{"id":"cd163989-bf5e-45af-8495-7ba3ae324fd8","arxiv_id":"1909.01034","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Local partial zero-forcing and its protective variant give closed-form spectral-efficiency expressions that outperform maximum-ratio and full zero-forcing and match regularized zero-forcing in cell-free massive MIMO.","lead":"This paper proposes two distributed precoding schemes, local partial zero-forcing and protective partial zero-forcing, for cell-free massive MIMO, and derives closed-form formulas for the spectral efficiency they achieve. The formulas allow tuning the trade-off between interference cancelation and signal boosting, and they enable power control that makes the schemes competitive with regularized zero-forcing.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Closed-form PZF/PPZF derivations appear correct; the load-bearing weakness is the RZF benchmark comparison, where PPZF-optimized power coefficients are substituted heuristically for RZF, so the abstract's 'optimal ... suitable for RZF' claim is not established as stated.","rationale":"The reader's weakest_assumption points to the Rayleigh/reciprocity modeling assumptions, which are real scope limitations but are also stated explicitly in the paper and are standard for this line of work; they do not threaten the internal correctness of the derivations. The more actionable concern is the RZF power-control overreach, which the reader also identified as issue (2). The threshold sensitivity noted as issue (1) is a second, milder concern: the v=95% grouping rule is tuned to the simulated topology, and Fig. 6 shows performance depends on it, but the paper acknowledges this and presents the schemes as offering an adaptable trade-off rather than a universal optimal threshold. Because the mathematical core survives scrutiny and the weaknesses are about scope and overstatement rather than validity, the appropriate verdict remains the reader's conditional one: the closed-form SE results are credible, but the abstract's strongest RZF-related claim should be reworded and ideally backed by an RZF-specific search or at least by a clear statement that the power coefficients are heuristic when applied to RZF.","tokens_in":27305,"tokens_out":17904,"duration_ms":189227,"concrete_test":"In the Fig. 4 setup (L=100, M=8, tau_P=7, K=10, D=500 m), perform a Monte-Carlo-based search over the RZF regularization scalars and over feasible long-term power coefficients (e.g., grid or coordinate ascent on the max-min objective), computing RZF's achievable SE by simulation. If a searched RZF configuration improves the 95%-likely SE by more than about 5% over the paper's RZF with PPZF-optimized coefficients, the 'comparable to RZF' claim needs qualification; if the PPZF-optimized coefficients are within a small margin of the searched optimum, the claim stands as stated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"No flaw found in the core derivations: the Wishart-based normalizations in (19), (24), and the SINR expressions (29) and (33) are internally consistent, and the Monte-Carlo overlap reported in Fig. 2 supports the algebra. The load-bearing concern is the abstract's comparative claim that these closed-form expressions can be used to devise optimal long-term power control strategies 'also suitable for RZF, whose closed-form expression for the SE is not available.' What the paper actually does in Section IV-A and Fig. 4 is to solve the max-min problem (39) with ps = PPZF, then use those PPZF-optimized coefficients for power allocation in the RZF evaluation (15) and (34), with HCD coefficients used only inside the regularization term of (34). There is no optimization over RZF's own achievable SE, because no closed form exists. Therefore the word 'optimal' is unsupported when applied to RZF; it is at best a heuristic transfer of PPZF power control to RZF. Since RZF is the headline benchmark, this makes the 'comparable to RZF' comparison potentially tilted: an RZF variant with differently chosen regularization or with powers optimized by search could narrow, match, or reverse the reported gap. The PZF/PPZF spectral-efficiency expressions themselves are not invalidated, but the strongest advertised performance claim goes beyond what the paper demonstrates.