{"id":"9c642267-e003-419c-addb-f63c540b633e","arxiv_id":"2507.09960","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A greedy, mutual-information-based RF chain selection framework for MIMO integrated sensing and communications achieves near-optimal performance with much lower complexity than exhaustive search.","lead":"This paper proposes three low-cost algorithms for choosing which radio frequency (RF) chains to keep active in MIMO systems that both communicate and sense. The methods preserve most of the mutual information for both tasks and get close to the best possible choice at much lower computational cost than exhaustive search.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central near-optimality claim is conditioned on an unvalidated proxy: sensing MI in Eq. (15) is never checked against MSE/CRLB ranking of chain subsets.","rationale":"I agree with the reader that the weakest point is the sensing-MI proxy, not the greedy algebra. The decompositions in Theorems 1-2 are correct and the DBS approximation is reasonable under the stated asymptotics; the mechanical issues (Algorithm 3's 'lowest' ranking and the dropped T in Eq. (52)) are fixable typos and do not affect the core claim. The claim that GES/GCS are near-optimal has two layers: exactness of each removal step for (17), and fidelity of (17) to sensing performance. The first is proven; the second is assumed. Since the introduction motivates the method precisely by the gap between MI and MSE/CRLB, the paper should demonstrate that the selected subsets are also near-optimal under one of those conventional metrics, or explicitly limit the claim to the MI-based formulation. This does not change the reader's CONDITIONAL verdict but sharpens the condition.","tokens_in":22396,"tokens_out":13345,"duration_ms":167048,"concrete_test":"Reuse the Fig. 3 setup (L=8, N_t=16, M=16, N_c=N_s=8, K=8, ω_c=ω_s=0.5, SNR=30 dB) and, for the same random channel/target realizations, compute for each candidate subset the sensing CRLB for angle estimation (or beam-pattern MSE). Compare GES/GCS-selected subsets against the subset that exhaustively optimizes that alternative metric. If the MI-selected subset's CRLB/MSE exceeds the alternative-optimal value by more than 10% in a substantial fraction of realizations, the Eq. (15) proxy is not faithful and the near-optimal claim must be restricted to the MI objective; if the gap is negligible, the concern does not land.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section II derives sensing MI as Eq. (15) from the wide-separation/independent-columns TRM model of [33] and the sample-covariance approximation (1/T)XX^H ≈ P I_K. Theorems 1-2 and the greedy algorithms are exact for the weighted objective (17), so the internal algebra is sound. What is load-bearing for the practical ISAC claim is the unstated assumption that the ordering over RF-chain subsets induced by (15) matches the ordering induced by sensing metrics that matter in practice, e.g., beam-pattern MSE and CRLB, which the introduction itself names as alternatives. Figures 3-6 validate only MI objectives; no experiment compares the chains selected by GES/GCS with the chains that would be optimal under MSE/CRLB, and no analysis shows MI is a surrogate for those metrics in the selection problem. If the proxy ranks chain subsets differently, a chain set that is near-optimal for (17) can be suboptimal for detection/estimation, and the paper's claim of practical effectiveness for MIMO ISAC outruns its evidence. This is an external-validity concern, not an internal contradiction; it should be stated as a condition on the main claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes greedy RF chain selection algorithms for MIMO ISAC systems under a unified MI-based objective. The authors formulate the transmit selection problem as maximizing the weighted sum of normalized communication MI and sensing MI (Eq. (17)), and show that the per-chain contribution can be isolated via the determinant lemma and Schur complement (Theorems 1 and 2). Based on these decompositions, they develop GES and GCS, two backward-elimination greedy algorithms with sequential rank-one updates (Corollaries 1 and 2), and extend the framework to beam selection in beamspace hybrid MIMO, including a simplified diagonal beam selection (DBS) under asymptotic conditions. Numerical results compare the proposed methods with exhaustive search, random selection, fixed selection, and full selection in terms of weighted MI, energy efficiency, and MI Pareto