{"id":"15ac0ff3-47c5-4c52-93ac-5cb1e7cc5bab","arxiv_id":"2501.00314","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Quantum-MUSIC recovers the complex multi-user wireless channel from magnitude-only Rydberg-atom measurements, then applies MUSIC to estimate each user's angle of arrival.","lead":"The paper proposes a signal processing algorithm that lets a Rydberg-atom radio receiver, which can only measure signal strength, locate several users at once. It pairs an existing phase-recovery step with the classic MUSIC direction-finding method and demonstrates the combination in simulation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Even with perfect channel recovery, the stated per-element random polarization gains break the ULA steering manifold assumed by MUSIC, so the reported AoA peaks cannot follow from the stated model.","rationale":"The reader's weakest assumption is that the modified Gerchberg-Saxton phase retrieval may fail to converge within 50 iterations. That is a legitimate concern, but it is not the single most load-bearing one: the claimed AoA results would still fail even if phase retrieval were exact, because the MUSIC stage assumes the recovered channel columns lie in the span of the ULA steering vectors of Eq. (13). Under the polarization model stated in Sec. IV-A, each channel column is pointwise multiplied by independent zero-mean random gains, randomizing both amplitude and sign across the array. The signal subspace is then a random K-plane, and the ideal steering vectors are generically not contained in it, so the Quantum-MUSIC peaks are not consistent estimators of the true AoAs. The simulation results in Figs. 4-6 are therefore evidence either that the model in the text was not followed (e.g., ε_{m,k} was actually constant across m) or that the reported performance is not reproducible. This is a concrete, checkable inconsistency in the central argument, not a matter of convergence tuning. The proposed test isolates this issue from the PR loop by using the true channel, and it settles whether the central claim holds under the stated assumptions. Because the paper could be repaired by explicitly restricting the polarization to per-user gains, a standard calibrated-array assumption, or by adding a calibration step, a conditional verdict is appropriate; if the test fails, the claim should be rejected.","tokens_in":9400,"tokens_out":20019,"duration_ms":219171,"concrete_test":"Take K=1, M=32. Generate the true channel exactly as in Sec. IV-A with ε_{m,1} ~ N(0,1/3) i.i.d. over m, form the noiseless covariance R = A A^H from this true A (bypassing the PR stage entirely), and compute the Quantum-MUSIC spectrum (14) on a fine θ grid for 2000 trials. Record the RMSE of the peak location relative to the true θ_1. If the RMSE is non-negligible, the MUSIC stage is mismatched even with perfect channel recovery, and the central claim fails under the stated model; if it is negligible, then the simulation must be using a per-user gain ε_k constant over m, and the text should say so.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Section II-B defines a_{m,k} = (1/ℏ)(μ^T_eg ε_{m,k})√P_k ρ_{m,k}e^{-jφ_{m,k}}, and Section IV-A says ε_{m,k} follows N(0,1/3). If ε_{m,k} is drawn independently per array element m, the user-k column of the channel matrix is v_k = diag(c_{1,k},...,c_{M,k}) a(θ_k), i.e., the ideal steering vector a(θ_k) of Eq. (13) multiplied elementwise by independent random gains rather than by a single scalar. The signal covariance R = (1/P)ˆAˆA^H then has a signal subspace spanned by these randomly weighted vectors. The noise subspace U_N found in Algorithm 1, step 13, is orthogonal to that random subspace, so for the ideal test vector a(θ) in Eq. (14), the denominator a^H(θ)U_NU_N^H a(θ) generally does not vanish at the true θ_k. Even in the noiseless limit with exact channel recovery, the Quantum-MUSIC spectrum peaks where the projection of a(θ) onto a random K-plane is maximal; with zero-mean per-element gains these peaks can be far from the true AoAs. The paper provides no calibration procedure and no statement that ε_{m,k} is constant across m, so the sharp peaks and low RMSE of Figs. 4-6 are inconsistent with the stated model. This is a model/algorithm mismatch that is independent of the phase-retrieval convergence question.