Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The topic is relevant and the framework is interesting, but several aspects need further improvement to strengthen the manuscript. The following comments should be addressed:

-The evaluation is fully simulation-based. The authors should provide stronger justification for the simulated data and, if possible, include real network or security traces.

-The theoretical analysis assumes bounded, zero-mean, and independent noise. The authors should discuss whether these assumptions are realistic under adversarial conditions.

-The service-class weights strongly affect the reported performance. A sensitivity analysis with different weight settings is needed.

-The baseline comparison is limited. More recent adaptive or learning-based trust-management methods should be included for a stronger comparison.

-The literature could be incorporated with more recent works such as "Link Traffic-Delay Mapping Model Learning Based on Multi-Class Samples in Software-Defined Networks," "Predefined-sequential-synchronized control with exploration of event-triggered mechanism," " Adaptive Fuzzy Control of Nonlinear Systems With Predefined Time and Accuracy," and so on.

-SENTRA performs less effectively under some conservative URLLC and cross-slice conditions. These limitations should be discussed more clearly.

-The improvement from inter-slice trust coupling appears relatively small. The authors should better justify its additional complexity and practical benefit.

-The scalability analysis is limited to a relatively small network. Larger numbers of devices and slices, together with computational overhead and memory requirements, should be evaluated.

Author Response

Comment 1

Reviewer comment: The evaluation is fully simulation-based. The authors should provide stronger justification for the simulated data and, if possible, include real network or security traces.

Response: Thank you for this suggestion. Section VII now explains why generated traffic, independent block Rayleigh fading, the FBL normal approximation, and deadline-based packet removal are used. This controlled setting allows all policies to be compared under the same assumptions. The evaluation is entirely simulation-based and is not calibrated to a particular deployment. No measured network or security traces are used. We acknowledge that validation with measured traffic and channel data remains necessary future work. Section VII also clarifies the admission cap and states that the reported reliability applies to admitted traffic.

Manuscript location: Section VII, simulation justification and traffic-generation paragraphs.

 

Comment 2

Reviewer comment. The theoretical analysis assumes bounded, zero-mean, and independent noise. The authors should discuss whether these assumptions are realistic under adversarial conditions.

Response: The PHY model uses AWGN and fading; it does not include adversarial perturbations. As clarified in Section II, Gaussian noise samples are not bounded. The boundedness and moment conditions used in the queue analysis instead concern arrivals, departures, control quantities, and transmit power. The revised text explicitly excludes jamming, adversarial interference, and malicious observations. We hope this distinction resolves the concern within the scope of the present model.

Manuscript location: Section II, discussion following Eq. (5); Section VI.

 

Comment 3

Reviewer comment: The service-class weights strongly affect the reported performance. A sensitivity analysis with different weight settings is needed.

Response: The model contains a single traffic class and therefore has no service-class weights. Section VII now distinguishes the reliability coefficient α, normalized-power coefficient β, urgency coefficient η, and control parameter V. The additional fixed-load experiment in Table II varies V while keeping α = 1.0, β = 0.35, and η = 2.0 fixed. It demonstrates sensitivity to V, but does not establish robustness to changes in the relative coefficients. We have kept that distinction explicit.

Manuscript location: Section VII, controller settings and V experiment; Section VIII, Table II.

 

Comment 4

Reviewer comment: The baseline comparison is limited. More recent adaptive or learning-based trust-management methods should be included for a stronger comparison.

Response: The present formulation concerns transmission control and contains no trust score or trust-management mechanism. Trust-management algorithms would therefore require a different model and evaluation criterion to serve as comparable baselines. Section VII now defines the four implemented transmission-control benchmarks more precisely. It also explains that Adaptive-FBL and EC-FBL are locally implemented policies motivated by prior adaptation principles, rather than exact reproductions of the complete algorithms in those studies. We have retained the comparison within the single-user FBL setting.

Manuscript location: Section VII, benchmark definitions.

 

Comment 5

Reviewer comment: The literature could be incorporated with more recent works such as "Link Traffic-Delay Mapping Model Learning Based on Multi-Class Samples in Software-Defined Networks," "Predefined-sequential-synchronized control with exploration of event-triggered mechanism," " Adaptive Fuzzy Control of Nonlinear Systems With Predefined Time and Accuracy," and so on.

