{"id":"f474db80-1e47-4b7f-84ee-578a52f1b8d9","arxiv_id":"2412.00215","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A DMA that jointly tunes its operating frequency and element resonances can match true-time-delay array gain, and a four-state subcarrier probing scheme estimates the user direction in one OFDM symbol.","lead":"Dynamic metasurface antennas use cheap tunable elements to steer signals, but their response changes with frequency. This paper shows how to choose both the operating frequency and antenna settings so a DMA can match a power-hungry true-time-delay array, and find a user's direction with a single training symbol.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The single-shot beam training claim rests on noiseless subcarrier selection (Eq. 48); the training-stage pilot SNR is roughly L^2 lower than the data-stage SNR, so Fig. 10 may overstate robustness unless operating SNR is high.","rationale":"I read the paper in good faith as a self-contained signal-processing analysis. The core frequency-selective beamforming result is well supported: Lemma 1 gives a closed-form Lorentzian-constrained beamformer; Lemma 2 shows N^2 gain is attainable when the normalized frequency p* is an integer; Lemma 3 gives explicit refractive-index and spacing formulas that make the integer condition hold throughout a specified angular range. I checked the algebra of Lemma 3 against the simulation parameters and found no internal inconsistency: for ft,min = 12 GHz, ft,max = 18 GHz, and phi_max = 30 deg, the formulas yield ng = 2.5 and dy = 0.42 lambda_c as used in Figs. 3-5. The bandwidth analysis in Lemma 4 and the 300 MHz 3 dB bandwidth are also consistent with Fig. 3. The single-shot beam training contribution, however, is qualitatively different: it requires an explicit estimation step, and the paper treats that step as a noiseless argmax over the beamforming gain. The reader identified this same weakness, and I agree it is the most load-bearing assumption. The concrete check proposed above would settle whether the concern actually degrades the headline rate claim; if the recommended operating SNR is high enough that selection errors are negligible, the conditional verdict may not need tightening, but the paper should still state and justify that regime explicitly.","tokens_in":21968,"tokens_out":19457,"duration_ms":187450,"concrete_test":"In the provided MATLAB code, add a noisy training stage before computing Fig. 10: transmit an OFDM pilot with per-subcarrier received sample y_k = sqrt(P_T/Ktr) * sqrt(G_ADMA(phi, f_k)) * s_k + n_k, with unit-energy pilot symbols s_k, noise n_k ~ CN(0, N0), select k* = argmax |y_k|^2, map k* to phi_hat, and recompute the angle-averaged achievable rate over phi ~ U(-30, 30) using the same P = 250 mW, r = 500 m, and bandwidth sweep. Repeat with transmit power reduced by factors of 10 (or with r increased) so that the pilot SNR at the best subcarrier, after the 20 log10 L training loss, spans roughly 0-30 dB. If the rate gap to the perfect-AoD curve exceeds 3 dB at any operating point shown in Fig. 10, the noiseless-selection assumption is load-bearing and the single-shot claim needs an SNR qualification.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing concern is that the beam training analysis and the Fig. 10 rate comparison assume the receiver selects the subcarrier index k* by maximizing the noiseless beamforming gain in Eq. (48). No noise, estimation error, or training pilot SNR is modeled at the selection step. During training, only Q = Nz/L subarrays are configured near the true angle, so the coherent pilot energy at the best subcarrier scales roughly as Q^2 Ny^2, while the data stage uses the full (Nz Ny)^2 gain; this is a 20 log10 L reduction, i.e., 12 dB for the L = 4 example. At sector edges the gains of adjacent sectors are deliberately within a factor delta of the peak, so even moderate noise can flip the selected index and shift the estimated angle. If subcarrier selection is imperfect, the achievable rate in Fig. 10 is an upper bound, and the claim that four resonant frequency states suffice for both training and beamforming is not yet established. The frequency-selective gain optimization in Lemmas 1-4 appears internally consistent, and the deterministic simulations support the N^2 operating-frequency result; the unresolved issue is specifically the unmodeled noisy training phase.