{"id":"f03c0e8a-4e6e-4abf-935c-6c9ab7d418f5","arxiv_id":"2412.13447","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":5,"one_line_summary":"A joint autocorrelation and cross-correlation scheme that first estimates wavefront curvature and then angle, enabling lower-complexity near-field channel and position estimation for XL-MIMO.","lead":"This paper proposes a scheme to estimate near-field wireless channels and user positions by separating two coupled signal parameters: the angle and the curvature of the incoming wave. The authors claim their gradient-descent algorithm runs in linear time and performs well at low signal-to-noise ratios, but the paper's complexity analysis and Cramer-Rao lower bound formulas contain apparent errors.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (17) and (23) use e^{+ik p1 x^2} to remove the quadratic phase, but h contains e^{+ik p1 x^2}; with perfect estimation the 'far-field' signal is e^{ik(2p1 x^2+p2 x)}, so the JAC decoupling step fails as written.","rationale":"The reader's weakest assumption concerns the sinc/Dirichlet approximation, which affects p1 estimation accuracy in certain regimes. The sign error in the compensation step is more fundamental: even in the ideal noiseless case with perfect p1, the scheme as written does not produce a far-field signal, so the claimed decoupling is not achieved by the described algorithm. This makes the central claim internally inconsistent rather than merely approximate. I therefore cannot endorse the paper as is; the reader's REJECT verdict is appropriate, though for a different primary reason than the stated weakest assumption. A corrected sign would make the core idea testable; without code, the simulations cannot confirm which sign was actually used. Hence the verdict remains REJECT/UNCHANGED, but the manuscript needs a substantive correction before any conditional acceptance.","tokens_in":17472,"tokens_out":12644,"duration_ms":105855,"concrete_test":"Simulate a noiseless single-user case (e.g., N=200, d=lambda/2, r=10 m, theta=30 deg). Generate h[n] = e^{ik(p1 n^2 d^2 + p2 n d)}. Apply Eq. (23) with \\hat p1 = p1 and run MUSIC: the spectral peak will not occur at p2 because the residual quadratic phase 2p1 n^2 d^2 remains. Repeat with the compensation factor e^{-ik \\hat p1 n^2 d^2}; only then does MUSIC peak at p2. This directly settles whether Eqs. (17)/(23) implement the claimed decoupling.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The JAC scheme's second stage, Eq. (17), claims that ~h(x) = e^{ik \\hat p1 x^2} h(x) is a far-field signal. Under the paper's own channel model h(x) = e^{ik(p1 x^2 + p2 x)} with p1 = -cos^2(theta)/(2r) < 0, the quadratic phase is e^{ik p1 x^2}. Multiplying by e^{+ik \\hat p1 x^2} with \\hat p1 = p1 doubles the quadratic phase, yielding e^{ik(2p1 x^2 + p2 x)}, which is not a plane wave. The correct compensating factor is e^{-ik \\hat p1 x^2}. The same plus sign appears in Eq. (23), Algorithm 1 step 6, and Algorithm 2 step 10. Thus, as written, the construction does not reduce the near-field problem to far-field AoA estimation, and the central decoupling claim fails even in the ideal noiseless case with perfect p1 knowledge. This is an internal algebraic inconsistency, not a numerical approximation issue, and it affects both proposed algorithms. If the simulations used the opposite sign in code, the paper's equations do not describe the simulated system, so the central claim as stated is unsupported.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper addresses near-field channel and position estimation for an extremely large MIMO ULA. It models the near-field channel phase as k(p1 x^2 + p2 x), where p1 (CoA) and p2 (AoA) are coupled. The proposed JAC scheme estimates p1 from the magnitude of the spatial autocorrelation of the received signal, exploiting that this magnitude depends only on p1, then compensates the quadratic phase to obtain an equivalent far-field channel from which p2 is estimated via MUSIC. Two p1 estimators are introduced: JAC-ISF (inverse-sinc) and JAC-GD (gradient descent). The paper also derives CRLBs for p1, p2, θ, and r, and claims O(TN) time complexity for the proposed algorithms.","tokens_in":17795,"tokens_out":9756,"duration_ms":82903,"significance":"If the decoupling construction and the complexity and CRLB analyses were correct, the contribution would be practically relevant: reducing a coupled two-parameter near-field search to two one-dimensional problems with linear-time scaling is a useful idea for XL-MIMO systems. The autocorrelation decoupling property in Eq. (15) is mathematically correct, and the simulation study is extensive. However, the central construction as written contains a systematic sign error that