{"id":"faa16e55-784b-439c-90d5-067d941ca8be","arxiv_id":"1908.09090","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"An ADMM-AltMin algorithm substitutes a low-complexity ADMM solve for the semidefinite relaxation step in alternating minimization for wideband DPA-MIMO hybrid precoding, matching SDR-AltMin spectral efficiency in simulations.","lead":"This paper proposes an ADMM-based algorithm for hybrid precoding in wideband millimeter-wave MIMO systems with distributed phased arrays. It claims similar spectral efficiency to a higher-complexity semidefinite relaxation method, using fewer computations and low-resolution phase shifters.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (17) is not the scaled ADMM update for Eq. (15): it drops the A1^T A1 + A2^T A2 Hessian term and the dual variable, so the central equivalence and complexity claims are unsupported as written.","rationale":"The reader's weakest assumption identifies exactly the load-bearing gap: Eq. (17) is presented without derivation and is inconsistent with the standard scaled ADMM update. My analysis confirms and strengthens this: beyond omitting the dual variable, Eq. (17) also drops the A1^T A1 + A2^T A2 Hessian term, which is not identity for this problem. This is not merely an aesthetic ADMM-orthodoxy issue; the scalar closed form is the basis for the claimed O(M_t^2 N_s^4) per-iteration complexity. Eq. (12) is a second manifestation of the same unstated non-orthogonality assumption. The simulations do show the claimed matching performance in one configuration, but no code or error bars are provided, so they do not independently establish that the printed algorithm (as opposed to a corrected variant) was actually run. Since the concern is concrete, central, and checkable, the reader's CONDITIONAL verdict is appropriate: the paper should not be accepted until the authors provide a correct derivation of Eq. (17) (or acknowledge and analyze the actual update) and rerun the comparisons with that corrected algorithm.","tokens_in":1090,"tokens_out":4169,"duration_ms":121035,"concrete_test":"Re-derive the x-update of Eq. (15) by completing the square, then test the printed Eq. (17) against the exact ADMM update on a minimal instance: K=1, Ns=1, Mt=2, Nsub=2, with F_RF = blkdiag([1,1]^T, [1,1]^T)/sqrt(2) so F_RF^H F_RF has off-diagonal 1, and a fixed random F_opt. Solve the sphere-constrained problem (14) globally by the Lagrange multiplier method (or SDR) and compare the fixed point of the printed recursion (17)-(18) with the exact ADMM recursion and with the global solution. If the fixed points differ, Eq. (17) does not solve the claimed subproblem, and the equivalence and complexity claims in Section IV need revision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (ADMM-AltMin matches SDR-AltMin at lower complexity) rests on the baseband ADMM recursion (17)-(18) solving the sphere-constrained least-squares problem (14)/(15). As written, Eq. (17) is incorrect. For g(x)=||A1 x - Re b||^2 + ||A2 x - Im b||^2, the exact scaled ADMM x-update is x^{i+1} = (2(A1^T A1 + A2^T A2) + rho I)^{-1} (2A1^T Re b + 2A2^T Im b + rho(y^i - nu^i)). Eq. (17) instead uses a scalar denominator 2+rho and replaces the dual variable nu^i by x^i. The scalar denominator is valid only if A1^T A1 + A2^T A2 = I, but here A = I_Ns (Kronecker product) F_RF, so this matrix is the real form of A^H A = I_Ns (Kronecker product) (F_RF^H F_RF), which is not identity because the constant-modulus columns of F_RF are generally non-orthogonal (e.g., two subarrays steering the same direction). The omitted dual variable is also not a typo-level detail: it changes the recursion. The same non-orthogonality assumption appears in Eq. (12), where ||F_RF F_BB||_F^2 is replaced by ||F_BB||_F^2. Consequently, the Section IV claim that the printed algorithm achieves the same spectral efficiency as SDR-AltMin has no sound analytical basis unless Eq. (17) is re-derived and shown to be a minimizer of (15).","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes ADMM-AltMin, an alternating-optimization algorithm for wideband hybrid precoding in distributed phased-array MIMO (DPA-MIMO) systems with OFDM. For a fixed RF precoder, the