{"id":"c1a58b9a-bbce-470e-973d-ab4d61437c4c","arxiv_id":"2501.00546","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A closed-form downlink spectral efficiency expression and a max-min fairness optimization are derived for STAR-RIS-aided cell-free massive MIMO under hardware impairments, phase noise, and RIS phase errors.","lead":"Many cooperating base stations serve users with the help of a smart surface that both reflects and transmits signals, and this paper derives a closed-form formula for the resulting download speed when hardware is imperfect. It also proposes an alternating optimization scheme that raises the worst user's speed and verifies the formula by simulation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Algorithm 3's power-control update uses the previous STAR-RIS coefficients, so the monotone-convergence proof in Section V-D does not apply to the algorithm as written; the AO convergence claim needs a concrete fix or verification.","rationale":"The reader's verdict was CONDITIONAL, citing the lack of a formal convergence guarantee for the AO algorithm and the heuristic, projection-free APSO update. My stress-test agrees with that overall direction but identifies a more precise and directly checkable internal mismatch: the pseudocode in Algorithm 3 updates the power coefficients using the previous iteration's passive beamforming, not the one just produced in the same iteration. The monotonicity argument in Section V-D requires the newly updated Φ to be passed into the power-control subproblem; as written, the algorithm is a Jacobi-style update and the chain of inequalities does not apply. This is a concrete, fixable flaw rather than a fundamental mathematical error, so it supports the CONDITIONAL verdict rather than REJECT. I also note the secondary feasibility gap in Algorithm 1: the unprojected particle update can leave the feasible STAR-RIS amplitude/phase box, so even the empirical improvement shown in Figs. 9-11 is not guaranteed to correspond to feasible parameters. The closed-form SE analysis in Proposition 1 was checked for internal consistency: the pilot-estimation covariance, the phase-noise factors in the desired-signal and beamforming-uncertainty terms, and the pilot-contamination interference terms all appear algebraically coherent under the stated Gaussian-product channel model. The model assumptions (Rayleigh AP-RIS, blocked/non-LoS direct links) are explicit and standard for this literature, so I do not treat them as a correctness defect. The main remaining risk is therefore the optimization contribution, and the concrete test above can settle whether the printed algorithm is merely misdescribed or actually non-monotone/infeasible.","tokens_in":27332,"tokens_out":32648,"duration_ms":313609,"concrete_test":"Correct Algorithm 3 step 5 to use the newly updated Φ^[tA] (Gauss-Seidel) and rerun the experiments underlying Figs. 9-11; if the worst-user SINR becomes monotone nondecreasing across AO iterations and every returned β_T,β_R lies in [0,1] with β_T+β_R=1, the printed pseudocode was simply missing the update. Also instrument Algorithm 1 to record the min/max of β_T, β_R and θ after each particle update, and run the literal Algorithm 3 as written: if F decreases in any AO iteration or any returned particle violates β_T,β_R ∈ [0,1], θ ∈ [0,2π), then the convergence proof and feasibility claim are false for the algorithm as stated.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central analytical result (Proposition 1) appears internally consistent under the stated Rayleigh AP-RIS / Rician RIS-UE model, so the main weakness is in the optimization half of the claim. In Algorithm 3 (Section V-D), iteration tA first updates the STAR-RIS passive beamforming to Φ^[tA] using the previous power coefficients z^[tA-1] (step 4), but then updates the power coefficients using the previous passive beamforming: step 5 reads 'Update z^[tA] by solving (19) with Algorithm 2 and Φ^[tA-1]_T and Φ^[tA-1]_R.' The monotonicity chain F(η^[tA],Φ^[tA]) ≤ F(η^[tA],Φ^[tA+1]) ≤ F(η^[tA+1],Φ^[tA+1]) in Section V-D is valid only for a Gauss-Seidel update in which the power subproblem is solved