{"id":"dd93bbd2-875d-409c-938e-2e1ba005a1a0","arxiv_id":"2608.04133","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Every optimal 1-bit RIS configuration aligns the channel vectors with a common direction, yielding an exact O(N^2) SISO algorithm and a structured MIMO sampling heuristic.","lead":"This paper proves that every optimal 1-bit RIS configuration must align each channel vector with a common direction, and uses this to cut the search space for single-antenna links from 2^N to at most N+1 candidate configurations. It also applies the same principle to an integrated sensing and communication scenario, where both tasks reduce to the same signed-vector maximization.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Sensing reduction Eq. (33) drops the reflection coefficient α; with complex α the norm-maximizing RIS configuration need not maximize the confidence η.","rationale":"Lemma 1 and the SISO enumeration are sound; I checked the sign-flip argument for the artificial variable x_{N+1} and it preserves optimality, and the sector-counting argument is valid because each sign pattern corresponds to a convex cone. The MIMO sampling method is a heuristic; in the d≥K regime the 'admissible' set may be all sign patterns, but the paper makes no theoretical guarantee there, so this is a limitation, not an error. The ISAC section, however, makes a concrete mathematical claim that the sensing phase design is the same norm-maximization problem, and that claim is false for the model as stated unless the reflection coefficient is assumed real and positive. This is exactly the reader's weakest assumption, and it is load-bearing because the title and abstract advertise sensing as one of the two applications of the framework. The rest of the paper's communication results can stand, so the appropriate verdict is conditional acceptance pending correction or qualification of the sensing equivalence.","tokens_in":14924,"tokens_out":14476,"duration_ms":135700,"concrete_test":"Keep α in the derivation of Eq. (33): substitute r_sen = α r_exp,sen into Eq. (32) and verify that η = Re(α)||r_exp,sen||. Then run the Section V-B ISAC experiment once with α = 0.8 and once with α = 0.8j (or -0.8); if the localization error and miss rate degrade in the latter cases, the norm-maximization objective of Eq. (34) is not the correct sensing objective for the stated model.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The load-bearing concern is the ISAC sensing reduction, not Lemma 1, which is correct. In Eq. (24), r_sen = α( h^T_{t,real}⊗h^R_{t,real} + Σ_{n=1}^N x_n C(ψ_n,ψ_{t,real}) h^T_n⊗h^R_{t,real} ), with α introduced as an unknown scalar target reflection coefficient. In Eqs. (31)-(32), the matched filter is H = r_exp,sen/||r_exp,sen||, where r_exp,sen omits α. At the true target position, r_sen = α r_exp,sen, so Eq. (32) gives η = Re(α)||r_exp,sen||, not ||r_exp,sen||. Eq. (33) drops α from the numerator and asserts η = ||r_exp,sen||. Consequently Eq. (34), which maximizes ||r_exp,sen|| over the binary sensing elements, is equivalent to maximizing the confidence only when Re(α)>0; for Re(α)<0 it maximizes the wrong quantity, and for generic complex α the real part couples to the phase of r_exp,sen. Since α is unconstrained in the stated model, the claimed unification of target localization with the communication norm-maximization problem is not established by the derivation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies the optimization of 1-bit RIS phase configurations for received-signal maximization and for an ISAC scenario. The communication problem is rewritten as maximizing ||Σ_k x_k a_k|| over x∈{±1}^K, where a_k are real channel-dependent vectors. Lemma 1 proves a necessary condition for global optimality: if x* is optimal and v* = Σ_k x*_k a_k, then x*_k = sgn(a_k·v*) for all k. The paper uses this condition to design an informed sampling algorithm for general MIMO settings (Algorithm 1) and an exact angular-enumeration algorithm for SISO settings (Algorithm 2), claiming that the SISO global optimum can be found by evaluating at most K candidates, with K ≤ N+1. The ISAC section partitions the RIS into communication and sensing subsets and claims that both subproblems have the same signed-vector