{"id":"8d2d0bb6-6d13-47fd-86c9-a5c5737d9875","arxiv_id":"2501.13025","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":4,"one_line_summary":"A sensing-triggered URLLC scheme using dirty-paper coding is analyzed in the finite blocklength regime, yielding a rate-reliability-detection trade-off that the authors claim outperforms power-sharing and time-sharing.","lead":"This paper designs a wireless system where a base station's radar-like sensing of a target automatically triggers urgent low-latency messages, and uses dirty-paper coding to keep these urgent messages from slowing down regular broadband traffic. The authors derive mathematical limits on how fast regular traffic can go while urgent messages and target detection stay reliable, and claim their method beats simpler alternatives.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 3's binomial formula (95) assumes i.i.d. URLLC detection events across blocks, but the sensing-triggered arrival process is serially dependent and block-varying, so the central rate bound lacks a valid joint distribution.","rationale":"The reader's weakest_assumption identifies exactly the same load-bearing gap that I find: the binomial product formula in eq. (95) is not justified by the system model. My independent reading of the proof confirms that Pdet is a block-dependent marginal that itself depends on whether a URLLC was transmitted in the current block, so the events {b in Bdetect} are neither identically distributed nor independent without an additional assumption. This is not a minor technical nuisance: Lemma 3's Ce and Ve are averages over the distribution of Bdetect, and eq. (95) is the only step that gives that distribution. Therefore Theorem 1, the central rate-reliability-detection trade-off, does not follow from the stated assumptions. The numerical claims in Section V optimize this theorem, so they inherit the gap. I note secondary concerns (the asserted but unproved power constraint for the DPC-transmitted signal, and the loosely specified baseline schemes in Fig. 2), but they are not needed to support the rejection. I agree with the reader's verdict and do not propose changing it.","tokens_in":15238,"tokens_out":5011,"duration_ms":55121,"concrete_test":"Run an exact finite-eta simulation of a stripped-down version of the model with two target-detection probabilities: p0 when no URLLC is transmitted in the current block and p1 when one is, and with per-block URLLC detection/decoding probabilities q0 (false alarm) and q1 (correct detection). Build the Markov chain for A_b and the conditional distribution of Bdetect for eta=4, and compare P[Bdetect=Bdt] with eq. (95) using Pdet = P_A q1 + (1-P_A) q0, where P_A is the marginal arrival probability. If the exact joint distribution differs from the binomial product for any Bdt, then Lemma 3's Ce and Ve are invalid. Also test the special case where P_{b-1,D} is forced constant and independent of A_b; if eq. (95) then holds, the concern is localized and a conditional revision with that assumption stated would be appropriate.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step is Appendix C, eq. (95): P[Bdetect = Bdt] = Pdet^{|Bdt|}(1-Pdet)^{eta-|Bdt|}. This product form requires the events {b in Bdetect} to be independent and identically distributed across blocks. The paper's own system model contradicts this. A_b, the URLLC-arrival indicator in block b, has probability P_{b-1,D} (Section II), and P_{b,D} from Lemma 1 depends on whether b in Barrival: the factors kappa1 and kappa2 in (27)-(28) switch between 'b not in Barrival' and 'otherwise'. Hence A_{b+1} depends on A_b: transmitting a URLLC in block b changes the sensing signal, which changes the target detection probability in that same block, which then determines the next URLLC arrival. Moreover, P_{b-1,D} is indexed by b and depends on the block's sensing channel eigenvalues, so even the marginal Pdet defined in (92)-(94) is not constant across blocks. No conditioning or Markov-chain calculation is provided to justify replacing the joint law of Bdetect by a product of identical Bernoulli marginals. Since eq. (95) feeds directly into