρ=0.005: the entire path from action B_t to Kalman gain (24), posterior covariance (23), control (27), and reward (34) is mediated by independent AWGN distortion with variance σ²_yk,t/(2^{2ρ D_k,t}−1). The paper acknowledges non-Gaussian quantization as a limitation in §V but provides no ablation that replaces (5) with an explicit quantizer, bit-level codec, or empirical rate–MSE curve while keeping the same plant and baselines. Without that check, the reported superiority in Figs. 6–9 and 13 is conditional on an untested abstraction that sits on the critical path from bandwidth to LQR cost. A minimal revision would re-run at least one main sweep under a concrete quantization model (or a non-Gaussian distortion family) and report whether the ranking versus sensing- and rate-oriented baselines is preserved.
  2. §II-A / Eq. (5): σ²_yk,t is treated as known when inverting mutual information, yet y_k,t = c_k^T x_t + w_k,t so its variance depends on the (unknown) state second moment and evolves with the closed loop. The manuscript does not specify whether σ²_yk,t is replaced by a prior-based estimate (e.g., from P_{t|t−1} and ĉ), a long-run average, or oracle knowledge. Because the distortion variance directly scales the measurement-noise matrix fW_t in (21), this choice affects both the Kalman recursion and the DRL observation–reward coupling. Please state the operational rule used in simulation and, if oracle variance is used, discuss the effect of replacing it by a filter-consistent estimate.
  3. §III-B.4, Eq. (34): training rewards use the true state cost x_t^T Q x_t, which is unavailable at the EIH under the POMDP observation (31). Offline simulation with privileged reward is standard, but the paper should make this explicit and address whether a critic trained only on observable quantities (or on estimated state cost using x̂_t) still yields a policy that retains the gains of Figs. 6–9. As written, the training setup is stronger than the information available at deployment, which weakens the claim that the same agent can be “deployed at the EIH” without further qualification (§III-E).
minor comments (6)
  1. §II-B, Eq. (12): finite-blocklength rate is written with packet error probability ε, then ε is set “very small” and packet drops are not simulated. Either drop the finite-blocklength terms for clarity or report sensitivity to a non-negligible ε with occasional erasures.
  2. Fig. 3 and §III-D: the three-path feature extractor is described only at a high level (dimensions 16/32/16 then 64 + ResBlock). A short statement of which path receives h_t, x̂_{t|t−1}, and P_{t|t−1} (and how the matrix P is flattened) would aid reproducibility.
  3. Table II lists entropy coefficient 0; the text still motivates an entropy term (39). Align the narrative with the actual hyperparameter choice or report a brief ablation with nonzero entropy.
  4. Notation: ey / eY / ewt and fW_t are hard to parse in plain text; consider standard tildes or hats consistently. Also fix “UA Vs” spacing and the duplicate Fig. 1 caption block early in the PDF.
  5. §IV-C baselines: “sensing-oriented scheme in [10]” is only loosely tied to Olfati-Saber’s distributed KF paper; a one-line definition of the exact SNR-weighted rule used would avoid ambiguity.
  6. Related work: position more sharply against the authors’ own SC3 / multi-loop LQR allocation results (e.g., [22]) so that the multi-sensor single-loop continuous-bandwidth novelty is explicit for non-specialist readers.

Circularity Check

1 steps flagged

No load-bearing circularity: LQR-minimizing PPO is an empirical policy search under an explicit rate-distortion model, not a result forced by its own inputs.

specific steps
  1. self citation load bearing [§II-A Eq. (5) and simulation settings (ρ=0.005); cf. [22]]
    "We introduce an information utilization ratio ρ∈(0,1] to capture that only a fraction of the raw sensing content is task-relevant to the control objective [22]. Then ... I(y_k,t; êy_k,t)=ρ D_k,t. The distortion noise variance can be calculated as σ²_nk,t = σ²_yk,t / (2^{2ρ D_k,t} − 1)."

