{"id":"cb9b17a4-1853-48df-8414-c9e4d103d46b","arxiv_id":"2501.07893","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"A weighted GLRT detector with joint power and weight optimization is proposed for multipath exploitation in OFDM-ISAC, but derivation errors in the weight update and false-alarm analysis undermine the algorithm.","lead":"This paper tries to improve target detection in OFDM-based integrated sensing and communication (ISAC) systems by combining multipath echoes with a weighted GLRT detector and jointly optimizing transmit power and detector weights. A generalist might read it because multipath is usually a nuisance in radar and communications, and turning it into a sensing advantage would be practically useful for 6G networks in cities.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Weighted GLRT derivation collapses without exact path orthogonality, so the claimed false-alarm invariance and the optimization objective are unsupported; a numerical orthogonality check and a single parameter-free Monte Carlo replication would settle it.","rationale":"The reader's weakest_assumption—exact channel orthogonality—is precisely the load-bearing assumption on which the rest of the detector and optimization rest. I independently re-derived the steps leading from (11)–(14) and found that the paper counts on ~H_l1 ~H_l2^H = 0 to (a) simplify the GLRT to (13), (b) write the weighted statistic as a sum of independent chi-squared-like terms for the F-distribution in (15), and (c) justify the objective f(A,w). This assumption is exact only on an ideal integer-tap grid; for the simulated geometry with continuous delays/Dopplers and the stated N=64, M=64, L=6, it is violated, and the paper provides no evidence that the violation is negligible. The reader's additional concerns—the 'false algebraic equality' in the weight optimization and the projector ignoring signal structure—are consistent with mine, but the orthogonality failure is the most load-bearing because it breaks both the detection-statistic derivation and the optimization claim. The central claim therefore lacks sufficient support: the detector may still work empirically in some regimes, but the paper does not prove the claimed diversity gain. I agree with the REJECT verdict because the theory is invalid as stated, and the simulations are not sufficient to rescue it given the mismatch between the assumed ideal grid and the simulated continuous-channel setup. If the authors could show the orthogonality error is negligible (e.g., by bounding the off-diagonal terms or by choosing N=M with on-grid delays), the paper could be repaired; as it stands, the rejection is appropriate.","tokens_in":8314,"tokens_out":1583,"duration_ms":16017,"concrete_test":"Run a Monte Carlo verification of the claimed false-alarm invariance using the paper's own simulation parameters (Table I, N=64, M=64, L=6, with the path delays and Dopplers implied by the geometry). Generate Z ~ CN(0, I_M ⊗ Σ) for a given Σ (e.g., σ_r² = -80 dBm), and compute the statistic ~η in (14) for the proposed 'joint design' A,w and for 'equal power/equal weights'. If the empirical distribution of ~η differs between the two settings—e.g., the 10^{-3} quantile shifts by more than a factor of 2—then the claimed dependence of P_fa on only N, M, and L is false for the paper's own configuration.","verdict_should_be":"REJECT","load_bearing_attack":"The central claim—that the weighted GLRT (14) achieves better detection than the LoS-only detector—rests on three load-bearing steps: (i) the projection matrices P_{s,l} are mutually orthogonal (Section III, after (11)); (ii) this orthogonality lets the paper apply Craig–Sakamoto to obtain (13), and then use (15) to assert that under H0 the statistic ~η0 follows an F-distribution whose parameters depend only on N, M, L, independent of A and w; (iii) the optimization objective f(A,w) in (17) is 'positively proportional to the detection probability.' Each step is only valid if the discrete-delay-Doppler channel vectors for different paths are exactly orthogonal, i.e., ~H_l1 ~H_l2^H = 0 for l1≠l2. The paper neither states nor verifies this condition. It holds only on an ideal grid with equal subcarrier and symbol counts and distinct integer taps—not for the continuous delay/Doppler values, arbitrary N and M, or the off-grid parameters used in the simulations. Without this exact orthogonality, (15) does not follow: the weighted sum ∑ w_l P_{s,l}Z is not a single projection of Z, its distribution is not a central F with the claimed degrees of freedom, and the false-alarm probability depends on A and w. Likewise, the non-centrality parameter in (16) is not ‖∑ w_l Λ̂_l H_l A X‖_F^2 = ∑ w_l²‖Λ̂_l H_l A X‖_F² unless the P_{s,l} are mutually orthogonal, so the objective f(A,w) is not a valid