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript introduces Quantum-MUSIC, a two-stage algorithm for angle-of-arrival (AoA) estimation in a multi-user uplink served by an atomic receiver array. Stage 1 treats the magnitude-only measurements as a biased phase retrieval problem, using spectral initialization and alternating Gerchberg-Saxton updates to recover the channel matrix \\hat A. Stage 2 forms the sample covariance R = (1/P)\\hat A\\hat A^H and applies conventional MUSIC to estimate the K user AoAs. Numerical results for M=32, P=100, K=1-4, and a 2,000-trial RMSE metric are reported, comparing the proposed method with classical RF MUSIC under different noise floors. The central claims are that channel recovery plus MUSIC enables multi-user sensing from magnitude-only data and that it outperforms the existing MUSIC algorithm.","tokens_in":9717,"tokens_out":7206,"duration_ms":71766,"significance":"If the model and recovery guarantees held, this would be a useful first step for multi-user quantum wireless sensing. The manuscript is self-contained, the algebraic form of the alternating updates is standard, and the simulation setup is described in enough detail to be reproduced. Its main limitations are that the polarization model in the simulation is inconsistent with the MUSIC steering manifold used for estimation, the phase-retrieval stage is not guaranteed or characterized for failure, and the performance comparison is made under a 15-dB noise-floor offset rather than at equal input SNR; these issues currently leave the central claim insufficiently supported.","major_comments":[{"comment":"The stated simulation model draws ε_{m,k} independently for each array element and user (Section IV-A: 'ϵm,k, ϵm,b follow N(0,1/3)'). With μ_eg = [0,1785.9qa0,0]^T, the m-th entry of the k-th channel column is therefore g_{m,k} e^{j m d sin θ_k/λ} with independent random g_{m,k}. The covariance R = (1/P)\\hat A\\hat A^H then spans these randomly weighted vectors, not the ideal manifold a(θ) of Eq. (13). Consequently the denominator a^H(θ)U_NU_N^H a(θ) in Eq. (14) does not vanish at θ_k even for exact channel recovery, and the sharp spectral peaks in Figs. 4-6 do not follow from the stated model. Please either justify that ε_{m,k} is constant in m for each user (a common-polarization assumption), or replace the test vector a(θ) with one that accounts for the per-element gains, or revise the model; the current manuscript cannot support the AoA claim as written.","section":"Section II-B and Section IV-A, Eqs. (13)-(14)"},{"comment":"The channel-recovery stage is the load-bearing first half of the algorithm, but no convergence guarantee or failure-rate analysis is provided for the spectral initialization of Eqs. (9)-(10) and the alternating updates of Eqs. (11)-(12). The simulations use a single configuration (P=100, N=50) and report RMSE averaged over trials; they do not report how often the phase-retrieval step fails or how sensitive the result is to P and N. Because an incorrect \\hat A directly invalidates the MUSIC covariance in Step 13, the paper should add either a theoretical justification (for example, a local-convergence or contraction result, or a reference that covers this biased variant) or a Monte Carlo characterization of failure probability as a function of P, N, and SNR.","section":"Algorithm 1, Section III-A"},{"comment":"The claimed 'outperforms' result is not an algorithmic comparison at equal input SNR. The simulation fixes σ_n^2 = 10^{-19.1} (QSN) and σ_t^2 = 10^{-17.6} (JNTN), a 15-dB offset, and then compares RMSE versus transmitted power. The conclusion that 'Quantum-MUSIC is superior to MUSIC' (Section IV-B) therefore conflates the noise-floor advantage of the atomic receiver with the performance of the proposed estimator. The abstract and conclusions should restate the claim as a system-level advantage, and ideally include a same-SNR comparison of Quantum-MUSIC against classical MUSIC to separate algorithmic gain from noise-floor gain.","section":"Section IV-B, Figs. 5-6"}],"minor_comments":[{"comment":"The Planck constant is printed as 6.6626×10^{-34} J·s; the standard value is 6.626×10^{-34} J·s.","section":"Section II-A"},{"comment":"The symbol b_m is used both as a scalar reference and as the P-vector [b_m,...,b_m]^H; a distinct notation such as b_m 1_P would avoid ambiguity in Eqs. (7) and (12).","section":"Section II-B and Algorithm 1"},{"comment":"The expression \\bar r = |v^H \\bar S| z_m / \\|\\bar S^H v\\|_2^2 is ambiguous as typeset because the elementwise product between the row vector |v^H \\bar S| and the column vector z_m is not indicated; please