{"id":"0788e419-eb82-4ac0-96dd-0e31fe2f0452","arxiv_id":"2508.04222","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"VR-HMM-P-SOMP interleaves HMM-based visibility region estimation via Viterbi decoding into polar-domain simultaneous OMP, improving near-field ELAA channel estimation in low-SNR and sparse scenarios.","lead":"This paper combines sparse recovery with a hidden Markov model to estimate which antennas on a very large array actually see each propagation path. The method improves channel estimation accuracy in noisy, near-field conditions while keeping computational cost close to the baseline.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"HMM emission mapping (Step 11) is an unvalidated heuristic with hand-tuned T=20; the algorithm's VR estimates and hence its NMSE advantage over P-SOMP rest entirely on this mapping.","rationale":"The reader's weakest assumption identifies the HMM emission mapping as ad hoc, and I agree that this is the most load-bearing concern. The central claim is that VR-HMM-P-SOMP outperforms P-SOMP, but the only algorithmic difference is the VR masking driven by the HMM. If the emission mapping is miscalibrated, the masks are wrong, and the algorithm could easily underperform the baseline. The paper provides no sensitivity analysis or theoretical justification for the sigmoid with T=20, leaving the result potentially overfitted to the specific simulation setup. While the missing comparison to TS-VRCE and Alternating MAP is also a notable weakness, it affects the breadth of the claim rather than its internal validity; the HMM emission concerns the core mechanism, making it more directly load-bearing. A simple parameter sweep and comparison to a data-driven emission model would settle whether the algorithm's advantage is robust or an artifact of the chosen heuristic. The reader's CONDITIONAL verdict already reflects this uncertainty, so my analysis does not change it.","tokens_in":9366,"tokens_out":7365,"duration_ms":88541,"concrete_test":"Re-run the Monte Carlo setup of Section IV with T swept over {1, 5, 20, 50, 100} and with the transition probabilities scaled by factors {0.5, 2}. Additionally, compute the empirical histograms of o conditional on true VR membership from the genie-aided simulations and replace the sigmoid with the empirically derived likelihood ratio. If the NMSE at 5 dB SNR (Fig. 3) varies by more than 1 dB across these settings, or if the HMM mask disagrees with the true non-binary mask (thresholded at 0.5) in more than 10% of antennas, the reported gains are not robust to the key design choice.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The algorithm's only new component relative to P-SOMP is the HMM-based VR estimation (Steps 7–13). The observation in Step 10 is the per-antenna average reduction in residual magnitude after subtracting the unmasked path contribution. Step 11 maps this scalar o to emission probabilities with a sigmoid of fixed steepness T=20. This mapping is introduced without derivation from the channel statistics (Eq. 3), without justification for the sigmoid form, and without an analysis of how T should scale with SNR, number of subcarriers/pilots, or VR geometry. Viterbi decoding then yields a binary mask that directly multiplies the steering vector in Step 13; any miscalibration therefore propagates into the support projection (Step 14) and the final channel estimate. If the true relationship between o and VR membership is not approximately sigmoidal with this steepness, the masks will be systematically wrong, and the masked steering vectors can be worse than the unmasked ones, potentially erasing the claimed 2 dB gain at 5 dB SNR. No sensitivity analysis is provided for T or the transition probabilities, and no comparison is made to a data-driven emission model. Since the paper's central claim is that VR-HMM-P-SOMP outperforms P-SOMP, and this advantage is the only place where the algorithm differs, the ad hoc emission model is the weakest link.