{"id":"983c3f71-446c-4c7c-b36d-1d3a8d7e8df2","arxiv_id":"2411.12672","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":1,"one_line_summary":"Lifted atomic norm minimization can jointly estimate radar targets and communication data in MIMO ISAC, with required samples set by dictionary coherence, but the main theorem has no proof in the paper.","lead":"An ISAC receiver uses lifted atomic norm minimization to estimate radar targets and communication symbols without a pilot link, with a theoretical sample-complexity bound that is stated but not proved. Its simulations show comparable performance to a pilot-aided benchmark across different dictionary matrices.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1, the paper's only sample-complexity guarantee, is stated without proof and even names 'problem (5)' as the recovery method although (5) is just the underdetermined measurement equation y = X(U); no derivation supports L4 ≥ CµKT log(10KT/δ).","rationale":"The reader's weakest assumption (SDR exactness) is one of several unproved steps in the theorem; my concern is broader and more basic: Theorem 1 is the sole source of the claimed sample-complexity scaling and it is not proved at all. I partially agree with the reader: the SDR issue is real, but it sits inside the missing proof. The article does provide simulations and phase-transition plots, which give some empirical support that the estimator works in selected settings, but those simulations do not validate the logarithmic scaling in K,T or the probability bound. There is no code or data release, so the numerical results are not independently reproducible as shipped. Since the theorem is asserted rather than derived, the paper is not suitable for acceptance; the reader's REJECT verdict should stand. I do not see grounds to move to a softer verdict, and no ad hominem is intended: the critique is purely that the main theoretical deliverable lacks a demonstration.","tokens_in":8200,"tokens_out":5025,"duration_ms":52744,"concrete_test":"Independently re-derive Theorem 1 from the atomic-norm program (8), not from the quoted statement: construct a feasible dual certificate for the atom set A in (6) under Assumptions 1–3 and show that the SDR in (15) attains the dual objective. As a numerical spot-check, for K=2, T=4, Hadamard D, and L=225, compare the SDR solution's feasibility in the dual (14) against a fine-grid discretization; if the SDR is not tight, the theorem's premise fails and the guarantee cannot be accepted.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is the theorem in Section II: with probability at least 1−δ, L4 ≥ CµKT log(10KT/δ) implies exact recovery of U. The theorem is asserted in a paragraph with no proof, no appendix, and no citation to a source where it is proved. The statement is also internally ambiguous: it says recovery is 'through problem (5)', but (5) is not a recovery problem; it is the linear observation model y = X(U). The actual estimator is the atomic-norm program (8), solved via the dual (9)/(14) and the SDR (15). The chain from (8) to (15) is entirely asserted: the paper never shows that the finite SDP (15) exactly represents the infinite-dimensional dual problem for the four-dimensional atom set A in (6), nor that a dual certificate exists under Assumptions 1–3 with the stated probability. Because Theorem 1 is the only quantitative support for the abstract's claims about sample complexity and dictionary coherence, the lack of proof is a load-bearing gap, not a stylistic omission. The simulations (Figs. 3–4) show only a fixed L=225 setting and phase transitions over L ∈ [120,170] for K,T up to 7; they do not test the predicted L4 scaling or the probability bound.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a lifted atomic norm minimization (LANM) receiver for a MIMO ISAC system that simultaneously estimates radar target parameters (AoD, AoA, delay, Doppler) and decodes communication symbols from the received signal without a pilot link. The transmit waveforms are assumed to lie in the range of a known dictionary matrix D, and the recovery problem is cast as atomic norm minimization over a four-dimensional parameter set; the dual problem is relaxed to an SDP. The paper states a sample-complexity theorem of the form L4 ≥ CµKT log(10KT/δ), discusses Gaussian, Hadamard, and Fourier dictionaries with different coherence parameters