proxy for detection probability in general. The paper's own simulations use a LoS path plus five NLoS paths with distinct delays and Dopplers; for the LoS path, ~H_l1 ~H_l2^H will generally be nonzero, so the exact orthogonality assumption fails in the very configuration shown in Fig. 2. The reader's verdict correctly identifies this as the load-bearing weakness, and I add that the paper's internal evidence is insufficient: the claimed F-distribution is not derived with the required independence assumptions stated, and no Monte Carlo verification of the P_fa invariance is reported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes a weighted GLRT detector for target detection in an OFDM-based ISAC system under multipath propagation, together with a joint design of subcarrier power allocation and detector weights via an MM-based alternating algorithm. The main theoretical claims are: (i) the detection statistic under H0 follows a parameter-free F-distribution independent of the power allocation and weights; (ii) the objective f(A,w) = sum_l w_l^2 ||H_l A X||^2_F is positively proportional to the detection probability; and (iii) the proposed design achieves significantly better target detection performance than the conventional detector, as supported by ROC simulations in Section V.","tokens_in":8777,"tokens_out":20308,"duration_ms":178832,"significance":"The problem of exploiting multipath for sensing in ISAC systems is relevant and timely, and the paper formulates a clean system model. If the theoretical results were correct, the proposed low-complexity joint design would be a useful contribution. However, the derivations contain multiple load-bearing errors: the GLRT statistic is not derived correctly, the false-alarm distribution claim is false, the weight design solves a different optimization problem than stated, and the proposed objective is not tied to detection probability. The simulation results cannot repair these gaps because the thresholds and comparisons rely on the invalid analysis. The claimed significance is therefore not established.","major_comments":[{"comment":"The reformulation of the GLRT as eta = ||Y||^2_F / ||Y - sum_l P_{s,l}Y||^2_F assumes that hat(Lambda)_l H_l A X = P_{s,l} Y for each l. From the given definition hat(Lambda)_l = P_{s,l} Y X^dagger A^dagger H_l^dagger, one obtains hat(Lambda)_l H_l A X = P_{s,l} Y (X^dagger A^dagger H_l^dagger H_l A X), and the factor in parentheses is not generally an identity. Since X is a block-diagonal symbol matrix that is not square and not isometric, and H_l A X differs from tilde(H)_l, the equality is not generally valid. This invalidates the derivation of the GLRT statistic on which the entire detector design is based.","section":"Section III, Eq. (12)"},{"comment":"The claim that the false-alarm statistic tilde(eta)_0 follows an F-distribution whose parameters depend only on N, M, and L is not correct. Under H0 with white Gaussian noise, the numerator equals ||sum_l w_l P_{s,l}Z||^2_F = sum_l w_l^2 ||P_{s,l}Z||^2_F, which is a weighted sum of independent chi-square variables with weights w_l^2; it is a chi-square only if all nonzero w_l^2 are equal. The ratio therefore does not have the stated central F-distribution, and the false-alarm probability depends on w. The constant false-alarm rate property asserted in Section III is thus unsupported, and the simulation thresholds in Section V are not justified.","section":"Section III, Eq. (15)"},{"comment":"The derivation assumes exact mutual orthogonality of the channel matrices of different paths, i.e., tilde(H)_{l1}^H tilde(H)_{l2} = 0 for l1 != l2. This holds only if the discretized delay and Doppler tap indices are distinct integers modulo N and M, respectively, which is neither stated nor verified. The continuous delay/Doppler model in (4) and the rounding in (5) do not guarantee this condition for the simulation parameters in Table I. Without exact orthogonality, the Craig-Sakamoto simplification in (13), the equivalence in (15), and the non-centrality parameter simplification in (16) all fail.","section":"Section III, after Eq. (11)"},{"comment":"The equality sum_l w_l^2 ||H_l A X||^2_F = |w^H d|^2 is algebraically false: the left-hand side is sum_l w_l^2 d_l^2, whereas the right-hand side equals sum_l w_l^2 d_l^2 + 2 sum_{l<k} w_l w_k d_l d_k. For nonnegative w and ||w||^2=1, maximization of sum_l w_l^2 d_l^2 is solved by a unit vector at the largest d_l, not by w = d/||d||. Consequently the closed-form solution in (24) and the alternating algorithm built on it do not solve the stated problem (18).","section":"Section IV-B, Eq. (23a)"},{"comment":"The statement that f(A,w) is 'positively proportional to the detection probability' is asserted without proof. The actual non-centrality parameter under H1 involves