form. Simulations compare the proposed sampling against uninformed binary sampling and a continuous-phase benchmark, and evaluate a communication–sensing trade-off.","tokens_in":15186,"tokens_out":14098,"duration_ms":136911,"significance":"Lemma 1 is a clean, self-contained structural result: the proof is correct, no parameters are fitted, and the necessary condition is genuinely restrictive. The SISO angular partition is the right approach and, once the pseudocode is corrected, would provide an exact polynomial-time algorithm. The simulation comparison is fair in that both binary methods use the same sampling budget, and the reported gains over direct sampling are plausible. The main weaknesses are concentrated in the ISAC section: Eq. (33) silently assumes a real positive target reflection coefficient, and the sensing measurement protocol is underspecified. These issues do not invalidate Lemma 1 or the communication-side algorithms, but they prevent the claimed communication–sensing unification from being accepted as derived.","major_comments":[{"comment":"Equation (33) is not correct as written. From Eq. (24), at the true target position r_sen = α r_exp,sen, so substituting into Eq. (32) gives η = Re(α) ||r_exp,sen||, not ||r_exp,sen||. The subsequent equivalence in Eq. (34), which maximizes ||r_exp,sen||, is therefore valid only when α is real and positive, or when the confidence is replaced by a magnitude-based quantity. Since α is introduced as an unknown scalar and the simulations set α = 0.8, the claimed reduction of sensing to the same geometric norm-maximization problem is not established for a generic complex or negative-real α. Please state the real-positive assumption explicitly, or change the confidence to a quantity such as |H^H r_sen| for which the norm-maximization equivalence holds for any nonzero α.","section":"IV-B, Eq. (33)"},{"comment":"The sensing measurement protocol is underspecified. The confidence η in Eq. (32) depends on the RIS configuration x through both the measured r_sen and the matched filter H = r_exp,sen/||r_exp,sen||. If x is optimized separately for each hypothesized position via Eqs. (34)-(35), then the measured r_sen and the matched filter H are not generated under the same configuration unless the RIS is reconfigured and a new measurement is taken for every candidate ψ_t. The paper does not describe such a protocol; if instead a single sensing configuration is used, Eq. (33) holds only at the position for which that configuration was designed. Please specify the actual measurement and reconfiguration procedure used in the localization step and in the simulations, and clarify how continuous-space Bayesian optimization over ψ_t is compatible with hypothesis-dependent RIS configurations.","section":"IV-B, Eqs. (27)-(35)"},{"comment":"As printed, Algorithm 2 cannot enumerate the claimed candidate set. Lines 3-4 append arg(a_k)+π/2 and arg(a_k)−π/2 modulo π; these two values are identical for every k, so after duplicate removal Θ has at most K entries, not 2K. The loop in line 9 then stops at LENGTH(Θ)/2, which for K = 1 evaluates zero configurations. The prose correctly describes the 2K rays and the reduction to K antipodal pairs, so the pseudocode should append angles modulo 2π and keep the LENGTH(Θ)/2 loop; alternatively, if Θ stores unoriented line angles, the loop should run over all LENGTH(Θ) sectors. With the printed version, the claimed exactness of the SISO algorithm is not delivered.","section":"III-C, Algorithm 2, lines 3-4 and 7-9"}],"minor_comments":[{"comment":"Algorithm 1 uses the symbol N for both the number of RIS elements and the number of samples; rename the sampling budget (e.g., N_trials) to avoid confusion.","section":"Algorithm 1, input/output"},{"comment":"In Eq. (13), the summation range is omitted; write Σ_{n=1}^{N+1} x_n a_n.","section":"III-A, Eq. (13)"},{"comment":"The notation for the true target position alternates between ψreal_t and ψt,real; unify it.","section":"IV, Eq. (24)"},{"comment":"In Algorithm 1, sgn(0) is undefined; for continuous sampling this has probability zero, but the implementation should specify a tie-breaking rule for the measure-zero event.","section":"Algorithm 1, line 4"},{"comment":"Section III-B states that the problem is