µ, and reports numerical NMSE/SER and phase-transition experiments. The main claimed contribution is the sample-complexity guarantee and the resulting dictionary-dependent trade-off in receiver complexity.","tokens_in":8491,"tokens_out":9161,"duration_ms":87525,"significance":"If the theorem and the SDR equivalence were established, the paper would be a useful contribution to the ISAC literature: it would provide an off-the-grid, pilot-free receiver with a quantitative sample-complexity bound that could be reduced by choosing low-coherence dictionaries. The paper does make a falsifiable qualitative prediction—Hadamard and Fourier dictionaries (µ=1) should outperform a Gaussian dictionary (µ=6logT)—and the simulations in Figs. 3–4 are consistent with that ordering, so the empirical section is not circular. The weakness is that the central theoretical claim is unproved, and the derivations connecting the atomic program to the implementable SDP are asserted rather than demonstrated. As it stands, the paper is an algorithmic proposal with illustrative simulations rather than a supported theoretical result.","major_comments":[{"comment":"Theorem 1 is the paper's only quantitative sample-complexity guarantee, but it is stated without proof and without a reference to a proof for this setting. The statement says recovery is 'through problem (5)', yet (5) is only the linear observation model y = X(U); the actual estimator is the atomic-norm program (8), whose dual is (9)/(14). No argument is given that a dual certificate exists under Assumptions 1–3, and the constant C is left unspecified. In addition, L is not defined before the theorem: the system model uses \\bar L for the sampled vector length, while L is later used as the number of observations in the simulations. Since the introduction claims that the paper proves the proportionality between the number of samples and CµKT log(10KT/δ), this missing proof is a load-bearing gap, not a stylistic omission.","section":"Section II, Theorem 1"},{"comment":"The conversion of the infinite-dimensional dual (14) into the SDR (15) is asserted with a reference to [29] but not derived. Exact equality between the SDP and the dual over [0,1)^4 is required for the solution of (15) to certify optimality of the atomic-norm minimizer in (8), and hence for Theorem 1 to apply to the algorithm actually implemented. The text does not specify the relaxation degrees, the dimensions of Q, or the conditions under which the sum-of-squares representation is exact for the four-dimensional atom set A in (6) coupled with the dictionary operator X. If (15) is a loose relaxation, the recovered U need not be the LANM solution, so the SDR exactness is a separate load-bearing premise.","section":"Section II, Eq. (15)"},{"comment":"The numerical section does not test the theorem. It shows NMSE/SER at L=225 and phase transitions for L in [120,170] with K,T up to 7, but no experiment varies L against the predicted L4 scaling, no experiment checks the probability bound 1−δ, and the unspecified constant C makes such a check impossible. The simulations are useful for demonstrating the qualitative dictionary ordering, but they cannot compensate for the absence of a proof of Theorem 1. The phase-transition study also uses only 20 Monte Carlo trials per setting, which is a small sample for a success-rate plot.","section":"Section III, Figs. 3–4"},{"comment":"The noisy dual program in (14) is stated as max_q ⟨q,y_w⟩_R − (σ/4)∥q∥_2 subject to ∥X*(q)∥*_A ≤ 1, but this does not follow from the constrained program (13) in the displayed form: the constraint ∥y_w−X(U)∥_2 ≤ σ^2 does not by itself produce the fixed penalty σ/4. Since the SDR (15) and all noisy simulations use this dual, the derivation should be provided or the program should be corrected. The inner product ⟨·,·⟩_R is also never defined.","section":"Section II, Eq. (14)"}],"minor_comments":[{"comment":"The affiliation line for the first author reads 'Tman Valiulahi'; this should be 'Iman Valiulahi'.","section":"Title page, author line"},{"comment":"The symbol τk is overloaded: τk is the delay parameter of target k and also the four-dimensional vector [θk,φk,τk,vk]^T. The atom index (r,s,l,k,1) also uses k both as a target index and as an array index, which makes the equation hard to parse.","section":"Section II, Eq. (3)"},{"comment":"The optimization in (16) contains