{"id":"b8c85fff-99b9-43b7-a89c-9dccaec5f06f","arxiv_id":"2502.01590","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A fractional-programming and block-coordinate-descent algorithm that jointly optimizes digital precoding and pinching-antenna locations achieves substantially higher multiuser downlink sum-rate than fixed-location antennas in Line-of-Sight simulations.","lead":"This paper designs a downlink beamforming scheme for pinching antennas, which are low-cost elements that slide along dielectric waveguides to get closer to users. In simulated indoor Line-of-Sight settings, the joint design of precoder and antenna locations delivers up to 6 dB higher throughput than fixed-location antenna arrays.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reported PAS gains are confounded by an unfair baseline: the conventional array is a compact half-wavelength ULA at the area center, while PAS elements can span the full service region, so the 6 dB gain largely reflects path-loss reduction rather than the joint beamforming and location optimization.","rationale":"The reader's weakest_assumption was the LoS-only channel model. That assumption is explicitly stated and scoped, and even under multipath, the proximity-based gains of PAS would partially survive; moreover, the paper's stated scenario is LoS-dominated indoor downlink. The more load-bearing issue is that the numerical evidence does not control for the physical aperture: the PAS has a distributed, movable aperture while the conventional baseline is a co-located half-wavelength ULA at the center. This confound directly affects the headline quantitative claim ('6 dB gain' and 'gains grow with side length') and, hence, the conclusion that the joint beamforming/location optimization is what delivers the benefit. My concern is testable by changing only the baseline geometry in the existing simulation code. The paper's core theoretical development (FP-BCD derivation, closed-form W update, Gauss-Seidel location search) appears sound, so a fair comparison may well preserve a meaningful gain; the verdict should remain CONDITIONAL, now explicitly requiring the fair-baseline comparison. I therefore leave the reader's verdict unchanged while flagging that the weakest assumption they identified is not the one I consider most load-bearing.","tokens_in":9714,"tokens_out":10922,"duration_ms":98821,"concrete_test":"Reproduce Fig. 2(a) (K=M=4, D=30 m) with a controlled fixed-location baseline that occupies the same physical aperture as the PAS: place the M fixed antennas at coordinates [D/2, (m-1)d, a] with d=D/(M-1), i.e., at the midpoints of the waveguides, and run the same FP-BCD algorithm with only the W update. Compare the sum-rate. If the PAS gain over this baseline is near 0 dB, the reported gain is an artifact of the compact baseline geometry; if the gain remains close to 6 dB, the joint location optimization has intrinsic value under the same aperture.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The numerical comparison in Section IV does not isolate the value of the proposed joint precoder/location optimization. The conventional baseline is described as 'M fixed-location antennas deployed along the y-axis, centered within the square region,' with half-wavelength spacing, i.e., a compact ULA at the area center. For M=4 at 28 GHz, this aperture is only about 1.6 cm. In contrast, the PAS waveguides span the full side length D (up to 50 m), and each pinching element can move along the entire x-range [0,D]. The channel model in (4) has a 1/D_{m,k} amplitude factor, so moving an element from the center to a location near the user yields a large path-loss reduction that is independent of any beamforming optimization. The paper's own explanation for Fig. 2(c) confirms this: it attributes the growing gain to 'the average distance between users and the center' being larger in the conventional setting. Thus the headline 6 dB gain (and the 11 dB against ZF) primarily reflects a comparison between a centralized co-located array and a distributed movable aperture, not the benefit of jointly optimizing W and l. A reader cannot distinguish how much of the claimed improvement comes from the algorithm's location optimization versus simply from deploying elements closer to users. This undermines the central claim that the proposed FP-BCD joint design itself is what yields the significant throughput boost.