NP-hard without proof or a precise pointer; please make the connection to the cited binary-quadratic-programming literature explicit.","section":"III-B"}],"recommendation":"major_revision","confidential_remarks":"The paper fits the journal's scope. The main structural result, Lemma 1, is correct and the SISO enumeration idea is sound, but the exact-algorithm pseudocode must be fixed and the ISAC section needs to specify the measurement protocol and the alpha assumption before the unification claim can be accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nYou should know two things about this paper. First, the SISO result is correct and genuinely useful: for a 1-bit RIS with a single-antenna pair, the global optimum of received-power maximization can be recovered by checking at most K sign patterns (K ≤ N+1) via an angular partition of the plane. Second, the ISAC section contains a real derivation error — Eq. (33) drops the target reflection coefficient α. That needs fixing before the sensing claims can be taken seriously.\n\nThe core lemma — every global maximizer of ||Σ x_k a_k|| satisfies x_k = sgn(a_k·v) for some common v — is proved cleanly, though it is a standard exchange argument. As the paper acknowledges, [26] already proved SISO polynomial solvability and [27] proposed sign alignment; the explicit candidate enumeration with O(K^2) complexity is the new concrete bit and deserves credit. The MIMO sampling that follows from Lemma 1 is a reasonable heuristic — sample directions v and evaluate the induced sign patterns — but it has no guarantee, and the paper does not compare it to cheaper local-search baselines.\n\nThe ISAC reduction is where the paper stumbles. At the true target, the matched-filter output is Re(α)||r_exp,sen||, not ||r_exp,sen|| as Eq. (33) claims. With complex α, the configuration that maximizes the norm need not maximize the confidence; with Re(α)<0 it maximizes the wrong quantity. The paper never states that α is real positive, and in general it is not. So the claimed equivalence between sensing and communication optimization is not established. A second concern is conceptual: the sensing procedure optimizes the RIS elements for each hypothetical target position ψ_t. For that to be a real protocol, you would need a codebook probed sequentially; as written, it can read as if the configuration is chosen with knowledge of the true target, which would be circular. The simulations fix α=0.8 (real positive), so the Eq. (33) error cancels there, but the theory is still wrong. Also, no code or data is released, so the simulation gains are illustrative.\n\nWho is this for? Anyone working on 1-bit RIS configuration, particularly SISO links, will find a clean exact algorithm worth citing. The ISAC part needs major revision. I would send it to a referee — the flaws are fixable and the core is solid — but I would ask the authors to correct the α algebra and to specify how the per-hypothesis sensing configurations are actually used.","headline":"The SISO exact algorithm is correct and useful; the ISAC equivalence in Eq. (33) drops the reflection coefficient, and the sensing protocol is underspecified.","tokens_in":15697,"tokens_out":10886,"would_cite":true,"duration_ms":100145,"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":"Every globally optimal 1-bit RIS configuration for received-signal maximization is induced by the signs of channel-dependent vector projections onto a common direction, and the paper turns this into efficient algorithms.","keywords":["Reconfigurable Intelligent Surfaces","1-bit phase quantization","binary optimization","signed sum","MIMO","single-input single-output","integrated sensing and communication","sign alignment"],"falsifier":"Run the SISO algorithm on a random $K=4$ instance in $\\mathbb{R}^2$ and compare its output against exhaustive search over all $2^4$ sign patterns; if it returns anything other than the true maximum, the angular-partition claim fails. For the ISAC equivalence, set the target reflection coefficient to a complex value with nonzero imaginary part and test whether the configuration maximizing $\\|r_{\\mathrm{exp,sen}}\\|$ also maximizes the confidence $\\eta$.","tokens_in":14746,"feed_emoji":"📡","tokens_out":12341,"duration_ms":98189,"temperature":0.7,"pith_summary":"This