Pith. sign in

REVIEW 4 major objections 5 minor 65 references

A Unified QoS-Aware Multiplexing Framework for Next Generation Immersive Communication with Legacy Wireless Applications

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

Pith's one-line read By learning slicing decisions from queue backlogs, two algorithms keep 6G immersive and legacy traffic on one RAN with sublinear regret and measured throughput and latency gains.

desk verdict Useful slicing framework, but the main regret bound rests on an unproved and generally false linearity assumption—worth referee time, not acceptance as is. read the letter →

arxiv 2504.21444 v2 pith:XVPE5G6U submitted 2025-04-30 cs.NI

classification cs.NI
keywords networkslicingimmersivecommunicationLyapunovdriftadversarialcontextualbanditQoS-awareresourceallocationnon-stationarychanneltrackingKalmanfilterdual-timescaleoptimization
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

The paper tries to show that next-generation immersive traffic (AR, VR, holographic telepresence) can be carried in the same radio access network as legacy eMBB and URLLC traffic, without the usual practice of isolating them into separate slices. It argues that statistical isolation causes throughput loss and increased queue backlog, and proposes instead to make slicing decisions adapt online, using queue lengths as context. The central claim is that two algorithms—Ad2S for stationary channels and Ad2S-NR with a Kalman tracker for non-stationary ones—achieve sublinear regret and so learn the near-optimal slicing configuration over time. If the paper is right, operators can mix 6G and legacy services on the same spectrum with lower latency and higher throughput than fixed slicing, while preserving strict URLLC delay guarantees.

What carries the argument

The load-bearing mechanism is a three-level control loop. (1) Virtual queues $G_n(k)$ are constructed from the real packet queues and the delay thresholds $\delta_{e,M}$, so that the long-term average-backlog constraint becomes a queue-stability problem. (2) The Lyapunov drift-minus-utility bound converts the original problem into a per-frame utility maximization that is solved by the PBRA algorithm, a penalty successive upper-bound minimization with block coordinate descent. (3) A non-stochastic contextual-bandit learner, of the LinEXP3 type, chooses the frequency-domain slice partition $\mathcal{F}_l$ at each super-frame, using the current virtual-queue and buffer-queue values (and, in Ad2S-NR, Kalman-extrapolated pathloss means $\hat{\mu}_n(l)$) as the context vector $X_l$.

What would settle it

Take a two-user, two-subchannel version, compute the optimal $F_k^*(\mathcal{F}_l)$ by exhaustive search over all binary allocations for a dense grid of backlog values, and fit $F_k^*$ against the context $X_l$: if the residual grows with $X_l$ or the slope depends on the allocation variables $b_n(t,f)$, the linearity assumption fails and the regret bound of Theorem 1 cannot be invoked. Alternatively, simulate Ad2S for large $L$ under i.i.d. channels and check whether the log-regret versus log-$L$ slope exceeds $2/3$ asymptotically.

Watch

Extended reading notes

Core claim

The paper's central claim is that the long-term throughput-maximization problem with mixed QoS constraints can be transformed, via the Lyapunov drift theorem, into a short-term utility maximization in which every slice choice is scored by a backlog-weighted rate sum. It then treats that utility as a non-stochastic reward and chooses slice configurations with a linear contextual-bandit learner whose context is the vector of virtual-queue and buffer-queue states, giving the Ad2S algorithm. The theorem states that with learning rate $\eta = L^{-2/3}(|\mathcal{F}|(1+4N))^{-1/3}(\log|\mathcal{F}|)^{2/3}$ and exploration parameter $\gamma = L^{-1/3}(|\mathcal{F}|(1+4N)\log|\mathcal{F}|)^{1/3}$, the cumulative regret after $L$ super-frames obeys $\mathrm{Regret}_L^{\mathrm{Ad2S}} \le 5 L^{2/3} (|\mathcal{F}|(1+4N)\log|\mathcal{F}|)^{1/3} F_{\max}$, and the non-stationary refinement Ad2S-NR, which adds ME-KF pathloss estimates to the context, satisfies the analogous bound with $1+8N$ in place of $1+4N$. Both bounds are sublinear in $L$, which is what it means for the slicing rule to be asymptotically optimal.

Load-bearing premise

The sublinear regret bound rests on the assumption that the per-frame utility $F_k^*(\mathcal{F}_l)$ produced by the resource-allocation solver is a linear function of the context vector fed to the bandit learner; the paper asserts this linearity rather than proving it, and the affine approximation it derives in the appendix has coefficients $\alpha_n(k)$, $\beta_n(k)$ that depend on the allocation decisions themselves.

Editorial extensions

If this is right

  • If the regret bounds hold, an online slicing rule can approach the optimal configuration without knowing the channel statistics in advance, because the sublinear regret guarantees that the time-averaged utility converges to the best fixed slice policy in hindsight.
  • Mixed 6G and legacy operation would no longer require hard statistical isolation: the simulations report a 3.86 Mbps throughput gain, a 63.96% latency reduction, and a 24.36% reduction in exploration time against the tested baselines.
  • The numerical study reports 92.8% URLLC QoS satisfaction, meaning the probability that the frame's delivered data clears the queue within the delay bound, which is higher than the EXP3, contextual-UCB, and non-adaptive baselines.
  • Operators can choose where to sit on the throughput-versus-latency trade-off by tuning the ratio $\omega_Q/\omega_T$ of the Lyapunov weights, with the suggested operating range $6\times 10^{-5}$ to $8\times 10^{-5}$.