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes two fully distributed downlink precoding schemes for cell-free massive MIMO, termed local partial zero-forcing (PZF) and local protective partial zero-forcing (PPZF). The schemes split each AP's users into strong and weak sets; strong users are zero-forced on the subspace spanned by their pilots, while weak users are served by MRT (PZF) or by MRT projected into the null space of the strong-user subspace (PPZF). Under independent Rayleigh fading, MMSE channel estimation, and pilot contamination, the authors derive closed-form achievable spectral-efficiency expressions (Eqs. (29) and (33)) using Wishart-based normalizations, and validate them against Monte Carlo simulations. They then formulate a max-min fairness power-control problem as a second-order cone program and compare the resulting performance against MRT, full-pilot zero-forcing (FZF), and regularized zero-forcing (RZF) in simulations. The paper claims that PZF and PPZF substantially outperform MRT and FZF while being comparable to RZF, and that the closed-form expressions enable optimal power control that is also suitable for RZF.","tokens_in":27588,"tokens_out":3515,"duration_ms":39092,"significance":"If the derivations are correct, the paper contributes a versatile, fronthaul-free precoding framework for cell-free massive MIMO with multi-antenna APs, together with tractable closed-form SE expressions that support power control. The Monte Carlo overlap in Figs. 2 and 3 gives strong evidence that the algebraic derivations are correct, and the paper provides explicit proofs in the appendices, which is a strength. The max-min power-control reformulation is a useful extension of existing cell-free massive MIMO results. The main advertised advantage over RZF—namely, the availability of closed-form SE expressions—is genuine and could be practically valuable. The comparative claim relative to RZF is, however, overstated in its current form, as explained in the major comments.","major_comments":[{"comment":"The claim that the closed-form expressions can be used to devise 'optimal' power control strategies 'also suitable for RZF' is not established. The optimization problem (39) is solved with ps = PPZF, and the resulting coefficients {rho^{MMF}_{l,k}|PPZF} are then plugged into the RZF evaluation (15) and (34); RZF's own achievable SE is never optimized, because no closed form exists. The notation RZF({rho^{HCD}}, {rho^{MMF}}^{PPZF}) in Fig. 4 makes this explicit. Consequently, the word 'optimal' is unsupported when applied to RZF; the RZF comparison is at best a heuristic transfer of PPZF-optimized powers. Since RZF is the headline benchmark, the authors should either rephrase the claim to state that the PPZF-optimal power coefficients can be heuristically applied to RZF, or provide evidence (e.g., a search over RZF regularization parameters or power coefficients) that the PPZF-optimal coefficients are near-optimal for RZF.","section":"Abstract and Section IV-A, Fig. 4"},{"comment":"The sentence 'PPZF performs as well as RZF (benchmark), suggesting that (33) might be a reliable closed-form expression to estimate the performance of RZF' is a speculation that goes beyond the demonstrated results. The simulation in Fig. 3 uses HCD power control and a particular simulation setup; the match between PPZF and RZF is not proven analytically, and RZF's performance depends on the regularization parameter inside (34). This claim should be softened or explicitly stated as an observation for the simulated scenario, not as a general reliability statement.","section":"Section VI-B, third paragraph"},{"comment":"The closed-form derivation for PZF is long, and the proof in Appendix B relies on several independence and zero-mean arguments (e.g., in (51)-(56)). I checked the key steps: the use of the Wishart normalization in (24), the decomposition into coherent and non-coherent interference, and the combination via (58) appear internally consistent. The Monte Carlo validation in Fig. 2 provides strong numerical support. I do not find a specific algebraic error, but given the complexity of the expression, it would help the reader if the authors added a short statement in the text or proof indicating which of the terms in (50) correspond to which physical interference components, and how the final compact form (29) groups them.","section":"Eq. (29) and Appendix B"}],"minor_comments":[{"comment":"Remark 1 states that co-pilot channel estimates are linearly dependent; the scaling in (5) is correct but the notation (using both k and t) may be slightly confusing. Suggest clarifying that the relation holds for any pair of UEs sharing the same pilot index.","section":"Section II-A, Eq. (5)"},{"comment":"The definition of lambda (the precoding vector) uses subscripts l,i_k; later the notation switches between w_{l,i_k} and