paper aims to show that the exponential configuration space of a 1-bit reconfigurable intelligent surface is structured rather than an unstructured pile of $2^N$ binary choices. It reformulates received-signal maximization as maximizing the norm of a signed sum $\\sum_k x_k a_k$ of channel-dependent real vectors, and proves that every globally optimal sign pattern must be induced by the signs of the projections $a_k \\cdot v$ onto a common vector $v$. The payoff is algorithmic: for single-antenna links the candidate set collapses to at most $N+1$ configurations and the global optimum is recovered in polynomial time, while for general MIMO systems a geometry-informed sampler concentrates its budget on admissible configurations. The same signed-sum form is also shown to govern an integrated sensing and communication scenario, where RIS elements are partitioned between communication enhancement and target localization.","feed_headline":"One direction shrinks the 1-bit RIS search from 2^N to N+1 checks","feed_subtitle":"One direction shapes every optimal RIS configuration; exponential search becomes polynomial checks.","key_machinery":"The load-bearing object is the signed sum $S(x) = \\sum_k x_k a_k$ together with the sign-projection rule of Lemma 1: a configuration can be globally optimal only if every vector $a_k$ is aligned, after multiplication by $x_k$, with a common direction $v$. In $\\mathbb{R}^2$ this turns the search into an angular partition: the lines $l_k = \\{p : a_k \\cdot p = 0\\}$ split the plane into at most $2K$ sectors, each sector codes one sign pattern, and antipodal sectors give the same objective value, so at most $K$ candidates need evaluation. The SISO algorithm sorts the boundary angles, reads off each sector's sign pattern from a midpoint vector, and returns the best norm; the dominant cost is sorting, giving $O(K \\log K)$ with parallel norm evaluation and $O(K^2)$ serially.","core_discovery":"On its own terms, the central discovery is Lemma 1: for the problem $\\max_{x \\in \\{\\pm1\\}^K} \\|\\sum_{k=1}^K x_k a_k\\|$ with nonzero $a_k \\in \\mathbb{R}^d$, every global maximizer $x^\\star$ satisfies $x_k^\\star = \\operatorname{sgn}(a_k \\cdot v^\\star)$ with $v^\\star = \\sum_k x_k^\\star a_k$. The proof flips a single sign and shows that unless the sign of $x_k^\\star$ agrees with the sign of $a_k \\cdot v^\\star$, the norm strictly increases, contradicting optimality. Therefore all global optima belong to the class of sign patterns that are constant on each side of a hyperplane through the origin in signal space. The paper calls these configurations structurally admissible, and their number depends on the signal-space dimension: in two dimensions the class is finite and small, while in higher dimensions it forms the basis of a sampling method.","pith_inferences":["Because $\\eta = \\mathrm{Re}(\\alpha)\\,\\|r_{\\mathrm{exp,sen}}\\|$ when the target is at the tested position, the paper's Eq. (33) equivalence only holds for a reflection coefficient with positive real part; a complex or negative-real $\\alpha$ would flip the sensing objective from maximization to minimization.","One could test Algorithm 2 exhaustively on many small random instances; a single counterexample would point to a boundary issue in the angular partition, since Lemma 1's proof is independent of any numerical implementation.","The half-space mechanism suggests a natural multi-bit analogue: choose each element's phase as the discrete value closest to aligning $a_k$ with the sampled direction $v$; the paper does not explore this, but the structure of Lemma 1 makes it a direct next question."],"forward_implications":["For single-antenna links, the global optimum of the 1-bit RIS problem is recovered exactly by checking at most $K \\le N+1$ configurations rather than $2^N$.","In MIMO systems, a fixed sampling budget restricted to structurally admissible configurations outperforms uninformed binary sampling by at least $5.36$ dB in the paper's simulations and lands within $3.12$ dB of a continuous-phase RIS.","In the ISAC scenario, the communication and sensing phase-design subproblems have the same signed-sum form, so one algorithm serves both subarrays.","The measured trade-off curve