Reading between the lines

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

  • Editors' extension: the same virtual-queue-as-context recipe could be applied to other shared infrastructures—edge clouds, fronthaul, or open RAN midhaul—where hard-QoS and elastic flows compete, and the regret analysis would carry over as long as a linear-utility approximation is plausible.
  • Editors' extension: a direct stress test is to run Ad2S on a scenario with a deliberately nonlinear utility (e.g., a fifth-order rate model); the regret is expected to degrade to linear, which would confirm that the linearity premise is the load-bearing part of Theorem 1.
  • Editors' extension: the ME-KF tracker is replaceable; feeding a learned channel predictor (say, a sequence model) into the context vector would test whether the sublinear bound survives when the extrapolation is no longer provably unbiased.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. This manuscript proposes a two-timescale radio access network slicing framework for multiplexing legacy eMBB/URLLC traffic with immersive MBBLL traffic. A Lyapunov drift transformation is used to convert the long-term throughput maximization into a per-frame weighted utility problem, a penalty-BCD algorithm (PBRA) is proposed for the frame-level mixed-integer resource allocation, and two contextual-bandit slicing algorithms (Ad2S and Ad2S-NR) are proposed for the super-frame-level slicing decisions, with an ME-KF channel tracker for non-stationary scenarios. The central theoretical claim is Theorem 1, which bounds the cumulative regret of both algorithms by O(L^{2/3}) under stationary and non-stationary channels. The paper also reports extensive numerical experiments showing throughput, latency, and convergence improvements over several baselines, including EXP3, contextual UCB, and a PPO-based scheme.

Significance. If the regret bounds were valid, the paper would provide a useful advance in non-stochastic network slicing with QoS awareness, and the numerical study is unusually thorough, covering multiple baselines, bursty traffic, varying traffic intensities, parameter sweeps, and an RL comparison. The ME-KF tracking component is a plausible engineering contribution, and the complexity analysis in Lemma 4 is useful. However, the theoretical guarantee is the paper's headline contribution, and it depends on a linear realizability condition that is neither proven nor generically true for the stated problem; the numerical results do not repair this gap because the regret plots are not accompanied by any alternative theoretical justification or by confidence intervals that would let the reader assess the variability of the reported gains.

major comments (4)
  1. [Section VI (Theorem 1), Appendix C, Appendix D, Eq. (35)] The claimed sublinear regret relies on the assertion that F*_k(F_l) is linear in the context vector X_l. This is not established and is generically false: F*_k is defined in Problem 3.1 as the maximum over {b_n(t,f), p(t,f)} of an affine objective, and the maximum of affine functions is convex and piecewise linear, with kinks at points where the optimal allocation changes. The context vector contains nonlinear features such as Q_n^2 and G_n Q_n, but the max operator breaks global linearity regardless of feature choice. Appendix D does not repair the gap: Eq. (35) gives r_n(k) = alpha_n(k) E[hat R_n(l)] + beta_n(k), with alpha_n(k) and beta_n(k) explicitly depending on the allocation variables b_n(t,f) and p(t,f), which are outputs of the optimization and therefore context-dependent. Hence the linear model required by the LinEXP3 regret bound in [45] is misspecified, and Theorem 1's sublinear regret bound does not follow.
  2. [Appendix A, proof of Lemma 1; Eq. (5), Eq. (15)] The proof drops the max operator in the queue dynamics (5) by asserting that r_n(k) is guaranteed to be less than Q_n(k). No constraint enforcing this inequality appears in Problem 1, Problem 2, or Problem 3.1, and the objective rewards large r_n(k), so the inequality is not guaranteed by the problem formulation. Without it, Eq. (30) and hence the drift bound (15) are not valid, and the claimed equivalence between Problem 1 and Problem 2 is not established. Even if the inequality did hold, minimizing the drift bound in (15) is a surrogate optimization, not an equivalent transformation, absent an optimality-gap analysis.
  3. [Appendix C, unbiased estimation step; Eq. (20)] The unbiasedness derivation assumes that there exists a true parameter theta_{l,f} such that the per-super-frame reward equals <X_l, theta_{l,f}>. Since the linearity assumption fails as described above, the estimator in Eq. (20) is the least-squares projection of the reward onto the span of X_l, not an unbiased estimator of a true parameter of the linear model. The importance-weighting factor I{...}/pi_l(f|X_l) does not restore the realizability condition required by the regret bound of [45], so the unbiased-estimation hypothesis of that theorem is not satisfied.
  4. [Algorithm 2, Eq. (19)] The softmax denominator in the policy is written as sum over f' != f of exp(...), omitting the current action f. As written, pi_l(.|X_l) is not a normalized probability distribution over slicing actions, so Algorithm 2 is not the LinEXP3 policy whose regret bound is cited. If this is a typographical error, the denominator should be corrected to sum over all f' in the action set; as it stands, the policy is not implementable as stated and the exploration probability gamma/|F| does not combine with the exponential term to give a valid distribution.
minor comments (5)
  1. [Section II-A, after Eq. (2)] The word "executively" should be "exclusively": each time-frequency element can be allocated to only one user.
  2. [Section III, first paragraph] The sentence "we first formulate ... and than discuss" contains a typo: "than" should be "then".
  3. [Theorem 1, Eq. (26) and Eq. (27)] The statement "approximately 1.3x additional regret" in the paragraph after Theorem 1 is only a numerical illustration; the ratio of the two bounds depends on N and |F| and is not a constant 1.3 in general.
  4. [Appendix B, Lemma 2 and Eq. (33)] The proof of Lemma 2 would be easier to follow if the notation for the penalty term and the surrogate function distinguished the iteration index of sigma from the block-coordinate inner iterations more clearly.
  5. [Fig. 4] The labels "1x" and "1.3x" in Fig. 4 are not defined in the caption; the reader has to infer that they refer to the regret overhead ratio claimed in the text.

Circularity Check

2 steps flagged · score 6.0 of 10

Theorem 1's sublinear regret bound rests on asserting—not proving—that the piecewise-defined optimum F*_k is linear in the context; Appendix C makes LinEXP3's central hypothesis true 'according to the definition' of F*.