with the just-updated Φ^[tA]. As printed, the algorithm performs a Jacobi-style update: each subproblem is optimized against stale variables, so F(η^[tA],Φ^[tA]) need not be nondecreasing and the stated convergence guarantee does not follow. In addition, Algorithm 1 updates particles as x^(t+1) = x^(t) + v^(t+1) without any projection or clamping to β_T,β_R ∈ [0,1] with β_T+β_R=1 and θ ∈ [0,2π); the returned 'best' particle can therefore be infeasible, and the claimed feasible max-min solution is not established. These issues do not invalidate the closed-form SE derivation, but they undermine the second central claim that the AO algorithm provably improves the worst-user SINR.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper investigates a STAR-RIS-aided cell-free massive MIMO system under spatially correlated fading, transceiver hardware impairments, time-varying phase noise, RIS phase errors, and pilot contamination. It derives a linear MMSE cascaded channel estimator (Section III) and a closed-form downlink ergodic spectral efficiency expression under MR precoding (Proposition 1, Eqs. (13)-(16)), with validation by Monte Carlo simulations in Section VI. It then formulates a max-min SINR problem for joint STAR-RIS passive beamforming and power control, and proposes an alternating optimization (AO) algorithm combining adaptive particle swarm optimization (APSO) and a bisection method. Numerical results demonstrate SE improvements over RIS-aided and conventional CF-mMIMO baselines and show convergence of the AO algorithm.","tokens_in":27669,"tokens_out":3980,"duration_ms":36645,"significance":"The closed-form SE derivation is a solid contribution: it handles finite numbers of APs, UEs, and RIS elements, incorporates realistic impairments in a structured way, and is validated by Monte Carlo simulation without curve fitting. The optimization part is more fragile: the convergence guarantee stated in Section V-D does not match Algorithm 3 as printed, and the APSO particle update in Algorithm 1 does not enforce feasibility. These are fixable within the manuscript's scope, so the paper merits revision rather than rejection. If the optimization claims are repaired (or downgraded to heuristic improvement with numerical evidence), the paper would be a useful addition to the STAR-RIS/cell-free MIMO literature.","major_comments":[{"comment":"The monotone-convergence proof is inconsistent with Algorithm 3 as written. Step 5 updates the power coefficients z^{[tA]} using Algorithm 2 with the previous passive beamforming Φ^{[tA-1]}_T and Φ^{[tA-1]}_R, whereas the proof's chain F(η^{[tA]},Φ^{[tA]}) ≤ F(η^{[tA]},Φ^{[tA+1]}) ≤ F(η^{[tA+1]},Φ^{[tA+1]}) requires the power subproblem to be solved with the just-updated Φ^{[tA]}. As printed, each subproblem is optimized against stale variables, so the objective value is not guaranteed nondecreasing and the stated convergence guarantee does not follow. Please either change step 5 to use Φ^{[tA]}_T and Φ^{[[tA]}_R, or provide a different convergence argument (or empirical verification) for the Jacobi-style update actually implemented.","section":"Section V-D and Algorithm 3"},{"comment":"The APSO particle update x^{[tP+1]}_{lP} = x^{[tP]}_{lP} + v^{[tP+1]}_{lP} is applied without any projection or clamping, even though the decision variables must satisfy β_{T,n} ∈ [0,1] (with β_{R,n}=1-β_{T,n}) and θ_{T,n},θ_{R,n} ∈ [0,2π). Consequently, particles can leave the feasible set during the search, and the 'global best' returned in step 15 may be infeasible. This undermines the claim that the AO algorithm provides a feasible max-min solution. Please add an explicit projection or repair step (e.g., clipping β to [0,1] and wrapping θ modulo 2π) after each velocity update, and state this in the algorithm.","section":"Algorithm 1 (Section V-B)"},{"comment":"The convergence proof depends on the statement that 'the updates of Φ and η can attain the max-min SINR when the other variable remains fixed.' This is not justified for the Φ-subproblem: APSO is a stochastic heuristic with no per-iteration improvement guarantee, and the cited reference [37] analyzes standard PSO dynamics, not monotone