{"id":"a3eda3df-0e93-4591-9e89-1a85b6ce583d","arxiv_id":"2411.17986","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For covert mmWave MIMO with finite-resolution DACs, the paper derives the warden's detection error probability and proposes AO and VSH hybrid beamformers to maximize the sum covert rate.","lead":"The paper designs hybrid beamforming for millimeter-wave covert communication where the transmitter uses low-resolution digital-to-analog converters, which add quantization noise that can help a warden detect the transmission. It proposes an alternating optimization scheme and a faster vector-space heuristic, claiming near-optimal covert rates as the antenna count grows, and shows simulation gains over two baselines.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 2 is proved for the unconstrained analog beamformer (60), while Algorithm 3 replaces it with the phase-only projection (83); no argument shows this projection preserves null-space, orthogonality, or power properties as N grows, so the VSH asymptotic claim is unproven for the…","rationale":"The paper has a useful system model and a plausible design pipeline; the AQN starting point is standard and I do not object to it as the primary risk. The reader's weakest_assumption identifies AQN, but the more immediate and internal problem is the proof-to-implementation mismatch in the VSH scheme: Proposition 2 proves an asymptotic result for the complex-valued beamformer (60), while the simulated algorithm uses the phase-only beamformer (83). No lemma quantifies the effect of this projection on the null-space property Ωw FR ≈ 0, on FR^H FR = I_K, or on user orthogonality. This directly affects the strongest claimed theoretical result and requires either a corrected proof, a restricted claim, or simulations of an ideal constant-modulus construction that provably preserves the required properties. The reader's secondary concern about the AO QCQP constraints is also valid: in (36b)–(36c), the Rq term present in (15) and (25) is dropped from Q1 and Q2, so Algorithm 2 may terminate at an FR that violates the original power or covertness constraints. Both issues are fixable, but they are load-bearing; a conditional verdict is appropriate until the VSH proof is reconciled with the implemented phase projection and the AO constraints are corrected.","tokens_in":20808,"tokens_out":13904,"duration_ms":132953,"concrete_test":"Run Algorithm 3 for N = 64, 128, 256, 512 with the channel model of Fig. 6 and fixed Willie/user angles, recording for each N: (i) Rsum from (13) using the final phase-projected bFR and normalized bFB, (ii) I(s;y) from (57) evaluated at the ideal eFR/eFB of (60)–(61), and (iii) the actual covertness LHS in (25) before and after the normalization (84). If Rsum_final/I does not approach 1 as N grows, or if |h_w^H bFR|² does not decay at the same rate as |h_w^H eFR|², then Proposition 2 does not transfer to the implemented VSH scheme.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's theoretical centerpiece is Proposition 2, used in the abstract to claim that the VSH scheme's SCR approaches channel mutual information as N → ∞. The proof applies to eFR = V_w^(0)[Vtilde]:,1:K in (60), an unconstrained complex matrix with orthonormal columns. It relies on three properties: eFR lies in the null space of Ωw (covertness), eFR^H eFR = I_K (power normalization), and the user channels are asymptotically orthogonal after this transformation. Algorithm 3, however, outputs bFR = exp(j∠eFR) via (83) and then rescales FB via (84). Phase projection generically destroys all three properties. For a generic unit-norm vector v with iid entries, the normalized correlation between v and exp(j∠v) tends to √(π/2)/2 ≈ 0.886, not 1, so 'approaching (60)' in (83) is not an asymptotic equivalence in any norm relevant to the proof. The text cites [34] for eliminating the constant-modulus constraints by doubling the number of RF chains, but N_RF is not doubled; only the phase projection is used. Consequently, Proposition 2 establishes the MI-equivalence for an intermediate design that is never simulated, while the VSH curves in Figs. 2–8 use (83), so the central asymptotic claim for the implemented VSH scheme is unsupported.