  1. self definitional [Appendix C.A.1, bullet 3 and 'Unbiased Estimation' paragraph (Proof of Theorem 1)]
    "According to the definition of F∗_k(Fl), for k∈[(l−1)×K̄+1,l×K̄], F∗_k(Fl) is linear with both {Gn(k)} and {Cn(k)}. ... E_l[θ̂_l,f] = E_l[ I{Fl=f}/π_l(f|X_l) (X_l^T X_l)^{-1} X_l (1/K̄) Σ F∗_k(Fl) ] = E_l[ E_l[I{Fl=f}/π_l(f|X_l)|X_l] θ_l,f ] = θ_l,f."

    F∗_k(Fl) is defined in Problem 2 as the maximum over {bn(t,f),p(t,f)} of a linear objective in the rates rn(k); the maximum of affine functions is convex and piecewise-linear, not globally affine. Linearity therefore does not follow from the definition; the proof asserts it as if it did. The subsequent unbiasedness computation replaces (1/K̄)ΣF∗_k(Fl) by X_l^T θ_l,f, i.e. it assumes the very linear model whose validity is in question. Since the LinEXP3 bound of [45] applies only under a true linear reward, Theorem 1's regret bound rests on a hypothesis made true by assertion ('according to the definition') rather than by the system equations. This is a circular verification of the external theorem's premise.

  2. other [Appendix D, Eq. (35) and the following coefficient definitions]
    "rn(k)=αn(k)E_l[R̂_n(l)]+βn(k), where αn(k)=Σ_{(t,f)∈T_k×F} {B, μ_n(l)≥ln(τ|F|/P_tot)/2; B b_n(t,f)p(t,f)e^{2N(0,σ_n^2)}, μ_n(l)<...}, βn(k)=Σ...{B log2(b_n(t,f)p(t,f))+2N(0,σ_n^2)log2(e), 0} are affine non-stochastic coefficients needed to be properly estimated."

    αn(k) and βn(k) are defined through b_n(t,f) and p(t,f), which are the outputs of the frame-scale optimization (Algorithm 1) and therefore depend on the queue/virtual-queue context and on the chosen slice Fl. Calling these coefficients 'non-stochastic' does not make them fixed parameters of a true linear reward model: they are functions of the same allocation decisions whose aggregate value F∗_k(Fl) the bandit is trying to predict. The asserted affine representation does not exhibit the fixed θ required by LinEXP3; it defines the 'linear model' in terms of the very optimization outputs. Thus the non-stationary regret bound (27) inherits the same by-assertion verification of the linearity hypothesis.

full rationale

The paper's numerical evaluation is self-contained against external baselines, and the LinEXP3 regret theorem [45] is genuine external support. No load-bearing self-citation chain was found, and no fitted parameter is renamed as a prediction. The circularity is confined to the theoretical bridge between the frame-scale optimization and the bandit regret analysis. Appendix C must verify two hypotheses of [45]: unbiased estimation and linear reward in the context. The unbiasedness algebra is correct only if a fixed θ exists with (1/K̄)ΣF∗_k(Fl)=X_l^T θ; the existence of such θ is exactly the linearity hypothesis. The 'proof' of linearity consists of asserting that F∗_k(Fl), a maximum over allocations of a linear objective, is linear with {Gn(k)} and {Cn(k)} 'according to the definition'. A finite maximum of affine functions is piecewise-linear and generally not affine, so the premise is not derived. Appendix D does not repair this: its affine coefficients αn(k),βn(k) contain b_n(t,f),p(t,f), which are themselves outputs of the optimization and hence context-dependent. The central regret bound therefore reduces, at the critical step, to an assertion that the external theorem's premise holds by construction of the context vector rather than by the equations of the system. This is partial circularity in the theoretical derivation, which is why the score is 6 rather than 0-2; the empirical claims and the system model itself are not circular.

Assumptions & free parameters 3 free parameters · 6 assumptions · 1 invented entities

The framework leans on standard Lyapunov and bandit tools, but introduces a crucial ad hoc assumption: the per-frame optimal reward F*_k(F_l) is linear in the engineered context vector. This assumption is not justified by the non-convex resource allocation problem and is the main source of correctness risk. The system model also relies on ergodicity and timescale separation, and on dropping the queue-dynamics max operator. The tunable weights omega_Q and omega_T are hand-picked from simulation contours, so they are free parameters for the reported tradeoff.