per-subcarrier baseband precoder is obtained by ADMM on a sphere-constrained least-squares problem; for a fixed baseband precoder, the RF precoder is updated in closed form with quantized phase shifters. The paper claims that the proposed method matches the spectral efficiency of the SDR-AltMin baseline at lower complexity, and it reports simulations with infinite and finite-resolution phase shifters and imperfect CSI.","tokens_in":10072,"tokens_out":7442,"duration_ms":77859,"significance":"If the claims were sound, the paper would provide a useful lower-complexity alternative to SDR-based wideband hybrid precoding for DPA-MIMO, with practical attention to phase-shifter quantization and imperfect CSI. The numerical study is benchmarked against external algorithms [20] and [25], uses concrete system parameters, and does not appear to fit constants to force the advertised result. However, the central ADMM derivation in Eq. (17) is incorrect as written, and this error is load-bearing: the claimed algorithm is not the ADMM recursion stated in Eq. (10), so the equivalence and complexity claims in Section IV lack a sound basis until the update is corrected and the numerical behavior is re-examined.","major_comments":[{"comment":"The x-update is not the scaled-ADMM minimizer for problem (15). For g(x) = ||A1 x - Re{b}||^2 + ||A2 x - Im{b}||^2, the minimizer of g(x) + (rho/2)||x - y^i + nu^i||^2 is (2(A1^T A1 + A2^T A2) + rho I)^-1 (2A1^T Re{b} + 2A2^T Im{b} + rho(y^i - nu^i)). The printed scalar update corresponds to assuming A1^T A1 + A2^T A2 = I and to replacing the dual variable nu^i with x^i, neither of which is stated or justified. With A = I_Ns (tensor product) F_RF, the relevant Gram matrix is I_Ns (tensor product) F_RF^H F_RF, which is not identity under the constant-modulus constraint (6c). Consequently, Algorithm 1 as written is not the ADMM recursion described by Eq. (10), and the Section IV claim that it matches SDR-AltMin has no analytical basis.","section":"III-B1, Eq. (17)"},{"comment":"The power-constraint reduction ||F_RF F_BB[k]||_F^2 = ||F_BB[k]||_F^2 is valid only when F_RF^H F_RF = I_Mt. The block-diagonal RF precoder with constant-modulus entries need not have orthogonal columns, and two subarrays can use similar or even identical steering vectors. This simplification changes the feasible set of the per-subcarrier problem (11) and is later embedded in problem (14), so it must either be proved from the DPA-MIMO assumptions or removed.","section":"III-B1, Eq. (12)"},{"comment":"The complexity comparison is incomplete and currently misleading. It compares a per-iteration cost O(M_t^2 N_s^4) for the proposed method with a SDR complexity expression that includes log(1/epsilon), without reporting the number of ADMM iterations required or the total complexity of the outer alternating loop. Moreover, if Eq. (17) is replaced by the correct matrix-inverse update, the per-iteration cost is no longer a scalar division; it requires solving a linear system of dimension 2 M_t N_s, so the stated complexity saving is not established.","section":"Section IV, complexity comparison"},{"comment":"The paper invokes ADMM for the non-convex sphere constraint but provides no convergence analysis. Standard ADMM convergence results such as those in [21] do not directly apply to the non-convex feasible set C in Eq. (8); the termination criteria in Eq. (19) are therefore not shown to guarantee convergence to a stationary point of problem (6). The equivalence claim in Section IV should be either supported by a convergence theorem for the corrected update or explicitly framed as an empirical observation.","section":"III-A and Algorithm 1"}],"minor_comments":[{"comment":"There are typographical errors that should be corrected, including 'hybird' in Section I, 'seperated' in Sections I and II, 'multi-dimentional' below Eq. (18), and 'combing' in Section II-C.","section":"Throughout"},{"comment":"The SDR complexity expression is described as 'per-iteration' but includes log(1/epsilon), which is characteristic of a total complexity estimate; please clarify what quantity is being compared with the ADMM per-iteration cost.","section":"Section IV"},{"comment":"The ADMM penalty