{"id":"4f2cd077-4f87-4215-a4bd-72e66ce30779","arxiv_id":"2505.01295","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":3.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A survey of network-level integrated sensing and communication, covering cooperative architectures, stochastic-geometry performance analysis, distributed signaling, and over-the-air synchronization.","lead":"Wireless networks that both communicate and sense their surroundings usually treat each base station on its own. This paper reviews the alternative: many base stations working together, sharing signals and timing, to sense and communicate more effectively across an entire network.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Unproven convexity of the S-C ASE region in §III-B may invalidate the boundary-search optimization and the Fig. 4(a) trade-off conclusions.","rationale":"The reader's verdict is CONDITIONAL with high confidence, and my read does not change that verdict. However, I identify a more specific load-bearing gap than the reader's stated weakest_assumption. The reader's weakest_assumption concerns stochastic-geometry and offset-reciprocity modeling assumptions; the rationale separately mentions the convexity claim. I focus on the convexity claim because it is an internally asserted mathematical fact, located in Section III-B, that the paper uses to justify a proposed optimization procedure. The one-sentence statement 'It is not difficult to prove that the S-C performance boundary is a convex region' is an omitted proof, and the structure of the constraint set suggests the claim may be false. If the region is not convex, the boundary-search method for P1 is invalid, and the ASE trade-off results and Fig. 4(a) would not be supported. This concern is concrete and testable by small exhaustive enumeration. It does not undermine the survey's overarching claim that network-level ISAC is an emerging paradigm, since the survey could be corrected by proving the claim or softening it. But it is exactly the kind of unsubstantiated assertion that justifies a cautious, conditional verdict. I agree with the reader's CONDITIONAL verdict, so no verdict change is needed; my concern is partial agreement because it differs from the reader's primary weakest_assumption but appears in the reader's rationale.","tokens_in":25433,"tokens_out":4137,"duration_ms":43680,"concrete_test":"Construct a small counterexample to the convexity claim using the model behind Fig. 4(a) (or a simplified proxy with T_ASE_c = K·log2(1+SINR_c(K,L,J,Q)) and T_ASE_s = J·log2(1+SINR_s(K,L,J,Q))). Enumerate all feasible integer tuples satisfying KL+J(Q-1)≤Mt and J≤Jmax for a few parameter settings (e.g., Mt=8, Jmax=4). Plot the achievable (T_ASE_c, T_ASE_s) pairs and test whether the convex hull of these pairs contains points not achievable by any integer tuple; equivalently, check whether a weighted-sum maximization over the boundary yields an objective value exceeding the maximum over all feasible integer tuples. If such a non-achievable convex combination exists, the region is not convex and the proof claim fails. Alternatively, attempt the claimed proof starting from the bilinear constraint and identify the step where the argument breaks.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's Section III-B asserts, without proof, that the S-C performance boundary (region C_c-s in Eq. 19) is convex, and then justifies optimizing P1 by 'searching the ASE of the boundary point.' This assertion is load-bearing because it underpins the interference-management analysis and the trade-off trends in Fig. 4(a). The region is generated by discrete allocation variables (K, L, J, Q) with a bilinear spatial-DoF constraint KL+J(Q-1)≤Mt. Even under continuous relaxation, the feasible set with this bilinear constraint is nonconvex: for Mt=4, the points (K=4,L=1,J=1,Q=1) and (K=1,L=4,J=1,Q=1) both satisfy KL≤4, but their convex combination (2.5,2.5) violates it. Additionally, the mapping from allocation variables to ASE pairs involves SINR/rate expressions from stochastic geometry whose concavity in these discrete variables is not established. Therefore, the claim that the boundary is convex is unsupported and likely false. If convexity fails, the boundary-search method can miss the true optimum, and the Fig. 4(a) conclusion that cooperation gain declines under stricter thresholds (L_th, Q_th) is not guaranteed by the given analysis. This is an omitted proof of a central analytical step, not merely an external modeling concern.