{"id":"71ce2430-e32b-4ea5-a69a-fc76cecc49e1","arxiv_id":"2509.07754","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"An enhanced coherent successive target cancellation (ECSTC) algorithm enables OFDM-ISAC with 64-QAM to reach sensing MSE comparable to constant-modulus signals in multi-target and scattering scenarios.","lead":"This paper shows that an enhanced successive-cancellation algorithm lets an OFDM system use high-order QAM for communications while keeping radar-like sensing accuracy close to what constant-modulus signals achieve. The method re-estimates each target after subtracting the estimated interference from all other targets, and is tested in multi-target ray-tracing scenes with diffuse scattering.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Simulation assumes known nearest-to-farthest target order and perfect detection; the 'always comparable' ECSTC claim is not yet validated for automatic detection/ordering.","rationale":"The paper's mathematical derivation (Sec. III) is clean: Eq. (19) correctly shows that modulation-induced interference power scales with the alphabet kurtosis and target amplitude, providing a solid theoretical basis for why constant-modulus is sensing-optimal and why cancellation is needed. The proposed ECSTC is a sensible enhancement: Eq. (22) re-estimates each target after removing all others, which should reduce the bias from cumulative interference. The simulation results are qualitatively consistent with the central claim in the idealized setting. However, the load-bearing weakness is that the simulation bypasses the detection and ordering problem entirely. The explicit statement 'The targets are processed from nearest to farthest' (Sec. VI-A) means the algorithm is given the true target order, and because all cubes have equal RCS, nearest-to-farthest is also strongest-to-weakest—an oracle ordering a real system would need to infer from the noisy RDM. If automatic detection misses a weak target, its interference is never subtracted in Eq. (22), so 64-QAM retains extra interference that QPSK (kappa=1) does not have, breaking the claimed parity. The paper even acknowledges scattering leaves residual reflections that CSTC cannot model, yet still claims ECSTC 'always' matches QPSK; that 'always' is unsupported without a detection-theoretic evaluation. The secondary issue—the CRB is computed with an average SNR per target rather than per-target SNR—weakens the 'close to CRB' statements but is not the main threat to the constant-modulus parity claim. My proposed test directly targets the gap by replacing the oracle ordering with an automatic detector and measuring whether the parity holds; if it does, the concern is resolved, and if not, the claim must be narrowed. This does not change the reader's conditional verdict; it reinforces it.","tokens_in":7780,"tokens_out":7853,"duration_ms":93797,"concrete_test":"Rerun the Sec. VI-B scattering scenario at SNR_Y = 20 dB, but replace the nearest-to-farthest ordering with an automatic detector: apply CA-CFAR (e.g., P_fa = 1e-3) to the first-pass range-Doppler matrix, order the detections by descending peak magnitude, and feed those detections into ECSTC. For each of K = 1000 trials, associate detected peaks to the 16 ground-truth targets (missed targets count as large error) and compute per-target MSE for distance and velocity. If the 64-QAM ECSTC MSE for any target exceeds the QPSK scattering MSE by more than 3 dB, the universal parity claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim in Sec. VI-B that ECSTC 'always leads to a performance comparable with the case of constant-modulus alphabets' rests on simulations where 'the targets are processed from nearest to farthest' (Sec. VI-A). This assumes the true target order is known or that an oracle provides correct detections and associations before the interference cancellation. In a real ISAC receiver, the first pass would run a detector (e.g., CA-CFAR) on the RDM and order peaks, likely by magnitude. If a weak target is missed in the first pass, its interference matrix A_l' is absent from Eq. (22), so its modulation-dependent interference—proportional to (kappa-1)|a_l|^2 per Eq. (19)—is never removed. Even if all targets are detected, ordering by peak magnitude rather than true distance can change which first-pass estimates are contaminated by residual interference, and the second-pass reconstruction of A_l' from biased first-pass parameters may not fully cancel. The paper provides no detection-theoretic analysis, no missed-detection/false-alarm study, and no sensitivity to processing order. Thus the 'always' claim goes beyond what the simulation actually demonstrates: it shows ECSTC works under oracle-ordered, fully detected targets, not necessarily under realistic automatic detection.