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies a single-user MISO system with a transmit dynamic metasurface antenna (DMA). The authors first address the frequency-selective beamforming gain maximization by jointly optimizing the DMA resonant frequencies and the operating frequency for a line-of-sight channel. They derive closed-form expressions for the optimal resonant frequencies (Lemma 1), characterize the gain achieved by optimizing the operating frequency (Lemma 2), and give designs for the waveguide refractive index and element spacing that extend the maximum-gain N^2 regime over a desired angular range (Lemma 3). They also derive the ν-bandwidth of the DMA frequency response (Lemma 4). The second contribution is a single-shot beam training scheme in which the DMA array is split into L subarrays, each tuned to a different sector angle and corresponding optimal frequency, and an OFDM pilot symbol is used to have the receiver select the best subcarrier; the selected frequency is then used for data transmission. Numerical simulations validate the gain analysis and compare the achievable rate of the proposed DMA configuration with fixed-frequency DMA benchmarks and true-time-delay arrays.","tokens_in":22182,"tokens_out":18650,"duration_ms":131597,"significance":"If the main claims hold, the paper offers a useful frequency-domain perspective on DMA beamforming: it shows that a DMA can match the N^2 beamforming gain of a true-time-delay array over a designable angular range by jointly choosing the operating frequency and the waveguide parameters, and that a finite set of resonant states can be used for both training and beamforming. The closed-form derivations in Lemmas 1–4 are internally consistent and are validated by deterministic simulations, and the MATLAB code is made publicly available. The frequency-selective gain analysis (Lemmas 2–3) is the strongest part of the paper. The beam training claim, however, is not yet fully established because the selection rule in Eq. (48) is assumed noiseless and the coverage analysis in Section IV-C is connected to the post-training gain rather than to the actual training-stage gain.","major_comments":[{"comment":"The single-shot beam training claim rests on the assumption that the receiver selects the best subcarrier without noise or estimation error. The paper does not model the training-stage SNR: during the pilot transmission only Q = Nz/L subarrays are configured near the true angle, so the coherent pilot energy at the best subcarrier scales as (Q Ny)^2 rather than (Nz Ny)^2, which is a 20 log10 L reduction (12 dB for L = 4). With the codebook keeping adjacent sector gains within δ of the peak (δ = 3 dB in the simulations), the selection is sensitive to noise at moderate SNR. The achievable rates in Fig. 10 are therefore upper bounds unless the operating SNR is high enough that the selection error is negligible. The paper should characterize the probability of correct selection as a function of the training SNR, or explicitly state the SNR regime for which the single-shot claim is intended.","section":"IV-B, Eq. (48)"},{"comment":"The codebook design and the coverage guarantee are based on the post-training gain GADMA(ϕ, fk*) in Eq. (49), where all Nz DMAs are reconfigured to the estimated frequency. However, the actual selection in Eq. (48) is performed on the training-stage gain of the L-sector sub-array configuration, which includes contributions from all L sectors (with only Q subarrays near the true angle). The paper does not prove that the subcarrier maximizing the training gain also ensures that the post-training gain is within δ of the maximum; the equality between the two objectives is not self-evident. The numerical examples in Figs. 8 and 9 verify the claim for two specific configurations, but the analytical statement in Section IV-C that the proposed codebook guarantees the δ-threshold for all angles is not connected to the selection rule. This should either be proven or explicitly presented as a numerically validated design.","section":"IV-C, Eqs. (48)–(49)"}],"minor_comments":[{"comment":"The receiver is described as computing k* = argmax_k GADMA(ϕ, fk), but a receiver can only measure received power, not gain. Please state explicitly that the transmit power spectral density is equalized across subcarriers (or that the receiver normalizes by the known PSD and free-space path loss), so that the argmax of the measured power coincides with Eq. (48).","section":"IV-B, Eq. (48)"},{"comment":"In Lemma 2, the case where p* is not an integer is solved numerically; this leaves a small gap in the 'closed-form' claim for the optimal operating frequency in that regime. It would help to state that the numerical maximization is a one-dimensional search over a bounded interval.","section":"III-B, Lemma 2"},{"comment":"The values of Ψδ(Ny) are quoted as approximately 0.448/Ny for δ = 0.5; please provide the numerical procedure or a reference for this constant, since it enters the codebook design.","section":"IV-C, Lemma 5"},{"comment":"Reference [2] is cited as 'submitted' for a core modeling assumption about the coupling between bandwidth and frequency selectivity; please update the citation if the work has been published, or note that it is a preprint.