Ce, Ve, and the Q^{-1} inversion in Lemma 3 and Theorem 1, the claimed eMBB rate bound, and hence the numerical outperformance, rests on an unjustified independence assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a bi-static MIMO ISAC system in which a sensing receiver detects the presence of a target and, upon detection, triggers the transmission of a URLLC message over the next block. The base station employs dirty-paper coding to precancel sensing and eMBB interference, and the authors derive a finite-blocklength rate-reliability-detection trade-off. The central result, Theorem 1, formulates a constrained maximization of an upper bound on the eMBB rate, with constraints on URLLC decoding error and target detection probability. Numerical results compare the proposed DPC-based scheme against power-sharing and time-sharing baselines and claim significant outperformance.","tokens_in":15658,"tokens_out":8208,"duration_ms":80308,"significance":"If the derivations were sound, the paper would offer a useful finite-blocklength framework for a sensing-triggered URLLC traffic model, which is a nontrivial extension of earlier works that assume Bernoulli or deterministic URLLC arrivals. The random-coding structure and the explicit closed-form expressions for the detection probability, URLLC error probability, and eMBB rate bound are a strength. However, the central rate bound relies on an independence assumption that is contradicted by the paper's own system model, and additional load-bearing steps are asserted rather than proved. As a result, Theorem 1 and the numerical comparisons do not currently provide a reliable achievability statement.","major_comments":[{"comment":"The binomial formula P[Bdetect = Bdt] = Pdet^{|Bdt|}(1-Pdet)^{eta-|Bdt|} requires the events {b in Bdetect} to be independent and identically distributed across blocks with common probability Pdet. This is not consistent with the system model. In Section II, A_1 = 0 and P(A_b = 1) = P_{b-1,D} for b >= 2, while Lemma 1 shows that P_{b,D} depends on whether b is in Barrival through the factors kappa1 and kappa2 in Eqs. (27)-(29). Hence A_{b+1} depends on A_b through the sensing detection probability, creating serial dependence, and the marginal probability Pdet defined in Eqs. (92)-(94) is not block-invariant because P_{b-1,D} varies with b and with the sensing channel eigenvalues. No conditioning or Markov-chain calculation is provided to replace the product form. Since Ce, Ve, and Delta_e in Eqs. (41)-(48) are weighted by this product, the objective in Theorem 1 and the numerical outperformance claims are not justified.","section":"Appendix C, Eq. (95)"},{"comment":"The expression for P[Ee,1|Bdetect = Bdt] is not a valid product of probabilities over the intended events. For a block b not in Bdt and not in Barrival, the required event is that no false alarm occurs, whose probability is 1 - P[EU,3]. The numerator in Eq. (100) instead uses P[EU,1](1-P_{b-1,D}) for the non-detected blocks, where P[EU,1] is the missed-detection probability defined in Eq. (37) for blocks in which a URLLC was actually transmitted. Thus the second factor in the numerator applies the missed-detection probability to blocks without a URLLC arrival, which is not the event of interest. This directly affects Delta_e in Eq. (48) and therefore the Q^{-1} inversion leading to the rate bound in Eq. (109).","section":"Appendix C, Eq. (100)"},{"comment":"The admissible-power condition Tr(XX^H) <= nP is asserted at the start of Section III but never proved for the proposed encoding. For a block b not in Barrival, the transmitted signal is X_{b,j} = S^{(1)}_{b,j}(m',s) + (1-alpha_{s,1}) X^{(s,1)}_{b,j}; its squared norm contains the cross term 2(1-alpha_{s,1}) Re<S^{(1)}, X^{(s,1)}>. The DPC condition in Eq. (13) constrains only the norm of S^{(1)} - alpha_{s,1} X^{(s,1)} and does not control this cross term. A similar issue appears for URLLC blocks through Eqs. (16)-(18). Without a proof that the generated codewords satisfy the power constraint, the finite-blocklength achievability bounds and the numerical optimization over beta and alpha in Theorem 1 are not grounded.","section":"Section