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes the sensing performance of OFDM-based ISAC when the communication symbols are drawn from an arbitrary, possibly non-constant-modulus constellation. The key theoretical contribution is a derivation showing that non-constant-modulus alphabets introduce an additive interference term whose per-target power is |a_l|^2(kappa-1), where kappa is the fourth-order moment of the constellation. The paper then evaluates an existing coherent successive target cancellation (CSTC) scheme and proposes an enhanced version (ECSTC) that performs a second parameter-estimation pass with all first-pass target interference removed. Simulations use a ray-tracing scene with 16 cubes, with and without diffuse scattering, and compare QPSK and 64-QAM operation. The central claim is that ECSTC makes a communication-centric OFDM system with 64-QAM achieve sensing accuracy comparable to that of constant-modulus QPSK.","tokens_in":8112,"tokens_out":6042,"duration_ms":73025,"significance":"If the central claim holds, the result is practically important: it would decouple the sensing accuracy from the constellation alphabet in OFDM-ISAC, allowing higher-order modulation without sensing loss. The theoretical derivation in Sec. III is clean and self-contained, and the ECSTC algorithm is simple and computationally cheap (a second addition of already-computed matrices). The use of Sionna ray tracing with diffuse scattering is a strength relative to idealized point-target papers. However, the load-bearing empirical claims are supported only under a favorable oracle setup: known target processing order and perfect detection. The paper would be significantly strengthened by a detection-theoretic evaluation or at least a robustness study. As it stands, the general claim 'always leads to a performance comparable with the case of constant-modulus alphabets' goes beyond the simulation evidence.","major_comments":[{"comment":"The central 'always comparable' claim in Sec. VI-B is supported only by simulations in which 'the targets are processed from nearest to farthest' (Sec. VI-A) and in which every target is detected before cancellation. Eq. (22) subtracts A_l' for all l' != l, which presupposes that the first pass found all targets and produced usable estimates. No CA-CFAR or automatic detection/association/ordering is simulated, and no missed-detection, false-alarm, or processing-order sensitivity analysis is provided. If a real system must detect and order targets from the noisy RDM, a missed target's modulation-dependent term (kappa-1)|a_l|^2 from Eq. (19) remains uncancelled. The paper should either replace the 'always' wording with a claim restricted to the simulated oracle scenario or add experiments with automatic detection and randomized/estimated target ordering.","section":"Sec. VI-A, Sec. VI-B, Eq. (22)"},{"comment":"The MSE results are averaged over K=1000 independent scene realizations, but no error bars, confidence intervals, or variance of the MSE estimate are reported. Many conclusions rely on the absence of a visible gap between the QPSK and 64-QAM ECSTC curves, and between the ECSTC curves and the CRB. Without interval estimates it is not possible to assess whether these differences are statistically significant. Please report standard errors or bootstrap intervals on the MSE values, at least for the key comparisons in Fig. 4.","section":"Sec. V-B, Figs. 3 and 4"},{"comment":"The CRB used in Figs. 3 and 4 is the AWGN-only, single-point-target bound of Eqs. (26)-(27). The comparison is made, however, under a 16-target channel and, in the scattering cases, under diffuse multi-ray reflections. For scattering, each cube generates a set of sub-rays, so the effective parameter space is different from that of a single specular point target; the displayed CRB may not be a valid lower bound for those cases. This does not invalidate the proposed algorithm, but it weakens the 'close to CRB' argument. Please clarify whether the CRB line is intended only for the specular cases, or provide a bound that accounts for the actual scattering model and multi-target interference.","section":"Sec. VI-B, Eqs. (26)-(27)"},{"comment":"The scattering parameters P_scatt = 10^-3 and scattering coefficient 0.9 are fixed without sensitivity analysis, and the conclusions in Sec. VI-B are stated as unconditional ('always'). Diffuse scattering strength is a free parameter that can strongly affect the residual interference after cancellation, since ECSTC reconstructs a single reflection per target. A sensitivity study over scattering coefficient/probability, or a clear boundary on the validity regime, is needed before the 'always' claim can be accepted.","section":"Sec. V-C, Sec. VI-B"}],"minor_comments":[{"comment":"Typo: 'We note hat the inter-carrier interference' should read 'We note that the inter-carrier interference'.","section":"Sec. II-B"},{"comment":"The normalization and dimension of the IFFT/FFT operations in Eq. (20) are not defined. Please specify whether these are the same N- and M-point transforms used in Eq. (10), and what normalization convention is applied.","section":"Sec. IV-A, Eq. (20)"},{"comment":"The legend entries for the scattering curves are difficult to parse (e.g., 'QPSK scattering 64-QAM MF with scattering 64-QAM ECSTC with scattering'). Please use a clear table or distinct line styles with a formal legend.","section":"Fig. 4"},{"comment":"Minor typo: 'The remainder of the back scattered still causes noticeable interference' should read 'The remainder of the backscattered signal still causes noticeable interference'.","section":"Sec. VI-B"},{"comment":"The statement that the D-term 'behaves as additional Gaussian interference in the RDM' is attributed to reference [5], which is a self-citation. Since this property is used to interpret the residual after cancellation, a short numerical verification or