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers downlink multiuser MIMO transmission from an access point equipped with pinching-antenna waveguides. The authors formulate a weighted-sum-rate maximization over the digital precoder W and the pinching-element locations ℓ, and propose a fractional-programming block-coordinate-descent (FP-BCD) algorithm that alternates between closed-form updates for the auxiliary variables and precoder (RZF) and a scalar grid search for each element location. Numerical experiments compare the proposed scheme against a conventional fixed-location antenna array and report substantial throughput gains, including a 6 dB power gain over the same algorithm with fixed antennas and an 11 dB gain over a zero-forcing baseline.","tokens_in":9976,"tokens_out":11020,"duration_ms":92579,"significance":"The paper addresses an important and timely topic: exploiting the flexibility of pinching antennas for multiuser downlink beamforming. The FP-BCD derivation is structurally sound, the RZF update in (33) is correct, and the complexity analysis is clearly presented. If the reported gains were properly isolated, the paper would be a useful step toward establishing PAS as a practical low-cost technology for LoS-dominated indoor deployments. However, the numerical validation currently confounds two distinct effects: the path-loss advantage of moving elements close to users and the benefit of jointly optimizing W and ℓ against a fixed aperture. The grid-search resolution is also insufficient to optimize the highly oscillatory location objective, which undermines the reliability of the reported convergence and the numerical results.","major_comments":[{"comment":"The comparison baseline is not an apples-to-apples control for the joint optimization claim. The conventional baseline is a compact half-wavelength ULA centered in the service area, with aperture roughly Mλ/2 (about 1.6 cm for M=4 at 28 GHz), whereas the PAS elements can move over the entire side length D (up to 50 m). Because the channel model in (4b) has a 1/D_{m,k}(ℓ_m) amplitude factor, the PAS gains in Fig. 2 largely reflect path-loss reduction from placing elements near users rather than the benefit of jointly optimizing W and ℓ. The paper's own discussion of Fig. 2(c) attributes the growing gain to the increased distance between users and the center in the conventional setting. To support the central contribution, the authors should compare against a fixed-location distributed array with the same aperture/coverage as the PAS, or against the same PAS with fixed (e.g., equally spaced) locations, so that the value of the joint design is isolated.","section":"Section IV (Experimental setting and Fig. 2)"},{"comment":"The scalar location update via grid search uses only 10^3 points over the interval [0, L_m], with L_m up to 50 m. The objective f_m(ℓ_m) in (39) contains a cosine whose argument has phase rate k0(∂D_{m,k}/∂ℓ_m + i_ref). Since ∂D_{m,k}/∂ℓ_m ∈ [-1,1], the local oscillation period at f=28 GHz and i_ref=1.44 lies between λ/(i_ref+1) ≈ 4.4 mm and λ/(i_ref-1) ≈ 24 mm. For D=30 m, the grid spacing is 30 mm, larger than the smallest period; the grid therefore cannot resolve the true maxima of (39). Consequently, the location update may not increase the objective, contradicting the convergence argument in Section III.D, and the numerical results may not reflect the actual performance of the proposed algorithm. The authors should either use a search resolution finer than the oscillation period, employ a local optimization method that exploits the structure of (39), or provide evidence that a coarse grid is sufficient for the reported configurations.","section":"Section III.C, Eq. (39) and Section IV (grid search)"},{"comment":"The entire optimization relies on the LoS-only channel model (2)-(4), with the explicit assumption that non-LoS paths are negligible. This assumption is load-bearing because the location optimization exploits the deterministic phase relation exp{-jk0(D_{m,k}(ℓ_m)+i_ref ℓ_m)}. In indoor environments with significant multipath, this relation no longer describes the channel and the benefit of optimizing ℓ_m could largely disappear. The paper offers no sensitivity analysis under Rician or measured channels, nor a discussion of the conditions under which the LoS assumption holds. The authors should add a robustness study or at least a quantitative discussion of the impact of multipath on the proposed design.","section":"Section II.A"}],"minor_comments":[{"comment":"The proof of Lemma 2 contains a typo: equation (19) reads 'SINR_k(˘W,l) = SINR_k(˘W,l)', where the two sides refer to