{"id":"3fc3314a-4de9-4ac0-bc3b-ace50c2f165e","arxiv_id":"2505.20012","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"The authors propose a joint optimization framework for hybrid RIS-aided integrated sensing and communication with imperfect target location, and simulation shows improved worst-case sensing SINR over passive and no-RIS baselines.","lead":"This paper designs a wireless system where a base station communicates with users and senses a target at the same time, assisted by smart surfaces that mix active and passive elements. It optimizes the transmitted signal and the surface settings to maximize sensing quality while keeping messages secure from an eavesdropping target.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"GFP optimality claim unsupported: Lemma III.1 requires concave fp, but the sensing SINR numerator in (26a) is convex in x and quartic in Θ; Algorithm 3's convergence to the optimum is not guaranteed.","rationale":"The reader's weakest-assumption is essentially correct. The paper invokes Lemma III.1 to claim that problem (26) belongs to GFP and that the iterative algorithm monotonically converges to the optimal value. The lemma's hypotheses fail for (26) because the sensing numerator is convex in x and quartic in Θ. The subsequent reformulation in (30) replaces the squared modulus with an affine term plus convex quadratic penalty, which is a valid lower-bound maximization only when the receive filter satisfies the MVDR alignment Re(w_p^H A_p x)=|w_p^H A_p x|; even then, the joint problem over Θ remains nonconvex and outside the lemma's scope. This is the most load-bearing concern because the abstract and Section IV claim that the design 'maximizes the worst-case sensing SINR' and 'improves over the state-of-the-art.' The latter empirical claim may survive as a heuristic statement; the former does not. I concur with the CONDITIONAL verdict: the paper has no formal verification, but the simulations and the subproblem derivations provide support for a heuristic algorithm. A revision that proves convergence to a stationary point (or explicitly claims a heuristic solution) and adds the missing comparison against the closest hybrid-RIS baseline would materially strengthen the paper. The exhaustive-search test directly targets the global optimality claim and would settle whether the mismatch is merely a proof gap or an actual performance gap.","tokens_in":981,"tokens_out":1713,"duration_ms":90810,"concrete_test":"Run an exhaustive-search benchmark on a small instance (e.g., Na=2, K=1, P=1, S=1, NI=2, b=2) with a fine grid over Θ and x under the power constraints, and compare the global optimum of (21)/(26) against Algorithm 3's output. If Algorithm 3 fails to reach the global optimum on any instance, the claim should be softened to 'heuristic improvement over baselines.' As an analytical cross-check, verify Lemma III.1's conditions by evaluating the Hessian of fp in (26a) at a non-MVDR iterate; if it is PSD rather than NSD, the lemma cannot be invoked.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that Algorithm 3 maximizes the worst-case sensing SINR rests on Lemma III.1 (Section III-A), which guarantees monotonic convergence to the GFP optimum when fp is non-negative concave and gp positive convex. In problem (26), fp(x,Θ)=|w_p^H A_p x|^2. For fixed w_p and Θ, this is x^H A_p^H w_p w_p^H A_p x, a convex quadratic (Hessian 2A_p^H w_p w_p^H A_p is PSD), not concave; with Θ varying it is quartic. The paper's replacement gp in (30) uses the affine term Re(w_p^H A_p x) rather than sqrt(fp(x)), so equivalence to (26) holds only when u_p equals the optimal ratio and Re(w_p^H A_p x)=|w_p^H A_p x|. The MVDR update (22) yields a real positive inner product at the updated point, but this property is not established for all iterates or for the joint Θ-subproblem. Therefore Lemma III.1's convergence-to-optimality conclusion does not apply; the monotone increase in Fig. 3 is only empirical evidence for a feasible-point algorithm. The abstract's 'maximizes the worst-case sensing SINR' is thus not established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies a multi-RIS-aided ISAC system in which a base station communicates with multiple SCUs while sensing an extended target that may act as an eavesdropper, under imperfect target-location knowledge. A joint optimization