{"id":"07161499-fe97-4c79-acd5-8717bb42894b","arxiv_id":"1908.03187","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"An alternating algorithm jointly optimizing uplink powers and CPU combining weights raises the minimum-user spectral efficiency in cell-free massive MIMO with local-MMSE receivers, compared with fixed-power combining only.","lead":"Cell-free massive MIMO spreads many small antennas across a coverage area to serve many users. This paper designs an algorithm that decides how strongly each user should transmit and how the central processor should combine signals, so that the worst-off user's connection is as good as possible, and shows in simulations that this beats a scheme that only tunes the combining.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The GP step optimizes an upper-bound SINR (19) instead of the exact SINR (11c), with no error control; whether Algorithm 1 actually maximizes the true min-user SINR is untested.","rationale":"The paper's strongest component is the weighting subproblem, which is exact for fixed powers and makes the alternation natural. The power subproblem, however, is not exact: (19) deletes a nonnegative desired-signal term from the denominator, so the GP solves a more optimistic problem. The paper provides no error bound, fixed-point analysis, or monotonicity argument, and no comparison with an existing max-min baseline such as [7]. The convergence plots cover only five channel realizations and six iterations, so they do not substitute for a proof. Additionally, the Appendix's c_k omits the sigma^2 multiplier that is present in (22), making the published GP constraint inconsistent with the approximate SINR it is meant to implement. These issues do not disprove the empirical claim, but they make it conditional: the published algorithm needs a corrected derivation and an exact-SINR evaluation before the headline comparison can be trusted.","tokens_in":9775,"tokens_out":12434,"duration_ms":141401,"concrete_test":"Run a small brute-force check (e.g., L=4 APs, N=2 antennas, K=3 users) over a uniform grid of p in [0,p_max]^K, computing for each p the exact min SINR from (11c) with the optimal weights from (16). Compare Algorithm 1's output with the grid optimum and with the fixed-power baseline (13). If Algorithm 1's exact min SINR falls below the grid optimum by more than 5%, or does not dominate the baseline under exact SINR, the GP approximation in (19) is load-bearing.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Algorithm 1's only non-exact step is the power update. Equation (19) replaces the exact denominator in (11c) by deleting the term p_k(E{|a^H g_{kk}|^2} - |a^H E{g_{kk}}|^2), which is nonnegative; hence the GP maximizes an upper bound on the true SINR. For the powers produced by the GP, the true min SINR — the quantity the paper's curves are intended to show — can be lower than the objective being optimized, and no bound on the gap is provided. The weighting step (16) is exact for fixed powers, so the whole empirical comparison against the fixed-power scheme of [9] rests on this approximation. The approximation is also not implemented cleanly in Appendix A: moving from (22) to (23) defines c_k without the sigma^2 factor carried by the noise term in (22), so the published GP constraint does not match even the approximate SINR (19). Unless the approximation error is quantified or the simulations are rerun with exact SINR, the central claim that the algorithm achieves higher min-user uplink SE is not established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers uplink max-min SINR optimization in a cell-free massive MIMO system with local MMSE combining. The authors decompose the nonconvex joint problem into a weighting-coefficient subproblem, solved exactly for fixed powers as a generalized eigenvalue problem, and a power-allocation subproblem, approximated as geometric programming using the SINR approximation in Eq. (19). An alternating algorithm (Algorithm 1) is proposed, and simulation results compare the minimum uplink SE against the fixed-power weighting scheme of [9]. The central empirical claim is that the proposed algorithm achieves higher min-user uplink SE than the fixed-power benchmark, and that it converges within a few iterations.","tokens_in":9968,"tokens_out":4953,"duration_ms":55820,"significance":"If the central claim is established, the paper offers a practical, low-complexity fairness mechanism for cell-free massive MIMO with local MMSE combining, going beyond the fixed-power weighting optimization of [9]. The weighting subproblem is cleanly derived and solved, and the overall algorithmic structure is plausible. The paper also includes useful numerical comparisons across pilot reuse factors and receiver types. However, the load-bearing power-update step relies on an unquantified approximation of the SINR, and the published GP constraint contains a missing noise-power factor. Because the main comparison is with a benchmark that is not optimized over powers, the validity of the empirical claim depends on closing this gap; the convergence evidence is also too limited to support a general conclusion. With corrections and additional verification, this would be a useful contribution.","major_comments":[{"comment":"The power allocation subproblem is solved using the approximate SINR (19), which is an upper bound on the exact SINR (11c): the approximation drops the nonnegative self-interference term p_k a_k^H (E{g_kk g_kk^H} - E{g_kk}E{g_kk}^H) a_k from the denominator. No error bound or fixed-point analysis is provided to show that increasing the approximate SINR increases the exact minimum SINR. As a result, the GP steps may not actually maximize the true worst-user SINR, and the empirical comparison against the fixed-power benchmark could be optimistic. Please quantify the approximation error for the simulation settings, or modify the algorithm to evaluate exact SINR, and report both the approximate and exact min-user SINR in the convergence and SE curves.","section":"Section III-B, Eq. (19)"},{"comment":"The GP constraint as printed does not match even the approximate SINR (19). The noise contribution in (22) is sigma^2 a_k^H D_k a_k, but after rearrangement c_k is defined without the sigma^2 factor, so (23) uses c_k/p_k instead of sigma^2 c_k/p_k. Since sigma^2 is not equal to 1 in the simulations (noise power is -96 dBm while powers are in mW), this is numerically significant and makes the published formulation inconsistent with the stated approximation. Please correct the definition of c_k or the GP constraint.","section":"Appendix A, Eq. (23)"},{"comment":"The convergence claim is supported only by five specific channel realizations and six iterations, with no statistical aggregation or theoretical monotonicity result. This does not establish that the alternating algorithm converges for typical realizations. Please provide convergence results over a larger sample, with averaged curves and variation ranges, or a proof that the sequence of min-user SE values is nondecreasing.","section":"Section V-C, Figs. 6 and 7"},{"comment":"Proposition 1 shows only that the approximate SINR constraint can be written as a posynomial constraint, not that the GP solution is related to the original max-min problem (15). This distinction should be stated explicitly, and the numerical results should clarify whether the displayed min-user SE is computed from the exact SINR (11c) or from the approximate SINR (19). If the displayed SE uses the approximation, the comparison with fixed-power scheme is not against the true SE.","section":"Proposition 1 and Problem (18)"}],"minor_comments":[{"comment":"The noise vector n_l is stated as CN(0, p_i), which appears to be a typo; it should be CN(0, sigma^2 I_N).","section":"Section II, Eq. (6)"},{"comment":"The definition of D_k as a tuple (E{||v_k1||^2}, ..., E{||v_kL||^2}) in C^{L x L} should be a diagonal matrix, since it is used in the form a_k^H D_k a_k.","section":"Section II, Eq. (11)"},{"comment":"The statement that the overall complexity is O(K^{7/2}) flops per iteration ignores the weighting subproblem complexity O(KL^3) stated in the same section; the per-iteration complexity should be given as a sum of both terms.","section":"Section IV"},{"comment":"The stopping criterion 'until required accuracy' is vague; please define a precise tolerance or a maximum number of iterations.","section":"Algorithm 1"},{"comment":"Several references, e.g., [6] and [12], have incomplete or inconsistent author names and accents; please standardize them.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The core idea is within the scope of the journal and is a reasonable incremental contribution. The main gate is the GP approximation and the missing sigma^2 in Eq. (23); if these are fixed with either an error analysis or a corrected formulation plus exact-SINR simulation results, the paper could be acceptable. I recommend that the editor require the authors to provide reproducible code or clearly specify how exact SINR was evaluated in the figures."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a conventional but genuinely useful optimization paper. The new thing is that it jointly optimizes transmit powers and CPU weighting coefficients for the uplink of a cell-free massive MIMO system with L-MMSE receivers. That combination is absent from the fixed-power weighting optimization of [9] and from earlier max-min work, so the paper does fill a small, real gap in the literature. The weighting-coefficient subproblem is cleanly derived as a generalized eigenvalue problem, and the alternating structure is natural. The complexity analysis is reasonable. The simulations compare against the fixed-power scheme and show consistent gains across pilot-reuse factors, which is the intended empirical contribution. I believe the paper as written does establish that the proposed algorithm improves min-user SE relative to that one benchmark.\n\nThe soft spots are concentrated in the power-allocation step, and they are proportionate to the paper's central claim. Equation (19) approximates the true SINR by deleting a nonnegative term from the denominator, so the GP is maximizing an upper bound on SINR. The paper gives no error bound and no fixed-point analysis connecting the GP solution to the true max-min SINR. The convergence section is also thin: five channel realizations, six iterations, no formal convergence argument. That is enough to suggest a trend, not to prove the method converges. A more serious concern is the Appendix: moving from (22) to (23) defines c_k without the sigma^2 factor that appears in the noise term of (22), so the published GP constraint does not even match the approximate SINR (19). That looks like a typo or an omitted factor, but it matters because a reader cannot reproduce the GP step as written.