parameter is fixed to rho = 1 and the stopping tolerances to 1e-6, but no sensitivity study is reported; a brief study of rho would make the empirical claims more robust.","section":"Section IV"},{"comment":"The outer stopping criterion is left as 'until some termination criterion is satisfied', which makes the number of outer iterations and the reported simulation behavior difficult to reproduce.","section":"Algorithm 1"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know: this is a modest algorithmic contribution with a load-bearing derivation error. The idea is to replace the SDR subproblem in the standard alternating minimization for wideband hybrid precoding with an ADMM step, and the paper applies this to a specific subarray architecture (DPA-MIMO). If the algorithm actually worked as described, it would be a useful complexity reduction for a narrow but real problem.\n\nWhat the paper does well: the problem is stated cleanly, the use of the alternating framework is standard, and the authors correctly attribute the RF-precoder update to [20]. They also test with quantized phase shifters, which is practical, and their simulations (if the algorithm is implemented as intended) show parity with SDR-AltMin. That is worth something.\n\nNow the soft spots, and the big one is not minor. Equation (17) is not the scaled ADMM update for (15). The correct x-update for that quadratic objective is a matrix inverse of (2(A1^T A1 + A2^T A2) + rho I), not the scalar denominator (2+rho), and it must include the dual variable nu^i. The printed version replaces nu^i with the previous iterate x^i. That changes the recursion. The scalar denominator is only valid if A1^T A1 + A2^T A2 = I, which is not true here because A = I_Ns ⊗ F_RF and F_RF columns are not orthogonal. The same incorrect assumption appears in Eq. (12), where they replace ||F_RF F_BB||_F^2 with ||F_BB||_F^2. Unless Eq. (17) is a typographical artifact (possibly from the extraction), the printed algorithm does not actually minimize the subproblem, and the Section IV equivalence claim has no analytical basis.\n\nOther soft spots: no code or data are provided, so I cannot tell whether the simulations use the correct but unwritten update. The complexity comparison gives per-iteration cost but no iteration counts or convergence rate, so the overall complexity claim is incomplete. They also cite prior ADMM applications in mmWave [22,23] but do not compare against them, which is a gap.\n\nWho is this for? Researchers in millimeter-wave hybrid precoding who might want to explore ADMM-based baseband updates. The paper deserves a serious referee because the error may be fixable and the idea is not silly, but as written it should not be accepted. I would send it to review with a clear request: verify Eq. (17), correct it or provide code, and either add convergence analysis or soften the complexity claim.","headline":"A plausible but unproven ADMM-AltMin algorithm for wideband DPA-MIMO; the core update in Eq. (17) looks wrong as written, and the equivalence claim needs a fix before the paper is usable.","tokens_in":10658,"tokens_out":2242,"would_cite":false,"duration_ms":22984,"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":"The paper claims an ADMM-based hybrid precoder matches SDR performance with lower per-iteration complexity in wideband DPA-MIMO.","keywords":["hybrid precoding","distributed phased arrays","wideband MIMO","OFDM","ADMM","millimeter-wave","quantized phase shifters","alternating optimization"],"falsifier":"Compare equation (17) with the general scaled ADMM update (10a) term by term. If the printed update is taken literally, the dual variable never enters the $x$-update; running the printed iteration on a small random instance (e.g., $M_t=2$, $N_s=1$) and checking whether it converges to the known constrained least-squares optimum of (14) would show whether the equivalence claim is real.","tokens_in":9490,"feed_emoji":"📡","tokens_out":9248,"duration_ms":87404,"temperature":0.7,"pith_summary":"The paper addresses hybrid precoding for wideband millimeter-wave systems in which the transmitter is a distributed phased array: several subarrays, each with its own RF chain, cooperate through