frontiers.","tokens_in":22506,"tokens_out":15413,"duration_ms":150545,"significance":"If the results hold, the paper gives a computationally efficient heuristic for a combinatorial RF chain selection problem, with exact decomposition identities that are likely useful beyond the specific setting. The determinant-lemma and Schur-complement derivations are clean, and the sequential updates avoid repeated matrix inversions, which is a genuine practical contribution. The paper is also careful to formulate the objective explicitly and does not fit parameters to the simulation outputs. The main caveat is that the practical relevance of the selection hinges on sensing MI being a faithful surrogate for the beam-pattern MSE or CRLB metrics that the introduction itself identifies; this is not tested, and the manuscript's claims should be scoped accordingly.","major_comments":[{"comment":"The central claim that GES and GCS achieve 'near-optimal performance' and demonstrate 'practical effectiveness for MIMO ISAC systems' is validated only against the weighted MI objective (17). The sensing MI in (15) is a surrogate metric, and the introduction explicitly contrasts it with beam-pattern MSE and CRLB as alternative sensing metrics. None of the experiments in Figs. 3-6 evaluates the selected RF-chain subsets under beam-pattern MSE or CRLB, so a subset that is near-optimal for (17) could be suboptimal for the detection/estimation task that those metrics describe. Please either add a simulation comparing the proposed selections against MSE/CRLB benchmarks, or restrict the conclusions to the MI-based objective.","section":"Section V, Eqs. (15) and (17)"},{"comment":"Line 5 of Algorithm 3 instructs the algorithm to 'Select K RF chains with the lowest contribution', but the surrounding derivation shows that DBS should keep the chains with the highest contribution: the objective after removing chain j is proportional to log((d~j)^omega_c * prod_n (e~n,j)^(omega_s/N_s)), so retaining the largest such terms maximizes (17). As printed, Algorithm 3 would select the worst chains and would not reproduce the DBS results in Figs. 3-5. Please correct the pseudocode to 'highest contribution' and ensure the released code, if any, matches.","section":"Algorithm 3"}],"minor_comments":[{"comment":"In Algorithm 1 line 9 and Algorithm 2 line 8, the condition 'if N^(i+1)_t = K' compares a set with an integer; the intended condition is 'if |N^(i+1)_t| = K'.","section":"Algorithms 1 and 2"},{"comment":"The simulation setup does not specify the number of time slots T, although the sensing MI in (15) contains the factor gamma*T. Please state the value of T used in Figs. 3-6.","section":"Section V"},{"comment":"The typesetting of Eq. (26) makes the exponent and product structure ambiguous; please rewrite with explicit parentheses and define the 'contribution' term consistently with Eq. (27).","section":"Eq. (26)"},{"comment":"The phrase 'without loss of generality' used to justify omitting Exh from subsequent simulations is imprecise; the omission is a computational choice, not a mathematical reduction. Adding Exh to Fig. 6, or at least noting its absence, would better support the Pareto-frontier claim.","section":"Section V, Fig. 6"},{"comment":"The asymptotic argument that array response vectors 'align closely with exactly one analog beamforming vector' is informal; please state it as an explicit assumption or approximation and give the regime in which it is expected to hold.","section":"Section IV"},{"comment":"The paper claims significantly lower complexity than exhaustive search but gives no quantitative complexity comparison; a brief table or big-O analysis for GES, GCS, DBS, and exhaustive search would make the contribution easier to assess.","section":"Complexity discussion"}],"recommendation":"major_revision","confidential_remarks":"I see no issues of novelty disclosure; the related work is appropriately cited. The main revision burden is the external-validity experiment for the sensing metric and the DBS pseudocode correction. The paper fits the scope of the journal, and the core algebraic derivations are sound."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a competent engineering paper that does what it says. The greedy backward-elimination framework for RF chain selection in MIMO ISAC, with the two MI decompositions (GES and GCS) and the asymptotic DBS variant, is a combination I have not seen in the cited prior work. The determinant-lemma and Schur-complement algebra in Theorems 1-2 and Corollaries 1-2 checks out, and the sequential updates genuinely avoid repeated inversions. The simulations show GES/GCS tracking exhaustive search closely over SNR and active-chain count, with clear gains over random and fixed baselines, plus sensible energy-efficiency and Pareto results. There is a real contribution here for the ISAC hardware-selection subfield.