    ρ and the task-relevant MI = ρD map are taken from overlapping-author work [22] rather than re-derived. This is framing/parameterization only: ρ is held fixed (0.005) and is not fitted to the LQR outcomes that the paper reports, so the DRL-vs-baseline ranking is not forced by that citation. Mild, non-load-bearing self-citation.

full rationale

The derivation chain is model → fixed LQG controller (Kalman + LQR) → POMDP with reward equal to negative instantaneous LQR cost → PPO training → Monte Carlo comparison to independent baselines under the same plant. Defining r_t = −(x_t^T Q x_t + u_t^T R u_t) so that maximizing return matches the stated objective is standard RL design, not a self-definitional prediction: the paper does not claim to derive the LQR cost from the reward; it claims a learned allocation policy beats sensing-oriented and sum-rate heuristics in simulation. The Gaussian rate-to-distortion map (Eqs. 3–5) is an explicit modeling ansatz, justified mainly by external LQG/rate-distortion work (Tanaka et al. [27]) and high-dimensional quantization approximations [28], [29], with ρ fixed as a simulation hyperparameter rather than fitted to the reported LQR curves. Self-citations to the authors’ SC3/EIH papers supply system framing and the ρ concept but do not import a uniqueness theorem that forces the bandwidth policy or the baseline ranking. There is no fitted-input-called-prediction, no renaming of a known empirical law, and no analytical “prediction” that reduces by construction to the training objective. Model-validity concerns (Gaussian n_k,t, fixed ρ) are assumption risks, not circularity. Score 1 only for non-load-bearing self-citation of the ρ framing.

Axiom & Free-Parameter Ledger

4 free parameters · 6 axioms · 1 invented entities

The central claim rests on standard LQG and wireless models plus several paper-specific modeling choices (Gaussian rate-limited distortion, ρ-scaled task-relevant rate, perfect downlink, continuous bandwidth). Free parameters (especially ρ and DRL/training knobs) shape the simulated rewards the agent optimizes. No new physical entity is required; EIH/SC3 are architectural framings from the authors’ program.

free parameters (4)
  • information utilization ratio ρ = 0.005
    Scales how much of the throughput counts as task-relevant mutual information in the distortion variance formula; set to 0.005 in simulations without a measurement-based fit.
  • LQR weights Q, R = 0.01 * I
    Define the cost the agent minimizes; set to 0.01 I without task-specific identification.
  • PPO / training hyperparameters = Table II defaults; l_th not numerically fixed in text
    Learning rate, rollout length, γ, λ, clip range, episode length, early-stop cost threshold l_th, etc., chosen for stable training and affect reported rewards.
  • packet error probability ε (ultra-reliable) = very small (not numerically specified)
    Set very small so links are treated as reliable bit pipes; drops random erasures from the state evolution.
axioms (6)
  • domain assumption Plant, sensing, process, and rate-limited distortion noises are Gaussian; linear sensing and dynamics hold so LQG separation applies.
    §II-A and §III-A use classical LQG Kalman + LQR; Gaussianity justifies the mutual-information formula for distortion variance.
  • ad hoc to paper Rate-limited reception equals y_k,t plus independent AWGN with variance from I(y; ŷ)=ρ D and Gaussian rate-distortion style inversion.
    Eqs. (3)–(5); motivated by [27]–[29] but adopted as the operational model for bandwidth→control coupling.
  • domain assumption Finite-blocklength achievable rate formula of Polyanskiy et al. gives effective throughput D_k,t = T R_k,t(B_k,t).
    Eq. (12)–(13), §II-B.
  • domain assumption Downlink control commands are received without distortion; uplink bandwidth is the bottleneck.
    §II-C; explicitly deferred joint UL/DL to future work.
  • domain assumption Orthogonal sensor channels, fixed known UAV/EIH position, suburban LoS/nLoS path-loss model.
    §II-B channel model and simulation geometry.
  • ad hoc to paper POMDP observation {h_t, x̂_{t|t−1}, P_{t|t−1}} is sufficient for effective RL despite partial observability of x_t.
    §III-B.2; justified as approximate Markov via Kalman sufficient statistics.
invented entities (1)
  • Edge information hub (EIH) as unified SC3 coordinator no independent evidence
    purpose: Architectural entity that co-locates communication, MEC, digital twin estimation, and control command generation for the closed loop.
    Framing device from the authors’ SC3 program; not a new physical particle/force, but a system construct the resource problem is defined around.

