{"id":"1f0b6272-9ee5-4204-af0b-d0ec2967be86","arxiv_id":"2505.04968","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"An artificial-noise dynamic precoder for near-field multiuser systems scrambles passive eavesdroppers in both angle and distance while preserving legitimate user rates.","lead":"This paper designs a wireless precoding method that scrambles messages for nearby eavesdroppers without knowing where they are, while keeping signals clean for intended users. It exploits near-field spherical waves and large antenna arrays to focus security in both direction and distance, which far-field methods cannot do.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The AN-nulling condition in Sec. III-D is stated at the wrong layer: because the analog precoder F is rank-deficient (N_RF < N), the correct condition for zero AN at an eavesdropper is F^H h(re) not in span(F^H h(ru_1),...,F^H h(ru_M)), not h(re) not in span(H_U); the 'any undesired position'…","rationale":"The reader's weakest assumption correctly points to the AN vanishing when the eavesdropper's channel lies in the legitimate-user span. However, the manuscript's stated condition is not merely incomplete; it is stated at the wrong layer. The hybrid architecture inserts AN in the digital domain after the rank-N_RF analog combiner F, so the quantity that matters is F^H h(re), not h(re) itself. Because F^H has a large null space, the set of channel vectors that are invisible to the AN is much larger than the user span. This is a genuine mathematical flaw in the proof of the 'any undesired position' claim, not just an unquantified edge case. At the same time, the paper's core construction and the algebraic steps in Propositions 1 and 2 appear coherent, and the corrected condition may still hold for most physical positions. The numerical test described above would settle whether the flaw has practical consequences for the simulated array geometry. Since the reader already recommended CONDITIONAL and our concern supports that conditional reading without changing the overall verdict, the verdict remains unchanged.","tokens_in":25319,"tokens_out":15226,"duration_ms":163869,"concrete_test":"Numerically evaluate the AN leakage metric gamma(r) = ||H_M^perp F^H h(r)|| / ||F^H h(r)|| over a dense grid of eavesdropper positions in the near-field region, using the paper's setup (N=1600, N_RF=40, M=2). Flag positions that are far from the user span, i.e. with ||h(r)-proj_{span(H_U)}h(r)|| large, but have gamma(r) approximately zero or below a small threshold such as 1%. If any such position exists, the 'any undesired position' guarantee is disproved for the simulated geometry; if gamma(r) is uniformly large away from a small neighborhood of the users, the practical impact of the rank-deficient projection is limited.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Sec. III-D, after Eq. (35), the paper asserts security when h(re_q) is not in span(h(ru_1),...,h(ru_M)), claiming then H_M^perp F^H h(re_q) != 0. This implication is false for the hybrid architecture. The AN component at the eavesdropper is h(re_q)^H F H_M^perp W0(k), and it vanishes for all W0(k) iff F^H h(re_q) lies in col(H_M) = span(F^H h(ru_1),...,F^H h(ru_M)). Since F is N x N_RF with N_RF < N, F^H has an (N-N_RF)-dimensional null space. Hence every vector h_out = h(ru_m)+z with z in null(F^H) satisfies F^H h_out in col(H_M) while generically h_out is not in span(H_U). The set of channel vectors annihilated by the AN is (N-N_RF+M)-dimensional, much larger than the M-dimensional user span. The paper gives no argument that the 3-D manifold of physical eavesdropper positions avoids this preimage, nor does it quantify the size of near-degenerate regions where the AN component is small. The secrecy maps in Sec. V-G already show outage probability close to 1 near the user positions, consistent with this failure mode. This does not invalidate the algebraic core (Propositions 1-2 are internally sound), but it removes the 'any undesired position' guarantee and requires the claim to be restated in terms of projected channels.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a dynamic hybrid precoding scheme for near-field MU-MISO downlink secrecy. The analog precoder is obtained from the dominant right singular vectors of the legitimate-user channel matrix, and the baseband precoder is a symbol-level ZF design in the equivalent (projected) channel domain, with random artificial noise injected into the null space of the equivalent legitimate-user channel. The authors derive a power-constrained choice of the AN amplitude, expressions for average SINR, achievable rate, secrecy capacity, secrecy outage probability, and a secrecy zone, and support the analysis with simulations including constellation diagrams, BER maps, secrecy maps, and secrecy-rate comparisons against ZF and WMMSE. The paper claims secure transmission at 'any undesired position' without eavesdropper CSI, in both angle and distance dimensions.","tokens_in":25672,"tokens_out":5782,"duration_ms":61871,"significance":"If the central claim were fully supported, the paper would be a useful contribution: it extends artificial-noise directional modulation to multi-user near-field systems with a low-complexity hybrid architecture, gives a closed-form average-SINR analysis, and characterizes secrecy outage and secrecy-zone metrics in the LoS case. The algebraic derivations in Propositions 1 and 2 are internally consistent, the power-constrained AN construction in Algorithm 1 is clearly specified, and the simulation study is reasonably extensive. However, the paper's headline security guarantee rests on a condition about eavesdropper channel vectors that is stated at the wrong layer of the hybrid architecture, and the theoretical secrecy-rate analysis relies on a ratio-of-expectations approximation whose accuracy is not quantified. These issues do not invalidate the algebraic core, but they require the main claims to be restated and further supported before the paper