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The frequency-selective beamforming gain analysis (Lemmas 1–4) is solid and is the paper's principal strength. The beam training contribution is the part that needs further work: the noiseless selection assumption and the gap between the training-stage and post-training gain analyses are load-bearing for the single-shot claim. The paper is within scope for the journal, and the issues appear fixable with additional analysis or a more restricted claim."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: the frequency-selective beamforming analysis is the real meat, and it holds up. The two-stage optimization—choosing the operating frequency, then the resonant frequencies—delivers a genuine N^2 gain over a designable angular range when the refractive index and spacing are designed via Lemma 3. That is new for DMAs and useful for wideband systems. The single-shot beam training idea is clever, but the paper overstates it: the rate comparison in Fig. 10 assumes the receiver picks the best subcarrier with no noise. During training only Q = Nz/L subarrays point at the true angle, so the pilot SNR is roughly 20 log10(L) lower than in the data stage. At sector edges adjacent beams are within the design threshold of each other, so a modest noise floor can flip the selection. The claim that four resonant states suffice for both training and beamforming is not yet established; Fig. 10 is an upper bound.\n\nWhat is well done: Lemmas 1-4 are internally consistent, the Lorentzian model is used correctly, and the closed-form expressions are plausible. Lemma 3's design criterion for N^2 coverage is a real contribution. The paper is self-contained, the MATLAB code is linked, and the deterministic simulations validate the gain formulas, including the 3 dB bandwidth prediction. The comparisons against fixed-frequency and binary-weight benchmarks are fair and show the proposed approach wins across the tuning range. Self-citation of [2] is minor; other prior work is cited properly.\n\nSoft spots beyond training noise: the analysis is ideal—LOS, no attenuation, single user—though they do show attenuation does not change the trend. The \"limited frequency states suffice\" conclusion is extrapolated from L = 4 examples; there is no general bound on L versus the gain threshold. That is a moderate gap, not fatal. Also, the operating-frequency optimization assumes the system can freely pick any frequency in the band, which may not be practical in a standards-based system, but the paper acknowledges this and motivates it via dynamic spectrum sharing.\n\nWho this is for: anyone working on DMA signal processing, wideband beamforming, or low-cost TTD alternatives. It deserves a serious referee; the main theoretical result is significant enough to warrant review even though the training analysis needs major revision. My recommendation: send it to review, but require the authors to either analyze noisy subcarrier selection or explicitly frame the training rate as an upper bound.","headline":"The two-stage frequency optimization is a solid, genuinely new result; the single-shot beam training rate plot overstates robustness because it assumes noiseless subcarrier selection.","tokens_in":22757,"tokens_out":2237,"would_cite":true,"duration_ms":21417,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"By optimizing operating frequency, a DMA reaches true-time-delay N-squared gain, and a single OFDM pilot trains its beam.","keywords":["dynamic metasurface antennas","frequency-selective beamforming","single-shot beam training","true-time-delay array","beamforming gain","resonant frequency tuning","OFDM pilot","sub-array codebook"],"falsifier":"Run the single-shot training with additive noise at the receiver at the SNR values of Fig. 10; if subcarrier selection errors push the post-training rate more than 3 dB below the perfect-AoD curve, the core claim that four frequency states suffice would fail. Alternatively, measure the beamforming gain at an angle midway between two adjacent codebook angles and check whether it falls below $\\delta N_y^2 N_z^2$, which would contradict Lemma 5.","tokens_in":21731,"feed_emoji":"📡","tokens_out":7684,"duration_ms":60868,"temperature":0.7,"pith_summary":"This paper shows that a dynamic metasurface antenna (DMA) transmitter can reach the same beamforming gain as a true-time-delay (TTD) array, $N^2$, by treating the operating frequency as an optimization variable along with the per-element resonant frequencies, and by designing the waveguide refractive index and element spacing for the desired angular coverage. The authors then exploit the same frequency-angle coupling to perform beam training in a single OFDM pilot symbol: sub-arrays of DMA elements point at different candidate angles on different subcarriers, the receiver reports the strongest subcarrier, and the array reconfigures to the corresponding optimal frequency. With only four resonant frequency states, the trained beam achieves a rate within 3 dB of the perfect-angle case. This matters