{"id":"a0f6c73a-909e-413a-8973-0cf562d631d0","arxiv_id":"2412.07245","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"An alternating optimization algorithm maximizes radar SCNR under target-direction uncertainty while respecting communication SINR constraints, using a single receive beamformer with a claimed convergence guarantee.","lead":"This paper designs transmit and receive beamformers for a combined radar and communication system when the target's direction is only known to be one of several possible angles. It reports a converging alternating optimization algorithm and shows that a single receive beamformer keeps radar detection performance almost unchanged under direction uncertainty.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Receive-beamformer 'global optimality' rests on an unverified rank-one relaxation; Toeplitz structure alone does not make (58)'s SDP tight.","rationale":"The reader's weakest assumption correctly identifies the receive-beamformer global optimality step as the load-bearing link in the convergence proof. My stress-test sharpens this: the paper's citation of [24] requires that the SDP relaxation of the rank-one constrained problem (58) be tight, i.e., that every Dinkelbach subproblem admit a rank-one optimal W. The Toeplitz property of B^l_{m,k} is true, but it does not imply that the extreme points of the relaxed feasible set (W⪰0, tr(W)=1, plus linear inequalities) are rank-one. A concrete finite-dimensional counterexample can be constructed with more target directions than receive antennas, where a mixed-state W (e.g., near the identity) achieves a higher max-min value than any single receive beamformer ww^H. If that gap exists, Dinkelbach applied to the relaxation returns an infeasible W, and inequality (61) is unsupported. This directly threatens the central claim that Algorithm 1 converges with guarantee. However, the paper's numerical contribution and the single-receive-beamformer idea remain plausible, and the rank-one tightness can be tested numerically for small instances. The reader's CONDITIONAL verdict is therefore appropriate; my concern does not move the verdict, but it should be an explicit condition for acceptance. I also note the paper's Section IV text itself flags the dependency on global optimality of the receive update, so the flaw is internal, not an external consensus dispute.","tokens_in":16106,"tokens_out":17967,"duration_ms":202364,"concrete_test":"Reproduce (58) with N_r=3, N_t=4, K=2, J=0 (no clutter), I=4 target directions such as θ = {-60°, -20°, 20°, 60°}, and fix transmit beamformers u_k so that the numerator coefficients |α_i|^2 Σ_k |a_t(θ_i)^H u_k|^2 are equal across i. Solve the SDP relaxation (drop rank(W)=1) and compute its optimal value. Separately solve the rank-one constrained problem (58) globally for N_r=3 by exhaustive or grid/random parameterization of w (4 real parameters) and compute its optimal value. If the SDP value exceeds the rank-one optimum, the Toeplitz-Dinkelbach argument is invalid for this class; if they match, the concern is refuted. Report both values and the rank of the SDP solution.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The convergence proof's key inequality (61) requires that solving (58) for fixed transmit beamformers yields a globally optimal receive beamformer. That assertion is supported only by the claim in Section III-B that every B^l_{m,k} is Toeplitz, so (58) becomes a 'generalized fractional program with Toeplitz quadratics' solvable by Dinkelbach via [24]. Even conceding that each B^l_{m,k} is Toeplitz, the argument omits a necessary step: Dinkelbach's method requires each parametric subproblem over W to have a rank-one optimal solution. Toeplitzness of the coefficient matrices does not by itself guarantee this. For instance, with N_r=3 and four target directions, the SDP relaxation of (58) (dropping rank(W)=1) can have a strictly larger value than the rank-one constrained maximum, because the feasible W is not constrained to be Toeplitz and a mixed state W can dominate every pure ww^H in the max-min objective. If such a gap occurs for finite arrays used in the paper, the Dinkelbach 'solution' returned by the SDP is not feasible for (58), so the nondecreasing chain (61) is not established and the claimed convergence guarantee for Algorithm 1 collapses. The numerical convergence in Section V-B may still occur for the tested instances, but it does not validate the theoretical claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers a dual-functional radar and communication (DFRC) system in which the target arrival direction is not exactly known, but is known to belong to a discrete set of possible angles. The authors formulate a max-min radar SCNR problem subject to per-user SINR constraints and a total power constraint, and propose an alternating