III-A, Eqs. (13)-(18)"},{"comment":"The paper states, 'For simplicity, we assume that at least one such a codeword exists' for the DPC encoding step. This is a non-trivial covering assumption: for every realization of the message and channel, there must exist an index s such that the precoded codeword falls in the required shell. The failure probability of this event under the random codebook construction is never bounded, and similarly for the index v in the URLLC encoding in Section III-B.2. Because the encoders in Definition 1 must be defined for every message, this assumption is part of the achievability argument and cannot be waived without quantification.","section":"Section III-B.1, paragraph after Eq. (13)"}],"minor_comments":[{"comment":"In the second term of the information density expression, the density f_{Y_{b,c}|S^{(2)}_b}(y_{b,c}|s^{(1)}_b) should condition on s^{(2)}_b, the codeword used in blocks b in Bdt, rather than on s^{(1)}_b.","section":"Section III-C.2, Eq. (25)"},{"comment":"The definition of Pdet is inconsistent between the main text and the appendix: Eq. (47) uses PU,2 as defined in Eq. (32), while Eq. (94) writes (1 - P_{b-1,D})(1 - (1 - tilde{epsilon}_{U,2})^{M_U M_v}) without the division by M_U M_v inside the exponent. Since Pdet enters Ce, Ve, and Delta_e, this discrepancy must be resolved.","section":"Appendix C, Eq. (94) and Lemma 3, Eq. (47)"},{"comment":"The phrase 'Our numerical analysis show' should be 'Our numerical analysis shows'.","section":"Abstract"},{"comment":"The numerical optimization procedure is described only verbally: it is not stated how the search over (alpha_u, beta_u) is performed, whether the constraints in Eqs. (49b)-(49c) are evaluated for every possible realization of Barrival or only for a stationary/averaged version, or what values are used for the auxiliary codebook sizes M_v and M_s. These details are needed to reproduce the figures.","section":"Section V"}],"recommendation":"reject","confidential_remarks":"The central achievability argument is not supported: the independence assumption in Eq. (95) is contradicted by the system model, and several other load-bearing steps (power constraint, shell-covering existence, and the conditional probability in Eq. (100)) are either unproved or incorrect. These are not local presentation issues; they invalidate the main theorem and the numerical comparisons. I recommend rejection, though a substantially revised version that models the Markov dependence of the arrival process and supplies rigorous power and shell arguments could be considered a new submission."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the paper's premise is attractive—let the sensing receiver's detection of a target trigger URLLC arrivals, then use DPC to keep both eMBB and sensing interference out of the URLLC stream—and the authors do a real service by putting this on a finite-blocklength footing. But the central rate bound in Lemma 3 assumes that URLLC detection events across blocks are i.i.d. Bernoulli with a common probability Pdet, and the system they define creates serial dependence: A_{b+1} is set by P_{b,D}, and P_{b,D} itself depends on whether a URLLC was transmitted in block b (through kappa1, kappa2). So equation (95) is not a valid joint distribution unless some extra 'sensing on separate resource' or 'fixed channel' assumption is added, and none is. Since (95) feeds directly into Ce, Ve, and the Q^{-1} inversion, Theorem 1 and the numerical outperformance are unsupported.\n\nWhat is genuinely new: the sensing-triggered arrival model (rather than deterministic or Bernoulli URLLC arrivals), and the two-tier DPC scheme that precancels sensing then eMBB interference for URLLC blocks. Lemma 1's LRT-based detection probability is derived carefully, and Lemma 2's threshold-metric bound is a reasonable finite-blocklength treatment. No fitting or circular parameter choices. Self-citations are to the authors' prior DPC/ISAC work and are relevant.