{"id":"8ee05538-99c9-42a7-bee6-25e0e8171125","arxiv_id":"2501.02252","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A two-timescale scheme exploits shared and independent scatterers among users to reuse pilots and jointly estimate channels, scatterers, and user positions.","lead":"This paper proposes a two-stage method that reuses pilot signals among users whose signal reflections come from separated objects, then jointly estimates channels and positions. It aims to cut pilot overhead in massive MIMO-OFDM systems, where the number of users can exceed the number of orthogonal pilots.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The pilot-reuse gain rests on overlap probabilities rho_k and sparsity lambda that the paper never shows how to estimate; if the simulations use oracle values, the reported 2.5-4x gain may not survive realistic prior estimation.","rationale":"The reader's weakest assumption correctly points at the sparse point-scatterer model and the reliance on prior overlap probabilities. I agree with the rho_k / lambda part, but I would de-emphasize the diffuse/multibounce mismatch as the primary risk because the paper's simulation appears to use a controlled L=6 scatterer environment. The sharper, internally checkable issue is that the short-timescale algorithm's message passing consumes rho_k and lambda as fixed inputs, and the paper never explains how they are obtained. This is a missing support rather than a demonstration of failure: the simulation may be using true support statistics, which would inflate the MU gain. The conditional verdict remains appropriate, but the acceptance conditions should include an ablation study where rho_k and lambda are estimated from the long-timescale detections and a sensitivity analysis over their errors. No machine-checked proofs or code are provided, so this concern cannot be dismissed from the text alone; a targeted simulation would settle it.","tokens_in":20178,"tokens_out":11865,"duration_ms":125767,"concrete_test":"Rerun the Section VI simulation with rho_k and lambda estimated from the long-timescale output rather than set to true values: for each user k, set rho_hat_k = |{detected scatterers associated with k}| / |{all detected scatterers}|, and set lambda_hat = |{all detected scatterers}| / Q. Then plot the Fig. 8 NMSE curves. If MU-NP at kappa=2 moves more than 3 dB away from the oracle-prior curve, or if the horizontal gap in Fig. 7 at fixed NMSE falls below 2x, the central claim fails under realistic prior estimation. A second arm: set rho_k=1 for all users while ground-truth L_k varies, to confirm sensitivity.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline result (Section VII, Fig. 7) is produced by SEA-LD-TurboCE, whose Module B message passing uses the joint-support prior in Eqs. (13)-(15). The conditional overlap probabilities rho_k = L_k/L in Eq. (14) and the sparsity rate lambda in Eq. (15) enter directly into the message updates (41)-(46). Nowhere in Section IV or Algorithm 2 is a procedure given for estimating rho_k or lambda from the long-timescale TST-MUSIC/DBSCAN output; the long-timescale phase estimates scatterer positions and associations only. If the simulations set rho_k and lambda from the true L_k, L, and Q, then the MU-OP gain over SU-OP and the MU-NP kappa=2 result in Fig. 8 include oracle knowledge of the support overlap. This is load-bearing because the entire rationale for MU pilot reuse is that nearby users share scatterers; a wrong rho_k (e.g., rho_k=1 when users are actually disjoint) biases the message passing to force common support and can create the very interference the scheme is supposed to avoid. The claimed 2.5-4x pilot expansion relative to SU-OP is therefore not yet established for a system that must estimate its own environment prior.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a two-timescale scheme for joint multi-user uplink channel estimation and localization in MIMO-OFDM systems. In the long-timescale phase, TST-MUSIC provides coarse per-user scatterer estimates, DBSCAN associates and refines shared scatterers across users, and a PSO-based graph-coloring algorithm groups users so that pilots can be spatially reused. In the short-timescale phase, the SEA-LD-TurboCE algorithm combines LMMSE estimation with location-domain message passing and EM-based grid/user-position refinement, exploiting a joint sparsity model of overlapping scatterers. Simulations with a QuaDRiGa UMa channel model compare the proposed MU-NP scheme with MU-OP, SU-OP, MUSIC+LS, and genie-aided LMMSE baselines, reporting pilot-reduction gains of 2.5-4x over SU-OP and near-genie NMSE for MU-OP.","tokens_in":20486,"tokens_out":12143,"duration_ms":108864,"significance":"If