a common baseband precoder across OFDM subcarriers. It proposes an alternating-optimization algorithm, ADMM-AltMin, that fixes the RF precoder and solves the per-subcarrier baseband precoding problem with ADMM-like updates, then updates the RF phases in closed form. The paper claims this scheme reaches the same spectral efficiency as the semidefinite-relaxation-based SDR-AltMin baseline while cutting the per-iteration cost to $O(M_t^2 N_s^4)$, and that it works with quantized phase shifters. A sympathetic reader would care because this is a concrete step toward practical low-complexity hybrid precoding for frequency-selective DPA-MIMO user equipment.","feed_headline":"ADMM hybrid precoder matches costly SDR in wideband MIMO","feed_subtitle":"A cheaper alternating method reaches the same spectral efficiency as SDR-AltMin for mmWave distributed phased arrays.","key_machinery":"The central object is the scaled ADMM iteration used on the baseband precoder. After transforming the complex problem into real variables $\\bar{x} = [\\Re\\{x\\}^T, \\Im\\{x\\}^T]^T$, the paper solves the sphere-constrained least-squares problem (15) by alternating between an $x$-update of the form (17) and a projection $y^{i+1} = \\sqrt{c}\\,(\\bar{x}^{i+1}+\\nu^i)/\\|\\bar{x}^{i+1}+\\nu^i\\|_2$ onto the sphere, with dual update $\\nu^{i+1} = \\nu^i + \\bar{x}^{i+1} - y^{i+1}$. This is the computational engine that replaces semidefinite relaxation; the rest of the machinery is the closed-form phase update $\\vartheta^\\star_{il} = \\angle(\\sum_k [F_{\\mathrm{opt}}[k]]_{i,:}[F_{\\mathrm{BB}}[k]]^H_{l,:})$, which yields the RF precoder and can be quantized to a low number of bits. The block-diagonal structure of the RF precoder makes both steps decomposable across subarrays.","core_discovery":"On the paper's own terms, the central discovery is that the hardest part of wideband hybrid precoding—the non-convex baseband subproblem that couples subcarriers through a single shared RF precoder and a per-subcarrier norm constraint—can be handled by a scaled ADMM update in the real domain (equations (17)–(18)) instead of a semidefinite relaxation. With that solver inside an alternating-minimization loop, the RF precoder has a closed-form phase update (21), optionally quantized to a few bits. The paper reports that the resulting ADMM-AltMin matches the average spectral efficiency of SDR-AltMin in simulation across SNR, and outperforms a fixed-subarray wideband hybrid precoder baseline. The claimed payoff is a per-iteration complexity of $O(M_t^2 N_s^4)$ against the much higher cost of solving an SDR each iteration.","pith_inferences":["The authors do not spell this out, but the same ADMM-style baseband solver generalizes beyond DPA-MIMO: any hybrid precoding formulation whose baseband subproblem is a norm-constrained least squares could substitute the SDR inner loop for this iteration, likely with a similar complexity saving.","A testable extension is to use the complexity slack not for one-shot optimization but for per-block or per-slot adaptation to time-varying CSI, replacing the current batch computation over $K$ subcarriers with streaming updates.","The 4-bit result hints that even coarser quantization or nonuniform phase-shifter bit allocation across subarrays might be acceptable; the paper does not optimize the bit allocation, so varying quantization levels per subarray is an open question."],"forward_implications":["If ADMM-AltMin is correct, wideband DPA-MIMO transmitters can get the same spectral efficiency as the SDR-based method with strictly cheaper per-iteration linear algebra.","The algorithm's per-step cost $O(M_t^2 N_s^4)$ makes it feasible to run baseband precoding updates repeatedly, for example during channel tracking, rather than once per channel realization.","Finite-resolution phase shifters, down to 4 bits in the simulations, appear sufficient to approach the infinite-resolution performance, so hardware cost can be reduced without leaving the DPA-MIMO architecture.","Because the RF precoder has a block-diagonal structure, the ADMM updates can be computed locally at each subarray, which suits distributed implementations."],"supporting_citations":[{"why":"Supplies the