Response: Thank you for suggesting these studies. The topics identified in the comment concern SDN traffic-delay mapping, predefined-time synchronization, and adaptive fuzzy control. Our contribution addresses queue-aware FBL transmission decisions and deadline expiration in a single-user wireless link. We have therefore retained a focused literature discussion rather than present these different control problems as direct transmission-control comparisons. The suggested references have not been added to the present revision.

Manuscript location: Section I, related-work discussion; no additional citation is claimed for this comment.

Comment 6

Reviewer comment: SENTRA performs less effectively under some conservative URLLC and cross-slice conditions. These limitations should be discussed more clearly.

Response: We agree that the limitations should be clear. SENTRA and cross-slice operation are not part of this manuscript. For the system actually evaluated, Section VIII explicitly states that the proposed controller saves transmit power but does not uniformly outperform the adaptive benchmarks in E2E failure or delay. It also explains that packet expiration becomes the main loss mechanism as traffic increases. Table II further quantifies the adverse reliability and delay effects of increasing V.

Manuscript location: Section VIII, concluding discussion and Table II.

 

Comment 7

Reviewer comment: The improvement from inter-slice trust coupling appears relatively small. The authors should better justify its additional complexity and practical benefit.

Response. The manuscript does not introduce inter-slice trust coupling or attribute any performance gain to such a mechanism. Its complexity analysis concerns the finite search over blocklength, rate, and power combinations for one transmitter. Consequently, there is no inter-slice coupling mechanism whose additional complexity can be assessed in the present study. We have retained the single-user scope..

Manuscript location. Section V, candidate-action search and complexity analysis.

Comment 8

Reviewer comment: The scalability analysis is limited to a relatively small network. Larger numbers of devices and slices, together with computational overhead and memory requirements, should be evaluated.

Response: The current study evaluates one transmitter–receiver link. Increasing the number of devices or slices would introduce scheduling and resource-sharing decisions beyond this model. Section IX now states this limitation explicitly and distinguishes candidate-action search complexity from network scalability. It also acknowledges that execution time, memory consumption, and compliance with a practical slot-processing deadline have not been measured. We therefore make no large-network validation claim.

Manuscript location: Section IX, paragraph beginning “The present evaluation is limited to a single-user link.”

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

1. Although the manuscript theoretically establishes the classical Lyapunov performance–backlog tradeoff, the simulation results do not examine the impact of the control parameter V. The authors should evaluate how different values of V affect transmit power, queue length, packet delay, and E2E reliability.

2. The reliability–energy tradeoff in Fig. 6 is obtained by varying the traffic arrival rate rather than V. The authors are encouraged to investigate the power–reliability and power–delay tradeoffs under different V values to better demonstrate the effectiveness and tunability of the proposed DPP framework.

3. The manuscript claims an (O(1/V)) optimality gap and (O(V)) queue-backlog scaling. These theoretical relationships should be supported by numerical results, particularly to clarify whether increasing V introduces unacceptable latency or reliability degradation in deadline-constrained URLLC systems.

4. Although the manuscript presents the theoretical Lyapunov performance–backlog tradeoff, its numerical validation is missing. The authors should investigate how varying the control parameter V affects transmit power, queue length, packet delay, and E2E reliability to substantiate the theoretical claims.

5. The derivation of the proposed drift-plus-penalty objective requires clarification. Specifically, the deadline-expiration term included in Eq. (40) is omitted from Eq. (29). The authors should justify this omission and ensure mathematical consistency between the queue dynamics and the final optimization objective.

6. The deadline-related reliability-risk component is insufficiently defined. The authors should explicitly explain how deadline violation risk is calculated based on packet urgency and queue conditions, and how it is incorporated into the proposed transmission decisions.

Author Response

Comment 1

Reviewer comment: 1. Although the manuscript theoretically establishes the classical Lyapunov performance–backlog tradeoff, the simulation results do not examine the impact of the control parameter V. The authors should evaluate how different values of V affect transmit power, queue length, packet delay, and E2E reliability.