different SINR definitions (17) and (11). Please clarify.","section":"Section III.A, Lemma 2 proof"},{"comment":"The description of the grid search does not state how the number of points 10^3 scales with the waveguide length L_m or the carrier frequency. Given the oscillation period in (39), the resolution should be tied to λ and i_ref; a brief note would help readers assess the complexity claims.","section":"Section IV (experimental setting)"},{"comment":"The convergence statement is terse: for a non-convex BCD scheme, global convergence typically requires exact or sufficiently improving block updates. The paper should state explicitly whether the current location is included in the grid set and under what conditions the grid search guarantees a non-decreasing objective.","section":"Section III.D (convergence)"},{"comment":"The square brackets in (39) make the expression difficult to parse; adding parentheses or defining the argument of the cosine explicitly would improve readability.","section":"Section III.C, Eq. (39)"},{"comment":"Figure 2(b) only shows two side-length values (D=5 m and D=20 m); presenting a wider range of D would better support the claim that the gain grows with area size.","section":"Section IV, Fig. 2(b)"}],"recommendation":"major_revision","confidential_remarks":"The paper is technically sound in its algorithmic derivation, but the numerical evidence for the central claim is not yet convincing. The grid-resolution issue and the baseline mismatch are both fixable and should be addressed before publication. The paper's self-citations to the earlier PAS literature are appropriate and do not raise concerns. I suggest the editor request a revision with a fair baseline and a resolution-aware location optimization."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is the first paper I know that formulates joint digital precoding and pinching-element location optimization for a general multiuser PAS downlink, and the FP-BCD machinery is applied cleanly. But the numerical evidence conflates two different things: the baseline comparison pits a compact half-wavelength ULA at the area center against pinching elements that can roam the full service region. So the headline 6 dB (and 11 dB vs ZF) mostly measures path-loss reduction from being near the users, not the value of the joint optimization. The paper needs a fixed-antenna baseline spread over the same aperture, or a PAS with unoptimized locations, before the algorithm's contribution can be assessed.\n\nWhat is actually good: the system model and problem formulation are clear, the derivations in Section III are mostly correct, the RZF update in (33) is standard and valid, and the complexity analysis is reasonable. The authors also state the LoS-only assumption explicitly rather than hiding it.\n\nSoft spots, in order: (1) the baseline unfairness above; (2) the grid search uses 10^3 points over up to 50 m, giving a spacing around 5 cm, while the location objective oscillates with spatial frequency up to about 2.4 times the wavenumber, i.e., a period of a few mm at 28 GHz. The 1-D search is too coarse to find the maximum of fm(·), so the claim that the algorithm 'optimizes' locations is not really supported. (3) The convergence guarantee says the objective is non-decreasing in each BCD step, but a grid search update doesn't exactly maximize the marginal, so monotonicity is not strict; the algorithm stops on a threshold rather than true convergence. (4) No error bars or sensitivity analysis for the LoS-only assumption.\n\nNone of this kills the central idea—joint design of digital precoding and movable pinching elements is a sensible problem and the algorithm is a reasonable first cut—but the numerical claims in Fig. 2 are not a clean demonstration of the algorithm's value. The paper deserves a serious referee who can insist on a fixed-aperture distributed baseline and a finer or smarter location search.