SDR-AltMin baseline and the closed-form RF phase solution (21) that ADMM-AltMin reuses.","marker":"[20]"},{"why":"Gives the scaled ADMM iterations (10a)-(10c) that the proposed baseband solver is meant to instantiate.","marker":"[21]"},{"why":"Provides the frequency-selective hybrid precoding formulation with a common RF precoder shared across subcarriers.","marker":"[18]"},{"why":"Defines the dynamic-subarray wideband hybrid precoding baseline whose spectral efficiency is compared in Fig. 3a.","marker":"[25]"},{"why":"Provides the alternating optimization framework that splits the optimization into baseband and RF subproblems.","marker":"[24]"},{"why":"Supplies the semidefinite relaxation complexity expression used to compare against the proposed $O(M_t^2 N_s^4)$ cost.","marker":"[29]"},{"why":"Establishes the matrix-approximation formulation, matching the Frobenius norm to the optimal digital precoder, used as the objective.","marker":"[9]"}],"fun_headline_variants":["ADMM-AltMin: wideband MIMO precoding at SDR performance","Lower-cost ADMM algorithm matches SDR for wideband hybrid precoding","ADMM hybrid precoder: SDR-level performance without SDR cost","Wideband MIMO: ADMM-AltMin matches SDR at reduced complexity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the update in equation (17) actually solves the constrained least-squares baseband problem (14)–(15); if the printed formula is wrong and the constraint-tracking term is missing, the claimed equivalence to SDR-AltMin does not follow.","fun_headline_variants_meta":{"raw":{"variants":["ADMM-AltMin: wideband MIMO precoding at SDR performance","Lower-cost ADMM algorithm matches SDR for wideband hybrid precoding","ADMM hybrid precoder: SDR-level performance without SDR cost","Wideband MIMO: ADMM-AltMin matches SDR at reduced complexity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000235,"raw_usage":{"total_tokens":1451,"prompt_tokens":850,"completion_tokens":601,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":466,"completion_tokens_details":{"reasoning_tokens":517}},"tokens_in":466,"tokens_out":601,"duration_ms":6047,"temperature":1.0,"reasoning_tokens":517,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:22:18.258631+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare equation (17) with the general scaled ADMM update (10a) term by term. If the printed update is taken literally, the dual variable never enters the $x$-update; running the printed iteration on a small random instance (e.g., $M_t=2$, $N_s=1$) and checking whether it converges to the known constrained least-squares optimum of (14) would show whether the equivalence claim is real.","supporting_citations":[{"cited_title":"Distributed optimiza- tion and statistical learning via the alternating directio n method of multipliers,","cited_arxiv_id":null,"evidence_quote":"Gives the scaled ADMM iterations (10a)-(10c) that the proposed baseband solver is meant to instantiate."},{"cited_title":"Frequency selective hybr id precoding for limited feedback millimeter wave systems,","cited_arxiv_id":null,"evidence_quote":"Provides the frequency-selective hybrid precoding formulation with a common RF precoder shared across subcarriers."},{"cited_title":"Dynamic subarra ys for hybrid precoding in wideband mmWave MIMO systems,","cited_arxiv_id":null,"evidence_quote":"Defines the dynamic-subarray wideband hybrid precoding baseline whose spectral efficiency is compared in Fig. 3a."},{"cited_title":"Near-optimal hybrid proce ssing for massive MIMO systems via matrix decomposition,","cited_arxiv_id":null,"evidence_quote":"Provides the alternating optimization framework that splits the optimization into baseband and RF subproblems."},{"cited_title":"Semidefinite r elaxation of quadratic optimization problems,","cited_arxiv_id":null,"evidence_quote":"Supplies the semidefinite relaxation complexity expression used to compare against the proposed $O(M_t^2 N_s^4)$ cost."},{"cited_title":"Spatially sparse precoding in millimeter wave MIMO systems,","cited_arxiv_id":null,"evidence_quote":"Establishes the matrix-approximation formulation, matching the Frobenius norm to the optimal digital precoder, used as the objective."}],"review_version":1}