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers covert multiuser mmWave MIMO downlink with a fully-connected hybrid beamforming architecture and finite-resolution DACs at the transmitter. Under the additive quantization noise (AQN) model, the authors derive an approximate detection error probability for a warden Willie, formulate a sum covert rate (SCR) maximization problem with power and covertness constraints, and propose two hybrid beamforming designs: an alternating optimization (AO) scheme based on fractional programming, quadratic transformation, and inner majorization-minimization, and a lower-complexity vector-space heuristic (VSH) scheme. The paper claims that, as the number of antennas grows to infinity, the SCR achieved by the VSH scheme approaches the channel mutual information, and that simulations show both proposed schemes outperform fully-digital and maximum-ratio baselines. The main theoretical results are Proposition 1 (warden KL divergence), Proposition 2 (asymptotic SCR-to-mutual-information equivalence for the VSH design), and the associated Algorithms 1-3.","tokens_in":21099,"tokens_out":18270,"duration_ms":167015,"significance":"If the claims are correct, the paper makes a useful contribution by introducing finite-resolution DAC hardware constraints into covert mmWave MIMO hybrid beamforming, a combination not treated in prior work. The AQN-based warden detection analysis and the AO algorithm are in line with current methodology in the field, and the VSH scheme offers a low-complexity alternative with a precise asymptotic performance target. The paper also provides reproducible simulation evidence for convergence speed, SCR, and energy efficiency. However, the central asymptotic claim and the constraint derivations contain load-bearing gaps: the power and covertness constraint matrices in the analog beamforming subproblem omit the quantization-noise covariance term, and the asymptotic proof for the VSH scheme applies to an unconstrained analog beamformer that is not the one implemented after phase projection. These issues affect the correctness of the proposed algorithms as stated and the support for the main theoretical claim.","major_comments":[{"comment":"The matrices Q1 and Q2 used in the analog beamforming subproblem omit the quantization-noise contribution. Substituting (3) into the power constraint (15) gives F_R[(1-β)^2 F_BF_B^H + β(1-β)diag(F_BF_B^H)]F_R^H ≤ P_max, and the vectorized quadratic form should therefore contain the term β(1-β)diag(F_BF_B^H)^T ⊗ I_N in Q1 and β(1-β)diag(F_BF_B^H)^T ⊗ Ω_w in Q2. Equations (37)-(38) keep only the first term, so the constraints enforced by (36b)-(36c) are weaker than those of the original problem (14). The same omission propagates into the majorization functions g1 and g2 in (42)-(43), so a solution returned by Algorithm 1 need not satisfy the true power or covertness constraints. Notably, the digital beamforming subproblem in (51), (54), and (55) does include the diag term, which indicates the omission in (37)-(38) is not an intentional modeling choice.","section":"Section III-B.1, Eqs. (36)-(38)"},{"comment":"Proposition 2 is proved for the unconstrained analog beamformer eFR = V_w^(0)[Ṽ]:,1:K in (60), which has orthonormal columns and lies in the null space of Ω_w. The VSH scheme actually simulated in Figs. 2-8 uses instead the phase-projected beamformer bFR = exp(j∠eFR) from (83), followed by rescaling of bFB in (84). No argument is supplied that phase projection preserves the null-space property Ω_w bFR = 0, the orthonormality of the columns, or the asymptotic channel-orthogonality properties used in the proof. The citation [34] removes constant-modulus constraints by doubling the number of RF chains, but N_RF is not doubled in Algorithm 3. Moreover, the power allocation optimization in (72)-(81) is derived under the assumption F_R = eFR; after (83) the effective user and warden channels change, so the optimized ζ is no longer tailored to the projected beamformer. Consequently the asymptotic \"SCR approaches mutual information\" claim is established only for an intermediate design that is never simulated, and the corresponding claim for the implemented VSH scheme is unsupported.","section":"Section IV-A, Proposition 2 and Algorithm 3"},{"comment":"The statement that bFR = exp(j∠eFR) is \"the optimal solution approaching (60)\" is not an asymptotic equivalence in any norm relevant to the proof. For a generic unit-norm vector v with iid entries, the normalized inner product between v and exp(j∠v) tends to approximately 0.886, not to 1, and the component of exp(j∠v) orthogonal to v has non-vanishing norm. More importantly for covertness, a vector that is exactly orthogonal to the steering vectors spanning Ω_w will, after phase projection, generally have an O(1) projection onto each such steering vector, so the left-hand side of the covertness constraint (25) is not inherited from eFR. The scaling in (84) can force feasibility, but only at the cost of a rate reduction that is not accounted for in Proposition 2. The paper should either prove the needed preservation properties for the projected beamformer, or explicitly state that Proposition 2 applies to the unconstrained design and treat the projected VSH as a heuristic supported only by simulation.","section":"Section IV-A, Eq. (83)"}],"minor_comments":[{"comment":"The sentence \"Pe = 1 means that Willie always makes error detection\" is inaccurate: P_FA + P_MD = 1 is equivalent to P(D1|H0) = P(D1|H1), i.e., Willie's decision is statistically independent of the true hypothesis, not