write it explicitly with an elementwise operator.","section":"Algorithm 1, line 4"},{"comment":"The angular range [30°, 150°] and the 2^14-point grid are not justified; a sentence on the expected resolution limit would help the reader interpret the RMSE values.","section":"Section IV-A"},{"comment":"The RMSE plots would be more informative with error bars or confidence bands, especially because the phase-retrieval stage may have occasional large failures that a mean RMSE can hide.","section":"Figures 4-6"}],"recommendation":"major_revision","confidential_remarks":"The central technical issue is the model/algorithm mismatch in the polarization model: independent per-element gains break the ULA steering manifold assumed by MUSIC, so the reported AoA peaks are not explained by the stated model. This is fixable by making a common-polarization assumption with explicit justification or by adapting the MUSIC test vector to the estimated per-element gains, but it must be addressed before the paper can be considered. The SNR comparison should also be reframed as a system-level comparison rather than an algorithmic one. There is no indication of any ethical problem; the issues are technical and within the scope of a major revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a plausible first application of phase retrieval plus MUSIC to atomic-array sensing, but the model as written has a load-bearing ambiguity. You should know going in that the random polarization gain ε_{m,k} is subscripted per array element. If those gains are independent across m, the channel matrix columns are not scalar multiples of the ULA steering vector a(θ_k); they are elementwise randomly weighted versions of it. Then the noise subspace found by MUSIC is orthogonal to a random K-plane, and the spectrum (14) will not peak at the true AoAs even with perfect channel recovery. The simulation figures could not have been produced from that model. If the intended reading is a single common gain per user (ε_k), the authors need to say so explicitly and adjust the notation. This is not a small typo: the whole MUSIC step stands or falls on it.\n\nWhat's genuinely new: the paper is the first to combine biased Gerchberg-Saxton phase recovery with MUSIC for magnitude-only Rydberg-atom sensing. The problem formulation is clean, the least-squares updates in Algorithm 1 are algebraically correct, and the physical noise-floor comparison (QSN vs. Johnson-Nyquist) is a sensible way to frame a quantum vs. classical receiver comparison.\n\nSoft spots, in order. (1) The polarization issue above. (2) The claimed modification of the biased GS algorithm from [6] is never specified; as written, Algorithm 1 looks like the same alternating updates. (3) The simulations are generated from the same model the algorithm assumes, so they only validate internal consistency, not real-world sensing; no code, no hardware, no Monte Carlo on the phase-retrieval convergence. (4) The performance gain over classical MUSIC is forced by the 15 dB lower noise floor of QSN; it's a valid point but not a surprising one.\n\nThe central idea is not broken if the model is cleaned up. But the paper needs a major revision before it can be trusted. A careful referee should ask for an explicit statement on ε, a corrected simulation with common gains (or a reworked MUSIC that accounts for per-element gains), and a real specification of the claimed modification. I'd send it to review rather than desk-reject, because the application is timely and the fix is likely achievable. For my own work, I wouldn't cite until the model is clarified.\n\nRecommendation: engage, but treat the polarization model as the gatekeeping issue.","headline":"Quantum-MUSIC is a plausible first step but the random polarization model breaks the MUSIC steering manifold; the authors must clarify whether the gain is common or per-element.","tokens_in":10253,"tokens_out":8180,"would_cite":false,"duration_ms":78680,"reading_group":"maybe","serious_thinker":"unclear","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94A12"],"pacs":[],"model":"deepseek-v4-flash","headline":"Quantum-MUSIC recovers channel phase from magnitude-only atomic-receiver measurements and applies MUSIC to estimate the angles of arrival of multiple users.","keywords":["quantum wireless sensing","atomic receiver","Rydberg atoms","MUSIC","angle-of-arrival estimation","phase retrieval","multi-user sensing","magnitude-only measurement"],"falsifier":"Run the