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers channel estimation for ELAA systems operating in the near field with spatial non-stationarity. It proposes a hybrid channel model that includes non-binary visibility-region (VR) masks generated by a knife-edge diffraction model. The main algorithmic contribution is VR-HMM-P-SOMP, which interleaves binary VR-mask estimation into the polar-domain simultaneous OMP framework. After each support selection, a two-state HMM over antennas is decoded via Viterbi to produce a VR mask; the mask is then applied to the selected steering vector before the next orthogonal projection. Simulations compare the proposed method with LS, P-SOMP, two subarray P-SOMP variants, and a genie-aided version, reporting NMSE gains particularly at low SNR and sparse settings. The paper claims the method preserves the asymptotic complexity of P-SOMP.","tokens_in":9792,"tokens_out":1934,"duration_ms":26724,"significance":"If the reported gains hold, the paper offers a low-complexity heuristic for incorporating spatial non-stationarity into greedy sparse channel estimation, an important problem for ELAA/6G. The paper is transparent about the oracle sparsity L_hat = 2L and about the heuristic nature of the HMM emission model, and it provides reproducible-looking algorithmic pseudocode and simulation details. However, the core advantage over P-SOMP depends entirely on the ad hoc sigmoid emission mapping with a fixed temperature T=20, and the validation is confined to the authors' own channel model with hand-set HMM parameters. The significance is therefore conditional on additional robustness and generalization evidence.","major_comments":[{"comment":"The emission model in Step 11 is an ad hoc sigmoid mapping o -> sigma(o) with fixed steepness T=20. This mapping is not derived from the channel model in Eq. (3), and no sensitivity analysis is provided for T, the transition probabilities (1-1/N, 1/N), or the initial probabilities (0.55, 0.45). Since Step 13 multiplies the steering vector by the resulting binary mask, any miscalibration directly propagates into the support projection and the final NMSE. The claimed 2 dB gain over P-SOMP at 5 dB SNR rests on this untested calibration. I request either a derivation of the emission mapping from the channel statistics or a systematic robustness study varying T and the HMM parameters across SNR, pilot length, subcarrier number, and VR geometry.","section":"Section III.C, Algorithm 1 Step 11 and Table I"},{"comment":"The simulations evaluate the algorithm only with the oracle stopping sparsity L_hat = 2L, as explicitly stated in Section III.A. This is a free parameter and is not available in practice. The performance comparison with P-SOMP and subarray methods could change materially when L_hat must be estimated from the data (e.g., via a threshold on residual energy or a sparsity penalty). The paper should either demonstrate that the reported gains persist with a realistic data-driven stopping rule or clearly state the sensitivity of the results to L_hat.","section":"Section III.A and Section IV"},{"comment":"The validation is entirely on the authors' synthetic channel model, including the non-binary VR masks generated by the knife-edge model in Eq. (5). The proposed algorithm estimates binary masks while the channel contains non-binary diffraction masks, so the benchmark comparison may be favorable because the competing methods are designed for stationary channels. To support the general claim of robustness, the paper should include at least one evaluation under a different or measurement-based channel model, or a model-mismatch analysis showing how performance degrades when the assumed VR statistics differ from the true ones.","section":"Section IV and channel model in Section II.B"}],"minor_comments":[{"comment":"The symbol T is used both for the pilot length in Eq. (1) and for the sigmoid temperature in Step 11 and Table I. This is confusing, especially in Step 10 where T appears in the averaging operation. Please rename one of them.","section":"Notation"},{"comment":"Table I lists 'Channel SNR 0 dB', but Fig. 3 sweeps SNR over -15 to 20 dB. This entry is inconsistent with the reported experiments and should be clarified or removed.","section":"Table I"},{"comment":"In the non-binary VR mask case, the mask depends on the subcarrier index m through lambda_m, but the model in Eq. (3) writes s_{m,l} without emphasizing this dependence in the notation. This makes the subsequent averaging in Step 10 over subcarriers less transparent. Please clarify.","section":"Section II.B"},{"comment":"The function freshape is described verbally but not defined precisely in the pseudocode or in the notation section. A concise definition would improve reproducibility.","section":"Algorithm 1"},{"comment":"The conclusion acknowledges that explicit