free parameters (3)
  • omega_Q = 5e-8 to 8e-8
    Lyapunov weight for the virtual queue; chosen by hand in Section VII-D based on simulation contours to meet operator KPIs (60 Mbps and 5 ms delay outage), defining the throughput-QoS tradeoff.
  • omega_T = 4e-4 to 1e-3
    Lyapunov weight for total throughput; tuned jointly with omega_Q in Section VII-D to select a working operating regime, with ratio omega_Q/omega_T in 6e-5 to 8e-5.
  • tau = 1 dB
    Threshold in Eq. (25) that switches the spectral efficiency approximation between high- and low-SNR regimes in the Ad2S-NR context vector; fixed in Table I and used in the linearity approximation.
assumptions (6)
  • standard math Lyapunov drift theorem provides an equivalent reformulation of the long-term throughput maximization as a drift-minus-utility minimization.
    Used in Lemma 1 and Problem 2. In fact the theorem provides an upper bound on the DMU, not an exact equivalence, so the paper relies on a standard result but states it too strongly.
  • standard math LinEXP3 regret bound from [45] holds when the reward model is linear in the context and the estimator is unbiased.
    Theorem 1 and Appendix C directly invoke this external result; the contentious part is verifying the linearity hypothesis for the MINLP-generated reward.
  • ad hoc to paper The reward F*_k(F_l) is linear in the context vector X_l.
    Assumed in Appendices C and D to apply [45]; not proven for the optimal value of a non-convex mixed-integer program. This is the weakest load-bearing premise.
  • domain assumption Channel magnitudes within each super-frame are i.i.d. and ergodic, with minimum ergodic period K_e < K.
    Section II assumptions; needed to segment the long-term constraint into per-super-frame form (29) and to make the Lyapunov drift time-average argument valid.
  • ad hoc to paper Throughput r_n(k) is always less than queue length Q_n(k), so the max operator in the queue dynamics can be dropped.
    Appendix A uses this claim to rewrite (5) as Q_n(k+1)=Q_n(k)-r_n(k)+Lambda_n(k); this is not guaranteed by the system model and is needed for the quadratic bound (30).
  • domain assumption The pathloss follows the AR(1) model (13) with Gaussian process and measurement noise.
    Section II; defines the non-stationary channel model that the ME-KF scheme is designed to track.
invented entities (1)
  • Virtual queue G_n(k) independent evidence
    purpose: Reformulates the long-term average backlog constraint as a queue stability constraint, enabling the Lyapunov drift approach.
    Standard construction from Lyapunov optimization [40], with externally established convergence properties; not an unsupported invention of the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Unified QoS-Aware Multiplexing Framework for Next Generation Immersive Communication with Legacy Wireless Applications." pith.science (2026). https://pith.science/paper/XVPE5G6U

@misc{pith2026250421444,
  author       = {Pith},
  title        = {Pith review of: A Unified QoS-Aware Multiplexing Framework for Next Generation Immersive Communication with Legacy Wireless Applications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XVPE5G6U}},
  note         = {Machine review of arXiv:2504.21444}
}
read the original abstract

Immersive communication, including emerging augmented reality, virtual reality, and holographic telepresence, has been identified as a key service for enabling next-generation wireless applications. To align with legacy wireless applications, such as enhanced mobile broadband or ultra-reliable low-latency communication, network slicing has been widely adopted. However, attempting to statistically isolate the above types of wireless applications through different network slices may lead to throughput degradation and increased queue backlog. To address these challenges, we establish a unified QoS-aware framework that supports immersive communication and legacy wireless applications simultaneously. Based on the Lyapunov drift theorem, we transform the original long-term throughput maximization problem into an equivalent short-term throughput maximization weighted by virtual queue length. Moreover, to cope with the challenges introduced by the interaction between large-timescale network slicing and short-timescale resource allocation, we propose an adaptive adversarial slicing (Ad2S) scheme for networks with invarying channel statistics. To track the network channel variations, we also propose a measurement extrapolation-Kalman filter (ME-KF)-based method and refine our scheme into Ad2S-non-stationary refinement (Ad2S-NR). Through extended numerical examples, we demonstrate that our proposed schemes achieve 3.86 Mbps throughput improvement and 63.96% latency reduction with 24.36% convergence time reduction. Within our framework, the trade-off between total throughput and user service experience can be achieved by tuning systematic parameters.

Figures

Figures reproduced from arXiv: 2504.21444 by the authors.

Figure 1
Figure 1. (a) System model; (b) Dual-scale resource grid for network slicing [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Slice configuration Fl and other system dynamics (e.g., Qe(k), QM(k) denoting average backlog for eMBB and MBBLL users, respectively, and their time-average E[Qe(k)], E[QM(k)]). After entering exploitation stage, as ⃝1 the E[Qe(k)] approaching δe, ⃝2 our proposed Ad2S algorithm adaptively slices in favor of honoring legacy QoS, ⃝3 Qe(k) decreases in response. gorithm 1 to obtain F ∗ k (Fl), the non-stationary condit… view at source ↗
Figure 3
Figure 3. Estimation error, e.g. |µˆn(l) − µn(l)| of a priori based [22] and proposed ME-KF scheme for each user. P ∗ ≜ liml→∞ E[P (l)] = F ∗ 1 P ∗F ∗ 1 T +  QA 0 0 Qµ  − F ∗ 1 P ∗  0 I   0 I [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Empirical cumulative regret. • Baseline 1: Non-adaptive slicing (NAdS) with dynamic resource allocation scheme (DRAS)5 [9], this scheme addresses Problem 3.1 following a heuristic QoS oriented scheme, NAdS indicates that fixed slicing configuration with equal sub-chann…
Figure 5
Figure 5. Figure 5: (a) Impact of different chunk sizes on exploration time and average reward for Ad2S and EXP3 under identical parameters (e.g., η = 1, γ = 0.5). (b) Slice configuration behavior of Ad2S under chunk size of 1 sub-channel. (c) Slice configuration behavior of EXP3 under ch…
Figure 6
Figure 6. Figure 6: Empirical Cumulative Distribution Function (CDF) of experienced [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Average transmission rate and cumulative regret versus total power budget. As shown in this figure, the transmission rate of our proposed schemes is superior to conventional schemes, as the total power budget increases. Cumulative regret of our proposed schemes is also…
Figure 10
Figure 10. Figure 10: Average latency versus number of sub-channels. As shown in this figure, the average latency of our proposed schemes is lower than baselines as well. Larger latency improvement is demonstrated under limited number of sub-channels, as our schemes dynamically utilize fra…
Figure 14
Figure 14. Figure 14: A 3D illustration of average latency outage and average transmission rate versus varying ωQ and ωT . As shown in this figure, the delay outage decrease as ωQ increases. While as ωT increases, higher average transmission rate are promised. meeting QoS requirements for …
Figure 12
Figure 12. Figure 12: Average latency versus number of users. As shown in this figure, the average user experienced latency of our proposed scheme is smaller as well, if compared with conventional schemes. First, we numerically verify the effectiveness of the pro￾posed Lyapunov-based frame…
Figure 15
Figure 15. Figure 15: Slice configuration Fl and other performance metrics including queueing backlog and cumulative regret, where the left and right parts of this picture are carried out under λa,M = 1 and λa,M = 5, respectively. As λa,M increases, the instantaneous queue length for MBBLL…
Figure 16
Figure 16. Figure 16: Average transmission rate versus total power budget. As shown in [PITH_FULL_IMAGE:figures/full_fig_p015_16.png]
Figure 18
Figure 18. Figure 18: Slice configuration Fl and other performance metrics including queueing backlog and cumulative regret, where the left and right parts of this picture are carried out under λa,M = 1 and λa,M = 5, respectively. As λa,M increases, the instantaneous queue length for MBBLL…
Figure 19
Figure 19. Figure 19: Average latency versus total power budget. As shown in the figure, the average latency of our proposed scheme is smaller than [PITH_FULL_IMAGE:figures/full_fig_p019_19.png]
Figure 20
Figure 20. Figure 20: Average transmission rate and latency versus total power budget. As shown in the figure, the average transmission rate and latency [PITH_FULL_IMAGE:figures/full_fig_p020_20.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