pith-pipeline@v1.1.0-grok45 · 25030 in / 3633 out tokens · 38427 ms · 2026-07-10T10:15:28.025765+00:00 · methodology

0 comments
read the original abstract

Robots are increasingly deployed in remote or hazardous areas for mission-critical control tasks. Due to their limited individual capabilities, they have to rely on other field sensors to obtain the state information of targets, and also a dedicated edge information hub (EIH) to enable information exchange, sensing data analysis and control command generation. Such configuration follows a sensing-communication-computing-control (SC3) closed loop. To optimize the whole closed-loop performance, this paper minimizes the linear quadratic regulator (LQR) control cost by designing the sensor-to-EIH bandwidth allocation. Specifically, we first model the distortion noise caused by limited communication data rate based on the mutual information theory. Next, under the control policy based on the Kalman filter and LQR controller, we formulate the control process as a partially observable Markov decision process (POMDP), and develop a deep reinforcement learning (DRL)-based sensor-to-EIH bandwidth allocation scheme. The proximal policy optimization (PPO) algorithm is utilized to train the DRL agent. Simulation results are provided to show the superiority of the proposed DRL-based scheme.

Figures

Figures reproduced from arXiv: 2607.08272 by Chengleyang Lei, Ning Ge, Shi Jin, Wei Feng, Yongxu Zhu, Yunfei Chen.

Figure 1
Figure 1. Figure 1: Illustration of the closed-loop control system where multiple sensors [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the proposed PPO-based bandwidth allocation framework. [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Convergence performance of the proposed DRL-based bandwidth [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Training performance of representative DRL algorithms. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Averaged LQR cost by different schemes varying with the bandwidth [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Averaged LQR cost by different schemes varying with the transmission [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 11
Figure 11. Figure 11: Averaged input cost by different schemes varying with the bandwidth [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Averaged data throughput by different schemes varying with the [PITH_FULL_IMAGE:figures/full_fig_p012_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Averaged LQR cost for different schemes with [PITH_FULL_IMAGE:figures/full_fig_p013_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Averaged LQR cost by two schemes varying with the bandwidth [PITH_FULL_IMAGE:figures/full_fig_p013_14.png] view at source ↗
Figure 16
Figure 16. Figure 16: Bandwidth allocation results with different state transition parameters [PITH_FULL_IMAGE:figures/full_fig_p014_16.png] view at source ↗

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

38 extracted references · 38 canonical work pages · 1 internal anchor

  1. [1]

    Collaborative multi-robot search and rescue: Planning, coordination, perception, and active vision,

    J. P. Queralta et al., “Collaborative multi-robot search and rescue: Planning, coordination, perception, and active vision,”IEEE Access, vol. 8, pp. 191617-191643, 2020

    work page 2020

  2. [2]

    Control-oriented deep space communications for unmanned space exploration,

    X. Fang, W. Feng, Y . Chen, N. Ge and G. Zheng, “Control-oriented deep space communications for unmanned space exploration,”IEEE Trans. Wireless Commun., vol. 23, no. 10, pp. 14466-14481, Oct. 2024

    work page 2024

  3. [3]

    Satellite-UA V networks for 6G control: A sensing-communication- computing-control closed loop perspective,

    C. Lei, X. Fang, W. Feng, Y . Chen, N. Ge, M. Xiao, and S. Mao, “Satellite-UA V networks for 6G control: A sensing-communication- computing-control closed loop perspective,”IEEE Network, vol. 39, no. 4, pp. 62-69, Jul. 2025

    work page 2025

  4. [4]