the random RCS matrices Lambda_l, which are omitted from f(A,w). Moreover, because the false-alarm threshold depends on w, maximizing f(A,w) does not necessarily maximize the detection probability at a fixed false-alarm rate. Thus problem (18) is not aligned with its stated objective.","section":"Section III, Eq. (17)"}],"minor_comments":[{"comment":"The definitions of the tap indices are swapped: the expression k_l = round[nu_l M T] cannot be a delay tap index and r_l = round[tau_l N Delta f] cannot be a Doppler tap index; they should be k_l = round[tau_l N Delta f] and r_l = round[nu_l M T].","section":"Section II, after Eq. (4)"},{"comment":"The exponential e^{j2*pi*(f0 + n*Delta*f*(t - m*T))} is ambiguous; it should be e^{j2*pi*(f0 + n*Delta*f)*(t - m*T)}.","section":"Eq. (1)"},{"comment":"The sentence 'Now the objective function is now expressed as a quadratic form' contains a duplicated 'now'; the text should be revised for clarity.","section":"Section IV-A, after Eq. (20)"},{"comment":"The text says 'Swelling-I type' RCS; the standard radar term is 'Swerling-I'. Also, Table I presents the beta values in an unclear format; it should specify which value corresponds to which path.","section":"Section V"},{"comment":"The paper does not state how the threshold xi in (14) is chosen in the simulations; this is important because the analysis claims a parameter-free false-alarm distribution, and the simulations should verify this claim directly.","section":"Section V"}],"recommendation":"reject","confidential_remarks":"The manuscript contains multiple load-bearing algebraic and statistical errors, and the simulations do not provide independent verification of the claimed detector characteristics. The paper does not include code or a reproducibility statement, so the numerical results cannot be checked. In my view, the errors are too fundamental to be addressed by a minor revision; the appropriate decision is reject."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the paper has the right shape—OFDM-ISAC, multipath diversity, weighted GLRT, joint power/weight design—but the mathematics doesn't close. I side with the reader: reject.\n\nWhat is genuine: the system model is clean, the problem formulation is sensible, and the MM-based alternating optimization is standard. The specific weighted-GLRT-with-optimization combination is new relative to Sen and Nehorai [5], and the simulation study is reasonably thorough. The idea of weighting path contributions to improve detection is plausible and worth pursuing.\n\nThe soft spots are load-bearing. First, equation (23a) is simply false. The left side is sum_l w_l^2 d_l^2, the right side is (sum_l w_l d_l)^2. These are not equal unless at most one d_l is nonzero. The claimed Rayleigh solution w = d/||d|| maximizes the wrong objective; the actual optimum for the stated problem is to put all weight on the path with the largest d_l. So the weight design is invalid, and the simulation results labeled 'proposed' are generated by a non-optimal weight vector.\n\nSecond, the false-alarm analysis depends on combining the weighted projections into a single norm squared in (15). That step requires the per-path projection matrices to be mutually orthogonal, which in turn requires exact orthogonality of the discrete delay-Doppler channel vectors for different paths. The paper never states or verifies that condition. It holds only on an ideal grid with equal subcarrier and symbol counts and distinct integer taps—not for the continuous, off-grid parameters used in their own simulations (Table I, Fig. 2). Without it, the F-distribution and the claimed invariance of P_fa to A and w do not follow. The stress-test observation is correct.\n\nThird, the objective f(A,w) is asserted to be 'positively proportional to detection probability.' The non-centrality parameter that connects them is again derived under the same orthogonality assumption. As written, the proxy is not justified.\n\nThe paper is not a fraud and the citation to [5] is fair. But these are not cosmetic issues; they are the core of the contribution. The authors could fix the weight solution (or change the constraint) and add a numerical check of the orthogonality condition or a Monte Carlo verification of the F-distribution. Until then, I would not cite it.