\n\nRecommendation: send to review, but expect the experimental section to be reworked.","headline":"Sensible first formulation of joint precoder/location optimization for multiuser PAS downlink, but the numerical baseline confounds distributed aperture effects with the algorithm's contribution.","tokens_in":10564,"tokens_out":2516,"would_cite":true,"duration_ms":22541,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94A05","90C26"],"pacs":[],"model":"deepseek-v4-flash","headline":"Jointly optimizing pinching-antenna locations and the digital precoder gives a 6 dB power gain over fixed-location arrays in multiuser downlink MIMO.","keywords":["pinching antenna systems","flexible antennas","hybrid beamforming","fractional programming","weighted sum-rate","multiuser MIMO","line-of-sight channels","block coordinate descent"],"falsifier":"Re-run the FP-BCD algorithm with the pure-LoS channel of Eq. (4b) replaced by a model that adds one reflected path (or Rayleigh fading) at about 10% of the LoS amplitude; if the weighted sum-rate gap between the pinching system and a fixed-location array falls below roughly 1 dB, the reported 6 dB gain depends on the LoS-only assumption rather than on location flexibility itself.","tokens_in":9492,"feed_emoji":"📡","tokens_out":9514,"duration_ms":77280,"temperature":0.7,"pith_summary":"Pinching antennas are low-cost radiators clipped onto dielectric waveguides, and this paper tries to establish that a multiuser downlink transmitter can exploit their movable locations as a real performance lever. The authors formulate the weighted sum-rate maximization as a joint optimization over the digital beamforming matrix and the element positions, and solve it with the FP-BCD algorithm, which alternates between a closed-form precoder update and a one-dimensional search over each pinching element's location. Under a line-of-sight indoor channel model, the optimized pinching system reaches the same sum-rate as a fixed-location antenna array with about 6 dB less transmit power, and the gain grows with the area being served. If the claim holds, pinching antennas offer a practical way to reduce transmit power and improve throughput without high-cost reconfigurable hardware.","feed_headline":"Pinching antennas beat fixed arrays by 6 dB in downlink MIMO","feed_subtitle":"Roving pinching elements on waveguides chase users, yielding up to 11 dB gain over conventional fixed antennas.","key_machinery":"The central object is the effective channel vector $\\mathbf{g}_k(\\boldsymbol{\\ell})$, whose $m$-th entry is the phase-shifted free-space response $\\xi \\alpha_{m,k} \\exp\\{-j k_0 (D_{m,k}(\\ell_m)+ i_{\\mathrm{ref}} \\ell_m)\\}/D_{m,k}(\\ell_m)$; it factors into a location-dependent diagonal matrix $\\mathbf{L}_k(\\boldsymbol{\\ell})$ times a location-independent vector $\\mathbf{g}_k^0$, which is why element positions can be optimized like tunable beamforming weights. The algorithmic machinery is the FP-BCD loop: the Lagrange dual transform and quadratic transform of fractional programming replace the sum-of-log-ratios objective with a quadratic surrogate; the precoder update then solves a regularized zero-forcing problem (a closed-form linear precoder balancing signal and interference) in closed form; and each pinching-element location is updated by a one-dimensional grid search over the scalar objective $f_m(\\ell_m)$, since the oscillating cosine term in that objective defeats gradient methods.","core_discovery":"The paper's central claim is that jointly tuning the digital precoder $\\mathbf{W}$ and the pinching-element locations $\\boldsymbol{\\ell}$ is what unlocks the gains of pinching-antenna systems in a general multiuser downlink, not just the two-user case studied earlier. With the line-of-sight channel model $g_{m,k}(\\ell_m) = \\xi \\alpha_{m,k} \\exp\\{-j k_0 (D_{m,k}(\\ell_m)+i_{\\mathrm{ref}} \\ell_m)\\}/D_{m,k}(\\ell_m)$, every element's position sets both a distance-dependent attenuation and a deterministic phase, so moving an element toward its served user acts like a passive beamforming adjustment. The FP-BCD algorithm turns the non-convex weighted-sum-rate problem into a sequence of tractable steps: fractional programming gives a quadratic surrogate, the digital update is closed-form regularized zero forcing, and each location update is a scalar maximization solved by grid search. Simulations show a 6 dB power gain over a same-algorithm fixed-array baseline and an 11 dB gain over a fixed-array zero-forcing baseline, with the advantage widening as the coverage area grows.","pith_inferences":["The same path-loss-reduction mechanism would likely carry over to uplink reception and to user scheduling, because what matters is placing elements near users; the paper does not simulate those settings.","In a multipath environment the coherent phase-alignment part of the gain would degrade, but a portion of the gain from simply shortening the distance to users may survive; a quantitative split would require new simulations.","A fixed-location baseline whose array position is also optimized, rather than fixed at the center of the