problem is formulated to maximize the worst-case sensing SINR over a grid of possible target locations, subject to constructive-interference QoS constraints, destructive-interference security constraints, discrete RIS phase-shift constraints, and power budgets. The proposed Algorithm 3 alternates closed-form MVDR receive-filter updates, convex feasibility updates for SCU beamformers, a GFP-based waveform update, and a PDD/PCCP-based RIS update followed by quantization to discrete reflection coefficients. Simulation results compare the method with no-RIS, fully passive, fully active, and random-RIS baselines.","tokens_in":1530,"tokens_out":1576,"duration_ms":103475,"significance":"If the theoretical claims were established, the paper would provide a fairly complete design framework for a relevant and nontrivial scenario: multiple hybrid RISs, an extended target with imperfect location, and both communication-QoS and security constraints. The system model is rich, and the algorithmic pipeline (GFP, PDD, PCCP, DRC projection) is plausible as a heuristic. However, the central optimality claim is not supported because the GFP lemma is applied outside its hypotheses, and the claimed improvement over the state of the art is not directly verified against the most relevant prior hybrid-RIS method. The manuscript is therefore best regarded as a heuristic design study whose comparative claims need both theoretical correction and additional baselines.","major_comments":[{"comment":"Lemma III.1 guarantees monotonic convergence to the GFP optimum only when the numerator functions fp are non-negative concave and the denominator functions gp are positive convex. In the application to problem (26), fp(x, Θ) = |w_p^H A_p x|^2 is a convex quadratic in x for fixed w_p and Θ (Hessian 2A_p^H w_p w_p^H A_p, which is positive semidefinite), and it is quartic in Θ; the paper does not verify the lemma's hypotheses. Moreover, the reformulated constraint gp in (30) replaces |w_p^H A_p x|^2 by 2u_p Re(w_p^H A_p x) - u_p^2 w_p^H(...)w_p, which equals the original numerator only when Im(w_p^H A_p x) = 0 after the u_p update. The MVDR update (22) makes w_p^H A_p x real and positive for the current x, but this property is not preserved during the x-subproblem (34) or the Θ-subproblem, and no constraint enforces it. Consequently, the statement that Algorithm 3 maximizes the worst-case sensing SINR is not established; Figures 3-14 provide only empirical evidence for a feasible-point algorithm. The sentence after (26), 'We prove that problem (26) is solvable,' is not followed by a proof.","section":"Section III-A, Eqs. (26)-(30)"},{"comment":"The abstract and Section IV claim that the proposed design improves the sensing and secure-transmission performance over 'the state-of-the-art RIS-aided ISAC approaches.' However, the simulations compare only against No-RIS, optimized passive RIS, random passive RIS, and optimized active RIS variants. The closest prior hybrid-RIS ISAC method, reference [20], is not included as a baseline, nor is the secure RIS-ISAC method [27] used in the comparison. As a result, the strong comparative claim in the abstract is not supported by the presented numerical evidence; the authors should either add these baselines or weaken the claim to comparisons against the investigated alternatives.","section":"Section IV, baseline selection"},{"comment":"The convergence guarantees of the overall alternating algorithm are not established. The PDD convergence conditions from [36] are not checked for the augmented Lagrangian problem (41), the BCD inner iteration is not shown to converge to a stationary point of (41), and the DRC projection step (47)-(48) together with the acceptance rule in Algorithm 3 makes the iteration discontinuous. The statement in Section III-C that 'the objective value of (41) converges to the limit value' is asserted without proof. If the authors intend to claim only monotone improvement of the computed objective, this should be stated explicitly and the theoretical claims adjusted accordingly.","section":"Section III-C, Algorithm 3"}],"minor_comments":[{"comment":"The discrete reflection-coefficient constraint mixes the phase index and the reflection-unit index: 'e^{j2πn/2^b}, n ∈ {0,...,2^b-1}, n ∈ NI' should be rewritten to state that each ϑ_{s,n} takes one of the 2^b phase values for each