invalidates the decoupling step, and the CRLB and complexity analyses contain separate load-bearing errors. The significance of the claimed contributions therefore cannot be credited in the present form.","major_comments":[{"comment":"The quadratic-phase compensation is implemented with the plus sign: Eq. (17) defines ~h(x) = e^{+ik\\hat p1 x^2} h(x), and Eq. (23) defines ~Y[n,t] = e^{+ik\\hat p1(nd)^2} Y[n,t]. Given the channel model h(x) = e^{ik(p1 x^2 + p2 x)} with p1 = -cos^2(θ)/(2r) < 0, this multiplication produces e^{ik(2p1 x^2 + p2 x)}, which is not a plane wave; the quadratic phase is doubled, not removed. The correct compensation factor is e^{-ik\\hat p1 x^2}. As written, the JAC scheme does not transform the near-field problem into a far-field AoA estimation problem even in the ideal noiseless case with perfect p1 knowledge. This is an internal algebraic inconsistency affecting both proposed algorithms; if the simulation code used the opposite sign, then the equations in the manuscript do not describe the simulated system.","section":"Section III-C (Eq. (17)), Section IV (Eq. (23)), Algorithm 1 step 6, Algorithm 2 step 10"},{"comment":"The CRLB formulas are dimensionally inconsistent. In Eq. (40), the bracketed numerator term d^2 sin^2(θ)(f(0)f(4)-f^2(2)) has units of length^6, the term -2 r d sin(θ)(f(1)f(2)-f(0)f(3)) has units of length^5, and the term r^2(f(0)f(2)-f^2(1)) has units of length^4, so these terms cannot be added. The denominator similarly mixes powers of length. Consequently, CRLB(r) as printed is not a valid variance expression with units m^2. The same dimensional problem appears in the intermediate expressions (35)-(36) and in CRLB(θ) in Eq. (39). Since these CRLBs are used as benchmarks in Fig. 11, the comparison against the proposed algorithms is not meaningful unless the FIM computation is corrected and re-verified.","section":"Section V-A, Eqs. (35)-(40)"},{"comment":"The claimed total complexity O(TN) omits the cost of the MUSIC stage used to estimate p2. MUSIC requires computing a sample covariance matrix from T snapshots, which costs O(T N^2), followed by an eigendecomposition, typically O(N^3), and a spatial spectrum search. Even a simplified maximum-eigenvalue variant does not reduce the eigen-decomposition cost to O(TN). Therefore the statement 'the process of estimating p2 is the MUSIC algorithm, so the time complexity is O(TN)' is incorrect, and the complexity comparison in Table I is not supported. Since low time complexity is a central advertised contribution, this requires a careful re-analysis of the second-stage cost.","section":"Section V-B and Table I"},{"comment":"The p1 estimator is based on the continuous sinc approximation of the autocorrelation in Eq. (15), but the actual sample autocorrelation computed in Eq. (19) is a finite sum (a Dirichlet-like kernel), not the continuous integral. The paper does not state the condition under which the continuous approximation is accurate. For the simulated array size (N=200) and distances (10-80 m), the argument k p1 η d^2 (N-ξ) can be large over part of the lag range, so the approximation can bias the p1 estimate. The authors should provide a validity condition or use the exact finite-sum autocorrelation model, and re-check the p1 estimation and localization results accordingly.","section":"Section III-C and Section IV-A, Eqs. (15), (19), (22)"}],"minor_comments":[{"comment":"The notation list refers to 'binary norm'; this should presumably be the Euclidean norm or ℓ2-norm.","section":"Notations paragraph"},{"comment":"The sentence 'To address these gaps, We need to search...' uses a capital 'We' mid-sentence; it should be lowercase.","section":"Section I, third paragraph"},{"comment":"The spatial Fourier kernel in the derivation of H(κ) appears with the opposite sign (e^{+ikκ(x-x0)}), whereas the stated result sinc(β(x0)-κ) requires the kernel e^{-ikκ(x-x0)}; as printed the exponent is inconsistent.","section":"Section III-B, Eq. (14)"},{"comment":"The text and Table I refer to the benchmark as P-SOWP, but Figs. 7 and 8 use P-SOMP; please unify the naming.","section":"Section VI-A and Figs. 7-8"},{"comment":"RMSE is defined as the squared ℓ2 norm ||p - \\hat p||_2^2, but the standard root mean-squared error should be the unsquared norm; the plotted units (m) indicate the square root was intended.","section":"Eq. (44)"},{"comment":"The section title uses 'Cramer's Lower Bound' and the text uses 'cramer-rao'; the standard spelling is 'Cramér-Rao lower bound'.","section":"Section V-A title"}],"recommendation":"reject","confidential_remarks":"The systematic plus-sign error in the core decoupling step appears in Eqs. (17) and (23) and in both algorithm listings, which means the written derivation does not define the scheme that is claimed. If the code used the correct opposite sign, the manuscript still fails to describe the simulated system. Combined with the dimensionally inconsistent CRLB formulas and the unsupported O(TN) complexity claim, the paper is not in a publishable state. I would not recommend a major revision unless the authors are prepared to re-derive the central equations, correct the CRLB, and re-run the experiments."