Response: Thank you for this suggestion. We have completed a fixed-load V sweep and included the numerical results in Table II. The experiment uses underlying Poisson intensity λ = 0.60 packets/slot, an admission cap of four packets per slot, and a deadline of three slots. For each V in {1, 5, 10, 20, 50, 100}, we use 100 independent runs of 30,000 slots; the original traffic-load comparisons use 40 runs. Increasing V from 1 to 100 reduces average transmit power from 2.273 to 1.193 mW, while average backlog rises from 0.9459 to 0.9886 packets, delivered-packet delay rises from 1.5327 to 1.5974 slots, and terminal E2E failure rises from 2.961% to 3.593%. The table and accompanying discussion report these four metrics together so that the power saving can be assessed alongside its latency and reliability cost.

Manuscript location: Section VII, V experiment settings; Section VIII, Table II and accompanying discussion.

 

Comment 2

Reviewer comment: 2. The reliability–energy tradeoff in Fig. 6 is obtained by varying the traffic arrival rate rather than V. The authors are encouraged to investigate the power–reliability and power–delay tradeoffs under different V values to better demonstrate the effectiveness and tunability of the proposed DPP framework.

Response: We agree that traffic-load variation and V variation answer different questions. The caption of Fig. 6 now identifies its operating points as comparisons across traffic loads. Separately, Table II reports power, E2E failure, and delivered-packet delay for each V at fixed underlying arrival intensity λ = 0.60 packets/slot. This presentation allows the power–reliability and power–delay tradeoffs to be compared directly without changing the offered-load parameter. Across the tested V range, power decreases by approximately 47.5%, while E2E failure increases by approximately 0.632 percentage points and delivered-packet delay increases by approximately 4.2%. We have presented the V results in a compact table rather than adding separate tradeoff figures.

Manuscript location. Section VIII, Fig. 6 caption and Table II.

 

Comment 3

Reviewer comment: 3. The manuscript claims an (O(1/V)) optimality gap and (O(V)) queue-backlog scaling. These theoretical relationships should be supported by numerical results, particularly to clarify whether increasing V introduces unacceptable latency or reliability degradation in deadline-constrained URLLC systems.

Response: We agree that the original scaling claim required qualification. Section VI now presents Eqs. (48)–(49) as conditional classical DPP benchmarks, rather than established bounds for the implemented urgency-weighted controller. The abstract and analytical discussion have been revised accordingly. With capped admissions and expiration, the real queue is bounded by A_max D_max = 12 packets from the stated empty initialization, independently of V. This bound does not establish virtual-queue stability or reliable delivery. Table II provides empirical sensitivity results, but neither estimates an optimality gap against a reference optimum nor establishes aggregate real-and-virtual queue scaling. It also shows that the tested settings yield packet-level E2E failure above 10⁻³. We therefore report the observed degradation without claiming that increasing V preserves a packet-level URLLC reliability target.

Manuscript location. Abstract; Section VI, Theorem 1 and discussion of Eqs. (48)–(49); Section VIII, Table II.

 

Comment 4

Reviewer comment: 4. Although the manuscript presents the theoretical Lyapunov performance–backlog tradeoff, its numerical validation is missing. The authors should investigate how varying the control parameter V affects transmit power, queue length, packet delay, and E2E reliability to substantiate the theoretical claims.

Response. This concern is addressed by the same fixed-load experiment reported in Table II. The results show lower average power as V increases, accompanied by higher backlog, delivered-packet delay, and terminal E2E failure. From V = 1 to V = 100, backlog increases by approximately 4.5% and delivered-packet delay by approximately 4.2%, while E2E failure increases from 2.961% to 3.593%. These empirical results clarify the practical tradeoff under the three-slot deadline. As explained in our response to Comment 3, they are not presented as numerical proof of the classical asymptotic bounds.

Manuscript location: Section VIII, Table II and the accompanying  sensitivity discussion.

 

Comment 5

Reviewer comment: 5. The derivation of the proposed drift-plus-penalty objective requires clarification. Specifically, the deadline-expiration term included in Eq. (40) is omitted from Eq. (29). The authors should justify this omission and ensure mathematical consistency between the queue dynamics and the final optimization objective.

Response: Thank you for identifying this missing step. The revised derivation uses the fact that Q(t) and the deadline-removal count L(t) are nonnegative. Therefore, the contribution −Q(t)E[L(t) | observed state, a(t)] is nonpositive. Dropping it gives a more conservative upper bound; it is not an algebraically equivalent expression, and deadline removals need not be action-independent. Section VI and Appendix B now state this distinction explicitly. Capped admissions, finite deadlines, and bounded control quantities permit a uniform finite bound on the quadratic terms. The revised discussion also identifies the additional bounded urgency preference in the implemented metric. Theorem 1 is limited to real-queue boundedness and the constraint implications of assumed virtual-queue mean-rate stability; it does not assert an unconditional optimality or virtual-stability guarantee for this policy.