\n\nI also note the paper does not compare against the closest max-min baseline, [7], which is a missed check: the abstract's claim of \"higher min-user uplink SE\" is only against fixed-power weighting, not against an existing max-min approach. That is a minor omission, not a fatal one.\n\nOn balance, the paper is honest and the core idea is sound enough to warrant engagement. The weighting step is exact and the approximation, while unquantified, is a known and fairly standard move in this subfield. I would send this to referees with a clear request to check the GP derivation and the missing sigma^2 factor, and to add either an error bound or an exact-SINR simulation. The paper is for researchers working on cell-free massive MIMO and practical resource allocation, and they will find it worth reading despite the rough edges. Recommended for peer review, conditional on those fixes.","headline":"A plausible, cleanly-presented alternating algorithm for max-min SINR in cell-free massive MIMO with L-MMSE, but the power-update step optimizes an unverified approximation and the paper's central empirical claim rests on that approximation.","tokens_in":10517,"tokens_out":1194,"would_cite":false,"duration_ms":15227,"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":"An alternating algorithm that jointly optimizes transmit powers and CPU weighting coefficients raises the minimum user SINR in cell-free massive MIMO with local-MMSE combining, beating the fixed-power weighting baseline.","keywords":["cell-free massive MIMO","max-min SINR","local-MMSE combining","geometric programming","generalized eigenvalue problem","uplink power control","user fairness","spectral efficiency"],"falsifier":"Run Algorithm 1 on many correlated Rayleigh fading user drops, then evaluate the true SINR (15) at the returned powers and weights: if the worst-user SINR at the output ever falls below the fixed-power baseline of [9] in a tested regime, or if the approximate SINR (19) deviates substantially from the true SINR at those points, the central claim is refuted.","tokens_in":9555,"feed_emoji":"📶","tokens_out":9852,"duration_ms":90073,"temperature":0.7,"pith_summary":"In a cell-free massive MIMO uplink, the users with the worst channels set the network's effective rate, so fairness is the performance bottleneck. The paper attacks this by maximizing the minimum SINR across users, optimizing both the transmit power of each UE and the weighting coefficients applied at the central processing unit, under per-UE power limits. Because the combined problem is not jointly convex, the authors split it into two subproblems: weighting coefficients come from a generalized eigenvalue problem, and power allocation is approximated as geometric programming. An alternating algorithm cycles between these steps, and numerical experiments show its min-user uplink spectral efficiency exceeds that of the fixed-power weighting scheme used as a baseline. The stated result is empirical, with convergence demonstrated only through simulations.","feed_headline":"Power-weight updates lift the weakest user's rate in cell-free MIMO","feed_subtitle":"Splitting the nonconvex max-min problem into two tractable steps lifts the weakest user's uplink spectral efficiency.","key_machinery":"The load-bearing object is the effective SINR expression (11a)-(11c), written as a generalized Rayleigh quotient in the CPU weighting vector $a_k$; this is what turns weighting design into a generalized eigenvalue problem for the matrix pair $(p_k\\mathbb{E}\\{g_{kk}\\}\\mathbb{E}\\{g_{kk}\\}^H, B_k)$. Since the numerator matrix is rank-one, the optimal weight takes the closed form $a_k = B_k^{-1}\\mathbb{E}\\{g_{kk}\\}$. The power-allocation half relies on the approximation (19), which removes the desired-signal term $-p_k\\mathbb{E}\\{g_{kk}\\}\\mathbb{E}\\{g_{kk}\\}^H$ from the denominator; the resulting constraint is a posynomial inequality, so the power subproblem becomes a geometric program. The alternating execution of these two steps, under the L-MMSE local combiners of (8), is the mechanism that produces the reported min-user SE gains.","core_discovery":"The paper's central claim is that the nonconvex max-min SINR problem for an uplink cell-free massive MIMO system with local-MMSE combining can be handled by alternating two subproblems, and that the resulting algorithm improves the worst-user experience. With transmit powers held fixed, the effective SINR of each user is a generalized Rayleigh quotient in the CPU weighting vector $a_k$, so the optimal weights are obtained as the generalized eigenvector $a_k = B_k^{-1}\\mathbb{E}\\{g_{kk}\\}$, where $B_k = \\sum_i p_i \\mathbb{E}\\{g_{ki}g_{ki}^H\\} - p_k \\mathbb{E}\\{g_{kk}\\}\\mathbb{E}\\{g_{kk}\\}^H + \\sigma^2 D_k$. With the weights held fixed, the power subproblem is converted into a geometric program by using the approximate SINR (19), which drops the term $-p_k \\mathbb{E}\\{g_{kk}\\}\\mathbb{E}\\{g_{kk}\\}^H$ from the denominator. Algorithm 1 iterates these two updates, and the numerical results show that the empirical CDF of the min-user uplink spectral efficiency sits to the right of the fixed-power