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new piece here is the JAC scheme: take the spatial autocorrelation of the received signal, note that its magnitude depends on the curvature p1 but not on the angle p2, estimate p1 first, then estimate p2 with MUSIC. That decoupling is correct in the continuous model, and it is a sensible way to turn a 2D near-field search into two 1D problems. The two algorithms, arcsinc inversion and gradient descent on the sinc loss, are natural implementations. If the scheme worked, the complexity savings over polar-domain codebook methods would be a real contribution.\n\nThe problem is that the scheme does not work as written. Equation (17) (and its discrete version, Eq. 23, plus Algorithm 1 step 6 and Algorithm 2 step 10) claims that multiplying h(x) by e^{+ik p1 x^2} removes the quadratic phase. But under the paper's own model, h(x) = e^{ik(p1 x^2 + p2 x)} with p1 < 0. With perfect p1, the product is e^{ik(2p1 x^2 + p2 x)}, which is still quadratic. The compensating factor has to be e^{-ik p1 x^2}. This is not an approximation issue; it is an algebraic sign error repeated consistently. Unless the simulations used the opposite sign in code, the equations do not describe the simulated system.\n\nThe other soft spots are serious but less fatal. The CRLB expressions in Eqs. (35)-(40) are dimensionally inconsistent; in Eq. (40), the bracket adds terms scaling as m^6, m^5, and m^4. The O(T N) complexity claim is not supported: computing the autocorrelation for all lags naively costs O(T N xi), so O(T N^2) when xi ~ N/2, and MUSIC on an N-antenna array needs an eigendecomposition that is not O(T N). The sampled autocorrelation is a Dirichlet-like sum, not exactly the sinc in Eq. (15), and no validity condition is given. There is no code and no error bars, so the simulation curves are hard to check.\n\nCredit where due: the paper compares honestly against polar-domain baselines, admits JAC-ISF saturates worse than P-SIGW at high SNR, and the citation pattern looks appropriate.\n\nWho should read this: anyone working on near-field XL-MIMO estimation. The decoupling idea deserves to be in the literature in corrected form. As submitted, I would not accept it. I would want a referee to see it, mainly to confirm the sign fix and the CRLB rederivation, but the authors need to fix the core algebra and rerun the simulations before this is publishable.","headline":"The autocorrelation decoupling trick is real and worth knowing, but a sign error in the far-field conversion step makes the central claim fail as written.","tokens_in":18317,"tokens_out":5986,"would_cite":false,"duration_ms":58380,"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":"Near-field channel and position estimation can be reduced to two sequential one-dimensional searches by estimating wavefront curvature from the spatial autocorrelation before estimating angle.","keywords":["near-field channel estimation","extremely large MIMO","curvature of arrival","angle of arrival","spatial autocorrelation","position estimation","spatial Doppler","Cramér–Rao lower bound"],"falsifier":"Run a noiseless simulation with a known user position, compute the exact sampled autocorrelation of the received ULA signal at several lags, and compare the location of its first null with the prediction $k p_1 \\eta d^2 (N-\\eta) = \\pi$ from the sinc model; a systematic mismatch at large array sizes or short distances would show that the decoupling identity used in equations (15) and (22) does not hold in the regime where the paper applies it.","tokens_in":17296,"feed_emoji":"📡","tokens_out":9645,"duration_ms":80546,"temperature":0.7,"pith_summary":"This paper tries to establish that the two parameters of a near-field channel—the curvature of the arriving wavefront (CoA) and its angle at the reference antenna (AoA)—do not need a joint two-dimensional search. The key statistic is the magnitude of the spatial autocorrelation of the received signal, which the paper shows equals an absolute sinc function whose argument contains the curvature parameter $p_1$ but not the angle parameter $p_2$. That lets the curvature be estimated first and then removed from the signal, reducing the remaining problem to a standard far-field angle estimation. The reported consequence is that near-field channel