channel-recovery step alone with a known channel matrix and compare $\\|\\hat{A}-A\\|$: if the relative recovery error does not decrease toward the noise floor as $N$ and $P$ grow, or if some Monte Carlo trials converge to a wrong local minimum, then the angle estimates from Eq. (14) are not trustworthy even though the RMSE curves look good.","tokens_in":9195,"feed_emoji":"⚛️","tokens_out":6192,"duration_ms":56133,"temperature":0.7,"pith_summary":"The paper claims that a Rydberg-atom receiver, which records only the magnitude of a received radio signal, can still support multi-user angle-of-arrival sensing. To do this, the authors recover the complex channel from magnitude-only measurements using a modified biased Gerchberg-Saxton phase-retrieval routine, then feed the recovered channel into the classical MUSIC subspace algorithm. In simulation, the proposed Quantum-MUSIC estimates the directions of one to four users and reports lower angle error than conventional MUSIC under comparable transmitted power, because the atomic receiver's quantum shot noise is far weaker than thermal noise in a radio-frequency antenna receiver.","feed_headline":"Magnitude-only atom receiver finds multi-user angles","feed_subtitle":"Atomic receivers see only signal strength; this paper recovers the channel phase and then finds each user's direction.","key_machinery":"The load-bearing object is the modified biased Gerchberg-Saxton iteration for channel recovery (Algorithm 1), which treats the holographic reference $b_m$ as an extra channel entry so that the magnitude-only observation $z_m = |\\bar{S}^H \\bar{a}_m + n_m|$ matches the standard phase-retrieval form. It is initialized by the principal eigenvector of the expanded covariance matrix $\\bar{R}$ from Eq. (9) and the magnitude estimate in Eq. (10), then alternates between phase updates in Eq. (11) and least-squares channel updates in Eq. (12). The recovered channel matrix $\\hat{A}$ feeds the eigenvalue decomposition whose noise subspace $U_N$ defines the MUSIC spectrum in Eq. (14).","core_discovery":"The central claim is that the magnitude-only constraint of an atomic receiver is not a barrier to subspace-based multi-user sensing: the received magnitudes $y = |A^H s + b + n|$ can be inverted into an estimate of the channel matrix $\\hat{A}$ by solving a phase-retrieval problem, after which the MUSIC spectrum $P_Q(\\theta) = 1/(a^H(\\theta) U_N U_N^H a(\\theta))$ built from $\\hat{A}$ resolves the per-user angles of arrival. The paper's simulation evidence is that Quantum-MUSIC outperforms classical RF-domain MUSIC across signal power levels and user counts, with the gap widening as the number of users grows.","pith_inferences":["A direct testable extension would be a convergence study of the channel-recovery step over the pilot count $P$ and iteration count $N$; the paper fixes $P=100$ and $N=50$ and does not report how close $\\hat{A}$ is to the true channel.","If phase retrieval is the bottleneck, then an error floor in the angle RMSE at high signal power would reveal a failure mode that the current simulations, which judge performance only by RMSE, do not separate from noise effects.","Because the model assumes a single line-of-sight path per user, extending the pipeline to multipath or joint range-angle estimation would require replacing the steering-vector dictionary, but the phase-recovery-plus-MUSIC structure itself is not tied to the single-path assumption."],"forward_implications":["If the channel recovery step works as simulated, the atomic receiver can estimate angles of arrival for multiple simultaneous users without any phase measurement, something previous Rydberg-sensor angle-of-arrival work did not do.","Because quantum shot noise is modeled roughly 15 dB below Johnson-Nyquist thermal noise at the chosen conditions, the quantum system's sensing accuracy exceeds an RF system at equal transmitted power.","The accuracy advantage of the atomic receiver grows with the number of users, making magnitude-only quantum sensing comparatively more attractive in denser multi-user scenarios.","The algorithm inherits MUSIC's subspace structure, so it can be used with any array geometry whose steering vector is known, not only the uniform linear array simulated in the paper."],"supporting_citations":[{"why":"Supplies the atomic MIMO receiver model and the biased Gerchberg-Saxton phase-retrieval method that Quantum-MUSIC modifies for its channel-recovery step.","marker":"[6]"},{"why":"Provides