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper is a survey of network-level integrated sensing and communication (ISAC), the paradigm in which multiple base stations cooperate to provide sensing and communication services at scale. It reviews cooperation schemes and distributed architectures, presents stochastic-geometry-based performance analysis for interference management and cooperative ISAC, discusses distributed signaling designs for coherent and non-coherent operation, reviews over-the-air synchronization techniques for bi-static and distributed deployments, and lists open challenges. The paper draws substantially on the authors' own prior work for quantitative claims such as localization scaling laws and synchronization gains, and illustrates the main trade-offs with numerical examples.","tokens_in":25773,"tokens_out":7286,"duration_ms":77506,"significance":"If taken as a state-of-the-art overview, the paper is timely and useful. It provides a clear taxonomy of cooperation levels (Table I), a concise catalog of performance metrics (Table II), and a readable discussion of synchronization methods for distributed ISAC, an area that is often scattered across the literature. The paper also explicitly connects network-level ISAC to practical constraints such as backhaul, synchronization, and hardware. However, the survey's quantitative conclusions rest heavily on the authors' own prior results, and at least one load-bearing analytical claim (the convexity of the S-C performance boundary in Section III-B) is not supported by the text. There are also several technical inaccuracies in the mathematical presentation. The organizational value of the survey is real, but the manuscript needs revision before it can serve as a reliable reference.","major_comments":[{"comment":"The statement that \"the S-C performance boundary is a convex region\" is unsupported and, as stated, likely false. The feasible set defined by KL + J(Q-1) <= M_t with integer K, L, J, Q is discrete, and even the continuous relaxation is nonconvex because of the bilinear term KL. For example, with M_t = 4, the points (K,L,J,Q) = (4,1,1,1) and (1,4,1,1) both satisfy KL <= 4, but their convex combination (2.5,2.5,1,1) gives KL = 6.25 > 4. The mapping from allocation variables to the ASE pair is not shown to be concave, so the boundary-search method used to solve P1 and the trade-off trends in Fig. 4(a) are not justified by the given analysis. Please provide a proof under explicit assumptions, or replace the boundary-search argument with an exact or approximate method that does not rely on this convexity claim.","section":"Section III-B, Eq. (19) and P1 (21)"},{"comment":"The constraint J <= Jmax appears in the definition of the S-C ASE region and in the optimization problem, but the quantity Jmax is never defined anywhere in the manuscript. Without a definition, the problem formulation is incomplete and the reader cannot reproduce the optimization. Please define Jmax explicitly, or remove the constraint if it is not essential.","section":"Section III-B, Eq. (19) and (21c)"},{"comment":"The stated CRLB for non-coherent MIMO sensing, CRLB_nc = tr(F_nc^{-1}(Theta_nc)), is not the localization CRLB. Since Theta_nc includes the nuisance parameters b_R and b_I, the trace of the full inverse FIM also includes the variance of those nuisance parameters. The localization accuracy should be the trace of the appropriate sub-block of the inverse FIM, typically tr([F_nc^{-1}]_{1:2,1:2}) after ordering the position parameters first. Because this CRLB is used as the sensing objective in the signaling design problem (29), the expression should be corrected.","section":"Section IV-B, Eq. (26)"}],"minor_comments":[{"comment":"There is a dimension mismatch: S_c^n is K x L, so E[S_c^n (S_c^m)^H] is K x K and cannot equal 0_{M_t x M_t} or I_{M_t x M_t}. The identity matrix in (3) should be I_K, and the text should clarify that the condition S_c^1 = ... = S_c^N is an assumption about the transmitted data, not a consequence of the covariance expression.","section":"Section II-C, Eqs. (2)-(3)"},{"comment":"The expression for F_A is garbled: the leading factor \"cos^2 theta_i d_j^2 d_i^2\" appears to be missing a fraction bar or other grouping, and the indices and symbols in the matrix entries are not defined carefully. Please check this equation against the cited derivation and rewrite it unambiguously.","section":"Section III-A2, Eq. (13)"},{"comment":"The section heading \"Interfernece Management\" contains a typo and should read \"Interference Management.\"","section":"Section III-B heading"},{"comment":"There are minor prose issues: \"the optimal target localization necessitates a dedicated per-subcarrier signal design, .