ascent on a given objective. Even if the stale-variable issue in Algorithm 3 is fixed, the inequality (a) requires at least that the APSO update does not decrease the worst-user SINR; without a repair strategy or a monotone variant, this should be stated as a heuristic claim supported by the numerical convergence plots rather than as a formal guarantee.","section":"Section V-D and Algorithm 1"}],"minor_comments":[{"comment":"There is a typo in 'opreate' (should be 'operate').","section":"Section II-C"},{"comment":"'Inter Core i9-14900HX' should be 'Intel Core i9-14900HX'.","section":"Section VI-C"},{"comment":"The row 'Number of UEs in the reflection space, KT 3' should read 'transmission space' rather than 'reflection space' for KT.","section":"Table II"},{"comment":"The sentence 'the global optimal solution of is nontrivial to attain' is missing a word; it should read 'of (18)' or similar.","section":"Section V-B, after Eq. (18)"},{"comment":"Minor wording: 'Updated particles' should be 'Update particles'.","section":"Algorithm 1, line 12"},{"comment":"The notation in the upper-level set condition mixes 'η_mk' and 'z_mk' in the same expression; please unify the notation (z_mk = sqrt(η_mk)) to avoid confusion.","section":"Appendix C, Eq. (35)"}],"recommendation":"major_revision","confidential_remarks":"The paper has two distinct halves of very different strength. The SE analysis (Sections II-IV) is careful and validated, and I would be comfortable with it after routine checks. The optimization half (Section V) currently overclaims: the convergence proof does not match the implemented algorithm, and the APSO feasibility issue is real. Both are localized and fixable, so I recommend major revision rather than rejection. The authors should also consider explicitly stating that the AO algorithm is a heuristic (with numerical convergence evidence) unless they can restore a genuine monotone ascent proof."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this paper for Proposition 1, not for the optimization. The closed-form downlink ergodic SE for STAR-RIS-aided cell-free massive MIMO under transceiver hardware impairments, phase noise, RIS phase errors, and spatially correlated fading is the core contribution. I checked the derivation at a level I trust: it uses standard MMSE estimation and use-and-then-forget bounding, and the Monte Carlo validation in the figures matches. That result is a useful, citable extension of the existing analysis framework.\n\nWhat is new: previous STAR-RIS cell-free work [22] assumed perfect hardware, and the hardware-impaired RIS-cell-free papers [32,33] use conventional RIS. The combination of STAR-RIS, imperfect hardware, and correlated fading appears to be new. But it is an expected extension; the techniques follow the established playbook. Significance is solid within the RIS-plus-cell-free subfield, not beyond it.\n\nThe soft spot is the optimization section. The convergence claim in Section V-D does not match Algorithm 3 as printed. The proof uses a Gauss-Seidel update chain F(η[tA],Φ[tA]) ≤ F(η[tA],Φ[tA+1]) ≤ F(η[tA+1],Φ[tA+1]), which requires the power control subproblem to be solved with the freshly updated Φ[tA]. Algorithm 3 step 5 explicitly solves (19) with Φ[tA-1]_T and Φ[tA-1]_R. So the monotonicity proof does not apply to the algorithm as written. The fix is straightforward—use Φ[tA] in step 5—but as it stands the provable convergence claim is not established. In addition, Algorithm 1 updates particles without any projection or clamping; the returned 'best' particle can leave the feasible set (β_T+β_R=1, θ in [0,2π)). That is an implementation gap. The numerical results may come from code that handles this, but the written algorithm does not.\n\nThese issues do not touch the closed-form SE derivation. The model assumptions—blocked direct links, Rayleigh AP-RIS product form—are stated clearly and limit applicability, but they are not hidden flaws. The paper is honest about what it does.