{"id":"ac5ba79e-d459-4f56-8bd2-cee979c93a7b","arxiv_id":"2605.26875","paper_version":2,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":4.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Proposes G-iMUSIC variants that integrate greedy pursuit with MUSIC using one initial EVD and FFT acceleration, claiming better multi-target DoA performance than OMP, OLS, or standard MUSIC in simulations.","lead":"The paper introduces OMP-iMUSIC and OLS-iMUSIC, two algorithms that combine greedy selection with subspace methods for estimating directions of multiple arriving signals. A smart generalist might read it for potential efficiency gains in radar, wireless sensing, or array-based localization systems.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Single initial EVD may fail to preserve accuracy for correlated or proximate targets without subspace updates","rationale":"The reader's weakest assumption directly identifies the same point: whether the unified framework truly allows one EVD without accuracy loss in the regimes the paper claims to improve. The concrete test isolates exactly that condition.","tokens_in":1768,"tokens_out":302,"duration_ms":23122,"concrete_test":"Re-run the Monte Carlo trials of Figure 4 (or equivalent) with target correlation \rho=0.9 and angular separation 2° using both the proposed single-EVD G-iMUSIC and a modified version that recomputes the EVD after each selection; if detection probability drops by >10% or RMSE increases by >15% relative to the recomputed version at SNR=10 dB, the single-EVD assumption does not hold.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim rests on a unified framework that lets OMP-iMUSIC and OLS-iMUSIC reuse the initial noise-subspace projector for all greedy iterations. This requires that the sample covariance EVD computed once remains an adequate basis for orthogonal projections even when the signal covariance is rank-deficient due to correlation or when closely spaced sources produce subspace leakage. If the initial estimate is perturbed (finite snapshots, high correlation), error propagation in the greedy selection step can occur without the corrective effect of recomputed decompositions used in prior iMUSIC variants.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper proposes two G-iMUSIC algorithms (OMP-iMUSIC and OLS-iMUSIC) for multi-target DoA estimation. These are derived from a unified framework linking greedy selection (OMP/OLS) with subspace methods (MUSIC), requiring only a single initial EVD rather than per-iteration decompositions, admitting FFT acceleration for ULAs, and claiming superior detection/precision over standard OMP, OLS, and MUSIC in Monte Carlo trials, plus diagnostic metrics for correlation and angular-proximity regimes.","tokens_in":1872,"tokens_out":460,"duration_ms":26882,"significance":"If the central claims hold, the work provides a computationally attractive bridge between greedy and subspace estimators that avoids repeated EVDs while retaining super-resolution capability. The FFT implementation path and diagnostic metrics for correlation/proximity regimes are concrete strengths that could aid practical deployment in OFDM radar and similar array-processing settings.","major_comments":[{"comment":"§3 (unified framework derivation): the assertion that a single initial noise-subspace projector suffices for all greedy iterations without recomputation must be supported by an explicit error bound or invariance condition when the signal covariance is rank-deficient (high correlation) or when sources are proximate; the skeptic concern about error propagation in the selection step is load-bearing for the complexity-reduction claim.","section":"§3"},{"comment":"§4 (Monte Carlo results): the reported gains in detection and precision over OMP/OLS/MUSIC are central, yet the text must specify the number of trials, exact SNR and snapshot counts, angular-separation grid, and correlation-coefficient sweep so that the single-EVD claim can be verified against the regimes where subspace leakage is expected.","section":"§4"}],"minor_comments":[{"comment":"Notation for the initial EVD and the reused projector should be introduced once with a clear equation reference rather than repeated descriptive phrases.","section":null},{"comment":"The diagnostic metrics are introduced in the abstract but their definitions and interpretation should appear in a dedicated subsection with explicit formulas.","section":null}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive comments, which help clarify the presentation of the unified framework and simulation details. We respond to each major comment below.","responses":[{"response":"Section 3 derives the fixed noise-subspace projector from the initial EVD under the standard array model, with greedy steps updating only the signal component. This yields the claimed single-EVD property. We agree an explicit invariance condition or error bound for rank-deficient covariance and proximate sources is not derived in the current text. We will add a brief discussion of the approximation's validity range and error-propagation considerations in the revised §3.","revision_made":"partial","referee_comment":"[§3] §3 (unified framework derivation): the assertion that a single initial noise-subspace projector suffices for all greedy iterations without recomputation must be supported by an explicit error bound or invariance condition when the signal covariance is rank-deficient (high correlation) or when sources are proximate; the skeptic concern about error propagation in the selection step is load-bearing for the complexity-reduction claim."},{"response":"We will revise the Monte Carlo section to list the exact number of trials, SNR range, snapshot count, angular-separation values, and correlation-coefficient sweep used in the experiments.","revision_made":"yes","referee_comment":"[§4] §4 (Monte Carlo results): the reported gains in detection and precision over OMP/OLS/MUSIC are central, yet the text must specify the number of trials, exact SNR and snapshot counts, angular-separation grid, and correlation-coefficient sweep so that the single-EVD claim can be verified against the regimes where subspace leakage is expected."