\" has a stray comma before the period, and some sentences in Section IV-B are run-ons. A careful copyedit would improve readability.","section":"Section IV-B/IV-C"},{"comment":"The scaling laws in Table III (1/ln N, 1/ln^2 N, and hybrid) are presented without explicitly listing the modeling assumptions under which they hold, such as Poisson point process BS locations, independent AOA and TOF measurement errors, and ideal synchronization. Since these laws are used to compare localization methods and to draw conclusions about cooperative gain, the assumptions should be stated in or near the table.","section":"Table III and Section III-C"},{"comment":"The offset reciprocity property is stated as an exact equality, but in practice it requires symmetric hardware and oscillator behavior between nodes. The manuscript should note that hardware asymmetries can break exact reciprocity, and that the cited method may need calibration in real deployments.","section":"Section V-B, Eqs. (33)-(34)"}],"recommendation":"major_revision","confidential_remarks":"The paper is primarily a survey and the quantitative claims rely heavily on the authors' own prior publications (e.g., scaling laws in [59], synchronization gains in [105]). This is not by itself a reason to reject, but the authors should be asked to clearly distinguish established textbook material from their own recent results, and to provide enough context for an independent reader to assess the latter. The convexity issue in Section III-B is the main technical concern; it should be fixed before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Best read as a map of the network-level ISAC subfield, not as a source of new results. The authors have pulled together a clean taxonomy (non-cooperative, interference management, decentralized cooperation, centralized cooperation), compare architectures in Table I, and give a genuinely useful tour of over-the-air synchronization methods, including their own offset-reciprocity work. If you need a quick orientation to this area, the paper works.\n\nThe main technical substance is a restatement of the authors' own prior stochastic-geometry results: the S-C ASE region, the CRLB scaling laws, and the synchronization gains. That is not a crime for a survey, but it means the paper's value is organizational, not evidential. There is no new theorem, data, or algorithm. The reader's conditional verdict seems right.\n\nThe soft spots are minor but real. The biggest is in Section III-B: the assertion that the S-C performance boundary is convex is simply stated (\"It is not difficult to prove...\") with no proof and no pointer to a proof in the cited papers. The feasible set in K,L,J,Q is discrete and the continuous relaxation of KL+J(Q-1)≤Mt is nonconvex; the stress-test example with (4,1) and (1,4) and midpoint (2.5,2.5) violating the constraint is correct. That does not automatically sink the survey, but it does mean the boundary-search optimization and the Fig. 4(a) trend are less solid than the text suggests. Either a citation to a proof in [28] or a one-paragraph convexification via time-sharing would fix it.\n\nAlso, Eq. (19) and (P1) use Jmax without defining it. The independence assumption between AOA and TOF errors (Section III-A2) and the offset-reciprocity assumptions (Section V-B) are repeated from prior work without much critical discussion. Those are worth flagging for readers.\n\nOn balance: a competent organizing survey. The taxonomy and synchronization review are the strongest parts; the scaling-law and interference-management sections are a faithful but uncritical summary of the authors' own papers. I'd send it to peer review and ask for the convexity claim to be either proved, cited, or softened. The paper should be accepted after a standard revision.","headline":"A useful organizing survey of network-level ISAC; the technical core is mostly the authors' own prior results, and one unproven convexity claim in Sec. III-B should be fixed before publication.","tokens_in":26215,"tokens_out":3284,"would_cite":true,"duration_ms":33874,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Network-level ISAC—cooperating base stations sharing signals and data—extends link-level joint sensing and communication to network scale, with gains gated by synchronization.","keywords":["integrated sensing and communication","network-level ISAC","distributed MIMO radar","coordinated beamforming","joint transmission CoMP","stochastic geometry","over-the-air synchronization","Cramér-Rao bound"],"falsifier":"Measure, in both directions, the time and frequency offsets between a pair of distributed ISAC nodes under non-line-of-sight and mobility conditions; if the measured offsets do not satisfy $\\Delta t_{n,m} \\approx -\\Delta t_{m,n}$ and $\\Delta f_{n,m} \\approx -\\Delta f_{m,n}$ at picosecond level, offset-reciprocity synchronization fails as described. Separately, collect base-station and target locations from a real deployment and compare the empirical localization error growth with the claimed $\\ln^2 N$ and $\\ln N$ curves; a systematic mismatch would show that the stochastic-geometry scaling laws do not transfer to