non-binary mask estimation is left for future research. This is an honest limitation, but it should be reflected earlier in the paper and in the abstract, as the title and contribution list emphasize non-binary VR modeling.","section":"Section V"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a solid engineering contribution. It interleaves an HMM-based visibility-region estimator into polar-domain SOMP, using a residual-reduction observation and Viterbi decoding to mask steering vectors. That combination is new and the idea is simple enough to be plausible. The authors are also transparent: they state the oracle sparsity (L_hat = 2L), give a full simulation configuration, and show asymptotic complexity stays O(hat L N S T M).\n\nWhat the paper does well is the evaluation within its own model. The proposed method consistently beats plain P-SOMP across SNR, with a 2 dB gain at 5 dB SNR and a 6.5 dB gain over subarray methods at -5 dB. The sparsity sweep in Figure 4 is also convincing. The genie-aided upper bound is a good sanity check.\n\nThe soft spots are real but not fatal. The HMM emission mapping in Step 11 is a sigmoid with hand-tuned temperature T=20, and there is no sensitivity analysis for T or the transition probabilities. The stress-test note is right: since this mapping is the only new component, the algorithm's advantage rests on it. But the simulations do test it across several SNR and sparsity conditions, so it is not unsupported—just under-analyzed. The bigger omission is the lack of comparison to the two most relevant VR-aware baselines, TS-VRCE and Alternating MAP. Those are cited but not simulated, and they are the natural competitors. The non-binary VR masks are used only for generation; the algorithm estimates binary masks, which the authors acknowledge.\n\nOne more limitation: the channel model is the one the algorithm was designed for, so the results are self-consistent but not evidence of robustness to model mismatch. That is typical for this subfield, so I do not treat it as a disqualifying flaw.\n\nWho should read this: anyone working on near-field ELAA channel estimation or spatial non-stationarity. It is an incremental but useful extension. My own verdict is that the central claim holds for the tested setting, but the paper needs a sensitivity analysis and the missing baseline comparison before I would fully trust it.\n\nRecommendation: send to peer review. The referees should ask for a sensitivity study of T and transition probabilities and a comparison with TS-VRCE and Alternating MAP. If those are added, the paper could be a solid journal contribution.","headline":"A clean, incremental extension of P-SOMP with HMM-based VR estimation; the gains are real in simulation but the emission model is hand-tuned and the most relevant baselines are missing.","tokens_in":10193,"tokens_out":2030,"would_cite":false,"duration_ms":26028,"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":"This paper proposes that interleaving HMM-based visibility-region estimation into polar-domain simultaneous OMP improves near-field, spatially non-stationary channel estimation in ELAA systems while keeping asymptotic complexity unchanged.","keywords":["channel estimation","extremely large aperture arrays","near-field","spatial non-stationarity","visibility region","hidden Markov model","orthogonal matching pursuit","6G"],"falsifier":"Run VR-HMM-P-SOMP on simulated channels where the true VR masks are known but the per-antenna residual reductions are deliberately decorrelated from VR membership, for example by adding two paths with overlapping support or strong noise correlation across antennas. If the Viterbi-decoded masks no longer match the true masks and the NMSE advantage over unmasked P-SOMP disappears, the central claim that residual reduction reveals visibility regions would be falsified. Alternatively, a measurement campaign with a controllable blocker could compare estimated masks against known blockage geometry.","tokens_in":9334,"feed_emoji":"📡","tokens_out":5834,"duration_ms":60727,"temperature":0.7,"pith_summary":"This paper is trying to establish that channel estimation for extremely large aperture arrays, where hundreds of antennas face near-field spherical wavefronts and different antennas see different propagation paths, can be made substantially more accurate by estimating, for each path, which antennas actually see it and masking the