a direct derivation would make the paper more self-contained.","section":"Sec. III, after Eq. (19)"}],"recommendation":"major_revision","confidential_remarks":"The paper is a reasonable incremental contribution: the kurtosis-based derivation is useful and the ECSTC idea is simple and likely effective. However, the manuscript overstates the strength of its empirical support. The core issue is that the simulations assume oracle-ordered, fully detected targets, while the conclusions are phrased unconditionally. I also note that the Gaussian-noise property of the modulation-dependent interference is delegated to a self-citation; an independent check would increase confidence. With additional robustness experiments and a more carefully scoped claim, the paper could be suitable for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Clear, competent within-field work. The kurtosis-based derivation of modulation-dependent interference is correct and compact, and the proposed second-iteration cancellation (ECSTC) is a simple, sensible enhancement that shows clear gains in simulation. The ray-tracing evaluation with diffuse scattering is a plus. The paper deserves a serious referee.\n\nWhat is actually new: the second iteration of CSTC, plus the demonstration that with it, 64-QAM matches QPSK sensing accuracy in their ray-traced scenes. The Sec. III derivation is sound: non-constant modulus adds an interference term per target with power |a_l|^2(kappa-1). That is a nice, usable result.\n\nSoft spots: the 'always' claim (Sec. VI-B) outruns the evidence. Targets are processed from nearest to farthest, which presumes the true order is known or that the first pass yields correct detections and ordering. No detection-theoretic analysis, no missed-detection/false-alarm study, no sensitivity to processing order. If a weak target is missed in the first pass, its interference is never cancelled in Eq. (22). That is a real limitation, though not a fatal flaw: the benefit is demonstrated under a favorable but not crazy assumption.\n\nAlso, no error bars on the MSE plots (K=1000 is decent but variance unreported), the CRB reference uses an average per-target SNR rather than the actual per-target SNR, and no code or data is released. The Sionna setup is described well enough to roughly reproduce but not exactly. The self-citation to [5] for the Gaussian-interference property is reasonable; I do not see a circularity problem.\n\nBottom line: useful for ISAC waveform people. It gives a clean explanation and a cheap fix. It does not change the physical picture but removes a practical barrier. Send it to review; ask the authors to test sensitivity to detection/ordering and soften the 'always' wording. I would cite the kurtosis result.","headline":"Clean derivation and a useful cancellation trick; the 'always' claim needs a detection-theoretic caveat.","tokens_in":8554,"tokens_out":3299,"would_cite":true,"duration_ms":32237,"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":"The paper argues that two-pass successive target cancellation lets OFDM-ISAC use arbitrary, non-constant-modulus alphabets like 64-QAM with sensing accuracy matching constant-modulus QPSK.","keywords":["OFDM-ISAC","integrated sensing and communications","interference mitigation","successive target cancellation","constant modulus","quadrature amplitude modulation","kurtosis","ray tracing"],"falsifier":"Run the same ray-tracing scene but order the targets by a CA-CFAR detector's detections from the noisy RDM instead of the true nearest-to-farthest sequence; if the ECSTC MSE at high SNR rises above the QPSK matched-filter baseline, the central claim fails.","tokens_in":7713,"feed_emoji":"📡","tokens_out":5628,"duration_ms":56791,"temperature":0.7,"pith_summary":"The paper tackles the sensing half of integrated sensing and communications (ISAC): when an OFDM communication waveform carries data, any constellation that varies in amplitude (like 64-QAM) creates a modulation-dependent interference in the radar range-Doppler map, with power scaling with the transmitted symbol kurtosis and the reflected target power. The authors derive this interference term explicitly, then show that coherent successive target cancellation (CSTC) removes much of it. Their enhancement, ECSTC, adds a second pass that subtracts the synthesized interference of every other target before re-estimating each target, bringing estimation error close to the Cramér-Rao bound in specular multi-target scenes and, in scattering scenes, to the level of constant-modulus QPSK. The practical point is that a 6G base station need not sacrifice high-order, high-data-rate constellations to keep its radar function accurate.","feed_headline":"One extra cancellation pass makes 64-QAM sense as sharply as QPSK","feed_subtitle":"An enhanced cancellation loop lets OFDM-ISAC use high-rate 64-QAM without losing radar accuracy.","key_machinery":"The central mechanism is the matched-filter range-Doppler map combined with per-target interference reconstruction. The paper shows that for a non-unit-modulus alphabet, |X|^2 = 1 + D, so after matched filtering the channel estimate carries an extra term D times the superposition of all target reflections; this term is the modulation-dependent interference. Each detected target is used to synthesize an interference matrix A_l by re-applying the