\n\nThe soft spots are mostly mechanical and one conceptual scope condition. Mechanical: Algorithm 3 selects the K lowest contributions, which would be wrong; it should select the highest (or equivalently remove the lowest). Eq. (52) drops the T in gamma*T. Algorithm 1's line 9 compares a set to a scalar. Appendix A has size sloppiness around the eigen-decomposition of R_T,n. These are typos, not load-bearing.\n\nThe conceptual caveat matters more. The near-optimality claim is only for the weighted MI objective in (17), and every experiment evaluates that same objective. The paper never checks whether the selected chain subset is also near-optimal for the sensing metrics it names in the introduction, beam-pattern MSE or CRLB. This is not circularity—the greedy algorithms are exact for the stated objective—but it is an external-validity condition. If the MI ordering over chain subsets diverges from the MSE/CRLB ordering, the 'practical effectiveness' claim outruns the evidence. A short experiment comparing the selected chains under MSE/CRLB, or at least an explicit statement that the claim is about MI, would fix it.\n\nAlso missing: code, error bars, and full simulation parameters for reproducibility. Given the small scale, that is a reasonable request for revision, not a reason to reject.\n\nVerdict: send it to peer review. It is a solid, useful subfield contribution with correct math and a clear limitation. A referee should ask for the sign fix, the T fix, the proxy-validation discussion, and reproducibility artifacts. I would not cite it for the sensing-MSE claim, but I would cite it for the MI-based greedy selection framework.","headline":"Solid, correct greedy MI-decomposition framework for RF chain selection in MIMO ISAC; the near-optimality claims are only validated against the same MI objective, so the practical sensing claim needs a caveat or one more experiment.","tokens_in":23135,"tokens_out":6390,"would_cite":true,"duration_ms":65901,"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":"Greedy MI-based chain selection reaches near-optimal MIMO ISAC performance while cutting complexity.","keywords":["MIMO integrated sensing and communication","RF chain selection","mutual information","greedy backward elimination","beam selection","beamspace MIMO","energy efficiency","antenna selection"],"falsifier":"Run exhaustive enumeration on a small instance (for example $N_t=8$, $K=4$) over one fixed channel and target realization, and compare the weighted-sum MI of the GES and GCS subsets with the exhaustive optimum; a single realization where the gap exceeds a few percent would falsify the near-optimality claim. Separately, compare the beam-pattern MSE or CRLB of the MI-selected chains against chains selected directly on those sensing metrics: if the MI-selected set is clearly worse, the unified-MI proxy is the wrong objective.","tokens_in":22100,"feed_emoji":"📡","tokens_out":7965,"duration_ms":83185,"temperature":0.7,"pith_summary":"The paper tackles the problem of choosing which radio-frequency (RF) chains to keep active in a MIMO system that simultaneously communicates with a user and senses targets. Because these two functions are traditionally measured with different metrics—mutual information for communication versus beam-pattern MSE or Cramér-Rao lower bound for sensing—the authors adopt sensing mutual information as a unified objective and maximize its weighted sum with communication MI. They show that the total MI can be decomposed into per-chain contributions, and on that basis they propose two greedy backward-elimination algorithms, GES (eigenvalue-based) and GCS (cofactor-based), which repeatedly remove the chain whose loss is smallest. Their simulations report that these methods match exhaustive search almost exactly while costing far less computation, and that the same machinery extends to beam selection in beamspace hybrid arrays, where a simplified diagonal variant (DBS) performs nearly as well. If the sensing-MI proxy is faithful, this gives a practical way to cut hardware cost and power in MIMO ISAC without sacrificing much performance.","feed_headline":"Greedy RF-chain selection matches near-optimal MIMO ISAC performance","feed_subtitle":"By removing the least-contributing RF chain at each step, GES and