the spectral-method initialization for phase retrieval that Algorithm 1 uses to build the expanded covariance matrix and starting point.","marker":"[17]"},{"why":"Defines the MUSIC subspace algorithm whose noise-subspace orthogonality is used to form the Quantum-MUSIC spectrum.","marker":"[18]"},{"why":"The previous Rydberg-atom angle-of-arrival work for a single source that Quantum-MUSIC extends to multiple users.","marker":"[9]"},{"why":"Frames the magnitude-only inversion as a phase-retrieval problem, justifying the recovery formulation in Eq. (8).","marker":"[16]"},{"why":"Supplies the quantum shot noise and thermal noise power values that set the comparison between the atomic receiver and the conventional RF receiver.","marker":"[20]"},{"why":"Used to compute the transition dipole moment of the chosen Rydberg levels for the simulation parameters.","marker":"[19]"}],"fun_headline_variants":["Phase recovery unlocks MUSIC for quantum sensors","Quantum-MUSIC: angle finding with magnitude-only data","From magnitude to angles: quantum MUSIC for multi-user","Quantum wireless sensing: beating the magnitude-only limit","Multi-user angles from magnitude-only atom signals"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The result stands or falls on whether 50 iterations of the alternating phase-retrieval updates reliably recover the true channel for $P=100$ pilots at the simulated noise levels; the paper does not prove convergence or test the recovery error directly.","fun_headline_variants_meta":{"raw":{"variants":["Phase recovery unlocks MUSIC for quantum sensors","Quantum-MUSIC: angle finding with magnitude-only data","From magnitude to angles: quantum MUSIC for multi-user","Quantum wireless sensing: beating the magnitude-only limit","Multi-user angles from magnitude-only atom signals"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00017,"raw_usage":{"total_tokens":1179,"prompt_tokens":770,"completion_tokens":409,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":386,"completion_tokens_details":{"reasoning_tokens":336}},"tokens_in":386,"tokens_out":409,"duration_ms":4400,"temperature":1.0,"reasoning_tokens":336,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:54:42.432192+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the channel-recovery step alone with a known channel matrix and compare $\\|\\hat{A}-A\\|$: if the relative recovery error does not decrease toward the noise floor as $N$ and $P$ grow, or if some Monte Carlo trials converge to a wrong local minimum, then the angle estimates from Eq. (14) are not trustworthy even though the RMSE curves look good.","supporting_citations":[{"cited_title":"Towards Atomic MIMO Receivers","cited_arxiv_id":"2404.04864","evidence_quote":"Supplies the atomic MIMO receiver model and the biased Gerchberg-Saxton phase-retrieval method that Quantum-MUSIC modifies for its channel-recovery step."},{"cited_title":"Phase retrieval via Wirtinger flow: Theory and algorithms,","cited_arxiv_id":null,"evidence_quote":"Provides the spectral-method initialization for phase retrieval that Algorithm 1 uses to build the expanded covariance matrix and starting point."},{"cited_title":"Multiple emitter location and signal parameter estimation,","cited_arxiv_id":null,"evidence_quote":"Defines the MUSIC subspace algorithm whose noise-subspace orthogonality is used to form the Quantum-MUSIC spectrum."},{"cited_title":"Determining the angle-of-arrival of a radio-frequency source with a Rydberg atom-based sensor,","cited_arxiv_id":null,"evidence_quote":"The previous Rydberg-atom angle-of-arrival work for a single source that Quantum-MUSIC extends to multiple users."},{"cited_title":"Phase retrieval: From computational imaging to machine learning: A tutorial,","cited_arxiv_id":null,"evidence_quote":"Frames the magnitude-only inversion as a phase-retrieval problem, justifying the recovery formulation in Eq. (8)."},{"cited_title":"Quantum shot noise limit in a Rydberg RF receiver compared to thermal noise limit in a conventional receiver,","cited_arxiv_id":null,"evidence_quote":"Supplies the quantum shot noise and thermal noise power values that set the comparison between the atomic receiver and the conventional RF receiver."},{"cited_title":"Arc: An open-source library for calculating properties of alkali Rydberg atoms,","cited_arxiv_id":null,"evidence_quote":"Used to compute the transition dipole moment of the chosen Rydberg levels for the simulation parameters."}],"review_version":1}