s and n.","section":"Eq. (21d)"},{"comment":"The notation '{θ_p, φ_p} ∈ card(χ̄, χ̂)' is unclear; the grid of P possible target locations should be defined explicitly, for example as a Cartesian product of the two angular intervals, with P denoting the number of grid points.","section":"Section II-B, around Eq. (15)"},{"comment":"The caption and legend of Fig. 3 do not clearly indicate which curves correspond to the CI-only case versus the CI+DI case and which uncertainty sets are used; the text should be more explicit so the reader can map the curves to the described scenarios.","section":"Fig. 3"},{"comment":"The complexity expression for Algorithm 2 contains terms such as P((3N+KN)SNI)^{1/2}2KN(SN_I^4) whose variables are not all defined in the surrounding text; please clarify the notation.","section":"Section III-C, complexity expression"},{"comment":"There are several typographical and formatting errors (e.g., 'optmization' in Section III-C, inconsistent use of 'reconfigurable' vs 'reconfigurable') that should be corrected in a final revision.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The GFP issue in Section III-A is the main technical obstacle: the core 'maximizes the worst-case sensing SINR' claim rests on a lemma whose hypotheses are not satisfied. This can likely be addressed in revision by either proving the required alignment condition or reframing the algorithm as a heuristic with empirical convergence. The baseline omission is straightforward to fix and should be required given the abstract's comparative claim."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"What should you know about this paper? It builds a fairly complete optimization framework for a multiple-hybrid-RIS ISAC system with an extended eavesdropping target under imperfect target location. That specific combination is new, as far as I can tell, and the authors handle a genuinely hard problem: the sensing SINR is quartic in the RIS coefficients, and they break it apart with PDD and PCCP into convex subproblems. The system model is carefully set up, the derivations in Appendix A are nontrivial, and the simulation study is extensive. On the engineering level, the paper is plausible and the simulation gains over the no-RIS, passive-RIS, and random-RIS baselines look credible.\n\nThe soft spots are concentrated in the theoretical claims. Lemma III.1 requires fp(x) to be non-negative concave and gp(x) to be positive convex for the GFP reformulation to converge to the optimal value. In problem (26), fp(x,Θ)=|w_p^H A_p x|^2 is a convex quadratic in x for fixed w_p and Θ, and quartic in Θ; the paper never checks the lemma's conditions. The reformulated gp in (30) uses the affine term Re(w_p^H A_p x) rather than sqrt(fp(x)), so equivalence to the original ratio only holds under extra conditions that are not established for all iterates. The sentence \"We prove that problem (26) is solvable\" is not backed by a proof in the text. This means the abstract's \"maximizes the worst-case sensing SINR\" is not supported. The monotone convergence in Fig. 3 is empirical evidence for a feasible-point algorithm, not a proof of optimality. This is a load-bearing gap, but it is fixable: either supply the missing convexity argument or soften the claim to \"a high-quality heuristic solution.\"\n\nThe other significant issue is baseline selection. The closest prior hybrid-RIS ISAC work, reference [20] by Liao et al., is cited but never compared in the simulations. The authors compare against passive RIS, active RIS, random RIS, and no-RIS, which show the value of hybrid RIS generally, but they do not show improvement over the existing hybrid-RIS optimization method. That weakens the \"over state-of-the-art\" statement. The rest is minor: a few typos, a rough complexity expression, and some notation that is hard to follow.