optimization algorithm: transmit beamformers are updated by a penalty-based successive convex approximation step, while the receive beamformer is updated by solving a fractional quadratic problem that they claim is a generalized fractional program with Toeplitz quadratics, solvable globally via Dinkelbach's method. The paper claims that the objective value is nondecreasing across iterations and hence that the algorithm converges, and presents numerical results showing convergence in about three iterations and robustness of the achieved SCNR to the number of possible target directions.","tokens_in":16372,"tokens_out":12771,"duration_ms":152593,"significance":"The problem is relevant and timely: target-direction uncertainty is a genuine limitation of prior DFRC beamforming designs, and the use of a single receive beamformer is an appealing complexity reduction relative to approaches that use one receive beamformer per candidate direction. The paper provides a clear problem formulation, a plausible algorithmic structure, and numerical experiments covering antenna counts, SINR thresholds, angular spread, and a comparison with dedicated receive beamformers. If the convergence and optimality claims can be rigorously supported, the contribution would be a useful design tool for DFRC receivers. However, as written, the central theoretical guarantee rests on two unproved steps: global optimality of the receive update and monotonicity of the true SCNR objective, rather than merely of a penalized auxiliary objective. These are load-bearing issues that prevent the convergence claim from being accepted as established.","major_comments":[{"comment":"The global optimality of the receive-beamformer update is not established. The paper argues that every B^l_{m,k} is Toeplitz and then invokes [24] to conclude that (58) is a generalized fractional program with Toeplitz quadratics for which Dinkelbach's algorithm gives a global solution. But the monotonicity chain (61) requires that, at each outer iteration, the update of w globally solves the rank-one constrained problem (58). Toeplitzness of the coefficient matrices alone does not imply that the SDP relaxation obtained by dropping rank(W)=1 has a rank-one optimal solution: the relaxed problem maximizes a concave function (a minimum of linear forms) over the spectrahedron {W ⪰ 0, Tr(W)=1}, and such maxima can occur at non-rank-one points. The paper should identify the specific result in [24], verify that its hypotheses hold for the finite-dimensional ULA steering-vector matrices used here, and prove the required rank-one tightness, or otherwise weaken the convergence claim accordingly.","section":"III-B, Eq. (58)"},{"comment":"The convergence proof tracks an auxiliary penalized objective rather than the SCNR objective of P1. The transmit update solves (46), which maximizes a surrogate objective that includes the penalty term -νb and uses surrogate SINR and power constraints, while χ in (59) is defined without -νb and is not shown to equal the true max-min SCNR min_i γ_r(θ_i, w, u) at the iterates. Consequently, showing u^H_{d,s} \\hat R u_{d,s} ≥ u^H_{d,s-1} \\hat R u_{d,s-1} does not by itself imply that the original max-min SCNR is nondecreasing. The authors need to supply the missing chain from the penalized objective to the true P1 objective, or state the convergence theorem as one about the penalized problem only.","section":"IV, Eqs. (59)-(62)"},{"comment":"The power constraint is enforced only asymptotically in the penalty parameter ν. Problem (46) replaces u^H u = P by u^H u ≤ P + b and 2Re(u^H u_0) - ||u_0||^2 ≥ P - b, with the objective penalized by -νb. For a fixed finite ν chosen at initialization, the optimal b need not be zero, so the returned u may violate the power budget and hence be infeasible for P1. The statement that b* → 0 for very large ν requires proof, and the algorithm as given has no rule for increasing ν or for certifying feasibility. An increasing penalty schedule with a feasibility test, or an exact penalty reformulation, is needed to justify the inner-loop updates used in the convergence argument.","section":"III-A, Eq. (46)"}],"minor_comments":[{"comment":"The notation is inconsistent: the terms in the numerator and denominator use α_i and α_j without the superscripts T and C introduced earlier, and the magnitudes should be written consistently as |α|^2. Please correct the notation in (9)-(10).","section":"Eq. (10)"},{"comment":"The transition from Eq. (10), where the noise term is σ_r w^H w, to Eq. (56), where the denominator contains I instead of σ_r I, silently assumes σ_r = 1. This normalization should be stated explicitly in the system model, since the numerical results depend on the noise scaling.","section":"II-A to III-B"},{"comment":"The name Dinkelbach is repeatedly misspelled as \"Dinkleback\"; please