can be accepted.","major_comments":[{"comment":"The security condition is stated incorrectly. The paper claims that secure transmission is ensured when h(re_q) is not in span(h(ru_1),...,h(ru_M)), and that then H_M^\\perp F^H h(re_q) \\neq 0. This implication is false for the hybrid architecture. The AN term at the eavesdropper is h^H(re_q) F H_M^\\perp W0(k), and it vanishes for all W0(k) if and only if F^H h(re_q) belongs to col(H_M) = span(F^H h(ru_1), ..., F^H h(ru_M)). Because F is N x N_RF with N_RF < N, the map F^H has an (N-N_RF)-dimensional null space, so every vector of the form h(ru_m) + z with z in null(F^H) yields zero AN at the eavesdropper while generically lying outside span(H_U). The set of channel vectors annihilated by the AN is therefore (N-N_RF+M)-dimensional, much larger than the M-dimensional user span. The paper gives no argument that the physical positions of eavesdroppers avoid this preimage, and the secrecy maps in Fig. 10 already show outage probability close to 1 near the user positions. The 'any undesired position' claim in the abstract and in Section III-D must be restated in terms of the projected channels, and the size of the protected region should be quantified.","section":"Section III-D, after Eq. (35)"},{"comment":"The average SINR is defined as a ratio of expectations, E[A]/(E[B]+\\sigma^2), rather than the expectation of the SINR ratio. This approximation is exact at the legitimate-user positions because the AN term cancels there and the interference vanishes deterministically, but it is not exact at eavesdropper positions, where the numerator and denominator depend on the same random W(k) and the same channel h(r). The subsequent secrecy-capacity approximation in Eq. (46) inherits this approximation, yet the paper does not provide a bound on the error or a discussion of the regimes in which the approximation is tight. Since the theoretical secrecy-rate claims are central to the paper, the authors should either justify the approximation more rigorously, provide a finite-N error bound, or clearly label the secrecy-rate expressions as heuristic approximations whose accuracy is only demonstrated by simulation.","section":"Section IV-A, Eq. (39)"},{"comment":"The multi-path analysis assumes that the channel at an unintended position r is independent of the legitimate-user channels and therefore independent of W(k), and that its covariance is the same R as that of the legitimate users. In a shared-scatterer geometry, which is the setup used in the simulations of Section V-D, the eavesdropper channels and the user channels are generated from the same scatterer positions, so these assumptions are questionable. The paper should either use a model in which the independence is explicitly enforced or discuss the error introduced when scatterers are shared. This is load-bearing for the claimed applicability of the proposed scheme to multi-path near-field channels.","section":"Section III-E and Eq. (43)"}],"minor_comments":[{"comment":"The phrase 'does not lie in the null subspace of the equivalent baseband channel matrix' is ambiguous; the paper should distinguish between the null space of H_M^H and the orthogonal complement of col(H_M).","section":"Section III-D, sentence before Remark 3"},{"comment":"The value of \\xi is set manually to 2\\times10^{-6} without checking that it satisfies the power constraint in Eq. (22) or explaining how it relates to the closed-form choices in Eqs. (27) and (34); please clarify.","section":"Section V-C3"},{"comment":"The doubly non-central F distribution is introduced without defining its parameters in the notation list; for readability, define the degrees of freedom and non-centrality parameters explicitly in the text rather than only in the PDF expression.","section":"Eq. (62)"},{"comment":"The comparison with ZF and WMMSE in Fig. 9(b) compares the proposed AN-aided scheme against baselines that do not include any AN or security mechanism; the authors should state that this is a comparison against non-secure benchmarks, otherwise the '20% higher secrecy rate' claim may be read as a comparison against secure alternatives.","section":"Section V-E1"},{"comment":"The notation for H_M^\\perp is not defined in the notation list; it is used before its formal definition in Eq. (20) and should be added for completeness.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper's algebraic core is sound, and the topic is timely for near-field physical-layer security. The main concern is that the security argument in Section III-D is stated for the wrong subspace: the AN is nulled whenever F^H h(re) lies in col(H_M), which is a much larger set than span(H_U). This undermines the 'any undesired position' claim in the abstract. This is a fixable issue if the authors restate the condition in the projected domain, quantify the protected region, and temper the corresponding claims. The other concerns (ratio-of-expectations approximation, multi-path independence) are secondary but should be addressed in revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take for your files.\n\nWhat's actually new: this extends AN-aided dynamic symbol-level precoding to multi-user near-field MU-MISO with a hybrid analog/digital architecture, and it adds a real performance analysis—average SINR, achievable rate, secrecy outage probability via a doubly non-central F distribution, and secrecy-zone maps. The algebraic core, Propositions 1 and 2, is correct. The idea of using the ELAA null space for symbol-level AN without eavesdropper CSI is reasonable, and the complexity claim is credible.\n\nThe main soft spot is Section III-D. The paper says security holds when h(re) is not in the span of user channels, then asserts H⊥_M F^H h(re)≠0. That implication is false for the hybrid architecture. The AN term at the eavesdropper is h^H(re) F H⊥_M W0(k), and it vanishes for all W0(k) when F^H h(re) lies in the column space of H_M = F^H H_U, i.e., in span(F^H h(ru_1),...,F^H h(ru_M)). Because F is N×N_RF with N_RF