practice.","tokens_in":25208,"feed_emoji":"📡","tokens_out":8176,"duration_ms":80428,"temperature":0.7,"pith_summary":"Integrated sensing and communication (ISAC) aims to let one wireless signal both carry data and perform radar sensing, and this survey argues that the capability reaches its full form only when many base stations cooperate instead of acting as isolated links. The paper's central thesis is that network-level ISAC—through coordinated beamforming, joint transmission, and distributed MIMO radar—relaxes per-node hardware demands, expands coverage, and turns many weak single-link observations into a large distributed aperture. It supports this with stochastic-geometry analyses in which localization error falls roughly as the square of the log of the number of cooperating nodes for time-of-flight and hybrid methods, while the practical bottleneck is synchronization: time and frequency offsets between nodes, if uncorrected, erase the cooperative gain. The review therefore organizes architectures, signaling strategies, and over-the-air synchronization methods around one design question: how much cooperation is worth its overhead in backhaul and clock alignment.","feed_headline":"Cooperative base stations beat single-link sensing — if synchronized","feed_subtitle":"Survey shows localization error falls as ln² N with cooperating nodes, and picosecond synchronization unlocks the gain.","key_machinery":"The analysis is carried by three connected objects. First, the signal model (Eq. (1)) splits each node's transmit signal into communication and sensing precoded components, whose correlation structure distinguishes coordinated beamforming (uncorrelated signals across nodes) from joint-transmission CoMP (identical signals across nodes); this determines whether inter-node interference is noise or constructive gain. Second, the stochastic-geometry CRLB framework (Eqs. (13)–(18)) turns random base-station locations into expected localization bounds, producing the $\\ln N$ and $\\ln^2 N$ scaling laws that quantify the value of cooperation. Third, offset reciprocity (Eqs. (33)–(34)) states that the time and frequency offsets between nodes $n$ and $m$ measured at $n$ are the negatives of those measured at $m$, the identity that lets over-the-air synchronization recover the cooperative gain without external references. These objects jointly define the paper's design spectrum: cooperation level sets the synchronization and backhaul price, while the CRLB and reciprocity results set the sensing payoff.","core_discovery":"The paper's central claim, stated on its own terms, is that network-level ISAC significantly extends link-level ISAC through distributed cooperation, enabling coordinated sensing and communication at an unprecedented scale. The quantitative core is a set of scaling results: with base stations distributed as a Poisson point process, the Cramér-Rao lower bound for target localization falls as $\\ln N$ for angle-of-arrival methods, $\\ln^2 N$ for time-of-flight methods, and $a\\ln^2 N + b\\ln N$ for hybrid methods, and a hybrid scheme is reported to reduce localization error to 1.3% of the angle-only error and 28.8% of the time-only error at optimal deployments. The other load-bearing result is offset reciprocity: two nodes sensing the same objects see time and frequency offsets that are equal in magnitude and opposite in sign, which allows distributed ISAC nodes to synchronize without dedicated reference signals, with super-resolution estimation reported at roughly 10 ps. On communication, the paper reports that cooperative interference nulling enlarges the achievable sensing-communication spectral-efficiency region and that coherent joint transmission converts inter-node interference into constructive signal power at the price of phase-level synchronization.","pith_inferences":["Extension beyond the paper: offset reciprocity could serve as a calibration-free synchronization primitive embedded in any pair of ISAC nodes, with the reciprocity error itself monitored as a network health metric.","Extension beyond the paper: the scaling-law comparison implies a deployment rule the paper does not state explicitly—under a fixed per-node power cap, AOA-based sensing favors a few concentrated sub-arrays, while TOF and hybrid sensing favor distributing antennas as thinly as possible.","Extension beyond the paper: the sensing-communication ASE region could be turned into a dynamic scheduler that selects the cooperation level per resource block based on instantaneous backhaul load and synchronization quality, a natural extension of the convex-boundary search described in Section III-B."],"forward_implications":["If the scaling results hold, operators can choose cooperation density deliberately: adding nodes improves localization but with diminishing returns, and hybrid AOA/TOF estimation extracts the most accuracy from a given