correct this in the abstract, body, and references.","section":"Throughout"},{"comment":"The algorithm initializes S, dmax, η, ν, µ, and ǫ, but Table 1 does not list their values, and the inner-loop count S is not described. Reporting these parameters and the stopping tolerance is necessary for reproducibility.","section":"Algorithm 1"},{"comment":"The footnote states that for a very large value of η the penalty term becomes zero, but for any finite η the penalty term is not exactly zero. The equivalence between (31) and (32) is therefore approximate, and the nature of this approximation should be stated explicitly.","section":"Footnote 2, after Eq. (33)"},{"comment":"The proof of Lemma 1 is abbreviated: the sentence \"Since the objective value does not decrease with each \\bar γ_r(θ_i)\" is unclear, and the proof should explicitly show that every individual SCNR term is nondecreasing in the common scaling of all transmit beamformers.","section":"Lemma 1"}],"recommendation":"major_revision","confidential_remarks":"The central concern is whether the result of [24] actually covers the rank-one relaxation in (58); I would recommend that the revised manuscript be checked by someone familiar with Toeplitz SDP tightness. The self-citations [10] and [17] are legitimate precursors and do not by themselves raise a circularity concern. The numerical study is useful, but it does not compensate for the missing proof steps in the convergence argument."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know: this is a legitimate extension of the DFRC beamforming literature—target direction uncertainty, max-min SCNR, and a single receive beamformer—and the numerics suggest the approach works. But the advertised convergence guarantee is not actually proven. The load-bearing step is the claim that the receive beamformer update in (58) is globally optimal because the coefficient matrices are Toeplitz, citing [24]. Toeplitzness alone does not make the rank-one relaxation tight. Dinkelbach needs each parametric subproblem to have a rank-one optimal solution; a mixed-state W can beat every pure ww^H in the max-min objective for the finite arrays used here. If that happens, the \"solution\" returned by the SDP is not feasible for (58), and the monotonicity chain in (61) collapses. The paper offers no argument for why the relaxation is tight in this setting. That is a real gap in the central theoretical claim.\n\nWhat is genuinely new: the problem formulation with uncertain target direction and a single receive beamformer, which avoids the multiple-beamformer complexity of [18]. The penalty-based SCA/MM transmit update is standard but sensibly assembled, and the numerical study does show SCNR stays flat as I grows, which is a useful practical takeaway. The paper is honest about its precursor [17], and the self-citations are background, not circular.\n\nWhere it is soft, in proportion: (1) The convergence proof, as above, is incomplete; the numerical convergence in Fig. 3 is the actual evidence, and it shows 3 iterations only for the tested cases. (2) No baseline comparisons—no runtime, no comparison to [18]'s scheme or a lower-bound scheme. The claim that a single beamformer is \"almost identical\" to dedicated beamformers is supported only by one beampattern plot and a linear-scale SCNR figure. (3) Notational inconsistencies (e.g., N_t vs N, the undefined N_r in (27), the garbled constraint in (45)) will slow any reader. (4) No code, so reproducibility is limited.\n\nBottom line: the idea is plausible and the practical conclusion is probably right, but the proof overreaches. This deserves a serious referee—conditional accept with a request to fix the proof or soften the claim, add baselines, and clean up the notation. I would bring it to a reading group if anyone cares about DFRC, but I would not cite the convergence guarantee as established.","headline":"Useful DFRC extension with a genuine proof gap: the claimed convergence relies on an unverified rank-one relaxation in the receive update.","tokens_in":16916,"tokens_out":2066,"would_cite":true,"duration_ms":77129,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"An alternating optimization algorithm maximizes worst-case radar SCNR under target-direction uncertainty with provable convergence, using a single receive beamformer.","keywords":["dual-function radar and communication","integrated sensing and communication","target direction uncertainty","signal-to-clutter-plus-noise ratio","beamforming optimization","alternating optimization","Dinkelbach algorithm","majorization-minimization"],"falsifier":"Solve the receive subproblem (58) for finite arrays with the paper's parameter table, reconstructing the optimal $W$ from Dinkelbach's solution, and check whether $\\operatorname{rank}(W)=1$; a case with rank greater than one would break the claimed