w_{l,k}. The paper would be easier to read if a single consistent subscript notation were used for pilot-index-based precoders.","section":"Section III-D, Eq. (23)"},{"comment":"The figure shows median SE curves, but the text refers to 'SE' without specifying that it is the median; adding 'median' to the axis label or caption would improve clarity.","section":"Section VI-B, Fig. 5"},{"comment":"The complexity expressions are useful, but the entries for PZF and PPZF count 'multiplications and divisions' without a row label saying which operations are counted; the caption should state explicitly that only complex multiplications and divisions are counted, as in the text.","section":"Table II"},{"comment":"There are minor typos and formatting issues, such as 'INTERDONA TOet al.' in the header and occasional inconsistent use of 'front-hauling' versus 'fronthaul'. A careful proofread is recommended.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Start with the punchline: the core derivations are solid, and the paper gives cell-free Massive MIMO a genuinely useful distributed precoding option. The closed-form SINR expressions for PZF and PPZF under MMSE estimation and pilot contamination are the real contribution; the Monte Carlo overlap in Figs. 2 and 3 confirms the algebra. Earlier partial zero-forcing work assumed perfect CSI, and that is what this paper fixes.\n\nThey do a few things well. The PZF/PPZF construction is simple: each AP uses a subset of spatial directions to cancel interference toward strong users, keeps array gain for the rest, and PPZF protects strong users by making the MRT component live in the orthogonal complement. The SE expressions are used to set up a max-min power control problem as a SOCP, which is clean and tractable. The complexity table is honest. The paper is also transparent about limitations: it states the M > tau_S condition, the independent Rayleigh assumption, and the reciprocity assumption in plain terms.\n\nNow the soft spots, in proportion. The load-bearing weakness is the abstract's claim that the closed-form expressions yield optimal power control 'also suitable for RZF.' What the paper actually does is solve max-min for PPZF, then use those PPZF-optimized coefficients in the RZF power allocation, with HCD coefficients only inside the regularizer. That is a heuristic transfer, not an optimization over RZF's own SE. The word 'optimal' is not supported for RZF. This should be reworded, and it mildly weakens the 'comparable to RZF' headline, though it does not touch the validity of the PZF/PPZF derivations. Second, the grouping threshold v=95% is tuned on the same simulation scenario. Fig. 6(a) shows some robustness, so this is a caveat, not a fatal flaw. The assumptions are standard for this literature; no invented entities or hidden circular steps. Citation pattern is appropriate, including self-citations to earlier FZF work.\n\nWho should read this: anyone working on cell-free Massive MIMO or distributed precoding. It deserves a serious peer review, with one requested change: temper the abstract and clarify that RZF uses PPZF-optimized powers heuristically. I would cite this paper if I were working on distributed downlink precoding.","headline":"The closed-form SE derivations are solid and Monte-Carlo validated; the abstract's 'optimal power control also suitable for RZF' claim exceeds what the paper actually demonstrates.","tokens_in":28128,"tokens_out":2676,"would_cite":true,"duration_ms":25163,"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 shows that local partial zero-forcing and its protective variant give cell-free Massive MIMO downlink rates comparable to regularized zero-forcing, with closed-form SINRs and no CSI exchange.","keywords":["cell-free massive MIMO","local partial zero-forcing","protective partial zero-forcing","distributed precoding","pilot contamination","spectral efficiency","max-min fairness power control","regularized zero-forcing"],"falsifier":"A direct test is to simulate or build a small cell-free setup with $M=4$ antennas per AP, $\\tau_P=3$ pilots, and two strong users sharing one pilot, then compare the time-averaged SINR to the prediction of (29) and (33). A consistent match under the paper's Rayleigh model would confirm the derivations, while a systematic mismatch under spatially correlated antennas (for example, an exponential correlation coefficient of 0.5) would show which independence assumption is doing the work.","tokens_in":27082,"feed_emoji":"📡","tokens_out":11908,"duration_ms":112748,"temperature":0.7,"pith_summary":"The paper