\n\nOverall, the central engineering message likely survives: a carefully designed multiple-hybrid-RIS system can improve robust sensing and secure communication. The paper deserves a serious referee, but the referee should push hard on the GFP claim and the missing comparison. I would send it to peer review rather than desk reject it, and I expect that with a corrected convergence claim (or a proof) and an added baseline it could become a solid contribution.","headline":"A competent engineering paper with a new system combination and a solid optimization pipeline, but the GFP convergence claim is unsupported as written and the baseline set misses the closest hybrid-RIS prior.","tokens_in":26377,"tokens_out":2320,"would_cite":true,"duration_ms":29071,"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":"A hybrid RIS with a few active and many passive units can maximize worst-case sensing SINR while forcing an eavesdropping extended target into a destructive decision zone in an ISAC downlink.","keywords":["integrated sensing and communication","secure communication","hybrid reconfigurable intelligent surface","extended target detection","generalized fractional programming","penalty dual decomposition","constructive interference","directional modulation"],"falsifier":"Run Algorithm 3 from a feasible start on a small instance, for example $N_a=4$, $K=2$, one hybrid RIS with $N_I=8$ and 2-bit phase resolution, and compare its worst-case sensing SINR against an exhaustive grid search over the transmit waveform and RIS phases; if any feasible grid point achieves a higher worst-case SINR, or if the algorithm's objective does not increase monotonically, the claimed optimality and convergence are not supported.","tokens_in":25382,"feed_emoji":"📡","tokens_out":8406,"duration_ms":83593,"temperature":0.7,"pith_summary":"The paper tries to establish that a hybrid reconfigurable intelligent surface—one with a few active amplifying units and many passive phase-shifting units—can simultaneously improve radar detection of an extended eavesdropping target and secure downlink communication in an integrated sensing and communication (ISAC) system. The key move is to design the transmit waveform, the base-station receive filters, the user receive beamformers, and all RIS weights together, maximizing the worst-case sensing SINR over an uncertain target-location set while forcing the eavesdropper's received symbols into a destructive decision zone. The authors argue that their generalized-fractional-programming, penalty-dual-decomposition, and penalty-convex-concave procedure handles the non-convex quartic problem, and simulations show the hybrid-RIS design outperforms passive-RIS, active-RIS, random-RIS, and no-RIS baselines. A reader would care because this addresses a realistic deployment constraint—target location is never known exactly—and quantifies how much sensing margin is lost to that uncertainty.","feed_headline":"Hybrid RIS lifts ISAC sensing SINR while blocking eavesdroppers","feed_subtitle":"Jointly optimizing waveform, filters, beamformers and RIS weights improves worst-case target detection and secures downlinks.","key_machinery":"The load-bearing machinery is the alternating optimization framework that splits the joint problem into four subproblems: closed-form minimum-variance-distortionless-response receive filters for each candidate target location, feasibility-style receive beamformers at the SCUs, a convex transmit-waveform update obtained via the generalized-fractional-programming quadratic transformation, and a hybrid-RIS update via penalty dual decomposition with an auxiliary copy $\\vartheta_1$ of the RIS coefficient vector $\\vartheta$. The auxiliary-copy trick turns the quartic sensing SINR into a quadratic form in $\\operatorname{vec}\\{\\vartheta\\vartheta_1^T\\}$, and the penalty convex-concave procedure handles the unit-modulus constraints; big-M relaxation with binary variables handles the discrete-phase security constraints. The key identity is the quadratic transform $g_p = 2u_p \\Re(w_p^H A_p x) - u_p^2 w_p^H(\\Pi_c+\\Sigma_p+\\sigma_R^2 I)w_p$, which converts a max-min ratio into a concave lower bound when the auxiliary variable $u_p$ is fixed.","core_discovery":"The paper claims that by jointly optimizing the transmit signal, the BS receive filter bank, the SCU receive beamformers, and the weights of multiple hybrid RISs, the worst-case sensing SINR for detecting an extended target can be maximized while guaranteeing constructive-interference communication quality and destructive-interference security against a single-antenna eavesdropping target whose location is only imperfectly known. The target is modeled as an extended scatterer through a target impulse response with clutter, and the design is formulated as a non-convex max-min problem with discrete RIS phase constraints and power