the typo 'prpr,∀r', and the matrix \\tilde D_j is not defined; the role of the index j in the dictionary and the sum over j needs clarification.","section":"Section II, Eq. (16)"},{"comment":"The caption of Fig. 4 says 'for k = 2' although the horizontal axis already varies the number of targets, and the text refers to 'Figs. 4(a), 4(a), and 4(b)' while the panels are labeled (a), (b), (c). These cross-references should be corrected.","section":"Section III, Fig. 4"},{"comment":"The sentence 'we use results from trigonometric polynomial theory [29] to propose an semidefinite relaxation' has an article error ('an semidefinite'); more substantively, the exact roles of the zero-padding and the relaxation degrees s', r', l', k' are not explained.","section":"Section II, after Eq. (15)"}],"recommendation":"reject","confidential_remarks":"I see no evidence of intentional circularity; the main problem is the absence of a proof of the central theorem and of the SDR exactness. If the authors can supply complete proofs of Theorem 1 and of the equivalence in (15), the paper may be suitable for a revised submission, but in its current form the central claim is unsupported."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe paper is worth a look for the problem framing: it extends lifted atomic norm minimization (LANM) to a bistatic MIMO ISAC receiver where the transmit signal is encoded via a known dictionary, and it makes the sensible point that the dictionary coherence µ controls the sample complexity. The comparison of Gaussian, Hadamard, and Fourier dictionaries in the simulations is consistent with that intuition, and the four-dimensional target parameterization (AoD, AoA, delay, Doppler) is a genuine extension over the 1D LANM work it builds on.\n\nThe soft spot is central and it is exactly where the stress-test note lands. Theorem 1 in Section II states the sample-complexity bound L^4 ≥ CµKT log(10KT/δ) but no proof is given anywhere in the paper, and no reference is supplied for this specific MIMO/4D setting. The theorem even says recovery is “through problem (5)”, but (5) is just the observation model y = X(U); the actual estimator is problem (8) (atomic norm minimization) and its SDR (15). That is not a minor typo, because the paper never shows that the SDR exactly represents the infinite-dimensional dual problem for the atom set in (6), nor that a dual certificate exists under Assumptions 1–3 with the claimed probability. The entire theoretical contribution rests on this unproved chain, so the paper is not yet a complete article.\n\nThe numerics are also narrower than the claims. The NMSE/SER experiments use a fixed L=225, and the phase transitions sweep L from 120 to 170 with K,T up to 7. That is enough to show the qualitative ranking of dictionaries, but it does not test the L^4 scaling or the probability bound δ. No code or data are provided.\n\nWho is this for: readers working on ISAC receiver design and super-resolution who want a substantive baseline for dictionary-aware LANM. The coherence ranking is a useful design heuristic. But as a paper, the missing proof of the main theorem is a load-bearing gap, not a style issue.\n\nRecommendation: send it to review, because the problem is relevant and the idea has merit, but the referee should require a complete proof of Theorem 1 (or a clear citation to an existing proof that covers this case), and a correction of the recovery problem reference. Without that, reject.","headline":"A useful extension of LANM to MIMO ISAC with a coherence-based sample-complexity claim, but the main theorem is stated without proof and references the wrong problem.","tokens_in":9008,"tokens_out":3228,"would_cite":false,"duration_ms":31167,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94A12","90C22"],"pacs":[],"model":"deepseek-v4-flash","headline":"A lifted atomic norm minimization receiver can recover radar target parameters and communication data from the same received signal, without any pilot link, provided the transmit dictionary is incoherent and L^4 ≥ C µ K T log(10KT/δ).","keywords":["integrated sensing and communication","lifted atomic norm minimization","semidefinite relaxation","off-the-grid estimation","dictionary coherence","joint radar-communication estimation","super-resolution","MIMO radar"],"falsifier":"Construct a dictionary D with coherence µ=1 (e.g., Hadamard), draw