the results hold, the paper addresses a practically important problem: pilot overhead reduction in massive MIMO by exploiting physical scatterer sharing among nearby users. The two-timescale pipeline, the combination of environment learning with turbo channel estimation, and the use of an independent QuaDRiGa channel model are notable strengths. The central claim of pilot expansion gain, however, depends on prior parameters whose estimation is not demonstrated; the reported gains may therefore overstate what is achievable without oracle knowledge. With additional ablations and corrected notation, the contribution could be solid and useful.","major_comments":[{"comment":"The support variables are defined as binary 0/1 in Section II.C, but Eqs. (12) and (14) use delta functions of the form delta(s+1) and delta(s-1), which correspond to a +/-1 encoding. In contrast, Table I and the message-passing updates in Eqs. (41)-(46) use delta(s) and delta(s-1), a 0/1 encoding. This inconsistency makes the derivation of Module B in Section V ambiguous; please adopt a single convention consistently throughout the paper and verify that the message update equations match that convention.","section":"Section II.C and Section III.C, Eqs. (12)-(15); Table I"},{"comment":"The joint-support prior depends on rho_k = L_k/L and the sparsity rate lambda, and these parameters enter directly into the message updates (41)-(46). The paper does not describe how rho_k or lambda are estimated from the long-timescale TST-MUSIC/DBSCAN output, nor does it report the values used in simulation. If rho_k and lambda are set from the true L_k, L, and support statistics, then the pilot-reuse gains in Figs. 7 and 8 include oracle prior knowledge. Please provide an estimation procedure or an ablation study with mismatched or estimated rho_k and lambda; without this, the central claim of 2.5-4x pilot expansion is not yet established for a system that must learn its own environment.","section":"Section III.C, Eqs. (14)-(15); Section V, Eqs. (41)-(46); Section VI"},{"comment":"Equation (22) models the received group signal as a sum over users of (Phi_k x_k + z_k), i.e., a separate noise vector per user. In a real receiver there is a single additive noise realization per received signal; as written, the noise variance is inflated by the group size and is inconsistent with the LMMSE update in Eq. (33), which uses a single sigma_z^2. Please correct Eq. (22) to y_g = sum_k Phi_k x_k + z_g or otherwise reconcile the noise model.","section":"Section III.C, Eq. (22)"},{"comment":"The proposed graph-coloring method is presented as solving the max-min intra-group distance problem in Eq. (10), but the fitness function in Eq. (32) only penalizes same-colored adjacent vertices and the number of colors used; it does not directly incorporate the minimum distance. Algorithm 1 decreases d_adj until G colors suffice, which yields a feasible coloring for some threshold but is not guaranteed to maximize the minimum distance. Please state this approximation explicitly, and additionally report the gap to an upper bound or compare against a direct optimization baseline for the grouping objective.","section":"Section IV.C, Eq. (10), Eq. (32), Algorithm 1"}],"minor_comments":[{"comment":"In Eq. (25), the square-root symbol appears as a stray \"q\" in the typeset expression, and the delay estimated by TST-MUSIC is written as tau_l(p_UE,k) rather than with a hat; please fix the notation and the formula.","section":"Section IV.A, Eq. (25)"},{"comment":"In the row for f_0^k, the functional form is written as CN(x_q; x_{B,q}^{k,pri}, v_{B,q}^{k,pri}) but it should involve x_0^k, not x_q. The table also mixes N and CN notation; please make it uniform.","section":"Table I"},{"comment":"The DBSCAN parameters epsilon and MinPts, the PSO acceleration coefficients c1 and c2, and the Algorithm 1 parameters Delta_d_adj and epsilon are not reported; please include these settings so that the simulations are reproducible.","section":"Section IV.B and Section IV.C"},{"comment":"Reference [21] appears to contain duplicated volume/page information (\"vol. 68, pp. 2598-2612, 2020\" followed by \"vol. 19, no. 5, pp. 3185-3198, 2020\"); please correct the bibliography entry.","section":"References"},{"comment":"There is a typo in the sentence \"Then we need to ulize ...\"; it should read \"utilize.\"","section":"Section IV.A"}],"recommendation":"major_revision","confidential_remarks":"The paper has a credible algorithmic pipeline and an independent QuaDRiGa evaluation, but the simulation section does not report how the joint-sparsity priors are obtained. If the reported gains rely on oracle rho_k and lambda, the practical claims are not yet supported. The support-variable notation inconsistency is also a correctness risk because the message-passing equations are derived from those priors. I recommend requesting an ablation study and a full consistency pass before further consideration."