\n\nWho this is for: readers working on RIS-aided cell-free massive MIMO who need a hardware-impaired performance analysis reference. It deserves a serious referee, but the optimization section needs to be fixed or explicitly de-emphasized before publication. The analytical part is worth publishing on its own.","headline":"Solid closed-form SE analysis for STAR-RIS cell-free MIMO under hardware impairments; the AO convergence proof does not match the algorithm as written.","tokens_in":28215,"tokens_out":5352,"would_cite":true,"duration_ms":48196,"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":"A closed-form downlink SE is derived for imperfect-hardware STAR-RIS cell-free MIMO, and joint design lifts the worst-user SINR.","keywords":["STAR-RIS","cell-free massive MIMO","imperfect hardware","spectral efficiency","channel estimation","max-min fairness","passive beamforming","phase noise"],"falsifier":"Set up the same system with a deterministic line-of-sight component in the AP-RIS channel (for example a Rician factor of 5-10 dB) while keeping all other parameters fixed, and compare the downlink ergodic spectral efficiency from Monte Carlo with the closed-form prediction of Eq. (13); a systematic gap that grows with the Rician factor would show that the Rayleigh-product assumption, not just the estimation bound, carries the result.","tokens_in":27126,"feed_emoji":"📶","tokens_out":8439,"duration_ms":76271,"temperature":0.7,"pith_summary":"This paper asks whether a STAR-RIS-aided cell-free massive MIMO network can be analyzed and optimized when radios and surfaces are imperfect. It answers yes: it builds an MMSE estimator for the cascaded channels and derives a closed-form expression for the downlink ergodic spectral efficiency under transceiver distortions, phase noise, RIS phase errors, correlated fading, and pilot contamination, with finite access points, users, and RIS elements. The formula matches Monte Carlo simulations and identifies which hardware imperfection hurts most. The paper then uses the formula to jointly optimize RIS transmission/reflection beams and power allocation, improving the worst-user spectral efficiency compared to random beamforming with equal power.","feed_headline":"Closed-form downlink SE derived for imperfect STAR-RIS cell-free MIMO","feed_subtitle":"New formula predicts spectral efficiency from channel statistics alone and a two-step optimizer raises worst-user rates.","key_machinery":"The load-bearing object is the linear MMSE cascaded-channel estimator and its covariance: $\\hat{\\mathbf{h}}_{mk}(0)=\\sqrt{\\gamma_T\\gamma_R p \\tau_p}\\mathbf{R}_{mk}\\mathbf{\\Psi}_{mk}^{-1}\\mathbf{z}_{mk}(0)$ with $\\mathbf{\\Psi}_{mk}$ in (5), built from the channel covariance $\\mathbf{R}_{mk}=\\mathbf{R}^d_{mk}+\\mathbf{R}_{A,m}\\mathrm{tr}(\\mathbf{R}^f_{mk})$ that folds in STAR-RIS phase errors through the characteristic-function coefficient $\\varsigma_p$. Proposition 1's SINR formula carries the argument: it turns every impairment--phase noise, RIS phase error, transceiver distortion, pilot contamination--into explicit trace terms that depend only on statistics. The optimization machinery is an alternating algorithm: adaptive particle swarm optimization for the $3N$ STAR-RIS amplitude and phase parameters, and a bisection method over second-order-cone feasibility for power control.","core_discovery":"The central claim is Proposition 1: for maximum-ratio precoding and arbitrary STAR-RIS passive beamforming, the downlink ergodic SINR of each user has the exact closed form in (15)-(16), with the desired-signal power $\\gamma_R\\gamma_T\\rho e^{-\\delta^2 t}|\\mathrm{tr}(\\boldsymbol{\\eta}_k^{1/2}\\boldsymbol{\\Omega}_k)|^2$ and a denominator $D_k(t)$ that collects beamforming uncertainty, pilot-contamination interference, transmitter and receiver distortions, and noise. Each term is written in channel covariance matrices, MMSE error covariances, hardware quality factors, and phase-noise variances, so the spectral efficiency in (13) is computable from channel statistics alone. The proof decomposes the received signal into desired, beamforming-uncertainty, inter-user interference, and hardware-distortion terms and evaluates each with Gaussian-moment identities. The paper further claims that the resulting expression is accurate against Monte Carlo, and that the