targets satisfying the separation condition (11), and solve the SDR (15) for an instance with L just above the theorem's bound; if the recovered U differs from the true U by more than 1e-3 in relative Frobenius norm, or if the dual polynomial from (15) exceeds magnitude 1 at any off-grid point, the theorem's SDR-equivalence premise fails.","tokens_in":8019,"feed_emoji":"📡","tokens_out":6277,"duration_ms":58889,"temperature":0.7,"pith_summary":"This paper proposes a receiver for integrated sensing and communication (ISAC) systems that estimates radar target parameters and decodes communication data from the same received signal, without relying on a pilot or reference link. The key move is to model the transmitted probing signals as lying in a known low-dimensional subspace spanned by a dictionary matrix D, and to write the received data as a linear map of a sparse low-rank matrix U. The paper claims a theorem: if the dictionary has coherence µ and the target parameters are separated, then with probability at least 1−δ, the condition $L^{4}$ ≥ C µ K T log(10KT/δ) guarantees exact recovery of U, and hence of both target parameters and transmitted symbols. The infinite-dimensional dual is solved by a semidefinite relaxation, and simulations with Hadamard, Fourier, and Gaussian dictionaries confirm the coherence-based sample-complexity predictions. A sympathetic reader would care because this offers a path to pilot-free, off-the-grid joint sensing and communication with a tunable cost/complexity trade-off.","feed_headline":"Blind receiver recovers radar targets and data without pilots","feed_subtitle":"Lifted atomic norm minimization needs samples set by dictionary coherence, letting Hadamard and Fourier dictionaries beat Gaussian in tests.","key_machinery":"The central object is the lifted matrix U and its atomic-norm gauge. The received vector y is written as y = X(U), where the atoms are a(τ) v^H with τ = (θ, φ, τ, v) ranging over the 4D torus and v an arbitrary unit-norm vector; X is a linear operator that couples the dictionary D and the array responses. Recovery is performed by minimizing the atomic norm of U subject to the observation constraint. The paper's proof machinery is the dual problem (9)/(14): it constructs a dual polynomial whose norm certifies that the true U is the unique atomic-norm minimizer, and the feasibility condition for that certificate produces the $L^{4}$ sample bound. The semidefinite relaxation (15), built from Toeplitz matrices via sum-of-squares theory, converts the infinite-dimensional search into a finite convex program.","core_discovery":"The paper establishes a sample-complexity guarantee for blind joint radar-communication estimation in a bistatic MIMO ISAC system. It shows that when the unknown transmit waveforms xk are expressed as D hk through a known dictionary D with coherence µ (Assumption 1), and the target parameters (angle of departure, angle of arrival, delay, Doppler) are separated as in Assumption 3, the received signal y is a linear map of a lifted matrix U that is a sparse combination of rank-one atoms. Theorem 1 states that with probability at least 1−δ, the condition $L^{4}$ ≥ C µ K T log(10KT/δ) ensures U can be recovered by solving the atomic-norm minimization problem (5). Because U encodes both the target parameters and the hk, this recovery simultaneously localizes the K targets and decodes the communication symbols. The paper further proposes solving the infinite-dimensional dual via a semidefinite relaxation (15), and shows numerically that different dictionaries, with Hadamard and Fourier giving coherence µ=1, require fewer observations than a Gaussian dictionary with µ=6 log T.","pith_inferences":["Beyond the paper: if the SDR duality is exact, the same lifting trick should generalize to estimating additional continuous parameters as long as the separation condition is measured in the product metric; this can be tested by extending the Toeplitz constraints in (15) to higher-order tensors.","The theorem's sample bound is stated for the noiseless case; for the noisy model (12)-(15) the paper gives no explicit error rate, so a natural test is to check whether NMSE decays as σ² with a constant that depends only logarithmically on K and T.","Because the bound depends on µ but not on the dictionary's detailed structure, deterministic low-coherence dictionaries should achieve