steering vectors accordingly. The proposed algorithm, VR-HMM-P-SOMP, does this by inserting a two-state hidden Markov model into the polar-domain simultaneous orthogonal matching pursuit loop: after each path is selected, the per-antenna reduction in the residual is converted into emission probabilities, Viterbi decoding yields the visibility-region mask, and the mask is applied before the next projection. In simulation the method beats the unmasked P-SOMP at every SNR tested and beats subarray-based estimators by about 6.5 dB at -5 dB SNR, while asymptotically keeping the same computational complexity. A reader should care because spatial non-stationarity and near-field propagation are two main obstacles to making ELAA systems work in 6G, and this is a low-complexity way to address both at once.","feed_headline":"Hidden Markov masks lift near-field channel estimates at low SNR","feed_subtitle":"VR-HMM-P-SOMP recovers which antennas see each path, beating subarray methods by 6.5 dB at -5 dB SNR.","key_machinery":"A two-state hidden Markov model over the antenna array, with states 'in VR' and 'out of VR,' whose emission probabilities are a sigmoid (steepness $T=20$) of the per-antenna average residual reduction caused by adding the newly selected path. Viterbi decoding then returns the most likely binary VR mask, which is applied to the selected near-field polar-domain steering vector before the least-squares re-projection. The HMM's transition probabilities encode spatial continuity (probability $1-1/N$ of staying in the same state between adjacent antennas), which is what lets the mask estimation use neighboring-antenna context rather than treating antennas independently.","core_discovery":"The central claim is that the visibility region of each propagation path can be recovered on the fly from the residual behavior of a greedy sparse-recovery algorithm, and that masking the near-field polar-domain steering vectors with that recovered mask yields a better channel estimate than either ignoring non-stationarity or partitioning the array into stationary subarrays. The paper models the channel with a physics-based hybrid model in which each path has a steering vector $b(\\theta_l, r_l)$ depending on angle and distance, multiplied element-wise by a VR mask $s_{m,l}$; masks may be all-ones, binary, or non-binary, with the non-binary case following a Fresnel knife-edge diffraction gain","pith_inferences":["The residual-reduction observation is not tied to OMP; the same HMM-VR masking idea could be bolted onto other greedy or message-passing sparse-recovery schemes, potentially improving them where spatial non-stationarity is present.","Because the emission mapping is an assumed sigmoid rather than derived from channel statistics, the reported gains may be sensitive to the steepness $T=20$ and to the transition probabilities; tuning these from data or learning the mapping could change the margins.","The paper itself notes that explicit non-binary mask estimation is left for future research, so diffraction-shaped power variations are currently approximated by a hard binary mask; estimating quantized gain levels directly is a natural extension.","The flattening of the NMSE curve at high SNR suggests a floor set by VR estimation error; the genie-aided comparison already indicates this ceiling, making closure of that gap an open target."],"forward_implications":["The algorithm improves NMSE over P-SOMP at all tested SNR levels, with roughly 2 dB gain at 5 dB SNR.","At low SNR, full-array VR masking beats subarray P-SOMP by up to 6.5 dB (at -5 dB SNR), because it avoids the reduced per-subarray aperture.","Adding HMM-based VR estimation does not change the asymptotic complexity order, $O(\\hat{L} N S T M)$, relative to P-SOMP.","In denser scattering (larger $L$), the method still outperforms the benchmarks, indicating resilience to reduced sparsity.","The approach works with non-binary (diffraction) VR masks in the channel model even though it estimates binary masks, leaving explicit non-binary mask estimation as future work."],"supporting_citations":[{"why":"Supplies the polar-domain near-field steering-vector basis and the baseline P-SOMP algorithm that the proposed method extends.","marker":"[24]"},{"why":"Supplies the non-binary (diffraction-based) VR mask model used to generate channels and motivates antenna-level VR modeling.","marker":"[25]"},{"why":"Supplies the