STAR-RIS system with imperfect hardware still beats reflection-only RIS and cell-free baselines, with receiver hardware quality the more sensitive parameter.","pith_inferences":["Editorial inference: Because the SINR formula is expressed purely in covariance and error statistics, the same expression can serve as an objective or constraint for other resource-allocation tasks, such as energy efficiency or user scheduling, without re-deriving the analysis.","Editorial inference: The numerical ordering that UE-side hardware quality dominates AP-side quality suggests a practical budget rule--improve user oscillators and RF chains before upgrading access points--though the paper itself stops short of stating such a rule.","Editorial inference: The closed form rests on a blocked direct AP-UE link and a Rayleigh AP-RIS product; if a deployment has a residual direct path or a line-of-sight AP-RIS component, the predicted STAR-RIS advantage could shrink or grow, so a natural extension is to re-derive the covariance with a Rician AP-RIS channel.","Editorial inference: The fast convergence of the proposed alternating algorithm, within the 1 ms coherence interval assumed here, suggests the approach is practical as a real-time passive-beamforming update rule rather than only an offline design tool."],"forward_implications":["Network performance can be predicted from large-scale fading statistics alone, without Monte Carlo simulation, for finite numbers of access points, users, and RIS elements.","The relative damage of each imperfection becomes quantifiable: receiver hardware quality affects spectral efficiency more than access-point hardware quality, and RIS phase errors degrade the system in a predictable way.","The closed-form SINR enables max-min fairness optimization, and the proposed alternating algorithm (APSO plus bisection) improves the worst-user spectral efficiency over random passive beamforming with equal power.","STAR-RIS with simultaneous transmission and reflection outperforms reflection-only RIS and cell-free massive MIMO baselines even under imperfect hardware and worst-case RIS phase errors.","The analytical formula also shows that extra RIS elements beyond a certain count yield diminishing returns under maximum-ratio precoding, while weighted MMSE precoding can keep exploiting them at higher complexity."],"supporting_citations":[{"why":"Defines the STAR-RIS cell-free massive MIMO system under spatially correlated channels and supplies the Rician RIS-UE channel model that this paper extends to imperfect hardware.","marker":"[22]"},{"why":"Derives the spectral efficiency of RIS-aided cell-free massive MIMO with MR processing, the analytical baseline this paper generalizes to STAR-RIS and hardware impairments.","marker":"[15]"},{"why":"Supplies the RIS phase-error model (Von Mises or uniform) and the imperfect-hardware treatment used in deriving the channel covariance.","marker":"[28]"},{"why":"Provides the RIS-cell-free massive MIMO performance analysis with transceiver hardware impairments that this paper extends to STAR-RIS and phase errors.","marker":"[32]"},{"why":"Quantifies the impact of hardware impairments on RIS-aided cell-free massive MIMO, supplying the EVM distortion model used for the downlink analysis here.","marker":"[33]"},{"why":"Models time-varying oscillator phase noise and hardware distortions for massive MIMO, giving the Wiener phase-noise process used in the SINR analysis.","marker":"[24]"},{"why":"Provides the spatially correlated RIS covariance model used in Eq. (1) for the RIS spatial correlation matrix.","marker":"[35]"},{"why":"Gives the second-order cone feasibility and bisection method used to solve the quasi-concave power-control sub-problem.","marker":"[38]"},{"why":"Supplies Lemma 1, the Gaussian-moment identity used to evaluate the channel covariance and the closed-form SINR terms.","marker":"[36]"}],"fun_headline_variants":["Closed-form SE for STAR-RIS MIMO with imperfect hardware","Exact SE prediction from channel stats in STAR-RIS MIMO","Joint design boosts worst user in imperfect STAR-RIS MIMO","STAR-RIS MIMO: SE formula and optimizer for hardware flaws","Imperfect