\n\nSoft spots, in order: (1) the binomial independence issue above—this is load-bearing; (2) the DPC transmit power constraint is asserted, not proved; with three superposed codewords on power shells the total norm can exceed ℓP unless the DPC slack parameters are carefully chosen; that is fixable but currently a gap; (3) the numerical comparison to 'power-sharing' and 'time-sharing' gives no equations or constants, so the claimed significant outperformance is not reproducible as written. These are not fatal to the idea, but they are fatal to the claims as stated.\n\nWho this is for: people working on ISAC/URLLC coexistence and finite-blocklength analysis. It deserves a serious referee: the idea is timely and the technical apparatus is mostly standard, so a competent reviewer could quickly identify the needed fixes. My recommendation to the editor: send it out, but with a clear request to address the independence issue and the baselines.","headline":"Good premise, but the central rate bound assumes an independence the system model itself breaks.","tokens_in":16148,"tokens_out":2139,"would_cite":false,"duration_ms":22397,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94A15","94A40"],"pacs":[],"model":"deepseek-v4-flash","headline":"A sensing-triggered MIMO ISAC scheme with dirty-paper coding promises higher eMBB rates while meeting URLLC and detection constraints.","keywords":["integrated sensing and communication","URLLC","dirty-paper coding","finite blocklength","MIMO","rate-reliability-detection tradeoff","target detection","eMBB"],"falsifier":"Run a Monte Carlo simulation of the sensing-triggered protocol: generate target echoes under Lemma 1's channel model, apply the likelihood-ratio test with the same false-alarm threshold, and record the set of blocks in which a URLLC message is actually transmitted and decoded. If the empirical joint distribution of this set diverges from the product formula \\(P[\\mathcal{B}_{\\mathrm{detect}} = \\mathcal{B}_{dt}] = P_{\\mathrm{det}}^{|\\mathcal{B}_{dt}|}(1-P_{\\mathrm{det}})^{\\eta-|\\mathcal{B}_{dt}|}\\) used in Lemma 3, the eMBB rate bound in Theorem 1 is not the true rate of the protocol at finite blocklength.","tokens_in":15045,"feed_emoji":"📡","tokens_out":6223,"duration_ms":60045,"temperature":0.7,"pith_summary":"This paper tries to show that a single MIMO base station can handle three jobs at once—carrying high-rate eMBB traffic, delivering randomly arriving ultra-reliable low-latency (URLLC) messages, and sensing a target—without dedicating separate resources to each. The proposed system detects the target with a sensing receiver, and that detection itself triggers the next URLLC transmission, so no external arrival model is needed. The key claim is that dirty-paper coding can cancel the interference of both the sensing waveform and the in-flight eMBB stream from the URLLC signal, yielding a finite-blocklength rate–reliability–detection tradeoff (Theorem 1) in which the eMBB rate bound is maximized subject to URLLC error and target-detection constraints. Numerical results indicate this DPC-based scheme supports a higher eMBB rate than power-sharing or time-sharing while meeting the same URLLC and sensing requirements.","feed_headline":"Dirty-paper coding beats power- and time-sharing in ISAC-URLLC","feed_subtitle":"A MIMO ISAC design uses target detection to trigger ultra-reliable messages while keeping high eMBB rates.","key_machinery":"The load-bearing mechanism is a two-stage dirty-paper coding (DPC) construction. In a block with no URLLC message, DPC with parameter \\(\\alpha_{s,1}\\) precancels the sensing signal from the eMBB codebook. In a block carrying URLLC, the transmitter first precancels the sensing signal from the eMBB signal with \\(\\alpha_{s,2}\\), then precancels the combined eMBB-plus-sensing interference from the URLLC codeword with \\(\\alpha_u\\) (equations 13–18). This makes the URLLC message see an almost interference-free channel despite riding on top of eMBB and sensing. The companion machinery is finite-blocklength Gaussian approximation: the information densities are shown to converge to Gaussians with means and variances built from the eigenvalues of the channel Gram matrices, so error probabilities are handled with Q-functions and Berry–Esseen corrections rather than