subarray P-SOMP benchmark whose trade-off the proposed full-array method avoids.","marker":"[27]"},{"why":"Provides the two-stage VR detection baseline that assumes a shared VR mask across paths, which this work generalizes to per-path masks.","marker":"[28]"},{"why":"Supplies the knife-edge diffraction gain formula used to construct non-binary VR masks in the channel model.","marker":"[30]"},{"why":"Supplies the Viterbi decoding algorithm used to infer the VR mask from HMM emission probabilities.","marker":"[31]"},{"why":"Provides an alternating MAP VR/channel estimation baseline that motivates the low-complexity design of the proposed method.","marker":"[19]"}],"fun_headline_variants":["VR-HMM masking sharpens near-field channel estimates","Hidden states reveal antenna visibility for better channel estimation","Adaptive VR masks boost sparse channel recovery at low SNR","Non-stationary near-field channels tamed by HMM-aided OMP"],"cache_read_input_tokens":2816,"weakest_assumption_plain":"The whole VR estimation rests on the assumption that the per-antenna average residual reduction is monotonically related to whether that antenna lies in the path's visibility region, with the relationship fixed as a sigmoid of steepness $T=20$; this mapping is introduced as a design choice and not derived from the channel statistics.","fun_headline_variants_meta":{"raw":{"variants":["VR-HMM masking sharpens near-field channel estimates","Hidden states reveal antenna visibility for better channel estimation","Adaptive VR masks boost sparse channel recovery at low SNR","Non-stationary near-field channels tamed by HMM-aided OMP"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000441,"raw_usage":{"total_tokens":2071,"prompt_tokens":739,"completion_tokens":1332,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":483,"completion_tokens_details":{"reasoning_tokens":1263}},"tokens_in":483,"tokens_out":1332,"duration_ms":11342,"temperature":1.0,"reasoning_tokens":1263,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T00:46:42.225520+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run VR-HMM-P-SOMP on simulated channels where the true VR masks are known but the per-antenna residual reductions are deliberately decorrelated from VR membership, for example by adding two paths with overlapping support or strong noise correlation across antennas. If the Viterbi-decoded masks no longer match the true masks and the NMSE advantage over unmasked P-SOMP disappears, the central claim that residual reduction reveals visibility regions would be falsified. Alternatively, a measurement campaign with a controllable blocker could compare estimated masks against known blockage geometry.","supporting_citations":[{"cited_title":"Channel estimation for extremely large-scale MIMO: Far-field or near-field?","cited_arxiv_id":null,"evidence_quote":"Supplies the polar-domain near-field steering-vector basis and the baseline P-SOMP algorithm that the proposed method extends."},{"cited_title":"Spatial non- stationary near-field channel modeling and validation for massive MIMO systems,","cited_arxiv_id":null,"evidence_quote":"Supplies the non-binary (diffraction-based) VR mask model used to generate channels and motivates antenna-level VR modeling."},{"cited_title":"Non-stationary channel estimation for extremely large-scale MIMO,","cited_arxiv_id":null,"evidence_quote":"Supplies the subarray P-SOMP benchmark whose trade-off the proposed full-array method avoids."},{"cited_title":"Joint visibility region and channel estimation for extremely large-scale MIMO systems,","cited_arxiv_id":null,"evidence_quote":"Provides the two-stage VR detection baseline that assumes a shared VR mask across paths, which this work generalizes to per-path masks."},{"cited_title":"Propagation by diffraction,","cited_arxiv_id":null,"evidence_quote":"Supplies the knife-edge diffraction gain formula used to construct non-binary VR masks in the channel model."},{"cited_title":"The viterbi algorithm,","cited_arxiv_id":null,"evidence_quote":"Supplies the Viterbi decoding algorithm used to infer the VR mask from HMM emission probabilities."},{"cited_title":"Joint visibility region detection and channel estimation for XL-MIMO systems via alternating MAP","cited_arxiv_id":null,"evidence_quote":"Provides an alternating MAP VR/channel estimation baseline that motivates the low-complexity design of the proposed method."}],"review_version":1}