hardware? Closed-form SE for STAR-RIS MIMO"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire analysis assumes the access-point-to-RIS channel is a Rayleigh product with no line-of-sight component and that direct AP-user links are completely blocked, so a real deployment with a strong direct AP-user path or a dominant line-of-sight AP-RIS component would fall outside the formulas and optimization conclusions.","fun_headline_variants_meta":{"raw":{"variants":["Closed-form SE for STAR-RIS MIMO with imperfect hardware","Exact SE prediction from channel stats in STAR-RIS MIMO","Joint design boosts worst user in imperfect STAR-RIS MIMO","STAR-RIS MIMO: SE formula and optimizer for hardware flaws","Imperfect hardware? Closed-form SE for STAR-RIS MIMO"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000355,"raw_usage":{"total_tokens":2008,"prompt_tokens":1105,"completion_tokens":903,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":721,"completion_tokens_details":{"reasoning_tokens":813}},"tokens_in":721,"tokens_out":903,"duration_ms":8459,"temperature":1.0,"reasoning_tokens":813,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:48:28.787066+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Set up the same system with a deterministic line-of-sight component in the AP-RIS channel (for example a Rician factor of 5-10 dB) while keeping all other parameters fixed, and compare the downlink ergodic spectral efficiency from Monte Carlo with the closed-form prediction of Eq. (13); a systematic gap that grows with the Rician factor would show that the Rayleigh-product assumption, not just the estimation bound, carries the result.","supporting_citations":[{"cited_title":"Intelligent reflecting surface-assisted MU-MISO systems with imperfect hardware: Channel estimation and beam- forming design,","cited_arxiv_id":null,"evidence_quote":"Supplies the RIS phase-error model (Von Mises or uniform) and the imperfect-hardware treatment used in deriving the channel covariance."},{"cited_title":"STAR-RIS assisted cell-free massive MIMO system under spatially-correlated channels,","cited_arxiv_id":null,"evidence_quote":"Defines the STAR-RIS cell-free massive MIMO system under spatially correlated channels and supplies the Rician RIS-UE channel model that this paper extends to imperfect hardware."},{"cited_title":"Reconfigurable intelligent surface-assisted cell-free massive MIMO systems over spatially-correlated channels,","cited_arxiv_id":null,"evidence_quote":"Derives the spectral efficiency of RIS-aided cell-free massive MIMO with MR processing, the analytical baseline this paper generalizes to STAR-RIS and hardware impairments."},{"cited_title":"Performance analysis of RIS-assisted cell-free mas- sive mimo systems with transceiver hardware impairments,","cited_arxiv_id":null,"evidence_quote":"Provides the RIS-cell-free massive MIMO performance analysis with transceiver hardware impairments that this paper extends to STAR-RIS and phase errors."},{"cited_title":"How much does reconfigurable intelligent surface improve cell- free massive MIMO uplink with hardware impairments?","cited_arxiv_id":null,"evidence_quote":"Quantifies the impact of hardware impairments on RIS-aided cell-free massive MIMO, supplying the EVM distortion model used for the downlink analysis here."},{"cited_title":"Massive MIMO with non-ideal arbitrary arrays: Hardware scaling laws and circuit-aware design,","cited_arxiv_id":null,"evidence_quote":"Models time-varying oscillator phase noise and hardware distortions for massive MIMO, giving the Wiener phase-noise process used in the SINR analysis."},{"cited_title":"Rayleigh fading modeling and channel hardening for reconfigurable intelligent surfaces,","cited_arxiv_id":null,"evidence_quote":"Provides the spatially correlated RIS covariance model used in Eq. (1) for the RIS spatial correlation matrix."},{"cited_title":"Two-timescale design for reconfigurable intelligent surface-aided massive MIMO systems with imperfect CSI,","cited_arxiv_id":null,"evidence_quote":"Supplies Lemma 1, the Gaussian-moment identity used to evaluate the channel covariance and the closed-form SINR terms."}],"review_version":1}