65 extracted references · 62 canonical work pages

  1. [45]

    Efficient and robust algorithms for adversarial linear contextual bandits,

    G. Neu and J. Olkhovskaya, “Efficient and robust algorithms for adversarial linear contextual bandits,” in Proceedings of Thirty Third Conference on Learning Theory . PMLR, Jul. 2020, pp. 3049–3068

  2. [1]

    Recommendation itu-r m.2160-0 (11/2023) - framework and overall objectives of the future development of imt for 2030 and beyond,

    C. Gachet, “Recommendation itu-r m.2160-0 (11/2023) - framework and overall objectives of the future development of imt for 2030 and beyond,” Nov. 2023. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 16

  3. [2]

    Survey on 6G frontiers: trends, applications, require- ments, technologies and future research,

    C. D. Alwis, A. Kalla, Q.-V . Pham, P. Kumar, K. Dev, W.-J. Hwang, and M. Liyanage, “Survey on 6G frontiers: trends, applications, require- ments, technologies and future research,” IEEE Open Journal of the Communications Society, vol. 2, p. 51, Apr. 2021

  4. [3]

    Toward low-latency and ultra-reliable virtual reality,

    M. S. Elbamby, C. Perfecto, M. Bennis, and K. Doppler, “Toward low-latency and ultra-reliable virtual reality,” IEEE Network , vol. 32, no. 2, pp. 78–84, Mar. 2018

  5. [4]

    3GPP work plan - version december 6th 2024,

    3GPP, “3GPP work plan - version december 6th 2024,” Tech. Rep., Dec. 2024

  6. [5]

    Joint spectrum reservation and on-demand request for mobile virtual network operators,

    Y . Zhang, S. Bi, and Y .-J. A. Zhang, “Joint spectrum reservation and on-demand request for mobile virtual network operators,” IEEE Trans. Commun., vol. 66, no. 7, pp. 2966–2977, Jul. 2018

  7. [6]

    Joint scheduling of URLLC and eMBB traffic in 5G wireless networks,

    A. Anand, “Joint scheduling of URLLC and eMBB traffic in 5G wireless networks,” IEEE/ACM Trans. Networking , vol. 28, no. 2, p. 14, Feb. 2020

  8. [7]

    Superposition- based URLLC traffic scheduling in 5G and beyond wireless networks,

    M. Almekhlafi, M. A. Arfaoui, C. Assi, and A. Ghrayeb, “Superposition- based URLLC traffic scheduling in 5G and beyond wireless networks,” IEEE Trans. Commun. , vol. 70, no. 9, pp. 6295–6309, Sep. 2022

Show all 65 references
  1. [8]

    Resource slicing for eMBB and URLLC services in radio access network using hierarchical deep learning,

    M. Setayesh, S. Bahrami, and V . W. Wong, “Resource slicing for eMBB and URLLC services in radio access network using hierarchical deep learning,” IEEE Trans. Wireless Commun. , vol. 21, no. 11, pp. 8950–8966, Nov. 2022

  2. [9]

    Resource allocation and slicing puncture in cellular networks with eMBB and URLLC terminals coexistence,

    Y . Zhao, X. Chi, L. Qian, Y . Zhu, and F. Hou, “Resource allocation and slicing puncture in cellular networks with eMBB and URLLC terminals coexistence,” IEEE Internet Things J. , vol. 9, no. 19, pp. 18 431–18 444, Oct. 2022

  3. [10]

    Temporal characterization and prediction of VR traffic: a network slicing use case,

    F. Chiariotti, M. Drago, P. Testolina, M. Lecci, A. Zanella, and M. Zorzi, “Temporal characterization and prediction of VR traffic: a network slicing use case,” IEEE Trans. Mob. Comput. , vol. 23, no. 5, pp. 3890–3908, May 2024

  4. [11]

    Service multiplexing and revenue maximization in sliced c-RAN incorporated with URLLC and multicast eMBB,

    J. Tang, B. Shim, and T. Q. S. Quek, “Service multiplexing and revenue maximization in sliced c-RAN incorporated with URLLC and multicast eMBB,” IEEE J. Select. Areas Commun. , vol. 37, no. 4, pp. 881–895, Feb. 2019

  5. [12]

    Multicast eMBB and bursty URLLC service multiplexing in a comp-enabled RAN,

    P. Yang, X. Xi, Y . Fu, T. Q. S. Quek, X. Cao, and D. Wu, “Multicast eMBB and bursty URLLC service multiplexing in a comp-enabled RAN,” IEEE Trans. Wireless Commun. , vol. 20, no. 5, pp. 3061–3077, May 2021

  6. [13]

    How should I orchestrate resources of my slices for bursty URLLC service provision?