aims to show that cell-free Massive MIMO downlink can run entirely on local channel estimates with precoders that are comparable to the regularized zero-forcing benchmark without any instantaneous CSI exchange. Its two schemes, local partial zero-forcing (PZF) and local protective partial zero-forcing (PPZF), let each access point split its users into strong and weak groups, cancel interference only toward the strong group, and steer the weak-group signals either with maximum-ratio transmission or with that transmission projected away from the strong directions. The payoff is twofold: the effective SINRs become closed-form expressions under independent Rayleigh fading, channel estimation error, and pilot contamination, and these expressions make max-min fairness power control a convex problem, which was previously not available for regularized zero-forcing. The paper thereby establishes that the trade-off between interference cancelation and array gain is tunable, and that APs with few antennas can still participate in interference suppression.","feed_headline":"Partial zero-forcing matches RZF rates in cell-free MIMO","feed_subtitle":"Closed-form SINRs let each access point cancel only its strongest users, no CSI sharing needed.","key_machinery":"The load-bearing object is the per-AP pilot-domain channel estimate $\\bar{H}_l = Y_l \\Phi$, a full-rank $M \\times \\tau_P$ matrix whose columns are the channel estimates associated with the orthogonal pilots. For PZF, the precoder to a strong user is built from the reduced matrix $\\bar{H}_l E_{S_l}$, which keeps only the $\\tau_{S_l}$ pilots used by strong users; a standard central-Wishart expectation turns the required normalization into $1/((M - \\tau_{S_l})\\theta_{l,k})$ and the effective SINR into the closed form in (29). For PPZF, the additional load-bearing object is the projection matrix $B_l = I_M - \\bar{H}_l E_{S_l}(E_{S_l}^H \\bar{H}_l^H \\bar{H}_l E_{S_l})^{-1} E_{S_l}^H \\bar{H}_l^H$ onto the orthogonal complement of the strong-user subspace, so that weak-user transmissions are invisible to strong users except for estimation error; this gives the second closed-form SINR, (33).","core_discovery":"On the paper's own terms, the central claim is that partial zero-forcing is not merely a compromise: each AP can choose the number $\\tau_{S_l}$ of pilots whose users it will zero-force, and the cost in array gain is exactly $M - \\tau_{S_l}$, the same cost full-pilot zero-forcing would pay for all $\\tau_P$ pilots. Because the strong-user matrix $E_{S_l}$ selects only a subset of the columns of the full-rank estimate matrix $\\bar{H}_l = Y_l \\Phi$, the PZF precoder in (23) has a normalization given by the Wishart expectation in (24), which yields the closed-form SINR in (29). The PPZF precoder in (32) projects the weak-user maximum-ratio transmission onto the orthogonal complement of the strong-user subspace, and its SINR is given in (33). The paper argues that these formulas are achievable lower bounds on the ergodic downlink capacity, that they agree with Monte Carlo simulations, and that PZF and PPZF substantially outperform maximum-ratio transmission and full-pilot zero-forcing while matching regularized zero-forcing.","pith_inferences":["If the closed-form SINRs are as accurate as the paper's Monte Carlo checks suggest, the strong-user threshold could be optimized per AP or per coherence interval, rather than fixed at the 95% used in the simulations, giving an extra tunable degree of freedom.","The same projection-based logic transfers to the uplink: an AP could combine partial zero-forcing toward strong users with projected maximum-ratio combining toward weak users, likely producing uplink SINR expressions of the same algebraic form.","Under spatial correlation or imperfect reciprocity, the independence-based Wishart normalization would likely understate the effective interference; estimating an effective number of independent antennas could yield a robust variant of (29) and (33)."],"forward_implications":["Each AP can serve weak users with maximum-ratio transmission while zero-forcing only a chosen subset, so the antenna requirement drops from $M > \\tau_P$ to $M > \\tau_{S_l}$, which is what makes implementation with very few antennas feasible.","Max-min fairness power control becomes a convex second-order cone program solvable by bisection for PZF and PPZF, and the PPZF solution can also be used to allocate power to regularized zero-forcing.","Only one precoding vector per pilot is needed rather than one per user, and the computational complexity is