that he always errs. This does not change the optimization because the same convention is used consistently, but the wording should be corrected.","section":"Section II-B, Eq. (11)"},{"comment":"The last diagonal entry in the expression for Υ should read ĕ_K R_q ĕ_K^H rather than ĕ_K F_R R_q ĕ_K^H; as printed, the dimensions do not match.","section":"Section IV-A, Eq. (68)"},{"comment":"The notation ϑ ∈ R^{K×K} appears to be a typo: ϑ_k is a K-dimensional vector for each k, so the statement should be ϑ_k ∈ R^K or ϑ ∈ R^K.","section":"Section IV-B, Eq. (73) and following"},{"comment":"In the term ϑ_k^T(ζ^T(Π_k+Θ_k)ζ + σ_k^2 I_K)ϑ_k, the quantity ζ^T(Π_k+Θ_k)ζ is a scalar, so the expression is a scalar times ||ϑ_k||^2; writing the scalar denominator explicitly would improve readability and avoid apparent dimension mismatch.","section":"Section IV-B, Eq. (76)"},{"comment":"After the phase projection and the scaling in (84), the digital beamformer bFB no longer satisfies the zero-interference property (69) exactly, and the power allocation from (81) is computed for a different effective channel. This should be acknowledged in the text surrounding Algorithm 3.","section":"Algorithm 3, steps 10-11"}],"recommendation":"major_revision","confidential_remarks":"The two main issues are load-bearing but appear fixable in principle: the missing diag term in Q1/Q2 can be added, and the VSH asymptotic claim can be either proved for the projected beamformer or explicitly restricted to the unconstrained design with the projected version presented as a heuristic. I would not recommend rejection, since the AO framework and the AQN-based warden analysis are useful and the simulation study is substantial. The paper should be sent back for a careful revision that reconciles the theoretical statements with the implemented Algorithm 3 and re-verifies the simulation conclusions under the corrected constraints."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper extends covert mmWave hybrid beamforming to finite-resolution DACs under the additive quantization noise model. That problem setup is genuinely new, and Proposition 1—the derivation of Willie's detection error probability—is algebraically clean. The AO scheme is a standard but competent assembly of fractional programming, quadratic transform, and inner majorization-minimization. The VSH heuristic is a reasonable low-complexity alternative.\n\nTwo soft spots matter. First, the transformation to the analog QCQP in (36)–(38) drops the R_q term from the power and covertness constraints. The original constraints (15) and (25) both include β(1−β)diag(F_BF_B^H), but (37) and (38) keep only the (1−β)^2(F_BF_B^H)^T part. For b=1 that is a large effect, so the analog beamformer subproblem is enforcing the wrong constraints. This needs a fix or a clear explanation.\n\nSecond, and more serious, Proposition 2 proves the asymptotic SCR-to-mutual-information equivalence for e_FR in (60), an unconstrained orthonormal matrix in the null space of Ω_w. Algorithm 3 outputs b_FR = exp(j∠e_FR) via (83) and then rescales F_B. Phase projection does not preserve the null-space, orthogonality, or power properties the proof relies on; the cited constant-modulus-elimination trick from [34] (doubling RF chains) is not actually used. So the theorem does not cover the implemented VSH scheme. The headline claim that the VSH scheme approaches channel mutual information as N→∞ is unproven for the simulated algorithm. This is a load-bearing gap, not a stylistic quibble.\n\nSmaller issues: the simulations compare only against FDBO and MRT, not the closest prior covert hybrid beamforming or low-resolution-DAC secure beamforming schemes, despite the abstract claiming improvement over \"existing schemes.\" The AQN model is an approximation, but that is a standard modeling assumption and I would treat it as minor.\n\nMy take: the problem is relevant and the warden-detection derivation is solid. The AO machinery is credible once the constraint bug is fixed. But the central VSH asymptotic claim is currently proved for a design that is never simulated. A serious referee should push for the corrected constraint transformation and either a proof that phase projection preserves the asymptotic properties or a revised claim. If those are addressed, this could be a solid JSAC-level paper; as written, I would not rely on its theoretical results.","headline":"The warden-detection derivation is clean and the problem is new, but the AO constraint transformation drops a quantization-noise term and the VSH asymptotic proof applies to a beamformer the algorithm never simulates.","tokens_in":21641,"tokens_out":5486,"would_cite":true,"duration_ms":49810,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94A15","94A40"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper shows that low-resolution DACs can be incorporated into covert mmWave MIMO beamforming, and