    P. Yang, X. Xi, T. Q. S. Quek, J. Chen, X. Cao, and D. Wu, “How should I orchestrate resources of my slices for bursty URLLC service provision?” IEEE Trans. Commun., vol. 69, no. 2, pp. 1134–1146, Feb. 2021

  7. [14]

    RAN slicing for massive IoT and bursty URLLC service multiplexing: analysis and optimization,

    P. Yang, X. Xi, T. Q. S. Quek, J. Chen, X. Cao, and d. Wu, “RAN slicing for massive IoT and bursty URLLC service multiplexing: analysis and optimization,” IEEE Internet Things J. , vol. 8, no. 18, pp. 14 258–14 275, Sep. 2021

  8. [15]

    Resource allocation for multi-traffic in cross- modal communications,

    L. Wang, A. Yin, X. Jiang, M. Chen, K. Dev, N. M. Faseeh Qureshi, J. Yao, and B. Zheng, “Resource allocation for multi-traffic in cross- modal communications,” IEEE Trans. Netw. Serv. Manage. , vol. 20, no. 1, pp. 60–72, Mar. 2023

  9. [16]

    Resource Allocation in an Open RAN System Using Network Slicing,

    M. Karbalaee Motalleb, V . Shah-Mansouri, S. Parsaeefard, and O. L. Alcaraz L ´opez, “Resource Allocation in an Open RAN System Using Network Slicing,” IEEE Trans. Netw. Serv. Manage., vol. 20, no. 1, pp. 471–485, Mar. 2023

  10. [17]

    Two-level soft RAN slicing for customized services in 5G-and-beyond wireless communications,

    W. Shi, J. Li, P. Yang, Q. Ye, W. Zhuang, X. Shen, and X. Li, “Two-level soft RAN slicing for customized services in 5G-and-beyond wireless communications,” IEEE Trans. Ind. Inf. , vol. 18, no. 6, pp. 4169–4179, Jun. 2022

  11. [18]

    Dynamic network slicing and resource allocation for heterogeneous wireless services,

    J. Kwak, J. Moon, H.-W. Lee, and L. B. Le, “Dynamic network slicing and resource allocation for heterogeneous wireless services,” in 2017 IEEE 28th Annual International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC) . Montreal, QC: IEEE, Oct. 2017, pp. 1–5

  12. [19]

    Deep Reinforcement Learning for Optimization of RAN Slicing Relying on Control- and User-Plane Separation,

    H. Tu, L. Zhao, Y . Zhang, G. Zheng, C. Feng, S. Song, and K. Liang, “Deep Reinforcement Learning for Optimization of RAN Slicing Relying on Control- and User-Plane Separation,” IEEE Internet Things J., vol. 11, no. 5, pp. 8485–8498, Mar. 2024

  13. [20]

    Performance vs. Cost Tradeoff for Network Slicing in Open RAN: An Intelligent Hierarchical Algorithm for Flexible Utility-Control,

    G. Zhou, L. Zhao, G. Zheng, S. Song, and K.-C. Chen, “Performance vs. Cost Tradeoff for Network Slicing in Open RAN: An Intelligent Hierarchical Algorithm for Flexible Utility-Control,” IEEE Trans. Veh. Technol., vol. 73, no. 11, pp. 17 697–17 713, Nov. 2024

  14. [21]

    Intelligent radio access network slicing for service provisioning in 6G: a hierarchical deep reinforcement learning approach,

    J. Mei, X. Wang, K. Zheng, G. Boudreau, A. B. Sediq, and H. Abou- Zeid, “Intelligent radio access network slicing for service provisioning in 6G: a hierarchical deep reinforcement learning approach,” IEEE Trans. Commun., vol. 69, no. 9, pp. 6063–6078, Sep. 2021

  15. [22]

    On estimating the autoregressive coefficients of time-varying fading channels,

    J. Vinogradova, G. Fodor, and P. Hammarberg, “On estimating the autoregressive coefficients of time-varying fading channels,” in 2022 IEEE 95th Vehicular Technology Conference: (VTC2022-Spring) , Jun. 2022, pp. 1–5, iSSN: 2577-2465

  16. [23]

    A unified channel estimation framework for stationary and non-stationary fading environments,

    Q. Shi, Y . Liu, S. Zhang, S. Xu, and V . K. N. Lau, “A unified channel estimation framework for stationary and non-stationary fading environments,” IEEE Trans. Commun. , vol. 69, no. 7, pp. 4937–4952, Jul. 2021

  17. [24]

    Deep Reinforcement Learning for Scalable Dynamic Bandwidth Allocation in RAN Slicing With Highly Mobile Users,

    S. Choi, S. Choi, G. Lee, S.-G. Yoon, and S. Bahk, “Deep Reinforcement Learning for Scalable Dynamic Bandwidth Allocation in RAN Slicing With Highly Mobile Users,” IEEE Trans. Veh. Technol., vol. 73, no. 1, pp. 576–590, Jan. 2024

  18. [25]

    Feeling of presence maximization: mmWave-enabled virtual reality meets deep reinforcement learning,

    P. Yang, T. Q. S. Quek, J. Chen, C. You, and X. Cao, “Feeling of presence maximization: mmWave-enabled virtual reality meets deep reinforcement learning,” IEEE Trans. Wireless Commun., vol. 21, no. 11, pp. 10 005–10 019, Nov. 2022

  19. [26]

    laco: a latency-driven network slicing orchestration in beyond-5G networks,

    L. Zanzi, V . Sciancalepore, A. Garcia-Saavedra, H. D. Schotten, and X. Costa-P ´erez, “laco: a latency-driven network slicing orchestration in beyond-5G networks,” IEEE Trans. Wireless Commun. , vol. 20, no. 1, pp. 667–682, Jan. 2021

  20. [27]

    Learning With Side Information: Elastic Multi-Resource Control for the Open RAN,

    X. Zhang, J. Zuo, Z. Huang, Z. Zhou, X. Chen, and C. Joe-Wong, “Learning With Side Information: Elastic Multi-Resource Control for the Open RAN,” IEEE J. Sel. Areas Commun. , vol. 42, no. 2, pp. 295–309, Feb. 2024

  21. [28]

    Can terahertz provide high-rate reliable low-latency communications for wireless VR?

