{"id":"a75ebdff-1a17-4afa-a19a-32a9e440dd55","arxiv_id":"2411.15570","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A single-antenna system localizes and tracks multiple moving RIS-equipped users by combining time-of-arrival, Doppler, and reflection-geometry measurements, beating a ToA-based baseline by up to 3x in simulation.","lead":"Vehicles carrying reconfigurable intelligent surfaces can be found and followed using a single transmitter antenna and several receiver antennas, by reading the reflected signals' timing, Doppler, and angle geometry. The proposed method reports up to three times lower localization error than a timing-only approach, and it is designed to track multiple vehicles at once.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Tracking results rely on oracle knowledge of acceleration and initial velocity; no estimator or adaptation rule for a_k is provided.","rationale":"Good faith: the paper's localization algorithm and CRLB are substantive; the factor-3 localization gain is supported by the simulations with the given geometry and does not depend on the tracking gap. The reader's CONDITIONAL verdict is reasonable. I did not find an internal inconsistency in the alpha/ToA estimation that would overturn the localization claim. The acceleration/initial-velocity issue is the weakest link in the central claim because the tracking results are quantified with unavailable oracle inputs. This does not require rejecting the paper; it requires either an acceleration estimator, a robustness study with mismatched a_k, or a softened claim that tracking assumes known motion inputs. Hence UNCHANGED relative to the reader's conditional verdict.","tokens_in":22316,"tokens_out":16502,"duration_ms":160924,"concrete_test":"Run the Fig. 11 scenario twice with Algorithm 3: (i) as published, with true a_k and true v_k[0]; (ii) with a_k=0 (constant-velocity prior) and v_k[0]=0, keeping all other parameters identical. Compare average tracking error and the CDFs in Fig. 11(b). If the oracle-free run degrades to roughly the localization-only error or diverges, the tracking gains in the paper are not achievable without an acceleration/initial-velocity estimator. As an analytic complement, derive the steady-state EKF bias caused by a constant acceleration mismatch of 2 m/s^2.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing weakness is in the tracking claim (Section IV). Algorithm 3's prediction step uses x_k[n|n-1] = A x_k[n-1|n-1] + B a_k (eq. 17) with the acceleration vector a_k as a known input, and the initialization line says only 'estimate x_k[0|0] employing Algorithm 2,' but Algorithm 2 estimates position only, not the velocity component of the 4D state. No procedure is given to measure, estimate, or adapt a_k or v_k[0]. The tracking simulations (Figs. 9-11) then report errors using true initial speeds (10/20 m/s) and true acceleration (2 m/s^2), so the demonstrated tracking gains are conditional on an oracle motion model. The text's claim that the filter 'can adapt itself ... by updating the value of acceleration' is not backed by any update rule, so a practitioner cannot run the tracker without externally supplied acceleration. Since the abstract's central contribution is simultaneous localization and tracking of vehicular users, this missing estimator is load-bearing for the tracking half of the claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers a scenario with a single-antenna transmitter, multiple single-antenna receivers, and multiple mobile users each equipped with a reconfigurable intelligent surface (RIS). It proposes a phase-shift design that alternates sign over time to cancel static scatterers and uses orthogonal codewords to separate signals from different RISs. From the received OFDM signals, the authors estimate time of arrival, Doppler frequency, and a geometric parameter alpha equal to the sum of the cosines of the angle of arrival and angle of departure at the RIS. These parameters are used in a least-squares localization problem and in an extended Kalman filter for tracking. The paper also derives Cramér-Rao lower bounds for the position, ToA, and alpha, and compares the localization error with the ToA-based benchmark of [31], reporting up to a factor of three improvement. Simulation results are provided for localization accuracy versus bandwidth, number of subcarriers, power, and number of receivers, and for tracking accuracy over several paths.","tokens_in":22506,"tokens_out":16321,"duration_ms":132697,"significance":"The paper addresses a timely problem—localization and tracking of RIS-equipped vehicles—and is, to my knowledge, the first to treat tracking of mobile RISs with a single-antenna transmitter and multiple single-antenna receivers. The proposed phase-shift design and the joint estimation of ToA and alpha are self-consistent, and the CRLB derivation provides a useful benchmark. The localization comparison against [31] is meaningful because the same phase-shift design and number of transmissions are used for both methods. If the two issues identified below are resolved, the work would be a useful contribution to RIS-aided