region, might capture part of the reported gain and give a fairer comparison.","Comparing PAS against a fixed array with more elements at comparable hardware cost would clarify whether the flexibility of pinching antennas is worth more than simply adding conventional antennas."],"forward_implications":["Under the paper's line-of-sight model, a pinching-antenna access point can hit a target multiuser rate with roughly 6 dB less transmit power than a same-sized fixed array using the same algorithm, and roughly 11 dB less than a fixed-array zero-forcing scheme.","The power and rate advantage grows with the side length of the coverage area, because movable elements shorten the distance to users and offset large-scale path loss that a centered fixed array cannot.","The algorithm's overall complexity per iteration is polynomial, dominated by $O(M^3)$ matrix work and $O(MLK)$ location search, so the approach is computationally plausible for modest array sizes and converges within about ten iterations.","Increasing the number of waveguides $M$ raises both the achievable weighted sum-rate and the pinching gain over the fixed baseline, matching the expectation of more spatial degrees of freedom."],"supporting_citations":[{"why":"Supplies the line-of-sight pinching-antenna channel model in Eq. (2) and the two-user PAS study that motivates extending the analysis to a general multiuser downlink.","marker":"[8]"},{"why":"Supplies the Lagrange dual transform of fractional programming that converts the weighted sum-of-log-ratios objective into a tractable variational form.","marker":"[14]"},{"why":"Supplies the quadratic transform and block-coordinate-descent machinery used for the inner precoding and location updates.","marker":"[15]"},{"why":"Introduces the pinching-antenna concept and its low-cost dielectric-waveguide implementation, grounding the practical premise of the paper.","marker":"[7]"},{"why":"Analyzes the array gain achievable by pinching-antenna systems, supporting the claim that element-location choice improves effective channel gain.","marker":"[10]"}],"fun_headline_variants":["Pinching antennas chase users for 6 dB downlink gain in MIMO","Roving antenna elements boost multiuser MIMO throughput by 11 dB","Joint optimization of pinching positions and precoder yields 6 dB gain","Pinching antennas beat fixed arrays: 6 dB gain, up to 11 dB","Moving antennas on waveguides cut power needs by 6 dB in MIMO"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the indoor channel is essentially line-of-sight only, so each user's channel phase is a known function of the element position; if significant multipath is present, that phase relation breaks and the location optimization would lose most of its leverage.","fun_headline_variants_meta":{"raw":{"variants":["Pinching antennas chase users for 6 dB downlink gain in MIMO","Roving antenna elements boost multiuser MIMO throughput by 11 dB","Joint optimization of pinching positions and precoder yields 6 dB gain","Pinching antennas beat fixed arrays: 6 dB gain, up to 11 dB","Moving antennas on waveguides cut power needs by 6 dB in MIMO"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000831,"raw_usage":{"total_tokens":3628,"prompt_tokens":943,"completion_tokens":2685,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":559,"completion_tokens_details":{"reasoning_tokens":2597}},"tokens_in":559,"tokens_out":2685,"duration_ms":16344,"temperature":1.0,"reasoning_tokens":2597,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T14:53:47.328451+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the FP-BCD algorithm with the pure-LoS channel of Eq. (4b) replaced by a model that adds one reflected path (or Rayleigh fading) at about 10% of the LoS amplitude; if the weighted sum-rate gap between the pinching system and a fixed-location array falls below roughly 1 dB, the reported 6 dB gain depends on the LoS-only assumption rather than on location flexibility itself.","supporting_citations":[{"cited_title":"Fractional programming for communication systems—Part II: Uplink scheduling via matching,","cited_arxiv_id":null,"evidence_quote":"Supplies the quadratic transform and block-coordinate-descent machinery used for the inner precoding and location updates."},{"cited_title":"Pinching antenna: Using a dielectric waveguide as an antenna,","cited_arxiv_id":null,"evidence_quote":"Introduces the pinching-antenna concept and its low-cost dielectric-waveguide implementation, grounding the practical premise of the paper."}],"review_version":1}