    C. Chaccour, M. N. Soorki, W. Saad, M. Bennis, and P. Popovski, “Can terahertz provide high-rate reliable low-latency communications for wireless VR?” IEEE Internet Things J. , vol. 9, no. 12, pp. 9712–9729, Jun. 2022

  22. [29]

    Online multi- user scheduling for XR transmissions with hard-latency constraint: performance analysis and practical design,

    X. Zhao, Y .-J. A. Zhang, M. Wang, X. Chen, and Y . Li, “Online multi- user scheduling for XR transmissions with hard-latency constraint: performance analysis and practical design,” IEEE Trans. Commun., pp. 1–1, 2024

  23. [30]

    Meta-reinforcement learning in non-stationary and dynamic environments,

    Z. Bing, D. Lerch, K. Huang, and A. Knoll, “Meta-reinforcement learning in non-stationary and dynamic environments,” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 45, no. 3, pp. 3476–3491, Mar. 2023

  24. [31]

    Little’s law,

    J. D. C. Little and S. C. Graves, “Little’s law,” in Building Intuition: Insights From Basic Operations Management Models and Principles , D. Chhajed and T. J. Lowe, Eds. Boston, MA: Springer US, 2008, pp. 81–100

  25. [32]

    Study on scenarios and requirements for next generation access technologies,

    “Study on scenarios and requirements for next generation access technologies,” Apr. 2022, the 3rd Generation Partnership Project, 3GPP document 38.801

  26. [33]

    NG-RAN; architecture description,

    3GPP, “NG-RAN; architecture description,” 3GPP, Technical Specifica- tion TS 38.401, Mar. 2021, release 16

  27. [34]

    Management and orchestration; concepts, use cases and require- ments,

    ——, “Management and orchestration; concepts, use cases and require- ments,” 3GPP, Technical Specification TS 28.530, Jun. 2023, release 18

  28. [35]

    Autoregressive modeling for fading channel simulation,

    K. Baddour and N. Beaulieu, “Autoregressive modeling for fading channel simulation,” IEEE Trans. Wireless Commun. , vol. 4, no. 4, pp. 1650–1662, Jul. 2005

  29. [36]

    Adaptive data-aided time-varying channel tracking for massive mimo systems,

    R. Chopra, C. R. Murthy, and K. Appaiah, “Adaptive data-aided time-varying channel tracking for massive mimo systems,” IEEE Trans. Commun., vol. 72, no. 9, pp. 5458–5472, Sep. 2024

  30. [37]

    J. D. Hamilton, Time series analysis . Princeton, N.J: Princeton University Press, 1994

  31. [38]

    A nonstationary poisson view of internet traffic,

    T. Karagiannis, M. Molle, M. Faloutsos, and A. Broido, “A nonstationary poisson view of internet traffic,” in IEEE INFOCOM 2004, vol. 3. Hong Kong, PR China: IEEE, 2004, pp. 1558–1569

  32. [39]

    Global optimization advances in mixed-integer nonlinear programming, minlp, and constrained derivative-free optimization, cdfo,

    F. Boukouvala, R. Misener, and C. A. Floudas, “Global optimization advances in mixed-integer nonlinear programming, minlp, and constrained derivative-free optimization, cdfo,” Eur. J. Oper. Res. , vol. 252, no. 3, pp. 701–727, Aug. 2016

  33. [40]

    Neely, Stochastic network optimization with application to com- munication and queueing systems , San Rafael, CA, USA: Morgan & Claypool, 2010

    M. Neely, Stochastic network optimization with application to com- munication and queueing systems , San Rafael, CA, USA: Morgan & Claypool, 2010

  34. [41]

    Dynamic programming,

    R. Bellman, “Dynamic programming,” Sci., vol. 153, no. 3731, pp. 4–37, 1966

  35. [42]

    The nonstochastic multiarmed bandit problem,

    P. Auer, N. Cesa-Bianchi, Y . Freund, and R. E. Schapire, “The nonstochastic multiarmed bandit problem,” SIAM J. Comput. , vol. 32, no. 1, pp. 48–77, Jan. 2002

  36. [43]

    S. P. Boyd and L. Vandenberghe, Convex optimization, version 29 ed. Cambridge New York Melbourne New Delhi Singapore: Cambridge University Press, 2023

  37. [44]

    Network slicing for service-oriented networks under resource constraints,

    N. Zhang, Y .-F. Liu, H. Farmanbar, T.-H. Chang, M. Hong, and Z.-Q. Luo, “Network slicing for service-oriented networks under resource constraints,” IEEE J. Select. Areas Commun. , vol. 35, no. 11, pp. 2512–2521, Nov. 2017. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 17

  38. [46]

    New York, NY: Wiley, 1987

    Anton and Howard, Elementary linear algebra, 5th ed. New York, NY: Wiley, 1987

  39. [47]

    Evaluation and analysis of the performance of the EXP3 algorithm in stochastic environments,

    Y . Seldin, C. Szepesv ´ari, P. Auer, and Y . Abbasi-Yadkori, “Evaluation and analysis of the performance of the EXP3 algorithm in stochastic environments,” in Proceedings of the Tenth European Workshop on Reinforcement Learning . PMLR, Jan. 2013, pp. 103–116, iSSN: 1938-7228