Manuscript location: Section IV, discussion of the decision metric; Section VI, drift derivation and Theorem 1; Appendix B.

 

Comment 6

Reviewer comment: 6. The deadline-related reliability-risk component is insufficiently defined. The authors should explicitly explain how deadline violation risk is calculated based on packet urgency and queue conditions, and how it is incorporated into the proposed transmission decisions

Response: The revised text defines the risk for the current HoL packet at its last admissible service opportunity. For a nonempty queue, equals when the observed HoL age reaches , and is zero otherwise. It is also zero for an empty queue. The expected service is for a nonempty queue and zero otherwise. The separate urgency weight is The implemented decision metric is , where Thus, backlog and age increase service priority, while the risk term penalizes failure at the HoL packet’s last opportunity. Section III also aligns the terminal-expiration indicator and normalized-power definitions with this surrogate. Because the surrogate excludes other expiring packets and is averaged per slot, its constraint is distinct from the measured packet-level E2E failure ratio.

Manuscript location. Section II, Eqs. (10)–(11); Sections III–V, surrogate objective, normalized power, and implemented metric

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The paper proposes a cross-layer optimization framework for Ultra-Reliable Low-Latency Communication systems operating under stochastic traffic arrivals and strict packet deadlines. It combines finite-blocklength communication, queue dynamics, packet expiration, and adaptive transmission control using Lyapunov drift-plus-penalty optimization.

The main contribution is an algorithm that jointly selects transmission blocklength, coding rate, and transmit power while considering reliability and energy constraints. Simulation results showed better performance on power savings compared with other existing methods. 

The topic is relevant and interesting.

The paper is has a good structure and it technically sounds.

The figures and tables are well presented and appropriate. 

The conclusions are generally consistent with the simulation results.

 

There are some observations:

• The references are relevant with no excessive self-citation, but only 10 of the 37 references were published within the last five years.  
• Some paragraphs and equations do not follow the template.
• The relationship between adaptive blocklength and fixed slot duration, the definition of deadline-related reliability risk, and the interpretation of queue stability under packet expiration should be clarified.
• The authors assume that each transmission slot has a fixed duration while the transmitter can select different blocklength values. How is this assumption physically implemented?
• Equation (18) specifies a hard deadline constraint, while the system model permits packet expiration and removal. The authors should distinguish between a constraint imposed on successfully delivered packets and a packet-discarding rule enforced by the simulator.
• The drift-plus-penalty metric includes the service indicator S(t), although the transmission outcome is not known when the action is selected. The authors should explicitly express the optimization metric using the conditional expected service probability.
• The manuscript introduces a Poisson arrival process but subsequently limits the number of admitted packets per slot to four. How are the capped arrivals treated? The excess packets are discarded or merged into the admitted arrivals?

Author Response

Comment 1

Reviewer comment: The references are relevant with no excessive self-citation, but only 10 of the 37 references were published within the last five years.

Response: Thank you for this observation. The current reference list retains the foundational FBL and Lyapunov sources needed for the derivations, together with recent studies relevant to transmission adaptation and wireless resource allocation. We agree that relevance and an accurate comparison are more useful than increasing the recent-reference count alone. We have not added references solely to change that count, and do not claim that the current revision includes a new literature update in response to this comment.

Manuscript location: Section I and the reference list.

 

Comment 2

Reviewer comment: Some paragraphs and equations do not follow the template.

Response: Thank you for drawing attention to the presentation. The revised equations remain editable in Word, with equation numbers retained, and the revised arrival and deadline definitions use consistent notation. We have also checked the rendered layout of the technical corrections. These checks do not establish compliance with the journal’s specific template; a final comparison of paragraph styles, equation alignment, captions, and table placement against that template remains outstanding.

Manuscript location: Sections II–III, revised Eqs. (1)–(2) and (18); presentation throughout. Journal-template verification remains outstanding.

 

Comment 3

Reviewer comment: The relationship between adaptive blocklength and fixed slot duration, the definition of deadline-related reliability risk, and the interpretation of queue stability under packet expiration should be clarified.