weighting baseline of [9] across pilot reuse factors and for both L-MMSE and MR combining, with convergence observed within a few iterations.","pith_inferences":["A direct test of the power-step approximation would replace (19) with the exact SINR (11c) and rerun Algorithm 1; comparing the two would isolate how much of the gain comes from the approximation rather than from jointly updating powers and weights.","The method's reliance on the approximate SINR suggests its advantage is most likely to hold in interference-limited regimes; in noise-limited or highly correlated channels, the omitted desired-signal term could matter more.","Because the paper demonstrates convergence only numerically, extending the analysis to a monotonicity or fixed-point argument on the true SINR would turn the empirical convergence into a guarantee.","The same generalized-eigenproblem-plus-GP split may transfer to other linear receivers or to downlink fairness problems, where similar quadratic SINR expressions appear."],"forward_implications":["The proposed alternating update gives higher min-user uplink SE than the fixed-power weighting scheme in the simulated cell-free setups with $L=100$, $K=40$, $N=4$ and $L=64$, $K=16$.","The improvement persists across pilot reuse factors $f=1,2,4$ and for both L-MMSE and MR combining, with orthogonal pilots producing the largest min-user SE.","The algorithm converges to a suboptimal solution within a few iterations (empirically after around two iterations in the tested configurations), so the extra computation over the fixed-power scheme is modest.","Because the power subproblem is a GP with $O(K^{7/2})$ per-iteration flops, the added cost of optimizing powers rather than fixing them scales polynomially in the number of users."],"supporting_citations":[{"why":"Supplies the L-MMSE combining scheme and the fixed-power weighting baseline whose min-user spectral efficiency is the comparison target.","marker":"[9]"},{"why":"Provides the generalized Rayleigh quotient formulation and Lemma B.10 used to derive the optimal weighting vector $a_k = B_k^{-1}\\mathbb{E}\\{g_{kk}\\}$.","marker":"[12]"},{"why":"Casts the weighting-coefficient subproblem as a generalized eigenvalue problem and supports the complexity estimate for matrix inversion.","marker":"[10]"},{"why":"Gives the geometric programming standard form and interior-point solution method that the power-allocation subproblem relies on.","marker":"[15]"},{"why":"Establishes the uplink max-min SINR formulation for cell-free massive MIMO that this paper extends to joint power and weight optimization.","marker":"[7]"}],"fun_headline_variants":["Alternating updates lift weakest user in cell-free MIMO","Two-step max-min SINR for L-MMSE cell-free massive MIMO","Power-weight alternation boosts worst-case uplink rate","Optimized fairness: alternating power and weights in cell-free MIMO"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The power-allocation step assumes that the approximated SINR (19), which drops the desired-signal energy term from the denominator, is accurate enough that maximizing it also raises the true minimum SINR (15); the paper provides no bound on the gap between the approximate and true SINR.","fun_headline_variants_meta":{"raw":{"variants":["Alternating updates lift weakest user in cell-free MIMO","Two-step max-min SINR for L-MMSE cell-free massive MIMO","Power-weight alternation boosts worst-case uplink rate","Optimized fairness: alternating power and weights in cell-free MIMO"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000248,"raw_usage":{"total_tokens":1587,"prompt_tokens":1025,"completion_tokens":562,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":641,"completion_tokens_details":{"reasoning_tokens":490}},"tokens_in":641,"tokens_out":562,"duration_ms":6203,"temperature":1.0,"reasoning_tokens":490,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:20:58.906620+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run Algorithm 1 on many correlated Rayleigh fading user drops, then evaluate the true SINR (15) at the returned powers and weights: if the worst-user SINR at the output ever falls below the fixed-power baseline of [9] in a tested regime, or if the approximate SINR (19) deviates substantially from the true SINR at those points, the central claim is refuted.","supporting_citations":[{"cited_title":"Making cell-free massive MIMO competitive with MMSE processing and centralized implementation,","cited_arxiv_id":null,"evidence_quote":"Supplies the L-MMSE combining scheme and the fixed-power weighting baseline whose min-user spectral efficiency is the comparison target."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Casts the weighting-coefficient subproblem as a generalized eigenvalue problem and supports the complexity estimate for matrix inversion."},{"cited_title":"A tutorial on geometric programming,","cited_arxiv_id":null,"evidence_quote":"Gives the geometric programming standard form and interior-point solution method that the power-allocation subproblem relies on."},{"cited_title":"Enhanced max-min sinr for uplink cell-free massive mimo systems,","cited_arxiv_id":null,"evidence_quote":"Establishes the uplink max-min SINR formulation for cell-free massive MIMO that this paper extends to joint power and weight optimization."}],"review_version":1}