and position estimation time complexity falls from multiplicative forms such as $O(T N S)$ to linear $O(T N)$, and the authors back this with two algorithms (closed-form inverse-sinc and gradient descent) and a Cramér–Rao lower bound for position. This matters because it offers a path to practical near-field beam training and localization in extremely large MIMO systems.","feed_headline":"Autocorrelation trick turns 2D near-field search into two 1D searches","feed_subtitle":"Autocorrelation depends only on wavefront curvature, so angle and distance are estimated one after the other.","key_machinery":"The load-bearing object is the spatial autocorrelation function of the received signal across the array, together with its amplitude identity $c(\\chi) = |\\operatorname{sinc}(k p_1 \\chi (D - \\tilde{D}))|$. Because the cross term containing $p_2$ cancels when $h(x)$ is multiplied by $h^*(x-\\chi)$, the autocorrelation magnitude is a function of the curvature $p_1$ alone; this is the mechanism that breaks the CoA–AoA coupling. The paper motivates the identity by drawing a duality between the near-field channel and the Doppler effect in time-domain signal processing, treating the array as a spatial frequency axis and the curvature as a spatial Doppler rate, and it uses the resulting coherence-space picture to justify estimating $p_1$ from the width of the autocorrelation main lobe.","core_discovery":"The central claim is that the amplitude of the spatial autocorrelation function of a near-field received signal, $c(\\chi) = |\\operatorname{sinc}(k p_1 \\chi (D - \\tilde{D}))|$, is independent of the AoA parameter $p_2$. Since $p_1$ controls the curvature of the spherical wavefront and $p_2$ controls its direction, this identity decouples the two coupled near-field parameters. The paper argues that $p_1$ can therefore be estimated first from the autocorrelation alone, after which multiplying the received signal by $e^{-jk\\hat{p}_1 x^2}$ converts the remaining problem into an equivalent far-field channel whose angle $p_2$ is found by a standard cross-correlation or MUSIC step. On the paper's own terms, this makes near-field channel estimation and user position estimation solvable by two sequential one-dimensional searches instead of one two-dimensional search, with time complexity $O(T N)$.","pith_inferences":["The same cancellation of the linear phase term should hold for uniform planar or circular arrays, where the curvature becomes a multi-dimensional parameter; the paper only treats a uniform linear array, so this is an extension, not a claim of the paper.","Because the sinc approximation is exact for the continuous integral but only approximate for the finite sum, there is a predictable small-argument regime beyond which JAC-ISF's closed-form estimate will drift; a lookup table built from the exact Dirichlet kernel would be a natural correction the paper does not develop.","In multi-user or multipath environments the autocorrelation of the received mixture contains cross terms between users, so the single-user LoS model would need an interference-suppression or orthogonal-access step before the JAC identity applies; this is not discussed in the paper.","The additive complexity structure suggests JAC could be composed with hierarchical beam training or compressed-sensing angle estimation to push near-field training latency even lower, a combination the paper leaves unexamined."],"forward_implications":["Near-field channel estimation complexity drops from multiplicative joint searches such as $O(T N S)$ to $O(T N)$, because the curvature is obtained from a one-dimensional autocorrelation and the angle from a one-dimensional far-field search.","Once the curvature is estimated, the phase-compensated signal is an equivalent far-field channel, so standard far-field angle estimators and codebooks can be reused for the near-field AoA step; the paper demonstrates this with MUSIC.","User position follows from the same two estimated parameters via $r = -\\cos^2\\theta/(2p_1)$ and $\\sin\\theta = p_2$, so channel estimation and localization are addressed by one decoupled estimation pipeline.","The derived Cramér–Rao lower bound for distance grows with range and diverges as $r\\to\\infty$, which is the formal statement that a finite array cannot resolve the distance of a far-field user.","The gradient-descent variant, which uses all lags of the autocorrelation in its loss function, is the one reported to track the CRLB closely and outperform polar-domain baselines at low SNR and few snapshots; the closed-form inverse-sinc variant is simpler but less accurate."],"supporting_citations":[{"why":"Supplies the Fresnel/Fraunhofer distance condition that justifies the quadratic (Fresnel) near-field phase