cluster.","Per-node hardware can be simplified in networked ISAC because resolution and power combining come from spatial distribution rather than from large arrays or very high-rate converters at every node.","Coherent joint transmission and coherent sensing are the high-gain regimes, but they require phase-level synchronization; until that is solved, non-coherent cooperation is the robust fallback.","Backhaul capacity bounds cluster size through the constraint $R_c + eN \\leq C_{\\text{backhaul}}$, so cluster dimensioning becomes an explicit network design variable rather than an afterthought.","Synchronization quality separates regimes: with coarse spectral cross-correlation (about 250 ps), distributed ISAC may not beat a single-node array with equal total antennas; with super-resolution estimation (about 10 ps), it does."],"supporting_citations":[{"why":"Establishes the link-level ISAC baseline and the limitations (inter-cell interference, limited coverage, restricted resolution) that motivate network-level cooperation.","marker":"[6]"},{"why":"Supplies the stochastic-geometry framework and unified coverage/rate metrics used to quantify ISAC network performance.","marker":"[27]"},{"why":"Defines the spatial-DoF-constrained sensing-communication ASE region and interference-nulling analysis that underlie the trade-off results.","marker":"[28]"},{"why":"Provides cooperative ISAC scaling laws (CRLB order) and the backhaul constraint model for cluster sizing.","marker":"[29]"},{"why":"Source of the AOA/TOF/hybrid localization CRLB scaling laws and the antenna-to-BS allocation comparison.","marker":"[59]"},{"why":"Defines the non-coherent distributed MIMO sensing CRLB and the joint TOF-AOA signaling design problem.","marker":"[37]"},{"why":"Introduces offset reciprocity and super-resolution offset estimation that enable over-the-air synchronization in distributed ISAC.","marker":"[105]"},{"why":"Provides the spectral cross-correlation baseline (on-grid FFT-based offset estimation) against which super-resolution synchronization is compared.","marker":"[106]"}],"fun_headline_variants":["Cooperative ISAC: localization error falls as log-squared N","Hybrid ISAC cuts localization error to 1.3% of angle-only","Synchronization unlocks cooperative sensing gains","Offset reciprocity: ISAC nodes sync without pilots","Network-level ISAC: distributed cooperation beats single-link limits"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The quantitative conclusions rest on modeling assumptions—base-station and target locations follow random point processes, angle and distance measurement errors are independent, and offset reciprocity holds between real node pairs—and if any of these fail in a deployed network, the reported scaling laws and synchronization gains do not transfer.","fun_headline_variants_meta":{"raw":{"variants":["Cooperative ISAC: localization error falls as log-squared N","Hybrid ISAC cuts localization error to 1.3% of angle-only","Synchronization unlocks cooperative sensing gains","Offset reciprocity: ISAC nodes sync without pilots","Network-level ISAC: distributed cooperation beats single-link limits"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.002459,"raw_usage":{"total_tokens":9468,"prompt_tokens":994,"completion_tokens":8474,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":610,"completion_tokens_details":{"reasoning_tokens":8393}},"tokens_in":610,"tokens_out":8474,"duration_ms":60100,"temperature":1.0,"reasoning_tokens":8393,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T04:20:24.507612+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure, in both directions, the time and frequency offsets between a pair of distributed ISAC nodes under non-line-of-sight and mobility conditions; if the measured offsets do not satisfy $\\Delta t_{n,m} \\approx -\\Delta t_{m,n}$ and $\\Delta f_{n,m} \\approx -\\Delta f_{m,n}$ at picosecond level, offset-reciprocity synchronization fails as described. Separately, collect base-station and target locations from a real deployment and compare the empirical localization error growth with the claimed $\\ln^2 N$ and $\\ln N$ curves; a systematic mismatch would show that the stochastic-geometry scaling laws do not transfer to practice.","supporting_citations":[{"cited_title":"Signaling Design for Noncoherent Distributed Integrated Sensing and Communication Systems","cited_arxiv_id":"2501.18264","evidence_quote":"Defines the non-coherent distributed MIMO sensing CRLB and the joint TOF-AOA signaling design problem."},{"cited_title":"Over-the-air synchronization for coherent digital automotive radar networks,","cited_arxiv_id":null,"evidence_quote":"Provides the spectral cross-correlation baseline (on-grid FFT-based offset estimation) against which super-resolution synchronization is compared."}],"review_version":1}