because DMAs use low-power varactor or PIN diodes instead of power-hungry phase shifters or true-time-delay lines, so matching TTD performance at lower hardware cost would make wideband beamforming more energy-efficient.","feed_headline":"One OFDM pilot trains a metasurface antenna's beam","feed_subtitle":"Tuning frequency along with element resonances lets a DMA hit N-squared gain and skip multi-shot training.","key_machinery":"The engine is the Lorentzian polarizability model of each DMA slot, $\\alpha_{M,n}(f) = F 2\\pi f^2 / (2\\pi f_{r,n}^2 - 2\\pi f^2 + j\\Gamma f)$, which couples the magnitude and phase of the tunable weight. The paper reparametrizes this constraint with a shift-of-origin transformation, $[f_{\\rm DMA}(f)]_n = (-j + e^{j\\tilde\\psi_n(f)})/2$, turning the coupled problem into a unit-modulus beamforming problem with closed-form optimal phases (Lemma 1). The load-bearing identity is the array-factor term $S(\\phi,f) = \\sin(\\pi N f d_y(n_g+\\sin\\phi)/c) / \\sin(\\pi f d_y(n_g+\\sin\\phi)/c)$, whose magnitude can equal $N$ when $p = f d_y(n_g+\\sin\\phi)/c$ is an integer; that condition links angle, frequency, waveguide index, and spacing, and it drives both the frequency-optimization lemma and the design rules for angular coverage. The beam-training mechanism is the same map read in reverse: different subcarriers carry different angles through $f_t^\\star(\\phi)$, so a sub-array codebook built from Lemma 5 lets one OFDM symbol sweep the whole angular range and report the best frequency.","core_discovery":"The central claim is that the frequency selectivity of a DMA is not merely a source of beam-squint loss but a resource: by choosing the operating frequency $f_t$ jointly with the element resonant frequencies $\\{f_{r,n}\\}$, the beamforming gain at any angle $\\phi$ in a designable range can be pushed to the TTD upper bound $N^2$. Lemma 2 gives the optimal frequency as $f_t^\\star(\\phi) = p^\\star c / (d_y(\\sin\\phi + n_g))$, with $p^\\star$ chosen to maximize the array factor $|\\sin(\\pi N p)/\\sin(\\pi p)|$; full gain is achieved exactly when an integer $p^\\star$ lies in the feasible band, in which case every slot resonates at the operating frequency. Lemma 3 provides closed-form waveguide refractive index $n_g^\\star$ and inter-element spacing $d_y^\\star$ that make this possible for every angle in a specified interval $[\\phi_{\\rm lw}, \\phi_{\\rm up}]$. For the sub-array architecture, the paper proves that a single OFDM pilot symbol suffices: each sub-group of DMAs probes a different sector angle at its own optimal frequency, the receiver selects the subcarrier with maximum gain, and reconfiguring all elements to that frequency keeps the beamforming gain within a factor $\\delta$ of $N_y^2 N_z^2$ over the whole sector (Lemmas 5 and the codebook recursion). The rate evaluation shows the trained configuration matches the perfect-AoD rate and tracks a TTD array up to roughly the DMA's 3 dB bandwidth.","pith_inferences":["Because the angle-to-frequency map is monotone and closed-form, the same resonant-frequency readout could in principle be used for passive angle-of-arrival estimation or for serving multiple users at different angles on different subcarriers without extra training, though the paper does not analyze multiuser or multipath settings.","The training analysis assumes error-free subcarrier selection. In a low-SNR training phase, selection errors will blur the angle estimate; a natural extension is to add a second pilot or to dither the sector frequencies to average out selection errors, which would test how much of the 3 dB margin survives.","The codebook recursion in Lemma 5 suggests a hardware-aware trade-off: for fixed total elements $N_y N_z$, narrower per-sector beams (larger $N_y$) require more sectors $L$ to cover the same range, so the number of resonant states can be traded against row length in a way the paper quantifies but does not optimize."],"forward_implications":["A DMA designed with $n_g = n_g^\\star$ and $d_y = d_y^\\star$ achieves the $N^2$ TTD-level beamforming gain at every angle in the target sector, provided the operating band is wide enough to contain an integer $p^\\star$ for each angle.","The optimal frequency strategy dominates the fixed-center-frequency DMA benchmark for every angle and every tuning range, with the gap shrinking as the tuning range grows (Figs. 4 and 11).","Four resonant frequency states suffice for both training and data transmission; continuous high-resolution tuning is not needed to stay within 3 dB of the ideal beamforming gain.","The DMA rate tracks the TTD rate up to about the 3 dB bandwidth of the gain response, roughly $\\Gamma/(2\\pi)$ in the element-limited regime, and falls off beyond it.","Binary (PIN-diode) DMA weights match continuous weights only at the single angle whose optimal frequency equals the center frequency; elsewhere they lose more than 