that a low-complexity heuristic reaches channel mutual information as the antenna array grows.","keywords":["Covert communications","Hybrid beamforming","Finite-resolution DAC","mmWave MIMO","Additive quantization noise","Sum covert rate","Multiuser communications"],"falsifier":"Run a Monte Carlo simulation of an actual $b$-bit uniform quantizer at Alice instead of the AQN approximation: generate Willie's observations under $H_0$ and $H_1$, apply the optimal likelihood-ratio test, and compare the empirical detection error probability $P_e$ with (11)-(25). A gap beyond the simulation margin would show that the derived covertness constraint is not the true constraint.","tokens_in":20529,"feed_emoji":"📡","tokens_out":9537,"duration_ms":77149,"temperature":0.7,"pith_summary":"The paper asks whether a millimeter-wave base station with cheap, low-resolution digital-to-analog converters can serve multiple users covertly while a warden tries to detect the transmission. Using the additive quantization noise model, it derives a closed-form expression for the warden's detection error probability and turns covertness into an explicit trace constraint. It then solves the resulting nonconvex beamforming problem with two schemes: an alternating optimization scheme and a much cheaper vector-space heuristic. The central theoretical result is that the heuristic's sum covert rate becomes equal to the channel mutual information as the number of antennas grows to infinity. Simulations indicate that both schemes outperform fully-digital and maximum-ratio baselines, especially when the DACs use very few bits.","feed_headline":"Low-bit DACs do not have to break covert mmWave links","feed_subtitle":"A low-complexity scheme reaches channel mutual information as antenna count grows, even with 1-bit converters.","key_machinery":"The load-bearing objects are the additive quantization noise model of Eqs. (2)-(3), which replaces the nonlinear DAC with a linear gain $(1-\\beta)$ plus Gaussian quantization noise with covariance $\\beta(1-\\beta)\\mathrm{diag}(F_BF_B^H)$; the SIQNR in Eq. (12) and the sum covert rate in Eq. (13); and, for the heuristic, the null-space construction of Eqs. (60)-(61), where the analog beamformer lies in the null space of Willie's channel covariance $\\Omega_w$ and the digital beamformer uses block diagonalization to remove multiuser interference. These objects convert covertness into a trace constraint, the rate objective into a quadratic form, and the asymptotic argument into a statement about orthogonal mmWave channel vectors.","core_discovery":"On the paper's own terms, the discovery is that finite-resolution DAC quantization noise does not have to be a covertness leak that forces a rate loss; it can be engineered around in the hybrid beamformer. The quantized transmit signal is modeled as $(1-\\beta)F_Bs+\\eta_q$ with $\\eta_q \\sim \\mathcal{CN}(0, \\beta(1-\\beta)\\mathrm{diag}(F_BF_B^H))$, which lets the authors compute Willie's optimal detection error probability and replace the covertness constraint with the trace bound in (25). The AO scheme iterates between analog and digital beamformers, using fractional programming, a QCQP formulation with constant-modulus constraints, and an inner majorization-minimization method. The VSH scheme projects the analog beamformer into the null space of Willie's channel covariance and block-diagonalizes the digital beamformer; Proposition 2 proves that for $N\\to\\infty$ the SCR in (13) equals the mutual information in (57). In simulations the AO scheme matches the fully-digital baseline at high DAC resolution and beats it under tight covertness with 1-bit DACs, while the VSH scheme serves as a fast initialization.","pith_inferences":["If the AQN model is accurate enough, the same null-space reasoning should extend to wideband mmWave systems, where frequency-dependent beams can also be placed in Willie's channel null space.","The asymptotic equivalence suggests that in massive arrays, covertness can come from array geometry rather than from artificial noise; a testable prediction is that the required jamming power drops to zero as $N$ grows.","Because the warden's optimal detector assumes Gaussian quantization noise, a stronger test is to let Willie run a data-driven detector on the true quantized waveform; under that threat model the trace constraint in (25) may need to be re-derived.","The power-allocation subroutine in (72)-(81) is separable from the beam directions and could be reused for other hybrid or fully-digital architectures with finite-resolution DACs."],"forward_implications":["With 1-bit DACs under a tight covertness requirement, hybrid beamforming can outperform a fully-digital beamformer because the analog stage reduces quantization-noise leakage toward Willie.","As the number of antennas grows, the VSH scheme's sum covert rate approaches the channel mutual information, so cheap