Response: We agree that these issues should be distinguished. Section II now explains the within-slot channel-use budget and unused time for shorter codewords. Sections II–V define the HoL deadline-risk surrogate and its use in action selection. Section VI explains that capped admissions and expiration give  from an empty initial queue. This is a bound on occupancy, not evidence that all packets are delivered. The virtual queues require separate mean-rate-stability conditions. Section VII and Table II therefore interpret backlog together with delivered-packet delay and terminal E2E loss.

Manuscript location: Section II, slot model and Eqs. (10)–(11); Sections IV–VI; Section VII, metric definitions; Section VIII, Table II.

Comment 4

Reviewer comment: The authors assume that each transmission slot has a fixed duration while the transmitter can select different blocklength values. How is this assumption physically implemented?

Response: The revised system model makes the resource interpretation explicit. Each fixed slot contains a data region of duration after reserved overhead. With channel-use duration , every candidate blocklength satisfies , and the data region is assumed to accommodate the largest candidate blocklength. A shorter codeword leaves the remaining portion unused in the single-packet-per-slot model. The channel is constant over the slot. This is a PHY abstraction rather than a deployment-specific timing calibration. Section VII also clarifies that reported power is the selected power averaged over slots, including zero power during idle slots. It is not weighted by transmission duration and should not be interpreted as burst energy PnT_sym or energy per delivered packet.

Manuscript location: Section II, opening slot-model paragraph; Section VII, performance-metric definitions; Section VIII, Table II note.

 

Comment 5

Reviewer comment: Equation (18) specifies a hard deadline constraint, while the system model permits packet expiration and removal. The authors should distinguish between a constraint imposed on successfully delivered packets and a packet-discarding rule enforced by the simulator.

Response. We agree. Equation (18) now applies only to packets successfully delivered in slot t, with admissible service ages from 1 through  It does not guarantee delivery of every admitted packet. Section II states the simulator’s event order: attempt the current HoL packet and remove it on success; append admitted arrivals at age zero; increment all remaining ages; and remove packets whose updated ages exceed  A new packet therefore has age one at its first subsequent service opportunity. Section VII defines delivered-packet delay as the age at successful service, conditional on delivery. It also distinguishes terminal E2E failure, , from the deadline-loss ratio, Unresolved packets at the finite horizon are excluded from the E2E denominator, and pre-admission excess arrivals are excluded from both ratios. These definitions separate successful-delivery timing, packet expiration, and finite-horizon loss accounting.

Manuscript location: Section II, slot event order; Section III, Eq. (18) and following paragraph; Section VII, metric definitions; Section VIII, Table II note.

Comment 6

Reviewer comment: The drift-plus-penalty metric includes the service indicator S(t), although the transmission outcome is not known when the action is selected. The authors should explicitly express the optimization metric using the conditional expected service probability.

Response: We agree. The revised decision-time metric uses conditional expected service, rather than an unobserved decoding outcome. For a nonempty queue, ; for an empty queue, it is zero. The observed information includes packet ages, channel state, and the real and virtual queues. Sections IV–V distinguish this quantity from the realized service used after transmission in queue evolution. The decision metric and Algorithm 1 use the expected-service formulation. The separate treatment of deadline removals in the drift bound is explained in our response to Reviewer 2, Comment 5.

Manuscript location: Sections III–V, observed information, Eqs. (29)–(30), and Algorithm 1; Section VI and Appendix B.

 

Comment 7

Reviewer comment: The manuscript introduces a Poisson arrival process but subsequently limits the number of admitted packets per slot to four. How are the capped arrivals treated? The excess packets are discarded or merged into the admitted arrivals?

Response: Excess potential arrivals are discarded at admission; they are not merged into admitted packets. Section II now distinguishes raw arrivals from admitted arrivals , with  in the evaluation. Equation (2) gives the capped distribution: probabilities for counts 0 through 3 are the corresponding Poisson probabilities, while . The admitted mean is  rather than λ. The queue dynamics use admitted  and  consistently denotes the underlying offered-load parameter. Section VII states that pre-admission excess arrivals are excluded from the reported loss ratios, so the reliability results apply to admitted traffic.

Manuscript location: Section II, Eqs. (1)–(2) and arrival-definition paragraphs; Section VII, traffic generation and metric definitions.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I have no more comments.

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have carefully addressed the reviewer’s comments.
The reviewer has no further comments or questions.