}],"tokens_in":1414,"tokens_out":370,"duration_ms":26916,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main thing here is that the paper shows how to run iterative MUSIC with only a single initial EVD by tying it to OMP and OLS steps in a unified framework. That avoids the repeated decompositions of earlier iMUSIC versions and opens the door to FFT acceleration on uniform linear arrays. The Monte Carlo results are presented as beating standard OMP, OLS, and MUSIC on detection and precision while cutting runtime, and they add diagnostic metrics that break performance down by correlation level and angular separation.\n\nWhat is actually new is the single-EVD reuse plus the FFT path for the ULA case. The diagnostics are a practical touch that lets readers see where the method holds or slips. For an OFDM radar setup this looks like a useful engineering step that keeps the super-resolution flavor of MUSIC while borrowing the selection logic from greedy algorithms.\n\nThe soft spot is exactly the one the stress-test flags: whether the fixed noise-subspace projector stays accurate enough once targets are correlated or close enough to cause leakage. The abstract says the diagnostics cover those regimes, so if the plots show stable gains there the concern does not land. If the results are mostly for well-separated, low-correlation cases then the single-EVD claim is narrower than it first appears. The paper does not appear to invent new math, just a cleaner combination, and the evidence is simulation-based rather than analytic bounds.\n\nThis is for people who already work on multi-target DoA in radar or communications and need lower complexity without giving up too much accuracy. It has enough of a concrete algorithmic change and some empirical backing to go to a serious referee rather than a desk reject. I would send it for review so the correlation plots and any derivation details get checked.","headline":"G-iMUSIC links greedy and subspace methods to cut iMUSIC down to one EVD with FFT options for ULAs, and the simulations claim gains, but the correlation handling is the part to verify first.","tokens_in":2364,"tokens_out":433,"would_cite":false,"duration_ms":32628,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"G-iMUSIC algorithms combine greedy selection with subspace methods using a single initial eigenvalue decomposition for multi-target DoA estimation.","keywords":["DoA estimation","MUSIC","greedy algorithms","OMP","OLS","array signal processing","eigenvalue decomposition","multi-target estimation"],"falsifier":"Monte Carlo trials on an OFDM radar array in which OMP-iMUSIC or OLS-iMUSIC show no gain in detection probability or root-mean-square error over standard MUSIC when two targets lie within one beamwidth and their signals are fully correlated.","tokens_in":2680,"feed_emoji":"📡","tokens_out":653,"duration_ms":24679,"temperature":0.7,"pith_summary":"The paper develops OMP-iMUSIC and OLS-iMUSIC to estimate directions of arrival for several targets at once. It creates a unified approach that joins the iterative selection of greedy algorithms with the angular resolution of MUSIC while needing only one eigenvalue decomposition at the start. This removes the need to repeat decompositions in each step and supports fast FFT versions on uniform linear arrays. Monte Carlo tests show higher detection rates and better location accuracy than plain OMP, OLS, or MUSIC, particularly when targets are close in angle or their signals are correlated.","feed_headline":"Single EVD powers greedy iterative MUSIC for multi-target DoA","feed_subtitle":"OMP-iMUSIC and OLS-iMUSIC improve detection of correlated or close targets while avoiding repeated decompositions and enabling FFT accelerat","key_machinery":"The unified framework linking subspace and greedy estimations, which permits a single initial eigenvalue decomposition to support the entire iterative selection process without recomputation.","core_discovery":"The proposed G-iMUSIC algorithms, namely OMP-iMUSIC and OLS-iMUSIC, are derived from a unified framework that links subspace and greedy estimations; unlike prior iterative MUSIC methods they require only one initial EVD, avoid eigendecomposition at each iteration, admit FFT-accelerated implementations for ULAs, and yield improved detection and precision over conventional OMP, OLS, and MUSIC in Monte Carlo simulations across signal correlation and angular proximity regimes.","pith_inferences":["The single-EVD property may extend the approach to arrays other than uniform linear ones if an equivalent fast transform exists.","The diagnostic metrics could serve as a general tool to predict when greedy-subspace hybrids outperform pure methods in other estimation tasks.","Reduced per-iteration cost could enable real-time operation in dynamic scenarios where targets move between snapshots."],"forward_implications":["Only one eigenvalue decomposition is performed instead of one per iteration.","FFT acceleration becomes available for uniform linear arrays, lowering computational cost.","Detection and localization improve relative to standalone OMP, OLS, or MUSIC for correlated or proximate targets.","Diagnostic metrics separate performance across correlation and angular-separation regimes."],"fun_headline_variants":["G-iMUSIC needs one EVD for multi-target DoA","Greedy iMUSIC avoids per-iteration eigendecomposition","Single EVD unifies OMP OLS and MUSIC for DoA","G-iMUSIC admits FFT acceleration for ULA arrays"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"A single unified framework can connect subspace and greedy methods so that one initial eigenvalue decomposition remains sufficient and accurate even when targets are angularly close or their signals are strongly correlated.","fun_headline_variants_meta":{"raw":{"variants":["G-iMUSIC needs one EVD for multi-target DoA","Greedy iMUSIC avoids per-iteration eigendecomposition","Single EVD unifies OMP OLS and MUSIC for DoA","G-iMUSIC admits FFT acceleration for ULA arrays"]},"model":"grok-4.3","cost_usd":0.0048,"raw_usage":{"total_tokens":2389,"prompt_tokens":723,"num_sources_used":0,"completion_tokens":69,"cost_in_usd_ticks":47999500,"prompt_tokens_details":{"text_tokens":723,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1597,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":723,"tokens_out":69,"duration_ms":15910,"temperature":1.0,"reasoning_tokens":1597,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-29T15:56:33.944166+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Monte Carlo trials on an OFDM radar array in which OMP-iMUSIC or OLS-iMUSIC show no gain in detection probability or root-mean-square error over standard MUSIC when two targets lie within one beamwidth and their signals are fully correlated.","supporting_citations":[],"review_version":1}