budgets. The authors solve it with an alternating framework based on generalized fractional programming, penalty dual decomposition, and a penalty convex-concave procedure. Their simulations report that the resulting hybrid-RIS design improves both detection and secure transmission over fully-passive, fully-active, random, and no-RIS benchmarks, with gains that grow as the number of RIS units and the power budget increase.","pith_inferences":["The same GFP/PDD/PCCP template could be applied to other quartic ISAC objectives, such as STAR-RIS or fully-active RIS variants, but only after the convexity conditions of Lemma III.1 are verified; the paper does not show that its reformulation satisfies the concave-numerator and convex-denominator requirements.","The simulations suggest a hardware-sizing rule: because most of the sensing gain comes from the first few active units, an operator could set the active-unit count from the power budget and the target-uncertainty size rather than maximizing active units.","The nominal-versus-worst-case SINR gap shown in the convergence plots could be used as a calibration curve: the slope of worst-case SINR versus uncertainty-set size quantifies how much sensing margin a given target-location error costs.","A natural testable extension would be to replace the fixed target-location uncertainty set with a Bayesian prior and compare the worst-case design against an expected-SINR design; the paper only treats the worst-case formulation."],"forward_implications":["In the single-RIS simulations, the proposed hybrid-RIS design outperforms the optimized passive RIS by roughly 77 percent in worst-case sensing SINR at $N_I=20$ and about 110 percent at $N_I=50$, indicating that the hybrid-RIS advantage grows with array size.","The design captures most of the sensing gain with only a small number of active units: moving from $A=0$ to $A=1$ gives a larger improvement than moving from $A=9$ to $A=10$, so active-unit count can be chosen near the knee of the performance curve.","Worst-case sensing SINR decreases as the target-location uncertainty set grows, and the loss is steeper when the angular separation between the user and the eavesdropper is small, meaning the robust design trades sensing margin for security margin.","With the destructive-interference constraint active, the eavesdropper's symbol error rate rises to about 0.8 under the constructive-interference-only design and approaches 1 under the combined CI plus DI design, so the eavesdropper cannot reliably decode.","The discrete-phase projection maintains feasibility and approaches the continuous-phase performance as phase resolution increases, whereas random discrete phases do not improve with resolution, showing that careful discrete-phase design is essential."],"supporting_citations":[{"why":"Supplies the constructive-interference symbol-level precoding baseline, including the SDR-based No-RIS method that the proposed design must beat.","marker":"[8]"},{"why":"Provides the optimized fully-passive RIS-aided DFRC design used as a benchmark and the communication-waveform sensing model.","marker":"[16]"},{"why":"Introduces the hybrid active-passive RIS model and its power-consumption structure, which the paper extends to multiple RISs.","marker":"[20]"},{"why":"Motivates the imperfect-CSI multi-RIS setting and the robust beamforming formulation under target-location uncertainty.","marker":"[23]"},{"why":"Establishes the joint waveform and reflective-beamforming secure ISAC formulation with CI and DI constraints that this work generalizes.","marker":"[27]"},{"why":"Supplies the extended-target impulse-response model, clutter model, and the SINR metric that drives detection performance.","marker":"[34]"},{"why":"Provides the big-M continuous relaxation used to convert the discrete DI security constraints into tractable form.","marker":"[35]"},{"why":"Supplies the penalty dual decomposition framework whose convergence underpins the alternating hybrid-RIS updates.","marker":"[36]"},{"why":"Provides the convex-concave procedure used to handle the unit-modulus constraints in the RIS subproblem.","marker":"[37]"}],"fun_headline_variants":["Hybrid RIS raises sensing SINR and blocks eavesdroppers","Secure ISAC with hybrid RIS: worst-case SINR maximized","Extended target sensing and secrecy improved by hybrid RIS","Hybrid RIS: joint optimization for secure ISAC detection","Robust hybrid-RIS beamforming secures ISAC against eavesdroppers"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the reformulated sensing SINR satisfies a generalized-fractional-programming lemma whose concave-numerator and convex-denominator conditions are never checked, and the raw numerator $|w_p^H A_p x|^2$ is actually a convex quadratic, so if those conditions fail the claimed monotone convergence is not assured and the algorithm may only return a heuristic solution.","fun_headline_variants_meta":{"raw":{"variants":["Hybrid RIS raises sensing SINR and blocks eavesdroppers","Secure ISAC with hybrid RIS: worst-case SINR maximized","Extended target sensing and secrecy improved by hybrid RIS","Hybrid RIS: joint optimization for secure ISAC detection","Robust hybrid-RIS beamforming secures ISAC against eavesdroppers"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000808,"raw_usage":{"total_tokens":3555,"prompt_tokens":965,"completion_tokens":2590,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":581,"completion_tokens_details":{"reasoning_tokens":2504}},"tokens_in":581,"tokens_out":2590,"duration_ms":23492,"temperature":1.0,"reasoning_tokens":2504,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:01:17.122476+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run Algorithm 3 from a feasible start on a small instance, for example $N_a=4$, $K=2$, one hybrid RIS with $N_I=8$ and 2-bit phase resolution, and compare its worst-case sensing SINR against an exhaustive grid search over the transmit waveform and RIS phases; if any feasible grid point achieves a higher worst-case SINR, or if the algorithm's objective does not increase monotonically, the claimed optimality and convergence are not supported.","supporting_citations":[{"cited_title":"Secure du al- functional radar-communication transmission: Exploitin g interference for resilience against target eavesdropping,","cited_arxiv_id":null,"evidence_quote":"Supplies the constructive-interference symbol-level precoding baseline, including the SDR-based No-RIS method that the proposed design must beat."},{"cited_title":"Joint transmit wa veform and passive beamforming design for RIS-aided DFRC systems,","cited_arxiv_id":null,"evidence_quote":"Provides the optimized fully-passive RIS-aided DFRC design used as a benchmark and the communication-waveform sensing model."},{"cited_title":"Beamf orming design for hybrid active-passive RIS assisted integrated s ensing and communications,","cited_arxiv_id":null,"evidence_quote":"Introduces the hybrid active-passive RIS model and its power-consumption structure, which the paper extends to multiple RISs."},{"cited_title":"Robust hybrid beamforming design for Multi-RIS assisted M IMO system with imperfect CSI,","cited_arxiv_id":null,"evidence_quote":"Motivates the imperfect-CSI multi-RIS setting and the robust beamforming formulation under target-location uncertainty."},{"cited_title":"STAR-RIS-enabled secure dual-func tional radar-communications: Joint waveform and reflective beamf orming op- timization,","cited_arxiv_id":null,"evidence_quote":"Establishes the joint waveform and reflective-beamforming secure ISAC formulation with CI and DI constraints that this work generalizes."},{"cited_title":"Robust transmit code and receive filter design for extended targets in clutter,","cited_arxiv_id":null,"evidence_quote":"Supplies the extended-target impulse-response model, clutter model, and the SINR metric that drives detection performance."},{"cited_title":"Joint network o ptimization and downlink beamforming for CoMP transmissions using mixe d integer conic programming,","cited_arxiv_id":null,"evidence_quote":"Provides the big-M continuous relaxation used to convert the discrete DI security constraints into tractable form."},{"cited_title":"Penalty dual decomposition method f or non- smooth nonconvex optimization-part I: Algorithms and conv ergence analysis,","cited_arxiv_id":null,"evidence_quote":"Supplies the penalty dual decomposition framework whose convergence underpins the alternating hybrid-RIS updates."},{"cited_title":"V ariations and extension of the con vex-concave procedure,","cited_arxiv_id":null,"evidence_quote":"Provides the convex-concave procedure used to handle the unit-modulus constraints in the RIS subproblem."}],"review_version":1}