  40. [48]

    The non-stationary stochastic multi-armed bandit problem,

    R. Allesiardo, R. F ´eraud, and O.-A. Maillard, “The non-stationary stochastic multi-armed bandit problem,” International Journal of Data Science and Analytics , vol. 3, no. 4, pp. 267–283, Jun. 2017

  41. [49]

    Incorporation of time delayed measurements in a discrete-time kalman filter,

    T. Larsen, N. Andersen, O. Ravn, and N. Poulsen, “Incorporation of time delayed measurements in a discrete-time kalman filter,” in Proceedings of the 37th IEEE Conference on Decision and Control , vol. 4. Tampa, FL, USA: IEEE, 1998, pp. 3972–3977

  42. [50]

    Linear estimation,

    T.Kailauth, A. H. Sayed, and B. Hassibi, “Linear estimation,” IEEE Trans. Inf. Theory, vol. 51, no. 6, pp. 2236–2240, 2000

  43. [51]

    Extrapolation of delayed measurements for fusion in a distributed sensor network,

    R. A. J. Chagas and J. Waldmann, “Extrapolation of delayed measurements for fusion in a distributed sensor network,” in 2016 24th Mediterranean Conference on Control and Automation (MED) . Athens, Greece: IEEE, Jun. 2016, pp. 1319–1324

  44. [52]

    Sequential multi-hypothesis testing in multi-armed bandit problems: an approach for asymptotic optimality,

    G. R. Prabhu, S. Bhashyam, A. Gopalan, and R. Sundaresan, “Sequential multi-hypothesis testing in multi-armed bandit problems: an approach for asymptotic optimality,” IEEE Trans. Inf. Theory , vol. 68, no. 7, pp. 4790–4817, Jul. 2022

  45. [53]

    Study on channel model for frequencies from 0.5 to 100 ghz,

    3GPP, “Study on channel model for frequencies from 0.5 to 100 ghz,” Apr. 2024, the 3rd Generation Partnership Project, 3GPP document 38.901

  46. [54]

    System architecture for the 5G system (5gs),

    ——, “System architecture for the 5G system (5gs),” Jun. 2024, the 3rd Generation Partnership Project, 3GPP document 23.501

  47. [55]

    A contextual-bandit approach to personalized news article recommendation,

    L. Li, W. Chu, J. Langford, and R. E. Schapire, “A contextual-bandit approach to personalized news article recommendation,” in Proceedings of the 19th international conference on World wide web . Raleigh North Carolina USA: ACM, Apr. 2010, pp. 661–670

  48. [56]

    Stochastic multi-armed-bandit problem with non-stationary rewards,

    O. Besbes, Y . Gur, and A. Zeevi, “Stochastic multi-armed-bandit problem with non-stationary rewards,” in Advances in Neural Information Processing Systems , vol. 27. Curran Associates, Inc., 2014

  49. [57]

    Poulkov, Ed., Future Access Enablers for Ubiquitous and Intelligent Infrastructures, ser

    V . Poulkov, Ed., Future Access Enablers for Ubiquitous and Intelligent Infrastructures, ser. Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering. Cham: Springer International Publishing, 2019, vol. 283

  50. [58]

    Lyapunov Drift-Plus-Penalty Optimization for Queues With Finite Capacity,

    L. Bracciale and P. Loreti, “Lyapunov Drift-Plus-Penalty Optimization for Queues With Finite Capacity,” IEEE Commun. Lett., vol. 24, no. 11, pp. 2555–2558, Nov. 2020

  51. [59]

    Multiagent Reinforcement Learning: Methods, Trustworthiness, Applications in Intelligent Vehicles, and Challenges,

    Z. Zhou, G. Liu, and Y . Tang, “Multiagent Reinforcement Learning: Methods, Trustworthiness, Applications in Intelligent Vehicles, and Challenges,” IEEE Trans. Intell. Veh., pp. 1–23, 2024

  52. [60]

    A survey of admm variants for distributed optimization: problems, algorithms and features,

    Y . Yang, X. Guan, Q.-S. Jia, L. Yu, B. Xu, and C. J. Spanos, “A survey of admm variants for distributed optimization: problems, algorithms and features,” Aug. 2022, arXiv:2208.03700 [cs]

  53. [61]

    Delay-oriented qos-aware user association and resource allocation in heterogeneous cellular networks,

    X. Luo, “Delay-oriented qos-aware user association and resource allocation in heterogeneous cellular networks,” IEEE Trans. Wireless Commun., vol. 16, no. 3, pp. 1809–1822, Mar. 2017

  54. [62]

    A unified algorithmic framework for block-structured optimization involving big data: with applications in machine learning and signal processing,

    M. Hong, M. Razaviyayn, Z.-Q. Luo, and J.-S. Pang, “A unified algorithmic framework for block-structured optimization involving big data: with applications in machine learning and signal processing,” IEEE Signal Process Mag. , vol. 33, no. 1, pp. 57–77, Jan. 2016

  55. [63]

    Strang, Introduction to Linear Algebra, Sixth Edition (2023) , 2011

    G. Strang, Introduction to Linear Algebra, Sixth Edition (2023) , 2011

  56. [64]

    Multiservice loss models in C-RAN supporting compound poisson traffic,

    I.-A. Chousainov, I. Moscholios, P. Sarigiannidis, and M. Logothetis, “Multiservice loss models in C-RAN supporting compound poisson traffic,” Electronics, vol. 11, no. 5, p. 773, Mar. 2022

  57. [65]

    Proximal policy optimization algorithms,

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Proximal policy optimization algorithms,” Aug. 2017, arXiv:1707.06347 [cs]. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 18 Supplementary Materials We demonstrate this supplementary results in respon...

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.