hardware plus a large array does not sacrifice the asymptotic rate.","The VSH scheme provides an initialization for the AO scheme that converges faster and to a higher converged sum covert rate than beam-training initialization.","DAC resolution has an energy-efficiency optimum: increasing the number of bits beyond about five yields little sum-rate gain while power consumption keeps rising."],"supporting_citations":[{"why":"Supplies the additive quantization noise model that expresses DAC output as a linearly scaled signal plus Gaussian noise.","marker":"[24]"},{"why":"Provides the underlying quantization-noise covariance result used in Eq. (3).","marker":"[25]"},{"why":"Supplies the square-root law and the optimal hypothesis-testing and Pinsker machinery used to bound Willie's detection error.","marker":"[4]"},{"why":"Gives the prior covert mmWave full-duplex setup whose detection-error and covertness-constraint derivation pattern is extended here.","marker":"[19]"},{"why":"Supplies the fractional programming and quadratic transformation used to make the sum covert rate objective tractable.","marker":"[29]"},{"why":"Provides the QCQP-with-constant-modulus theory and the inner majorization-minimization method used in the analog beamformer step.","marker":"[31]"},{"why":"Shows how to relax constant-modulus constraints by doubling the number of RF chains, used in the VSH analog beamformer construction.","marker":"[34]"},{"why":"States the information-theoretic fact that the sum rate is at most the mutual information, the target of Proposition 2.","marker":"[27]"}],"fun_headline_variants":["Low-bit DACs don't break covert mmWave links","Covert mmWave with finite DACs: a beamforming fix","Quantized beams keep mmWave covertness intact","Beamforming design tames low-bit DACs for covert mmWave","Covert mmWave survives 1-bit DAC quantization"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole design assumes the additive quantization noise model, in which a $b$-bit DAC behaves like a linear scaling plus independent Gaussian noise; if the true quantization error is correlated with the signal or non-Gaussian, the derived detection probability and the power and covertness constraints are not the true ones.","fun_headline_variants_meta":{"raw":{"variants":["Low-bit DACs don't break covert mmWave links","Covert mmWave with finite DACs: a beamforming fix","Quantized beams keep mmWave covertness intact","Beamforming design tames low-bit DACs for covert mmWave","Covert mmWave survives 1-bit DAC quantization"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000282,"raw_usage":{"total_tokens":1700,"prompt_tokens":1010,"completion_tokens":690,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":626,"completion_tokens_details":{"reasoning_tokens":607}},"tokens_in":626,"tokens_out":690,"duration_ms":6042,"temperature":1.0,"reasoning_tokens":607,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:38:35.362859+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a Monte Carlo simulation of an actual $b$-bit uniform quantizer at Alice instead of the AQN approximation: generate Willie's observations under $H_0$ and $H_1$, apply the optimal likelihood-ratio test, and compare the empirical detection error probability $P_e$ with (11)-(25). A gap beyond the simulation margin would show that the derived covertness constraint is not the true constraint.","supporting_citations":[{"cited_title":"Achievable rate and energy efficiency of hybrid and digital beamforming receivers with low resolution ADC,","cited_arxiv_id":null,"evidence_quote":"Supplies the additive quantization noise model that expresses DAC output as a linearly scaled signal plus Gaussian noise."},{"cited_title":"Robust predictive quantization: Analysis and design via convex optimization,","cited_arxiv_id":null,"evidence_quote":"Provides the underlying quantization-noise covariance result used in Eq. (3)."},{"cited_title":"Covert rate optimization of millimeter wave full-duplex communications,","cited_arxiv_id":null,"evidence_quote":"Gives the prior covert mmWave full-duplex setup whose detection-error and covertness-constraint derivation pattern is extended here."},{"cited_title":"QCQP with extra constant modulus constraints: Theory and application to SINR constrained mmWave hybrid beam- forming,","cited_arxiv_id":null,"evidence_quote":"Provides the QCQP-with-constant-modulus theory and the inner majorization-minimization method used in the analog beamformer step."},{"cited_title":"Hybrid beamforming for multi-user massive MIMO systems,","cited_arxiv_id":null,"evidence_quote":"Shows how to relax constant-modulus constraints by doubling the number of RF chains, used in the VSH analog beamformer construction."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"States the information-theoretic fact that the sum rate is at most the mutual information, the target of Proposition 2."}],"review_version":1}