the same guarantee; verifying this would require removing the i.i.d. assumption in Assumption 1."],"forward_implications":["With Hadamard or continuous Fourier dictionaries (µ=1), the required observation count is minimized, and the receiver can cut antennas or sampling rate while keeping the same recovery guarantee.","The estimator removes the need for a direct transmitter-receiver reference link, avoiding the bandwidth waste and demodulation errors that pilots introduce.","The same LANM formulation handles AWGN through the denoised problem (13)-(15); simulations show NMSE and SER improve with SNR across all dictionaries.","Choosing a dictionary controls receiver complexity: larger coherence µ increases the required L, so in the paper's experiments the Gaussian dictionary performs worst."],"supporting_citations":[{"why":"Establishes the lifting-based convex approach to blind sparse spikes deconvolution that LANM builds on.","marker":"[25]"},{"why":"Applies off-the-grid joint channel and data estimation, a direct predecessor for noncoherent transmission.","marker":"[26]"},{"why":"Provides the separation-condition framework for super-resolution radar parameter recovery.","marker":"[23]"},{"why":"Extends super-resolution recovery to MIMO radar, the baseline the paper generalizes to joint data estimation.","marker":"[24]"},{"why":"Supplies the positive trigonometric polynomial theory used to derive the semidefinite relaxation (15).","marker":"[29]"},{"why":"Presents atomic norm minimization for MIMO radar, used as the simulation benchmark.","marker":"[20]"}],"fun_headline_variants":["Dictionary choice cuts samples for blind ISAC recovery","Hadamard and Fourier dictionaries beat Gaussian for blind ISAC","Low-coherence dictionaries slash samples for blind ISAC","Blind ISAC estimator's sample need tied to dictionary coherence","Dictionary coherence sets sample count for blind radar-comm recovery"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper assumes without proof that the semidefinite relaxation in (15) exactly represents the infinite-dimensional dual problem (14); if that relaxation is loose, the recovered matrix need not be the atomic-norm minimizer, and Theorem 1's guarantee no longer applies.","fun_headline_variants_meta":{"raw":{"variants":["Dictionary choice cuts samples for blind ISAC recovery","Hadamard and Fourier dictionaries beat Gaussian for blind ISAC","Low-coherence dictionaries slash samples for blind ISAC","Blind ISAC estimator's sample need tied to dictionary coherence","Dictionary coherence sets sample count for blind radar-comm recovery"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000941,"raw_usage":{"total_tokens":3995,"prompt_tokens":891,"completion_tokens":3104,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":507,"completion_tokens_details":{"reasoning_tokens":3024}},"tokens_in":507,"tokens_out":3104,"duration_ms":18999,"temperature":1.0,"reasoning_tokens":3024,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T17:16:04.901526+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct a dictionary D with coherence µ=1 (e.g., Hadamard), draw targets satisfying the separation condition (11), and solve the SDR (15) for an instance with L just above the theorem's bound; if the recovered U differs from the true U by more than 1e-3 in relative Frobenius norm, or if the dual polynomial from (15) exceeds magnitude 1 at any off-grid point, the theorem's SDR-equivalence premise fails.","supporting_citations":[{"cited_title":"Noncoherent ofdm transmission via off-the-grid joint channel and data estimation,","cited_arxiv_id":null,"evidence_quote":"Applies off-the-grid joint channel and data estimation, a direct predecessor for noncoherent transmission."},{"cited_title":"Super-resolution radar,","cited_arxiv_id":null,"evidence_quote":"Provides the separation-condition framework for super-resolution radar parameter recovery."},{"cited_title":"Super-resolution mimo radar,","cited_arxiv_id":null,"evidence_quote":"Extends super-resolution recovery to MIMO radar, the baseline the paper generalizes to joint data estimation."},{"cited_title":"Range-angle decoupling and estimation for fda-mimo radar via atomic norm minimization and accelerated proximal gradient,","cited_arxiv_id":null,"evidence_quote":"Presents atomic norm minimization for MIMO radar, used as the simulation benchmark."}],"review_version":1}