model used throughout.","marker":"[19]"},{"why":"Introduces the curvature-of-arrival concept as the additional near-field parameter beyond angle.","marker":"[25]"},{"why":"Documents the coupling of distance and angle parameters that prevents orthogonal decomposition of near-field space, the problem JAC addresses.","marker":"[26]"},{"why":"Provides the Fisher-information formula used to compute the Cramér–Rao lower bounds for $p_1$, $p_2$, $\\theta$, and $r$.","marker":"[30]"},{"why":"Provides the parameter-transformation rule used to convert the CRLB from channel parameters to position parameters.","marker":"[31]"},{"why":"Supplies the polar-domain on-grid/off-grid baseline algorithms whose $O(T N S)$ complexity is the multiplicative-cost comparison point.","marker":"[32]"},{"why":"Documents the range-angle coupled beam-focusing model that motivates the claim that joint near-field estimation ordinarily has multiplicative angle-distance complexity.","marker":"[33]"}],"fun_headline_variants":["Autocorrelation decouples angle and curvature in near-field estimation","Autocorrelation decoupling for efficient near-field estimation","Two-step near-field estimation: autocorrelation then cross-correlation","JAC-GD offers low-complexity near-field position and channel estimation","Autocorrelation trick reduces near-field estimation complexity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The scheme hinges on treating the sampled autocorrelation magnitude as a continuous sinc function; on a finite uniform linear array the true quantity is a Dirichlet-like ratio of sines, and the approximation is accurate only when $k p_1 \\eta d^2$ stays small, so for close users or large arrays the estimated curvature can be biased.","fun_headline_variants_meta":{"raw":{"variants":["Autocorrelation decouples angle and curvature in near-field estimation","Autocorrelation decoupling for efficient near-field estimation","Two-step near-field estimation: autocorrelation then cross-correlation","JAC-GD offers low-complexity near-field position and channel estimation","Autocorrelation trick reduces near-field estimation complexity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000835,"raw_usage":{"total_tokens":3670,"prompt_tokens":997,"completion_tokens":2673,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":613,"completion_tokens_details":{"reasoning_tokens":2588}},"tokens_in":613,"tokens_out":2673,"duration_ms":18891,"temperature":1.0,"reasoning_tokens":2588,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T13:07:04.783674+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a noiseless simulation with a known user position, compute the exact sampled autocorrelation of the received ULA signal at several lags, and compare the location of its first null with the prediction $k p_1 \\eta d^2 (N-\\eta) = \\pi$ from the sinc model; a systematic mismatch at large array sizes or short distances would show that the decoupling identity used in equations (15) and (22) does not hold in the regime where the paper applies it.","supporting_citations":[{"cited_title":"Fraunhofer and fresnel d istances: Uni- fied derivation for aperture antennas,","cited_arxiv_id":null,"evidence_quote":"Supplies the Fresnel/Fraunhofer distance condition that justifies the quadratic (Fresnel) near-field phase model used throughout."},{"cited_title":"Localization of signals in the near-fi eld of an antenna array,","cited_arxiv_id":null,"evidence_quote":"Introduces the curvature-of-arrival concept as the additional near-field parameter beyond angle."},{"cited_title":"Parametric near-field channel estimation for extremely large aperture arrays,","cited_arxiv_id":null,"evidence_quote":"Documents the coupling of distance and angle parameters that prevents orthogonal decomposition of near-field space, the problem JAC addresses."},{"cited_title":"Fisher information for a complex gaussi an random variable: beamforming applications for wave propagation i n a random medium,","cited_arxiv_id":null,"evidence_quote":"Provides the Fisher-information formula used to compute the Cramér–Rao lower bounds for $p_1$, $p_2$, $\\theta$, and $r$."},{"cited_title":"Channel estimation for extremely lar ge-scale MIMO: far-field or near-field?","cited_arxiv_id":null,"evidence_quote":"Supplies the polar-domain on-grid/off-grid baseline algorithms whose $O(T N S)$ complexity is the multiplicative-cost comparison point."},{"cited_title":"An analytical ra nge-angle dependent beam focusing model for Terahertz linear antenna array,","cited_arxiv_id":null,"evidence_quote":"Documents the range-angle coupled beam-focusing model that motivates the claim that joint near-field estimation ordinarily has multiplicative angle-distance complexity."}],"review_version":1}