localization with low-complexity infrastructure. At present, however, the tracking results are conditional on oracle knowledge of acceleration and initial velocity, and the default simulation parameters make the Doppler frequency unobservable in the proposed estimator.","major_comments":[{"comment":"Table II sets NT = 2, and this value is used in the default localization and tracking simulations (Figs. 3, 5–11). In Section III-B, Algorithm 1 estimates tau_{k,nr} and f_{d,k,nr} by taking a 2D FFT of the matrix R_{k,nr} in C^{N x NT/2} formed from (6). With NT = 2, R_{k,nr} has a single column, so the Doppler dimension of the 2D FFT has length one and f_d is unobservable; the IFFT variant in Algorithm 1 provides no additional information. The later Doppler-removal step in (7) and the estimate of alpha in (12) both rely on the estimated f_d. Consequently, the localization and tracking results shown for the default NT = 2 settings are not supported by Algorithm 1 as written. Please either run the default simulations with NT > 2 or give a Doppler estimation procedure that works when NT/2 = 1.","section":"Table II, Section III-B, Algorithm 1"},{"comment":"The EKF prediction step x_k[n|n-1] = A x_k[n-1|n-1] + B a_k treats the acceleration a_k as a known input, and the initialization step of Algorithm 3 says only 'estimate x_k[0|0] employing Algorithm 2,' while Algorithm 2 estimates position only and does not provide the initial velocity component of the 4D state. No estimator or adaptation rule for a_k or v_k[0] is given; the statement in Section IV that the filter 'can adapt itself ... by updating the value of acceleration' is not supported by any update equation. The tracking simulations (Figs. 9–11) use the true initial speeds (10/20 m/s) and true acceleration (2 m/s^2), so the reported tracking gains are conditional on an oracle motion model. This is a load-bearing gap for the tracking half of the central claim; the paper should either provide an acceleration/velocity estimator or clearly re-scope the contribution as tracking with externally provided motion inputs.","section":"Section IV, eq. (17), Algorithm 3"},{"comment":"The expression for alpha_{k,nr} in (14) uses absolute values of coordinate differences, e.g., |p_{k,1} - p_{t,1}| and |p_{k,1} - p_{r,nr,1}|. For the simulated geometry (Tx at (0,0), Rxs on the x-axis, RIS at (7,7)), these absolute values change the sign of the departure terms with respect to the standard steering-vector definition: for Rx1, the correct cos(theta) term is (p_{r,1} - p_{k,1})/||p_r - p_k|| = 5/sqrt(74), whereas (14) evaluates to -5/sqrt(74). Because (14) is used in the LS problem (15), the EKF measurement model (18), and the CRLB derivatives in (21), this potential sign inconsistency affects the correctness of the proposed estimators. Please verify the derivation of (14), state the angle convention explicitly, and ensure the simulation code is consistent with the text.","section":"Section III-D, eq. (14)"}],"minor_comments":[{"comment":"In the second arg max, the matrix subscript uses n_{1,fd} instead of n_{2,fd}; the peak should be sought over both the ToA and Doppler dimensions after the IFFT.","section":"Algorithm 1, line 5"},{"comment":"The expression '22.4N^{1.4}' appears to be a typesetting error; it should likely be '2^{2.4} N^{1.4}' or a similar complexity expression.","section":"Section IV, complexity statement"},{"comment":"The contribution bullet claiming that the localization approach 'works even with a single Rx' is not demonstrated by any simulation or identifiability analysis; all localization figures use Nr = 3 or more.","section":"Section I, contributions"},{"comment":"The standard deviation sigma0 of the initial localization error is introduced in the caption, but the mechanism for generating this initial error is not described; please clarify how the initial estimate with a given sigma0 is produced.","section":"Figure 10 caption"}],"recommendation":"major_revision","confidential_remarks":"The manuscript addresses a relevant problem and contains substantial derivations. The two main concerns (NT=2 and oracle acceleration) are correctable in revision, but the sign ambiguity in eq. (14) should be resolved by the authors before acceptance; a quick numerical check of the alpha value for the geometry in Fig. 2 would settle it. I recommend major revision rather than rejection, as the localization approach may be sound after these corrections."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: there is a genuine idea here—single-antenna TX, multiple single-antenna RXs, unknown-location moving RISs, and a clever phase-shift design that cancels static scatterers and separates users. The combination of Doppler handling, the alpha geometric parameter, and CRLB analysis is not in the earlier RIS-localization papers. The localization comparison against [31] shows a meaningful gain, and the derivations are substantial. This is an incremental but useful step, not a paradigm shift.\n\nThe soft spots are real and two are load-bearing.