global optimality and the monotonicity chain (61), and a subsequent decrease in the objective across an outer iteration would directly falsify the convergence claim.","tokens_in":15879,"feed_emoji":"📡","tokens_out":6165,"duration_ms":62959,"temperature":0.7,"pith_summary":"This paper tackles the realistic case of a dual-function radar-communication (DFRC) system where the target's direction is known only as a set of possible angles. It formulates a max-min optimization problem that maximizes the worst-case radar signal-to-clutter-plus-noise ratio (SCNR) while keeping every communication user's SINR above a required threshold. The authors propose an alternating algorithm: transmit beamformers are updated with a penalty-based majorization-minimization step, and a single receive beamformer is updated globally via Dinkelbach's method for generalized fractional programs. They prove convergence by showing the objective is nondecreasing across iterations, and their numerics indicate convergence in about three iterations with SCNR almost unchanged as the number of candidate target directions grows. The practical payoff is that one receive beamformer can replace a bank of direction-specific beamformers without meaningful detection loss.","feed_headline":"Single receive beamformer handles uncertain target direction","feed_subtitle":"The scheme converges in about three iterations and keeps radar SCNR almost flat as target-angle candidates multiply.","key_machinery":"The mechanism that carries the argument is the reduction of the receive-beamformer subproblem to a generalized fractional program with Toeplitz quadratics. A Toeplitz matrix has constant entries along each diagonal, and each $B^l_{m,k}=A(\\theta^l_m)u_k u_k^H A^H(\\theta^l_m)$ has this structure, so problem (58) with its rank-one constraint can be solved globally by Dinkelbach's algorithm. On the transmit side, the machinery is a penalty-based majorization-minimization update: the nonconvex power equality is relaxed with a slack variable and penalty, and the quadratic objective and SINR constraints are replaced by first-order surrogate functions that are tight at the current iterate, yielding a convex problem. Auxiliary rotation matrices $Q_i$ are updated in closed form by aligning the clutter-whitened vector with the target-whitened vector. Alternating these updates produces a nondecreasing objective sequence, which is the convergence argument.","core_discovery":"The central claim is that the nonconvex problem P1, which maximizes the minimum SCNR over a finite set of possible target directions subject to per-user SINR constraints, a total power budget, and unit-norm receive combining, can be solved by alternating optimization with a provable convergence guarantee. For a fixed receive beamformer, the transmit subproblem is rendered convex through a penalty reformulation and successive convex approximation; for fixed transmit beamformers, the receive subproblem is shown to be a generalized fractional program whose matrices $B^l_{m,k}$ are Toeplitz, so Dinkelbach's algorithm yields a global optimum. The paper argues that because each subproblem is solved optimally or with a valid surrogate, the achieved objective value never decreases, and since it is bounded, Algorithm 1 converges. The numerical evidence further claims that one receive beamformer suffices: SCNR is almost flat in the number $I$ of possible target directions, and the receive beampattern is nearly identical for different angular spreads, so only the two extreme angles of the spread are needed for design.","pith_inferences":["An implicit consequence is that the algorithm's per-iteration cost is largely independent of the number of candidate target directions, since the receive beamformer is updated once per outer iteration; this could make the method attractive for tracking scenarios where the uncertainty set is refined over time.","The same Toeplitz-based global-solution step could be applied to other max-min beamforming problems in integrated sensing and communication, such as worst-case localization under angle uncertainty, by replacing SCNR with an estimation-theoretic objective.","Because the two extreme angles appear to determine the receive beampattern, a continuous uncertainty interval could be handled by sampling its endpoints, which is a testable bridge between the paper's discrete model and a continuous angular uncertainty set."],"forward_implications":["A DFRC base station can use a single receive beamformer for target detection even when the target angle is uncertain, reducing receiver hardware and signal-processing load compared with dedicated beamformers per candidate direction.","Designing the receive beamformer from the two extreme angles of the target's angular spread is enough; adding more candidate directions inside the spread barely changes