\n\nFor a journal, I would still send it to reviewers—the topic is relevant and the flaws are identifiable and correctable—but I would expect a major revision or rejection in current form. For a reading group, it's a decent case study in how an orthogonality assumption can silently break a GLRT derivation.","headline":"The multipath diversity idea is worth a look, but the paper's weight solution and false-alarm invariance rest on false equalities and an unverified orthogonality assumption, so the claimed gains are not established.","tokens_in":9300,"tokens_out":4978,"would_cite":false,"duration_ms":48714,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper argues that multipath propagation, normally a nuisance in radar sensing, can be exploited as diversity: a weighted generalized likelihood ratio test on OFDM echo data, with jointly optimized subcarrier powers and detector…","keywords":["integrated sensing and communication","OFDM","multipath exploitation","target detection","generalized likelihood ratio test","delay-Doppler diversity","power allocation","majorization-minimization"],"falsifier":"Run the detector in a simulated two-path scenario in which the two paths land on the same discretized delay tap (or the same Doppler tap), so the projection matrices are no longer orthogonal, and measure the empirical false-alarm probability for several different power-allocation and weight choices; if the measured $P_{\\mathrm{FA}}$ shifts with the design variables, the central claim that the threshold depends only on $N$, $M$, and $L$ is falsified.","tokens_in":8153,"feed_emoji":"🎯","tokens_out":9626,"duration_ms":84574,"temperature":0.7,"pith_summary":"Most radar sensing treats multipath echoes as clutter or ambiguity; this paper argues that in an OFDM system that communicates and senses from the same waveform, the reflected paths between base station and target can be harvested as extra diversity. The authors develop a weighted generalized likelihood ratio test (a statistical decision rule for target present versus absent) that combines echoes from all resolvable paths, each with its own delay and Doppler shift. They then jointly design the subcarrier transmit powers and the per-path detector weights, maximizing detection probability subject to the communication user's SNR requirement and a total power budget. Simulation results show that the joint design outperforms a conventional detector that uses only the line-of-sight echo, and that the gain grows when the non-line-of-sight reflections carry more energy.","feed_headline":"Multipath echoes sharpen OFDM-ISAC target detection","feed_subtitle":"Non-line-of-sight echoes become diversity gain, lifting detection probability without raising false alarm.","key_machinery":"The central object is the weighted GLRT statistic built from rank-one projection matrices $P_{s,l} = \\tilde{\\mathbf{H}}_l(\\tilde{\\mathbf{H}}_l^H \\tilde{\\mathbf{H}}_l)^{\\dagger} \\tilde{\\mathbf{H}}_l^H$ onto each path's delay-Doppler channel vector. The load-bearing property is that these projections are mutually orthogonal, which lets the authors apply the Craig-Sakamoto theorem and write the detector as a ratio of independent F-distributed pieces; this is what makes the false-alarm distribution depend on only $N$, $M$, and $L$. The optimization machinery is an alternating loop: with weights fixed, the non-concave power-allocation objective is replaced by a concave first-order surrogate (the majorization-minimization step), and with power fixed, the weight vector is the normalized vector of path strengths, $w = d/\\|d\\|_2$, which solves a Rayleigh quotient.","core_discovery":"The central claim is that a GLRT detector on multipath OFDM echoes decomposes into a sum of per-path projection terms, so that detection can be improved by weighting those terms and optimizing transmit power. Specifically, with $L$ separable delay-Doppler paths, the weighted statistic is $\\tilde{\\eta} = 1 + \\sum_{l=1}^{L} w_l^2 \\|P_{s,l}Y\\|_F^2 / \\|P_n Y\\|_F^2$, where $P_{s,l}$ projects onto the $l$-th path's channel vector and $P_n$ onto the complementary null space. The paper derives that under the no-target hypothesis this statistic follows an F-distribution determined only by $N$, $M$, and $L$, independent of the transmit power allocation and the weights, so the false-alarm threshold does not move during design. It then maximizes the non-centrality parameter $\\sum_l w_l^2 \\|H_l A X\\|_F^2$ in place of the unknown RCS-weighted term, and shows by simulation that the resulting joint power-and-weight design achieves materially better ROC performance than the LoS-only conventional detector.","pith_inferences":["Because the false-alarm invariance relies on exact projection orthogonality, a natural robustness test would be to simulate fractional or clustered delay/Doppler taps and watch whether $P_{\\mathrm{FA}}$ stays flat; the paper does not report such a test.","The same weighted-combining idea likely extends to MIMO-OFDM ISAC, where multiple transmit and receive antennas provide additional diversity branches beyond the delay-Doppler paths, though the paper does not analyze that case.","The optimization objective drops the RCS coefficients $\\Lambda_l$ and uses path-strength surrogates, so the weights favor geometrically strong