model, and ECSTC subtracts all matrices except the target of interest before the final parameter estimation. The kurtosis of the alphabet appears as the single scalar controlling interference power.","core_discovery":"The paper claims that the sensing loss caused by arbitrary (non-constant-modulus) modulation alphabets in OFDM-ISAC can be almost entirely undone by an enhanced successive-cancellation procedure. For a matched-filtered OFDM radar, the transmitted symbol power fluctuation |X|^2-1 multiplies the channel and acts as additive interference with power |a_l|^2 (kappa-1), where kappa is the fourth moment of the alphabet. CSTC detects and subtracts the strongest targets one by one; ECSTC re-estimates each target after removing all other targets' reconstructed interference matrices. In ray-tracing simulations with 16 extended targets, ECSTC pushes 64-QAM range and velocity MSE down to the QPSK level a","pith_inferences":["If a real receiver must discover the target set and ordering from the noisy RDM (e.g., via CFAR) rather than using the known nearest-to-farthest order, the first-pass estimates feeding ECSTC would contain detection and ordering errors, so the reported MSE gains could shrink. This is our inference, not stated in the paper.","The same cancellation principle might transfer to MIMO-ISAC, where the interference matrices would be per transmit-receive path but the kurtosis-derived interference term remains.","The scattering simulation uses a single configurable scattering probability; testing ECSTC on measured or varied scattering surfaces would show how robust the cancellation is to model mismatch."],"forward_implications":["64-QAM (or any high-order QAM) can be used for ISAC sensing without the usual accuracy penalty, as long as the receiver performs ECSTC.","The extra computation is modest: one sum of already-computed interference matrices per target, with no new matched-filter passes.","In low-SNR conditions the enhancement is unnecessary; in high-SNR conditions it closes the dominant interference gap.","The derived kurtosis scaling gives a design target: constellations with lower fourth moment generate less sensing interference, so shaping schemes can be co-optimized."],"supporting_citations":[{"why":"Supplies the anticipated 6G OFDM system parameters (carrier frequency, subcarrier spacing, bandwidth, frame size) used in the simulations.","marker":"[3]"},{"why":"Demonstrates that the matched filter is unbiased and that the non-constant-modulus term behaves as additive Gaussian interference in the range-Doppler map.","marker":"[5]"},{"why":"Proposes the original coherent successive target cancellation algorithm and provides the Cramér-Rao bounds for distance and velocity estimates.","marker":"[8]"},{"why":"Provides the ray-tracing engine used to generate the multi-target scenes with specular and diffuse scattering.","marker":"[10]"}],"fun_headline_variants":["Effective cancellation lets 64-QAM match QPSK sensing precision","ECSTC: 64-QAM sensing matches QPSK in OFDM-ISAC","One extra cancellation pass: 64-QAM radar accuracy equals QPSK","Enhanced successive cancellation erases modulation penalty in OFDM-ISAC","Arbitrary modulations meet sensing exactness via enhanced cancellation"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The simulations process the known 16 targets from nearest to farthest, so the true target order and count are not inferred from the noisy range-Doppler map; a deployed system that must detect and order targets first would feed less accurate first-pass estimates into the cancellation loop.","fun_headline_variants_meta":{"raw":{"variants":["Effective cancellation lets 64-QAM match QPSK sensing precision","ECSTC: 64-QAM sensing matches QPSK in OFDM-ISAC","One extra cancellation pass: 64-QAM radar accuracy equals QPSK","Enhanced successive cancellation erases modulation penalty in OFDM-ISAC","Arbitrary modulations meet sensing exactness via enhanced cancellation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000772,"raw_usage":{"total_tokens":3208,"prompt_tokens":651,"completion_tokens":2557,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":395,"completion_tokens_details":{"reasoning_tokens":2469}},"tokens_in":395,"tokens_out":2557,"duration_ms":20445,"temperature":1.0,"reasoning_tokens":2469,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T21:41:27.999371+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same ray-tracing scene but order the targets by a CA-CFAR detector's detections from the noisy RDM instead of the true nearest-to-farthest sequence; if the ECSTC MSE at high SNR rises above the QPSK matched-filter baseline, the central claim fails.","supporting_citations":[{"cited_title":"Mandelli, M","cited_arxiv_id":null,"evidence_quote":"Supplies the anticipated 6G OFDM system parameters (carrier frequency, subcarrier spacing, bandwidth, frame size) used in the simulations."},{"cited_title":"Geiger, F","cited_arxiv_id":null,"evidence_quote":"Demonstrates that the matched filter is unbiased and that the non-constant-modulus term behaves as additive Gaussian interference in the range-Doppler map."},{"cited_title":"Braun, C","cited_arxiv_id":null,"evidence_quote":"Proposes the original coherent successive target cancellation algorithm and provides the Cramér-Rao bounds for distance and velocity estimates."}],"review_version":1}