SCNR.","Radar detection performance degrades gracefully as communication SINR requirements tighten, with larger antenna arrays absorbing most of the loss.","The convergence guarantee makes the algorithm suitable for online adaptation when the set of possible target directions changes over time."],"supporting_citations":[{"why":"Supplies the joint radar-communication system model and the observation that the matrices $B^l_{m,k}$ are Toeplitz, which underpins the global receive-beamformer solution.","marker":"[10]"},{"why":"The prior DFRC transceiver design that uses as many receive beamformers as target directions; the paper's single-beamformer scheme is positioned against this baseline.","marker":"[18]"},{"why":"The earlier iterative waveform-design approach whose monotonicity failure motivates the new alternating scheme and its convergence guarantee.","marker":"[19]"},{"why":"Provides the closed-form update for the auxiliary rotation matrices $Q_i$ used in the transmit-side penalty reformulation.","marker":"[20]"},{"why":"Supplies the majorization-minimization surrogate-function properties used to convexify the transmit beamforming subproblem.","marker":"[21]"},{"why":"Establishes that generalized fractional programs with Toeplitz quadratics can be solved globally by Dinkelbach's algorithm, the key step enabling global optimality for the receive beamformer.","marker":"[24]"}],"fun_headline_variants":["One receive beamformer handles target angle uncertainty","Three iterations conquer direction uncertainty in DFRC","Radar SCNR stays flat as target directions multiply","Single beamformer robust to uncertain radar directions","DFRC solves uncertain angle beamforming in 3 steps"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The convergence guarantee depends on the receive subproblem always being solved globally, which the paper obtains from the Toeplitz structure and Dinkelbach's method; if that global-solution or rank-one claim fails for a particular finite array and noise normalization, the monotonicity proof no longer applies.","fun_headline_variants_meta":{"raw":{"variants":["One receive beamformer handles target angle uncertainty","Three iterations conquer direction uncertainty in DFRC","Radar SCNR stays flat as target directions multiply","Single beamformer robust to uncertain radar directions","DFRC solves uncertain angle beamforming in 3 steps"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000272,"raw_usage":{"total_tokens":1658,"prompt_tokens":995,"completion_tokens":663,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":611,"completion_tokens_details":{"reasoning_tokens":590}},"tokens_in":611,"tokens_out":663,"duration_ms":7925,"temperature":1.0,"reasoning_tokens":590,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T18:57:02.432526+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Solve the receive subproblem (58) for finite arrays with the paper's parameter table, reconstructing the optimal $W$ from Dinkelbach's solution, and check whether $\\operatorname{rank}(W)=1$; a case with rank greater than one would break the claimed global optimality and the monotonicity chain (61), and a subsequent decrease in the objective across an outer iteration would directly falsify the convergence claim.","supporting_citations":[{"cited_title":"Joint optimization of radar and communications performan ce in 6G cellular systems,","cited_arxiv_id":null,"evidence_quote":"Supplies the joint radar-communication system model and the observation that the matrices $B^l_{m,k}$ are Toeplitz, which underpins the global receive-beamformer solution."},{"cited_title":"Efficient transcei ver design for MIMO dual-function radar-communication systems,","cited_arxiv_id":null,"evidence_quote":"The prior DFRC transceiver design that uses as many receive beamformers as target directions; the paper's single-beamformer scheme is positioned against this baseline."},{"cited_title":"MIMO radar waveform de sign with constant modulus and similarity constraints,","cited_arxiv_id":null,"evidence_quote":"The earlier iterative waveform-design approach whose monotonicity failure motivates the new alternating scheme and its convergence guarantee."},{"cited_title":"Grab-n-Pull: A max-min fractional quadratic programming framework with applications in signal and information proc essing,","cited_arxiv_id":null,"evidence_quote":"Provides the closed-form update for the auxiliary rotation matrices $Q_i$ used in the transmit-side penalty reformulation."},{"cited_title":"New results on g eneralized fractional programming problems with Toeplitz quadratics ,","cited_arxiv_id":null,"evidence_quote":"Establishes that generalized fractional programs with Toeplitz quadratics can be solved globally by Dinkelbach's algorithm, the key step enabling global optimality for the receive beamformer."}],"review_version":1}