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a genuine engineering contribution, not a repackaging. The new thing is the assembly: long-timescale scatterer localization (TST-MUSIC), DBSCAN-based association across users to find shared scatterers, graph-coloring/PSO to group users so they can reuse pilots, and a short-timescale turbo message-passing estimator that exploits the resulting location-domain joint sparsity. Each component exists, but the integrated two-timescale pipeline with location-domain spatial pilot reuse is, as far as I can tell, new. The paper also does most things right in evaluation: Quadriga UMa channels, K=200 users, sensible baselines (SU-OP, MUSIC+LS), and a genie-aided LMMSE bound. The message-passing derivation is detailed and looks self-consistent; the MU-OP curve approaching the genie bound is a useful sanity check.\n\nThe soft spots are real but not disqualifying. Most important: the prior parameters rho_k = L_k/L and lambda are load-bearing in the Module B updates (Eqs. 41-46), and the paper never says how they are obtained in practice or in the simulations. If the simulations set them from the true scatterer counts and support sizes, the reported 2.5-4x pilot reuse gain may be optimistic for a system that has to estimate its own environment. The long-timescale DBSCAN output could plausibly provide these estimates, but that step isn't described. This needs to be addressed.\n\nAlso: hyperparameters (DBSCAN eps/MinPts, PSO c1/c2, lambda, sigma_UE) aren't reported; there are no error bars, only 100 realizations; Eq. (25) has a typo (it minimizes over r_l but the expression uses r_l inside and has no apparent dependence on the estimated range), and Table I copies the wrong functional form for f_k^0. The graph-coloring fitness is a heuristic relaxation of the stated max-min problem, so the gap between (10) and (32) should be discussed.\n\nNet: the central idea holds up and the evidence is mostly consistent with it, but the paper needs a revision that discloses the prior setup and adds robustness checks. I think it deserves a serious referee; a conditional acceptance with a request for the missing parameter settings and a discussion of rho_k/lambda estimation would be appropriate. If the authors can show the gain survives with estimated priors, this is a solid journal paper.","headline":"A well-put-together two-timescale MU channel estimation/localization scheme with real pilot-reuse promise; the main open question is whether the sparsity priors (rho_k, lambda) are estimated or oracle-fed in simulation.","tokens_in":20995,"tokens_out":3131,"would_cite":true,"duration_ms":32100,"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":"Spatially reused pilots can double pilot capacity in massive MIMO with little channel-estimation loss.","keywords":["MIMO-OFDM","multi-user channel estimation","localization","pilot reuse","spatial division multiplexing","scatterer association","location-domain sparsity","turbo channel estimation"],"falsifier":"Run the proposed pipeline on measured or realistically simulated channels that contain dense diffuse scattering, multi-bounce paths, or non-point scatterers, and compare the multi-user non-orthogonal-pilot NMSE against the genie-aided LMMSE bound at high SNR; if the NMSE floor stays far above the bound even with orthogonal pilots, or if the gap between the non-orthogonal and orthogonal variants grows much larger than the simulated small loss, then the discrete-scatterer joint-sparsity model is the point of failure. A simpler check: compute the miss-detection probability of shared scatterers as SNR increases; if it does not fall with SNR, the physical premise of shared discrete scatterers is violated.","tokens_in":19993,"feed_emoji":"📡","tokens_out":5772,"duration_ms":53648,"temperature":0.7,"pith_summary":"Massive MIMO systems face a pilot bottleneck: every user needs a distinct pilot to avoid interference, so the number of users is capped by the pilot budget. This paper argues that the bottleneck can be loosened by treating the scattering environment as a shared resource: nearby users see the same scatterers, while distant users see nearly orthogonal channels. It proposes a two-timescale scheme that first maps each user's scatterers, merges shared scatterers across users, and groups users so that the same pilot can be reused for spatially separated channels. The short-timescale estimation then jointly recovers channels and locations under these reused pilots. Simulations