GCS match exhaustive search at a fraction of the cost.","key_machinery":"The load-bearing object is the per-RF-chain MI contribution. For a candidate chain $j$, the contribution to communication MI is $T\\log_2\\big((1-\\gamma\\alpha_j)^{-1}\\big)$ with $\\alpha_j=\\mathbf{h}_j^{\\mathsf{H}}\\mathbf{A}\\mathbf{h}_j$, and the contribution to sensing MI is $\\sum_n \\log_2\\big((1-\\gamma T\\beta_{n,j})^{-1}\\big)$ with rank-reduced $\\mathbf{G}_n$ factors; GCS expresses the same losses as $\\log_2(\\delta_j^{-1})$ and $\\log_2(\\varepsilon_{n,j}^{-1})$ using diagonal entries of inverse matrices. These identities turn subset selection into repeated scalar comparisons, and Corollaries 1 and 2 give inversion-free updates of the needed matrices via the Woodbury identity and Schur complement. The same decomposition, with the analog beamforming matrix folded into the channel and target-response matrices, carries the method from antenna-level chain selection to beamspace beam selection; under asymptotic orthogonality it collapses to the one-pass diagonal beam selection.","core_discovery":"On the paper's own account, the central discovery is that the weighted-sum RF-chain selection problem (17), though NP-hard in general, can be solved near-optimally by sequential elimination: at each step, the algorithm removes the RF chain whose deletion causes the smallest drop in the weighted sum of normalized communication and sensing MI. The mechanism is an exact decomposition—Theorems 1 and 2—of each MI term into a baseline term and a scalar per-chain loss, so that ranking chains reduces to comparing scalars rather than recomputing determinants. GES computes the loss through eigenvalue decomposition of the sensing covariance and the Woodbury-based updates of Corollary 1; GCS computes it from diagonal entries of inverse matrices updated through the Schur complement in Corollary 2. In the beamspace hybrid-array regime, the same theorems reduce to a diagonal approximation, DBS, which selects chains in one pass. The paper's simulations show GES and GCS tracking exhaustive search across SNR, active-chain count, and Pareto weighting, with DBS close behind.","pith_inferences":["The backward-elimination formulation implicitly favors removing redundant chains: GCS's cofactor criterion is the determinant of the matrix with that chain's row and column deleted, which measures orthogonality of the remaining vectors, so the method may generalize to other submodular subset-selection objectives with similar curvature.","If sensing MI is later found to diverge from beam-pattern MSE or CRLB in some regime, the same greedy framework could be rerun with those metrics substituted wherever the chain contribution has a similar determinant form; the theorems would still supply the ranking.","The near-diagonality that DBS exploits suggests a cheap diagnostic: when the off-diagonal entries of $\\tilde{\\mathbf{D}}$ and $\\tilde{\\mathbf{E}}_n$ are small relative to the diagonal, one-pass selection is safe, so a system could adaptively switch between GCS and DBS based on measured channel statistics.","Because the algorithms need only matrices built from channels and target covariances, they could be applied online as the channel changes, with the low-rank updates in Corollaries 1 and 2 tracking the evolving best subset."],"forward_implications":["GES and GCS can replace exhaustive search in MIMO ISAC chain selection, giving essentially the same weighted-sum MI at substantially lower complexity, so larger antenna arrays become tractable.","Selecting a small carefully chosen subset of RF chains improves energy efficiency relative to full selection; the paper reports up to 7.79 percent EE gain in the tested setup.","Because the MI decomposition is agnostic to the communication-versus-sensing weight, the same greedy loop traces out a wide Pareto frontier between communication MI and sensing MI, so one algorithm serves any operating point on the trade-off.","The extension to beamspace MIMO means hybrid-architecture systems can use GES and GCS, and DBS offers a one-pass alternative that stays within a few percent of the full greedy methods even when its asymptotic assumptions are not met."],"supporting_citations":[{"why":"Supplies the wide-separation target-response model with independent columns that the sensing MI derivation in (14)-(16) is built on.","marker":"[33]"},{"why":"Provides the sensing mutual-information expression and its link to MSE and CRLB that makes the unified MI objective credible.","marker":"[37]"},{"why":"Supports the large-$T$ sample-covariance approximation (13) and the MI-based characterization of Gaussian ISAC channels.","marker":"[11]"},{"why":"Establishes the beamspace MIMO beam-selection formulation with low RF complexity that the paper's hybrid-array extension follows.","marker":"[14]"},{"why":"Introduces mutual information as an information-theoretic sensing performance metric, the foundation for treating sensing MI as a unified objective.","marker":"[30]"},{"why":"Motivates the unified MI-based performance framework for ISAC that the weighted-sum objective (17) adopts.","marker":"[38]"},{"why":"Supplies the hybrid analog-digital beamforming and DFT-beam model underlying the beam-selection extension.","marker":"[42]"},{"why":"Provides the asymptotic orthogonality of array response vectors that makes $\\tilde{\\mathbf{H}}_c$ and $\\tilde{\\mathbf{R}}_{T,n}$ nearly diagonal, grounding DBS.","marker":"[43]"}],"fun_headline_variants":["Greedy RF chain selection for MIMO ISAC hits near-optimal","Greedy selection nearly matches exhaustive RF search in MIMO ISAC","RF chain selection: greedy near-optimal at low complexity","Greedy RF selection achieves near-optimal for MIMO ISAC"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that sensing mutual information, computed with the wide-separation independent-columns target model and the sample-covariance approximation of (13), faithfully captures what the chosen RF chains deliver for sensing; if that MI proxy diverges from beam-pattern MSE or CRLB in the operating regime, chains picked to maximize weighted MI may not preserve actual sensing capability even though the theorems remain exact for the stated objective.","fun_headline_variants_meta":{"raw":{"variants":["Greedy RF chain selection for MIMO ISAC hits near-optimal","Greedy selection nearly matches exhaustive RF search in MIMO ISAC","RF chain selection: greedy near-optimal at low complexity","Greedy RF selection achieves near-optimal for MIMO ISAC"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000393,"raw_usage":{"total_tokens":2080,"prompt_tokens":980,"completion_tokens":1100,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":596,"completion_tokens_details":{"reasoning_tokens":1023}},"tokens_in":596,"tokens_out":1100,"duration_ms":8866,"temperature":1.0,"reasoning_tokens":1023,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T17:44:52.268169+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run exhaustive enumeration on a small instance (for example $N_t=8$, $K=4$) over one fixed channel and target realization, and compare the weighted-sum MI of the GES and GCS subsets with the exhaustive optimum; a single realization where the gap exceeds a few percent would falsify the near-optimality claim. Separately, compare the beam-pattern MSE or CRLB of the MI-selected chains against chains selected directly on those sensing metrics: if the MI-selected set is clearly worse, the unified-MI proxy is the wrong objective.","supporting_citations":[{"cited_title":"Spectrally constrained MIMO radar waveform design based on mutual information,","cited_arxiv_id":null,"evidence_quote":"Supplies the wide-separation target-response model with independent columns that the sensing MI derivation in (14)-(16) is built on."},{"cited_title":"Unified ISAC Pareto boundary based on mutual information and minimum mean- square error estimation,","cited_arxiv_id":null,"evidence_quote":"Provides the sensing mutual-information expression and its link to MSE and CRLB that makes the unified MI objective credible."},{"cited_title":"On the fundamental tradeoff of integrated sensing and communications under gaussian channels,","cited_arxiv_id":null,"evidence_quote":"Supports the large-$T$ sample-covariance approximation (13) and the MI-based characterization of Gaussian ISAC channels."},{"cited_title":"Low RF-complexity millimeter-wave beamspace-MIMO systems by beam selection,","cited_arxiv_id":null,"evidence_quote":"Establishes the beamspace MIMO beam-selection formulation with low RF complexity that the paper's hybrid-array extension follows."},{"cited_title":"Information theory and radar waveform design,","cited_arxiv_id":null,"evidence_quote":"Introduces mutual information as an information-theoretic sensing performance metric, the foundation for treating sensing MI as a unified objective."},{"cited_title":"Integrated sensing and communications: A mutual information-based framework,","cited_arxiv_id":null,"evidence_quote":"Motivates the unified MI-based performance framework for ISAC that the weighted-sum objective (17) adopts."}],"review_version":1}