2 dB because some slots are turned off."],"supporting_citations":[{"why":"Supplies the Lorentzian polarizability model of DMA slots and the no-attenuation waveguide model used throughout.","marker":"[4]"},{"why":"Provides the true-time-delay array benchmark whose $N^2$ gain the paper matches.","marker":"[5]"},{"why":"Supplies the shift-of-origin/common phase rotation approach and the prior fixed-frequency DMA optimization that this paper extends with frequency optimization.","marker":"[9]"},{"why":"Models DMA frequency selectivity in MIMO-OFDM, the prior wideband DMA treatment this paper generalizes.","marker":"[17]"},{"why":"Demonstrates single-shot beam training with TTD arrays, the idea the paper adapts to DMAs.","marker":"[20]"},{"why":"Provides fast beam training with TTD arrays as a benchmark for single-shot training.","marker":"[21]"},{"why":"Shows single-shot path discovery with a leaky waveguide antenna, prior art on frequency-swept training.","marker":"[22]"},{"why":"Compares beam-squint degradation of phased arrays versus TTD arrays, the baseline the DMA analysis extends.","marker":"[26]"}],"fun_headline_variants":["Frequency tuning turns DMA loss into N² gain","One OFDM pilot trains DMA beam to full gain","Metasurface antenna: one pilot, N² beam gain","DMA frequency reconfigurability enables single-shot training","N² gain from one OFDM pilot via frequency-tuned DMA"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The single-shot beam training claim rests on the receiver identifying the strongest subcarrier without error; the paper does not model noise or estimation error, and during training only $Q = N_z/L$ sub-arrays point at the true direction, so the training-phase SNR is lower than in the final beamforming stage.","fun_headline_variants_meta":{"raw":{"variants":["Frequency tuning turns DMA loss into N² gain","One OFDM pilot trains DMA beam to full gain","Metasurface antenna: one pilot, N² beam gain","DMA frequency reconfigurability enables single-shot training","N² gain from one OFDM pilot via frequency-tuned DMA"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001038,"raw_usage":{"total_tokens":4441,"prompt_tokens":1088,"completion_tokens":3353,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":704,"completion_tokens_details":{"reasoning_tokens":3272}},"tokens_in":704,"tokens_out":3353,"duration_ms":20053,"temperature":1.0,"reasoning_tokens":3272,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T05:37:20.483498+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the single-shot training with additive noise at the receiver at the SNR values of Fig. 10; if subcarrier selection errors push the post-training rate more than 3 dB below the perfect-AoD curve, the core claim that four frequency states suffice would fail. Alternatively, measure the beamforming gain at an angle midway between two adjacent codebook angles and check whether it falls below $\\delta N_y^2 N_z^2$, which would contradict Lemma 5.","supporting_citations":[{"cited_title":"Analysis of a waveguide-fed metasurface antenna,","cited_arxiv_id":null,"evidence_quote":"Supplies the Lorentzian polarizability model of DMA slots and the no-attenuation waveguide model used throughout."},{"cited_title":"Nonuniform true time delay precoding in wideband MISO systems,","cited_arxiv_id":null,"evidence_quote":"Provides the true-time-delay array benchmark whose $N^2$ gain the paper matches."},{"cited_title":"Hierarchical codebook design with dynamic metasurface antennas for energy-efficient arrays,","cited_arxiv_id":null,"evidence_quote":"Supplies the shift-of-origin/common phase rotation approach and the prior fixed-frequency DMA optimization that this paper extends with frequency optimization."},{"cited_title":"Dynamic metasurface antennas for MIMO-OFDM receivers with bit-limited ADCs,","cited_arxiv_id":null,"evidence_quote":"Models DMA frequency selectivity in MIMO-OFDM, the prior wideband DMA treatment this paper generalizes."},{"cited_title":"Wideband millimeter-wave beam training with true-time-delay array architecture,","cited_arxiv_id":null,"evidence_quote":"Demonstrates single-shot beam training with TTD arrays, the idea the paper adapts to DMAs."},{"cited_title":"Fast beam training with true-time-delay arrays in wideband millimeter-wave systems,","cited_arxiv_id":null,"evidence_quote":"Provides fast beam training with TTD arrays as a benchmark for single-shot training."},{"cited_title":"Single shot single antenna path discovery in thz networks,","cited_arxiv_id":null,"evidence_quote":"Shows single-shot path discovery with a leaky waveguide antenna, prior art on frequency-swept training."},{"cited_title":"A wideband generalization of the near-field region for extremely large phased-arrays,","cited_arxiv_id":null,"evidence_quote":"Compares beam-squint degradation of phased arrays versus TTD arrays, the baseline the DMA analysis extends."}],"review_version":1}