indicate the multi-user non-orthogonal-pilot scheme supports more than twice the users per pilot resource as a single-user orthogonal-pilot baseline, with a modest loss in NMSE.","feed_headline":"Shared scatterers let pilots serve twice the users","feed_subtitle":"Massive MIMO scheme reuses pilots among nearby users while keeping channel estimates close to the ideal bound.","key_machinery":"The load-bearing object is the 3-D location-domain sparse representation of the multi-user channel, together with the joint sparsity prior that ties users together. Each user's NLOS channel is written as a sparse combination of basis vectors indexed by candidate scatterer locations, so users who see the same physical scatterer share non-zero entries in their support vectors; the overlap probability $\\rho_k = L_k/L$ quantifies how much of the global scatterer set a given user sees. This representation turns the physical fact of common scattering into a signal-processing structure that the turbo algorithm can exploit, and it is what makes pilot reuse possible: users whose scatterer sets are far apart have nearly orthogonal columns and can safely share one pilot sequence. The two-timescale pipeline (TST-MUSIC for coarse angles and delays, DBSCAN for associating scatterer estimates across users, graph-coloring grouping, and SEA-LD-TurboCE for joint recovery) is the mechanism that converts this representation into a working estimator.","core_discovery":"The central claim is that channel estimation and localization in a massive MIMO-OFDM uplink can be performed jointly, accurately, and with far fewer pilot resources by exploiting the spatial structure shared among users. The paper models each user's channel in a 3-D location domain, where a channel is a line-of-sight component plus a few single-bounce paths through discrete scatterers, each described by its own position. Because nearby users are assumed to share scatterers, their support vectors in this domain overlap, and that overlap is parameterized by known probabilities. The proposed long-timescale phase uses TST-MUSIC to obtain coarse scatterer positions per user, DBSCAN to associate and refine shared scatterers, and a graph-coloring particle-swarm algorithm to group users whose channels are quasi-orthogonal in the location domain; users in the same group then reuse the same pilots. The short-timescale phase runs the SEA-LD-TurboCE algorithm, a turbo-style LMMSE/message-passing estimator with EM-based grid and position refinement, to estimate all channels and positions from the spatially reused pilots. Simulation results show the orthogonal-pilot variant approaching the genie-aided LMMSE bound and the non-orthogonal variant achieving more than double the pilot expansion gain of the single-user scheme with only a small NMSE loss.","pith_inferences":["The same location-domain joint-sparsity idea could be carried over to FDD downlink estimation, since scatterer geometry is reciprocal even when channel coefficients are not; the paper does not address this direction.","If user density grows so that most users share scatterers, the pilot-reuse factor might be pushed higher, but the roughly 10% of same-group users closer than 50 m in the simulations suggests estimation error in scatterer positions is the limiting factor.","A natural experimental test would be to run the grouping-plus-turbo pipeline on measured urban channels with rich scattering; if diffuse and multi-bounce energy is large, the sparse single-bounce model may need extension to keep the promised gains.","The scheme could feed integrated sensing and communication systems, since the refined scatterer map it produces is itself a map of the physical environment, not only a channel estimate."],"forward_implications":["Pilot overhead per user can be cut by a factor of 2 to 4 while keeping channel estimation error close to the orthogonal-pilot baseline, so a fixed pilot budget can support proportionally more users.","Scatterer localization accuracy improves with the number of users, because shared scatterers are observed and averaged across several users' estimates, reducing miss-detection probability and RMSE.","The user-grouping decision is made once per long-timescale phase and reused for many short-timescale transmissions, so the pilot-reuse gain comes at a low recurring computational cost.","Even with imperfect prior user-location information (simulated with standard deviation up to 20 m), the multi-user scheme degrades less than a MUSIC-based baseline, because fusing shared-scatterer observations partially cancels location errors.","The scheme applies specifically to TDD massive MIMO-OFDM uplinks, where reciprocity lets the estimated uplink channel serve as downlink