    Structured satellite- UA V-terrestrial networks for 6G Internet of Things,

    W. Feng, Y . Wang, Y . Chen, N. Ge, and C.-X. Wang, “Structured satellite- UA V-terrestrial networks for 6G Internet of Things,”IEEE Network, vol. 38, no. 4, pp. 48-54, Jul. 2024

    work page 2024

  5. [5]

    Digital twin in industry: State-of-the-art,

    F. Tao, H. Zhang, A. Liu and A. Y . C. Nee, “Digital twin in industry: State-of-the-art,”IEEE Trans. Ind. Informat., vol. 15, no. 4, pp. 2405- 2415, Apr. 2019

    work page 2019

  6. [6]

    Digital twins: A survey on enabling technologies, challenges, trends and future prospects,

    S. Mihai et al., “Digital twins: A survey on enabling technologies, challenges, trends and future prospects,”IEEE Commun. Surveys Tuts., vol. 24, no. 4, pp. 2255-2291, 4th quarter 2022. 15

    work page 2022

  7. [7]

    Edge information hub: Orchestrating satellites, UA Vs, MEC, sensing and communications for 6G closed-loop controls,

    C. Lei, W. Feng, P. Wei, Y . Chen, N. Ge, and S. Mao, “Edge information hub: Orchestrating satellites, UA Vs, MEC, sensing and communications for 6G closed-loop controls, ”IEEE J. Sel. Areas Commun.vol. 43, no. 1, pp. 5-20, Jan. 2025

    work page 2025

  8. [8]

    Environmental wireless sensor networks,

    P. Corke, T. Wark, R. Jurdak, W. Hu, P. Valencia and D. Moore, “Environmental wireless sensor networks,”Proc. IEEE, vol. 98, no. 11, pp. 1903-1917, Nov. 2010

    work page 1903

  9. [9]

    A systematic review on fusion techniques and approaches used in applications,

    S. Jusoh and S. Almajali, “A systematic review on fusion techniques and approaches used in applications,”IEEE Access, vol. 8, pp. 14424-14439, 2020

    work page 2020

  10. [10]

    Distributed Kalman filtering for sensor networks,

    R. Olfati-Saber, “Distributed Kalman filtering for sensor networks,” in Proc. 46th IEEE Conf. Decis. Control, New Orleans, LA, USA, 2007, pp. 5492-5498

    work page 2007

  11. [11]

    A novel kalman filter for human motion tracking with an inertial-based dynamic inclinometer,

    G. Ligorio and A. M. Sabatini, “A novel kalman filter for human motion tracking with an inertial-based dynamic inclinometer,”IEEE Trans. Biomed. Eng., vol. 62, no. 8, pp. 2033-2043, Aug. 2015

    work page 2033

  12. [12]

    Deep coupling autoencoder for fault diagnosis with multimodal sensory data,

    M. Ma, C. Sun and X. Chen, “Deep coupling autoencoder for fault diagnosis with multimodal sensory data,”IEEE Trans. Ind. Informat., vol. 14, no. 3, pp. 1137-1145, Mar. 2018

    work page 2018

  13. [13]

    Multi-Modal Fusion Transformer for End-to-End Autonomous Driving,

    A. Prakash, K. Chitta and A. Geiger, “Multi-Modal Fusion Transformer for End-to-End Autonomous Driving,” inProc. IEEE/CVF Conf. Com- puter Vis. Pattern Recognit. (CVPR), Nashville, TN, USA, 2021, pp. 7073-7083

    work page 2021

  14. [14]

    Partially-connected hybrid beamforming design for integrated sensing and communication systems,

    X. Wang, Z. Fei, J. A. Zhang and J. Xu, “Partially-connected hybrid beamforming design for integrated sensing and communication systems,” IEEE Trans. Commun., vol. 70, no. 10, pp. 6648-6660, Oct. 2022

    work page 2022

  15. [15]

    Joint transmit and receive beamforming design in full-duplex integrated sensing and communica- tions,