\n\nFirst, Table II sets NT=2, but Algorithm 1 needs at least NT/2 > 1 to estimate Doppler via the 2D FFT. With NT/2=1, the Doppler dimension has length one, fd is unobservable, and the FFT/IFFT sign selection in Algorithm 1 becomes arbitrary. The main localization and tracking simulations use these default settings, so either the authors silently used a larger NT or the numbers do not come from the stated algorithm. That needs to be fixed and clarified before the results are trustworthy.\n\nSecond, the EKF tracking (Section IV, Algorithm 3) takes the acceleration vector a_k as a known input. No estimation or adaptation rule is given. The initialization step estimates only position, not velocity. The simulations feed in the true acceleration and true initial speed, so the reported tracking accuracy rests on oracle knowledge that a real system would not have. The text says the filter can adapt by updating acceleration, but no update rule is provided. This directly undermines the tracking half of the abstract's claim.\n\nAlso, the claim that this is the first study of RIS tracking is contradicted by the paper's own Table I, where [30] has Tracking=Y. The specific combination may be new, but not the problem itself. Minor: simulation curves lack error bars, and the comparison uses a single baseline.\n\nBottom line: worth a serious referee. The core idea deserves attention, but the paper needs major revision. Send it to review rather than desk reject.","headline":"Real idea, but the tracking results rely on oracle acceleration and the default NT=2 makes Doppler unobservable.","tokens_in":23088,"tokens_out":6376,"would_cite":false,"duration_ms":57291,"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":"Coded smart-surface reflections let one antenna localize and track several vehicles, cutting localization error by up to three times versus delay-only methods.","keywords":["reconfigurable intelligent surface","RIS localization","multi-user tracking","extended Kalman filter","time-of-arrival estimation","Doppler estimation","Cramér-Rao lower bound","vehicular networks"],"falsifier":"Run Algorithm 3 with the same geometry and parameters as Section VI but withhold the true acceleration from the tracker and insert an abrupt turn mid-path; if tracking error rises above the error of repeating the static localization at each step, then the reported tracking accuracy depends on knowing the vehicle's maneuvers in advance.","tokens_in":22113,"feed_emoji":"📡","tokens_out":8679,"duration_ms":79411,"temperature":0.7,"pith_summary":"This paper sets out to show that the location and velocity of several vehicles, each carrying a reconfigurable intelligent surface (RIS), can be estimated and tracked using only one transmit antenna and a few single-antenna receivers, even with fixed scatterers present. The key idea is to design the RIS phase shifts so that reflections from different vehicles become separable codes and scatterer reflections cancel, then to read out, per vehicle and receiver, the total path delay and the sum of the cosines of the angles at the RIS. These two quantities are enough to localize each RIS through a least-squares problem and to track it with an extended Kalman filter. If the method works as reported, it offers a low-complexity radio localization option for vehicular RIS systems and reduces localization error by up to a factor of three compared with a time-of-arrival-only approach.","feed_headline":"Smart-surface reflections let one antenna track vehicles","feed_subtitle":"A phase design isolates each car's signal, cutting localization error up to threefold versus delay-only methods.","key_machinery":"The load-bearing object is the phase-profile matrix $\\Omega_k(t,n_T)=\\omega_{k,t}\\Omega_k^{(n_T)}$ programmed into each RIS. The paper sets $\\omega_{k,2t}=-\\omega_{k,2t-1}$ so that subtracting even-slot from odd-slot received matrices $\\mathbf{Y}'_{n_r,n_T}=\\frac{1}{2}(\\mathbf{Y}_{n_r,n_T,o}-\\mathbf{Y}_{n_r,n_T,e})$ cancels the fixed scatterer contributions; it chooses the per-RIS code vectors $\\gamma_k$ as orthogonal columns of an FFT matrix to separate the $K$ users; and it toggles only the second RIS element between odd and even blocks so that the nonlinear system for $\\alpha_{k,n_r}$ collapses to a single-exponent least-squares solution $\\hat{\\alpha}_{k,n_r}=\\frac{\\lambda}{2\\pi d}\\arg\\!\\left(\\frac{1}{N_T}\\sum_{n_T=1}^{N_T/2}\\frac{\\hat{s}_{k,n_r,2n_T-1}-\\hat{s}_{k,n_r,2n_T}}{\\Omega_{k,2,2}^{(2n_T-1)}-\\Omega_{k,2,2}^{(2n_T)}}\\right)$. The measurement vector used by both the initialization and the extended Kalman filter is the concatenation of the estimated path sums $\\hat{\\xi}_{k,n_r}=c\\hat{\\tau}_{k,n_r}$ and the estimated $\\hat{\\alpha}_{k,n_r}$, related to the RIS position by equations (13) and (14) of the paper.","core_discovery":"The paper's central claim is that a moving RIS is not just a reflector to be calibrated away but a target whose position and velocity can be read directly from the reflected signal. With a single-antenna transmitter and $N_r$ single-antenna receivers, the paper estimates, for each RIS $k$ and receiver $n_r$, the delay $\\tau_{k,n_r}$ of the Tx\\textendash RIS\\textendash Rx path and the geometric parameter $\\alpha_{k,n_r}=\\cos\\phi_k+\\cos\\theta_{k,n_r}$, the sum of the cosines of the angle of arrival at the RIS and the angle of departure from it. Because $\\tau_{k,n_r}$ fixes the ellipse of points whose Tx\\textendash RIS\\textendash Rx path length equals $c\\tau_{k,n_r}$, and $\\alpha_{k,n_r}$ fixes a second curve that depends on the RIS orientation, their intersection localizes the RIS; the paper solves this as a least-squares problem and feeds the same measurements into an extended Kalman filter that tracks position and velocity over time. The reported result is that this geometry-augmented estimator outperforms a ToA-only baseline by up to a factor of three in localization error, and that the tracking filter further reduces the error along a trajectory.","pith_inferences":["A natural extension not studied here is estimating the acceleration online, for example with an interacting-multiple-model filter; that would show how much of the reported tracking accuracy survives when the known-acceleration input is removed.","The same sign-alternation and orthogonal-code construction should transfer to any backscatter tag with a controllable reflection phase, not only vehicular RISs, so the separation principle may extend to low-power IoT localization.","Because the gap between the simulated localization error and the PEB is attributed to FFT resolution, super-resolution delay and Doppler estimators are a direct, testable route to closing that gap."],"forward_implications":["A roadside node with one transmit antenna and a handful of single-antenna receivers can simultaneously separate and track several RIS-equipped vehicles in an environment with fixed scatterers.","For a target localization accuracy of 0.1 m, the method needs roughly 20% less transmit power than the ToA-only benchmark, or it can reach the same accuracy with half the number of subcarriers.","The same OFDM waveform used for communication carries the delay and $\\alpha$ measurements, so localization can be layered onto an existing data link without extra ranging pulses.","The derived PEB heat map gives a design rule: receivers should be placed so that vehicle paths stay near the receivers, where the bound is lowest, rather than near the transmitter region where the bound rises."],"supporting_citations":[{"why":"The ToA/TDoA-based benchmark for localizing multiple RIS-enabled users; the proposed localization approach is compared against it and reports up to a threefold error reduction.","marker":"[31]"},{"why":"Supplies the linear-RIS array model and phase-profile programming convention assumed in the system model.","marker":"[37]\\textendash[40]"},{"why":"Provides the ML invariance property, the extended Kalman filter recursions, and the Cram\\'er-Rao lower bound formulas on which the estimation, tracking, and bound derivations are built.","marker":"[41]"}],"fun_headline_variants":["One antenna, smart reflections: track vehicles 3x better","RIS reflections let a single antenna localize cars, error cut 3x","Single antenna tracks vehicles via RIS, tripling accuracy","One Tx antenna + RIS reflections: 3x better vehicle tracking","RIS-equipped cars tracked from one antenna, 3x less error"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The tracking filter assumes the receiver is given each vehicle's acceleration vector at every prediction step, and the paper supplies no way to estimate or adapt it.","fun_headline_variants_meta":{"raw":{"variants":["One antenna, smart reflections: track vehicles 3x better","RIS reflections let a single antenna localize cars, error cut 3x","Single antenna tracks vehicles via RIS, tripling accuracy","One Tx antenna + RIS reflections: 3x better vehicle tracking","RIS-equipped cars tracked from one antenna, 3x less error"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000986,"raw_usage":{"total_tokens":4192,"prompt_tokens":962,"completion_tokens":3230,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":578,"completion_tokens_details":{"reasoning_tokens":3141}},"tokens_in":578,"tokens_out":3230,"duration_ms":20360,"temperature":1.0,"reasoning_tokens":3141,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T14:12:19.951151+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run Algorithm 3 with the same geometry and parameters as Section VI but withhold the true acceleration from the tracker and insert an abrupt turn mid-path; if tracking error rises above the error of repeating the static localization at each step, then the reported tracking accuracy depends on knowing the vehicle's maneuvers in advance.","supporting_citations":[{"cited_title":"Semi-passive 3D positioning of multiple RIS-enabled users,","cited_arxiv_id":null,"evidence_quote":"The ToA/TDoA-based benchmark for localizing multiple RIS-enabled users; the proposed localization approach is compared against it and reports up to a threefold error reduction."},{"cited_title":"AI/ML in 5G","cited_arxiv_id":null,"evidence_quote":"Provides the ML invariance property, the extended Kalman filter recursions, and the Cram\\'er-Rao lower bound formulas on which the estimation, tracking, and bound derivations are built."}],"review_version":1}