asymptotic capacity arguments.","core_discovery":"On its own terms, the paper establishes the rate-reliability-detection tradeoff for a bi-static MIMO ISAC system with sensing-triggered URLLC. The central result, Theorem 1 (equation 49), maximizes the upper bound on the eMBB rate, \\(R_e \\leq C_e - \\sqrt{V_e/n}\\,$Q^{{-1}}$(\\epsilon_e - \\Delta_e) - K_e \\frac{\\log\\,n}{n} - \\frac{\\log\\,M_s}{n}\\), over the DPC power-splitting parameters \\(\\beta_u, \\beta_{s,1}, \\beta_{s,2}\\) and \\(\\alpha_u, \\alpha_{s,1}, \\alpha_{s,2}\\), subject to per-block URLLC error probability at most \\(\\epsilon_U\\) and per-block detection probability at least \\(P_D\\). This bound is assembled from Lemma 1 (the detection probability expressed through generalized chi-square statistics of the echo), Lemma 2 (the URLLC error probability decomposed into missed-detection, misdecoding, and false-alarm events), and Lemma 3 (the eMBB rate bound conditioned on which blocks were detected as carrying URLLC). The claim that the scheme outperforms power-sharing and time-sharing is that this optimized bound lies strictly above the corresponding bounds of those schemes at the same constraints.","pith_inferences":["If block-level detection outcomes are serially correlated because a detection in block \\(b\\) changes the probability of a URLLC transmission in block \\(b+1\\), the product-form probability in Lemma 3 may overstate how often the UE's detection set matches the true arrival set; a Markov-chain extension could be tested against the paper's binomial formula.","The same trigger loop could be inverted: instead of sensing triggering URLLC, a URLLC arrival could be used as a sensing cue, turning the scheme into a communication-driven radar setup.","One can test DPC gain with imperfect channel state information by treating estimation error as additional interference to be precanceled; the present framework already supplies the SNR-like terms where such error would enter.","Because the sensing receiver and the base station are linked by an ideal backhaul in the paper, a practical latency budget would need to fold backhaul delay into the trigger decision, turning the current per-block tradeoff into a latency-aware scheduling law."],"forward_implications":["If Theorem 1 is correct, the same power budget can serve eMBB, URLLC, and sensing simultaneously, with no orthogonal resource split, so the eMBB rate is not cut by the sensing duty.","Tightening the URLLC reliability requirement \\(\\epsilon_U\\) lowers the maximal eMBB rate; the paper's numerical curves show the scheme absorbs this loss more gently than power- or time-sharing.","Raising the required detection probability \\(P_D\\) also lowers the eMBB rate; at a fixed \\(P_D\\) the DPC scheme holds a rate margin over both baselines.","Because URLLC transmissions are triggered by physical detection rather than by an assumed Bernoulli arrival process, the scheme extends to environments where URLLC events are deterministic or unpredictable.","The detector threshold can be set to meet a desired false-alarm probability while the DPC parameters are re-optimized per block, decoupling sensing robustness from communication reliability."],"supporting_citations":[{"why":"Supplies the dirty-paper coding technique for consecutive messages with heterogeneous decoding deadlines in the finite blocklength regime, the foundation for the URLLC precancellation.","marker":"[15]"},{"why":"Establishes the classic dirty-paper writing-on-dirty-paper result that makes precancellation of known interference capacity-optimal.","marker":"[16]"},{"why":"Provides the broadcast-channel-style joint communications and sensing coding framework used to generate dual-function signals.","marker":"[17]"},{"why":"Lays out integrated sensing and communication in the finite blocklength regime, the framework the paper extends to sensing-triggered URLLC.","marker":"[18]"},{"why":"Gives the likelihood-ratio detection theory used to define the target detection and false-alarm probabilities.","marker":"[19]"},{"why":"Supplies the Berry-Esseen central-limit-theorem result for information densities that yields the Gaussian