    Z. Liu, S. Aditya, H. Li and B. Clerckx, “Joint transmit and receive beamforming design in full-duplex integrated sensing and communica- tions,”IEEE J. Sel. Areas Commun., vol. 41, no. 9, pp. 2907-2919, Sep. 2023

    work page 2023

  16. [16]

    Near-field integrated sensing and commu- nications,

    Z. Wang, X. Mu and Y . Liu, “Near-field integrated sensing and commu- nications,”IEEE Commun. Letters, vol. 27, no. 8, pp. 2048-2052, Aug. 2023

    work page 2048

  17. [17]

    Collaborative energy and information transfer in green wireless sensor networks for smart cities,

    W. Lu, Y . Gong, X. Liu, J. Wu and H. Peng, “Collaborative energy and information transfer in green wireless sensor networks for smart cities,” IEEE Trans. Ind. Informat., vol. 14, no. 4, pp. 1585-1593, Apr. 2018

    work page 2018

  18. [18]

    Wireless powered sensor networks for Internet of Things: Maximum throughput and optimal power allocation,

    Z. Chu, F. Zhou, Z. Zhu, R. Q. Hu and P. Xiao, “Wireless powered sensor networks for Internet of Things: Maximum throughput and optimal power allocation,”IEEE Internet Things J., vol. 5, no. 1, pp. 310-321, Feb. 2018

    work page 2018

  19. [19]

    Resource allocation for data gathering in UA V-aided wireless sensor networks,

    H. Zanjie, N. Hiroki, K. Nei, O. Fumie, M. Ryu and Z. Baohua, “Resource allocation for data gathering in UA V-aided wireless sensor networks,”Proc. 4th IEEE Int. Conf. Netw. Infrastructure Digit. Content, Beijing, China, 2014, pp. 11-16

    work page 2014

  20. [20]

    Resource allocation for underwater acoustic sensor networks with partial spectrum sharing: When optimization meets deep reinforcement learning,

    R. Tang, R. Zhang, Y . Xu, C. Liu, C. Huang and C. Yuen, “Resource allocation for underwater acoustic sensor networks with partial spectrum sharing: When optimization meets deep reinforcement learning,”IEEE Trans. Netw. Service Manag., vol. 22, no. 3, pp. 2715-2730, Jun. 2025

    work page 2025

  21. [21]

    Quantized Kalman filtering,

    S. Sun, J. Lin, L. Xie and W. Xiao, “Quantized Kalman filtering,” in Proc. IEEE 22nd Int. Symp. Intell. Control, Singapore, 2007, pp. 7-12

    work page 2007

  22. [22]

    Sensing-communication-computing-control closed-loop optimization for 6G digital twin-empowered robotic systems,

    X. Fang, C. Lei, W. Feng, Y . Chen, M. Xiao, N. Ge, and C. Wang, “Sensing-communication-computing-control closed-loop optimization for 6G digital twin-empowered robotic systems,”IEEE J. Sel. Areas Commun. vol. 43, no. 10, pp. 3330-3346, Oct. 2025

    work page 2025

  23. [23]

    Deep reinforcement learning based transmission scheduling for sensing and control coordination,

    T. Jin, C. Chen, Y . Ma and X. Guan, “Deep reinforcement learning based transmission scheduling for sensing and control coordination,” IEEE Trans. Signal Inf. Processing Netw., vol. 11, pp. 1409-1421, 2025

    work page 2025

  24. [24]

    Joint secure offloading and resource allocation for vehicular edge computing network: A multi-agent deep reinforcement learning approach,

    Y . Ju et al., “Joint secure offloading and resource allocation for vehicular edge computing network: A multi-agent deep reinforcement learning approach,”IEEE Trans. Intell. Transp. Syst., vol. 24, no. 5, pp. 5555- 5569, May 2023

    work page 2023

  25. [25]

    Energy-Efficient Joint Task Offloading and Resource Allocation in OFDMA-Based Collaborative Edge Computing,