lower than full-pilot zero-forcing because $\\tau_{S_l} \\le \\tau_P$.","All precoding is fully distributed: no instantaneous CSI exchange between APs and the central unit is required, and only long-term statistical coefficients are needed for optimal power control."],"supporting_citations":[{"why":"It defines the cell-free Massive MIMO system model, TDD pilot training, and the max-min power control problem that the paper extends to PZF and PPZF.","marker":"[12]"},{"why":"It supplies the earlier full-pilot zero-forcing evaluation that PZF generalizes and the source of the $M > \\tau_P$ constraint.","marker":"[20]"},{"why":"It introduces full-pilot zero-forcing for multi-cell Massive MIMO, the structure PZF inherits for its strong-user subset.","marker":"[21]"},{"why":"It provides the hardening-bound argument used to define the achievable spectral efficiency and the complexity accounting method.","marker":"[27]"},{"why":"It supplies the central Wishart expectation lemma that turns the precoder normalizations and SINR denominators into closed form.","marker":"[30]"},{"why":"It gives the maximum-ratio transmission SINR expression and the descending large-scale-fading grouping rule adapted for the strong and weak user sets.","marker":"[29]"}],"fun_headline_variants":["Partial ZF matches RZF rates in cell-free MIMO, closed-form SINR","Local PZF: zero-force only strong users, match RZF","Partial ZF precoding: closed-form SINR, RZF-level performance","Cell-free MIMO: PZF beats MRT/ZF, matches RZF"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole closed-form analysis stands on the assumption that the channels from an AP to a user are independent Rayleigh fading with known large-scale coefficients and that uplink and downlink are perfectly reciprocal, because only then are the estimated channel columns independent complex-Gaussian with a Wishart normalization; if antennas are spatially correlated or calibration is imperfect, equations (29) and (33) are not guaranteed to hold.","fun_headline_variants_meta":{"raw":{"variants":["Partial ZF matches RZF rates in cell-free MIMO, closed-form SINR","Local PZF: zero-force only strong users, match RZF","Partial ZF precoding: closed-form SINR, RZF-level performance","Cell-free MIMO: PZF beats MRT/ZF, matches RZF"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000642,"raw_usage":{"total_tokens":2982,"prompt_tokens":1003,"completion_tokens":1979,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":619,"completion_tokens_details":{"reasoning_tokens":1895}},"tokens_in":619,"tokens_out":1979,"duration_ms":13932,"temperature":1.0,"reasoning_tokens":1895,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:29:00.055192+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct test is to simulate or build a small cell-free setup with $M=4$ antennas per AP, $\\tau_P=3$ pilots, and two strong users sharing one pilot, then compare the time-averaged SINR to the prediction of (29) and (33). A consistent match under the paper's Rayleigh model would confirm the derivations, while a systematic mismatch under spatially correlated antennas (for example, an exponential correlation coefficient of 0.5) would show which independence assumption is doing the work.","supporting_citations":[{"cited_title":"Downlink spectral efficiency of cell-free massive MIMO with full-pil ot zero- forcing,","cited_arxiv_id":null,"evidence_quote":"It supplies the earlier full-pilot zero-forcing evaluation that PZF generalizes and the source of the $M > \\tau_P$ constraint."},{"cited_title":"Massive MIM O for maximal spectral efficiency: How many users and pilots shoul d be allocated?","cited_arxiv_id":null,"evidence_quote":"It introduces full-pilot zero-forcing for multi-cell Massive MIMO, the structure PZF inherits for its strong-user subset."},{"cited_title":"Massive M IMO networks: Spectral, energy, and hardware efficiency,","cited_arxiv_id":null,"evidence_quote":"It provides the hardening-bound argument used to define the achievable spectral efficiency and the complexity accounting method."},{"cited_title":"Random matrix theory and wireless commu- nications,","cited_arxiv_id":null,"evidence_quote":"It supplies the central Wishart expectation lemma that turns the precoder normalizations and SINR denominators into closed form."},{"cited_title":"On the total energy efficiency of cell-free massive MIMO,","cited_arxiv_id":null,"evidence_quote":"It gives the maximum-ratio transmission SINR expression and the descending large-scale-fading grouping rule adapted for the strong and weak user sets."}],"review_version":1}