paths; a stochastic version that accounts for path-dependent RCS fluctuations could behave differently under Swerling-type targets.","If the communication symbols $x_{n,m}$ are random rather than known pilots, the power-allocation objective would need to be averaged over the data; the paper treats $X$ as given, leaving a data-dependent extension open."],"forward_implications":["In a multipath-rich cell, an OFDM-ISAC base station can detect targets whose direct line-of-sight echo is weak by pooling non-line-of-sight reflections.","The false-alarm threshold can be set once from the system dimensions $N$, $M$, and $L$, so transmit power and detector weights can be optimized without re-tuning the detection threshold.","Jointly optimizing subcarrier powers and detector weights yields better ROC performance than optimizing either one alone, and the improvement is driven mainly by the weight design.","The optimized detector removes the ghost targets that the unoptimized detector produces on the range-Doppler grid.","Communication service is protected during the optimization because every subcarrier's SNR at the user is kept above the required level."],"supporting_citations":[{"why":"supplies the delay-Doppler channel representation and SFFT mapping used to build the per-path channel matrices $H_l$.","marker":"[8]"},{"why":"supplies the OFDM radar multipath detection setup and the non-central F-distribution link between the non-centrality parameter and detection probability.","marker":"[5]"},{"why":"provides the GLRT/Neyman-Pearson detection framework that defines the likelihood ratio in the binary hypothesis test.","marker":"[9]"},{"why":"provides the generalized MANOVA formulation used to express the GLRT as the energy-ratio statistic in (12).","marker":"[10]"},{"why":"supplies the Craig-Sakamoto theorem that decouples the orthogonal projection terms into the compact F-distributed detector form.","marker":"[11]"}],"fun_headline_variants":["Multipath diversity boosts OFDM-ISAC detection","GLRT weights exploit multipath for ISAC detection","Joint power-weight design lifts OFDM-ISAC detection","Multipath echoes power GLRT in OFDM-ISAC","Non-LoS paths sharpen target detection in ISAC"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument depends on the discrete delay-Doppler channel vectors of different paths being exactly orthogonal, so the per-path projection matrices decouple and the false-alarm distribution is F; this exact orthogonality holds only on an idealized grid of distinct integer delay and Doppler taps, and the paper does not test how performance changes when real channels leave that grid.","fun_headline_variants_meta":{"raw":{"variants":["Multipath diversity boosts OFDM-ISAC detection","GLRT weights exploit multipath for ISAC detection","Joint power-weight design lifts OFDM-ISAC detection","Multipath echoes power GLRT in OFDM-ISAC","Non-LoS paths sharpen target detection in ISAC"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00018,"raw_usage":{"total_tokens":1311,"prompt_tokens":963,"completion_tokens":348,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":579,"completion_tokens_details":{"reasoning_tokens":281}},"tokens_in":579,"tokens_out":348,"duration_ms":3737,"temperature":1.0,"reasoning_tokens":281,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T20:31:20.500571+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the detector in a simulated two-path scenario in which the two paths land on the same discretized delay tap (or the same Doppler tap), so the projection matrices are no longer orthogonal, and measure the empirical false-alarm probability for several different power-allocation and weight choices; if the measured $P_{\\mathrm{FA}}$ shifts with the design variables, the central claim that the threshold depends only on $N$, $M$, and $L$ is falsified.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the delay-Doppler channel representation and SFFT mapping used to build the per-path channel matrices $H_l$."},{"cited_title":"Zhang, M","cited_arxiv_id":null,"evidence_quote":"supplies the OFDM radar multipath detection setup and the non-central F-distribution link between the non-centrality parameter and detection probability."},{"cited_title":"Rihan, E","cited_arxiv_id":null,"evidence_quote":"provides the GLRT/Neyman-Pearson detection framework that defines the likelihood ratio in the binary hypothesis test."},{"cited_title":"Wiesbeck and L","cited_arxiv_id":null,"evidence_quote":"provides the generalized MANOVA formulation used to express the GLRT as the energy-ratio statistic in (12)."},{"cited_title":"Hadani, S","cited_arxiv_id":null,"evidence_quote":"supplies the Craig-Sakamoto theorem that decouples the orthogonal projection terms into the compact F-distributed detector form."}],"review_version":1}