approximations in Lemmas 2, 3, and 5.","marker":"[21]"},{"why":"Provides the finite-blocklength channel-coding bounds and the threshold decoding lemma used to control URLLC and eMBB error probabilities.","marker":"[22]"}],"fun_headline_variants":["Dirty-paper coding lifts eMBB rate in sensing-triggered URLLC","MIMO ISAC with DPC: Higher eMBB rate under URLLC and sensing","Target-triggered URLLC via dirty-paper coding in MIMO ISAC","DPC-based ISAC beats power and time sharing for URLLC","Sensing-triggered URLLC: Dirty-paper coding wins over sharing"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The derivation assumes that whether the UE thinks a URLLC message arrived in a block is an independent coin flip with one fixed probability, even though in the protocol a detection in one block changes how likely the next block carries a URLLC message.","fun_headline_variants_meta":{"raw":{"variants":["Dirty-paper coding lifts eMBB rate in sensing-triggered URLLC","MIMO ISAC with DPC: Higher eMBB rate under URLLC and sensing","Target-triggered URLLC via dirty-paper coding in MIMO ISAC","DPC-based ISAC beats power and time sharing for URLLC","Sensing-triggered URLLC: Dirty-paper coding wins over sharing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000234,"raw_usage":{"total_tokens":1561,"prompt_tokens":1075,"completion_tokens":486,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":691,"completion_tokens_details":{"reasoning_tokens":373}},"tokens_in":691,"tokens_out":486,"duration_ms":5444,"temperature":1.0,"reasoning_tokens":373,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T16:31:51.911199+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a Monte Carlo simulation of the sensing-triggered protocol: generate target echoes under Lemma 1's channel model, apply the likelihood-ratio test with the same false-alarm threshold, and record the set of blocks in which a URLLC message is actually transmitted and decoded. If the empirical joint distribution of this set diverges from the product formula \\(P[\\mathcal{B}_{\\mathrm{detect}} = \\mathcal{B}_{dt}] = P_{\\mathrm{det}}^{|\\mathcal{B}_{dt}|}(1-P_{\\mathrm{det}})^{\\eta-|\\mathcal{B}_{dt}|}\\) used in Lemma 3, the eMBB rate bound in Theorem 1 is not the true rate of the protocol at finite blocklength.","supporting_citations":[{"cited_title":"Dirty paper coding for con- secutive messages with heterogeneous decoding deadlines in the finite blocklength regime,","cited_arxiv_id":null,"evidence_quote":"Supplies the dirty-paper coding technique for consecutive messages with heterogeneous decoding deadlines in the finite blocklength regime, the foundation for the URLLC precancellation."},{"cited_title":"Writing on dirty paper (Corresp.),","cited_arxiv_id":null,"evidence_quote":"Establishes the classic dirty-paper writing-on-dirty-paper result that makes precancellation of known interference capacity-optimal."},{"cited_title":"A broadcast channel framework for joint communications and sensing-part II: Superposition coding,","cited_arxiv_id":null,"evidence_quote":"Provides the broadcast-channel-style joint communications and sensing coding framework used to generate dual-function signals."},{"cited_title":"Integrated sensing and communication in the finite blocklength regime,","cited_arxiv_id":null,"evidence_quote":"Lays out integrated sensing and communication in the finite blocklength regime, the framework the paper extends to sensing-triggered URLLC."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the likelihood-ratio detection theory used to define the target detection and false-alarm probabilities."},{"cited_title":"A Finite-Blocklength Perspective on Gaussian Multi-Access Channels","cited_arxiv_id":"1309.2343","evidence_quote":"Supplies the Berry-Esseen central-limit-theorem result for information densities that yields the Gaussian approximations in Lemmas 2, 3, and 5."},{"cited_title":"Channel coding rate in the finite blocklength regime,","cited_arxiv_id":null,"evidence_quote":"Provides the finite-blocklength channel-coding bounds and the threshold decoding lemma used to control URLLC and eMBB error probabilities."}],"review_version":1}