    L. Tan, Z. Kuang, L. Zhao and A. Liu, “Energy-Efficient Joint Task Offloading and Resource Allocation in OFDMA-Based Collaborative Edge Computing,”IEEE Trans. Wireless Commun., vol. 21, no. 3, pp. 1960-1972, Mar. 2022

    work page 1960

  26. [26]

    Bubnicki,Modern Control Theory

    Z. Bubnicki,Modern Control Theory. Berlin, Germany: Springer, 2005

    work page 2005

  27. [27]

    LQG Control with minimum directed information: Semidefinite programming approach,

    T. Tanaka, P. M. Esfahani and S. K. Mitter, “LQG Control with minimum directed information: Semidefinite programming approach,”IEEE Trans. Auto. Control, vol. 63, no. 1, pp. 37-52, Jan. 2018

    work page 2018

  28. [28]

    Gaussian approximation of quantization error for estimation from compressed data,

    A. Kipnis and G. Reeves, “Gaussian approximation of quantization error for estimation from compressed data,”IEEE Trans. Inf. Theory, vol. 67, no. 8, pp. 5562-5579, Aug. 2021

    work page 2021

  29. [29]

    Analysis of quantization noise and state estimation with quantized measurements,

    J. Xu, J. Li, and S. Xu, “Analysis of quantization noise and state estimation with quantized measurements,”J. Control Theory Appl., vol. 9, no. 1, pp. 66–75, 2011

    work page 2011

  30. [30]

    Channel coding rate in the finite blocklength regime,

    Y . Polyanskiy, H. V . Poor, and S. Verdu, “Channel coding rate in the finite blocklength regime,”IEEE Trans. Inf. Theory, vol. 56, no. 5, pp. 2307–2359, May 2010

    work page 2010

  31. [31]

    Optimal LAP altitude for maximum coverage,

    A. Al-Hourani, S. Kandeepan, and S. Lardner, “Optimal LAP altitude for maximum coverage,”IEEE Wireless Commun. Lett., vol. 3, no. 6, pp. 569–572, Dec. 2014

    work page 2014

  32. [32]

    Efficient 3-D placement of an aerial base station in next generation cellular networks,

    R. I. Bor-Yaliniz, A. El-Keyi and H. Yanikomeroglu, “Efficient 3-D placement of an aerial base station in next generation cellular networks,” inProc. IEEE International Conference on Communications (ICC), Kuala Lumpur, Malaysia, May 2016, pp. 1-5

    work page 2016

  33. [33]

    D. E. Kirk,Optimal Control Theory: An Introduction. New York: Wiley, 1967

    work page 1967

  34. [34]

    The role and use of the stochastic linear-quadratic-Gaussian problem in control system design,

    M. Athans, “The role and use of the stochastic linear-quadratic-Gaussian problem in control system design,”IEEE Trans. Auto. Control, vol. 16, no. 6, pp. 529-552, Dec. 1971

    work page 1971

  35. [35]

    “An introduction to the Kalman filter

    G. Bishop and G. Welch, ““An introduction to the Kalman filter” Dept. Comput. Sci., Univ. North Carolina Chapel Hill, Chapel Hill, NC, USA, Rep. TR 95-041, 1995

    work page 1995

  36. [36]

    Proximal Policy Optimization Algorithms

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Prox- imal policy optimization algorithms” arXiv preprint arXiv: 1707.06347, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  37. [37]

    Stable-baselines3: Reliable reinforcement learning implemen- tations

    A. Raffin, A. Hill, A. Gleave, A. Kanervisto, M. Ernestus, and N. Dormann, “Stable-baselines3: Reliable reinforcement learning implemen- tations.”J. Mach. Learn. Res., vol. 22, no. 268, pp. 1–8, 2021

    work page 2021

  38. [38]

    Control-oriented power allocation for integrated satellite-UA V networks,

    C. Lei, W. Feng, J. Wang, et al. “Control-oriented power allocation for integrated satellite-UA V networks,”IEEE Wireless Commun. Lett., vol. 12, no. 5, pp. 883-887, May 2023

    work page 2023