CSI."],"supporting_citations":[{"why":"Supplies the TST-MUSIC algorithm that gives the coarse angle and delay estimates from which scatterer positions are derived.","marker":"[26]"},{"why":"Provides the DBSCAN-based clustering approach used to associate scatterer observations across users and identify shared scatterers.","marker":"[30]"},{"why":"Establishes the previous single-user turbo channel-estimation framework that SEA-LD-TurboCE extends to the multi-user location-domain setting.","marker":"[13]"},{"why":"Reports channel measurements showing that physically proximate users share local scattering clusters, the empirical premise for the joint sparsity model.","marker":"[25]"},{"why":"Introduces graph-coloring-based pilot allocation that the user-grouping algorithm adapts for spatial pilot reuse.","marker":"[10]"},{"why":"Defines the standardized channel-model simulation setup used to generate the multi-path channels in the numerical results.","marker":"[31]"}],"fun_headline_variants":["Shared scatterers let pilots serve many users","Pilot reuse via scatterer-aware user grouping","Scatterer overlap enables pilot sharing in MIMO","Location-domain modeling reduces pilot overhead","Joint channel and location with reused pilots"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole pipeline assumes a real uplink channel is well approximated by one line-of-sight path plus a small number of single-bounce paths from discrete point scatterers, and that the degree of scatterer overlap between users ($\\rho_k = L_k/L$) is known or can be estimated before estimation begins.","fun_headline_variants_meta":{"raw":{"variants":["Shared scatterers let pilots serve many users","Pilot reuse via scatterer-aware user grouping","Scatterer overlap enables pilot sharing in MIMO","Location-domain modeling reduces pilot overhead","Joint channel and location with reused pilots"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000287,"raw_usage":{"total_tokens":1759,"prompt_tokens":1095,"completion_tokens":664,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":711,"completion_tokens_details":{"reasoning_tokens":598}},"tokens_in":711,"tokens_out":664,"duration_ms":7064,"temperature":1.0,"reasoning_tokens":598,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:14:06.934183+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the proposed pipeline on measured or realistically simulated channels that contain dense diffuse scattering, multi-bounce paths, or non-point scatterers, and compare the multi-user non-orthogonal-pilot NMSE against the genie-aided LMMSE bound at high SNR; if the NMSE floor stays far above the bound even with orthogonal pilots, or if the gap between the non-orthogonal and orthogonal variants grows much larger than the simulated small loss, then the discrete-scatterer joint-sparsity model is the point of failure. A simpler check: compute the miss-detection probability of shared scatterers as SNR increases; if it does not fall with SNR, the physical premise of shared discrete scatterers is violated.","supporting_citations":[{"cited_title":"TST-MUSIC for joint DOA delay estimation,","cited_arxiv_id":null,"evidence_quote":"Supplies the TST-MUSIC algorithm that gives the coarse angle and delay estimates from which scatterer positions are derived."},{"cited_title":"An Entropy-Based Adaptive DBSCAN Clustering Algorithm and Its Application in THz Wireless Channels,","cited_arxiv_id":null,"evidence_quote":"Provides the DBSCAN-based clustering approach used to associate scatterer observations across users and identify shared scatterers."},{"cited_title":"Robust Multi-user Channel Tracking Scheme for 5G New Radio,","cited_arxiv_id":null,"evidence_quote":"Establishes the previous single-user turbo channel-estimation framework that SEA-LD-TurboCE extends to the multi-user location-domain setting."},{"cited_title":"Channel measurements for large antenna arrays","cited_arxiv_id":null,"evidence_quote":"Reports channel measurements showing that physically proximate users share local scattering clusters, the empirical premise for the joint sparsity model."},{"cited_title":"Graph coloring based pilot allocation to mitigate pilot contamination for multi-cell massive MIMO systems, IEEE Communications Letters , 2015, 19(10): 1842-1845","cited_arxiv_id":null,"evidence_quote":"Introduces graph-coloring-based pilot allocation that the user-grouping algorithm adapts for spatial pilot reuse."},{"cited_title":"Study on channel model for frequencies from 0.5 to 100 GHz (3GPP TR 38.901 version 16.1.0 release 16),","cited_arxiv_id":null,"evidence_quote":"Defines the standardized channel-model simulation setup used to generate the multi-path channels in the numerical results."}],"review_version":1}