shows an operating region in which added communication elements raise received power without a proportional rise in localization misses, for example a $16\\times16$ RIS with $160$ communication and $96$ sensing elements keeps a $1\\%$ miss ratio."],"supporting_citations":[{"why":"Establishes that the SISO 1-bit RIS problem is solvable in polynomial time, the tractability result the paper's exact SISO algorithm makes explicit.","marker":"[26]"},{"why":"Introduces a sign-alignment rule for 1-bit RIS phase configuration; Lemma 1 provides the exact geometric condition behind that rule.","marker":"[27]"},{"why":"Supplies the RIS-aided MIMO radar method used as the continuous-phase benchmark and as the basis for the sensing confidence construction.","marker":"[8]"},{"why":"Places the general MIMO binary problem in the context of NP-hard combinatorial optimization and its polynomially solvable special cases.","marker":"[34]"},{"why":"Represents the quantize-after-optimization strategy whose loss of optimality motivates solving the binary problem directly.","marker":"[19]"}],"fun_headline_variants":["One direction shrinks 1-bit RIS search from 2^N to N+1","Geometry cuts 1-bit RIS optimization to N+1 checks","All optimal 1-bit RIS configurations come from one direction","Unified geometry for 1-bit RIS in communication and sensing"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The characterization depends on the optimization being exactly the Euclidean norm of a signed sum of nonzero real channel vectors; in the ISAC sensing half, the paper further assumes the target reflection coefficient $\\alpha$ is a real positive number so that maximizing the norm is the same as maximizing detection confidence.","fun_headline_variants_meta":{"raw":{"variants":["One direction shrinks 1-bit RIS search from 2^N to N+1","Geometry cuts 1-bit RIS optimization to N+1 checks","All optimal 1-bit RIS configurations come from one direction","Unified geometry for 1-bit RIS in communication and sensing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000903,"raw_usage":{"total_tokens":3922,"prompt_tokens":1017,"completion_tokens":2905,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":633,"completion_tokens_details":{"reasoning_tokens":2829}},"tokens_in":633,"tokens_out":2905,"duration_ms":17057,"temperature":1.0,"reasoning_tokens":2829,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T14:43:23.479520+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the SISO algorithm on a random $K=4$ instance in $\\mathbb{R}^2$ and compare its output against exhaustive search over all $2^4$ sign patterns; if it returns anything other than the true maximum, the angular-partition claim fails. For the ISAC equivalence, set the target reflection coefficient to a complex value with nonzero imaginary part and test whether the configuration maximizing $\\|r_{\\mathrm{exp,sen}}\\|$ also maximizes the confidence $\\eta$.","supporting_citations":[{"cited_title":"Optimal discrete beam- forming for intelligent reflecting surface,","cited_arxiv_id":null,"evidence_quote":"Establishes that the SISO 1-bit RIS problem is solvable in polynomial time, the tractability result the paper's exact SISO algorithm makes explicit."},{"cited_title":"Asymp- totically optimal closed-form phase configuration of 1-bit RISs via sign alignment,","cited_arxiv_id":null,"evidence_quote":"Introduces a sign-alignment rule for 1-bit RIS phase configuration; Lemma 1 provides the exact geometric condition behind that rule."},{"cited_title":"Foundations of MIMO radar detection aided by reconfigurable intelligent surfaces,","cited_arxiv_id":null,"evidence_quote":"Supplies the RIS-aided MIMO radar method used as the continuous-phase benchmark and as the basis for the sensing confidence construction."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Places the general MIMO binary problem in the context of NP-hard combinatorial optimization and its polynomially solvable special cases."},{"cited_title":"Energy efficient multi-user MISO communication using low resolution large intelligent surfaces,","cited_arxiv_id":null,"evidence_quote":"Represents the quantize-after-optimization strategy whose loss of optimality motivates solving the binary problem directly."}],"review_version":2}