{"id":"5be93a95-41ef-482e-a828-d2a8987a1aab","arxiv_id":"2505.13392","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A first experimental BD-RIS prototype, built from an antenna array and a switchable load network, is characterized and shown via model-based optimization to benefit from inter-element connections.","lead":"This paper reports the first working prototype of a reflective beyond-diagonal reconfigurable intelligent surface, where the antenna elements can be linked through tunable connections. Using measured component responses and a physics-based model, the authors find that these inter-element connections improve wireless gains, though the full system's performance is evaluated in software rather than measured directly.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"All claimed performance gains are computed from an unvalidated model: H is synthesized via Eq. (1) from a Virtual VNA estimate of S, with no direct end-to-end measurement of any realized configuration.","rationale":"The reader's verdict is already CONDITIONAL and identifies exactly this missing direct validation. My stress-test pass does not find a different, more severe internal inconsistency; the hardware prototype and component characterizations are genuine, and the logic of the MNT model is standard. The single load-bearing weakness is that Eq. (1) is used as a substitute for measurement without any validation of the estimated S or of the assembled S_L. Because the paper's headline contribution is explicitly the experimental validation of BD-RIS benefits, this missing check is necessary before the performance claims can be accepted as experimental results. I therefore leave the verdict unchanged rather than moving it, while emphasizing that the conditional should explicitly require the direct H measurement described above or a rephrasing of the claims as model-based predictions.","tokens_in":7740,"tokens_out":9466,"duration_ms":100955,"concrete_test":"Directly measure the end-to-end channel with the same 8-port VNA for a subset of realized configurations. Connect the VNA to the seven TX/RX ports while the eight RIS ports remain terminated by the physical tunable load network, and measure the 7-port scattering matrix over 700-900 MHz for: (i) the optimized '123' configurations for the four KPIs, (ii) the OC and D-RIS benchmark configurations, and (iii) several random configurations. Extract the measured 3x4 TX-RX block and compare it element-wise with H predicted by Eq. (1) from the estimated S and S_L. Report complex error and gain error (dB) per frequency point.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that inter-element connections are beneficial rests entirely on software-computed channels. In Sec. IV (first paragraph), the authors state that H 'for any realizable BD-RIS configuration' is computed from the estimated static S and the measured load scattering matrix S_L using Eq. (1), and all KPIs in Sec. IV-D are evaluated from these computed H matrices. No direct measurement of H for any realized load configuration is reported anywhere in the manuscript. The two inputs to Eq. (1) are therefore the load-bearing assumption. S_L is assembled from separate one-port/two-port VNA measurements of individual and coupled loads (Sec. III), assuming those characteristics fully describe the eight-port load network without parasitic coupling, switch leakage, or in-situ detuning after SMA connection. S is estimated by the Virtual VNA gradient-descent procedure in the Appendix; the manuscript reports no convergence residual, uncertainty, or validation of this estimate, and explicitly notes a block-wise sign ambiguity that is resolved by a single auxiliary measurement. Under exhaustive optimization, model errors are not averaged out: the optimizer will preferentially select configurations in which Eq. (1) predicts large gains, so even modest model bias could produce the reported 24.5%, 50.3%, 13.8%, and 1.7% improvements and the coupled-load advantages over D-RIS. Until H is measured directly for at least the optimized configurations, the abstract's statement that the evaluations 'validate' the benefit of inter-element connections is stronger than the evidence supports.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a reflective beyond-diagonal RIS (BD-RIS) prototype built by connecting eight commercial antennas to a PCB-realized tunable load network. The load network offers three individual loads per port and two switchable connections to adjacent ports, realizing a tridiagonal load scattering matrix. The static 15-port scattering matrix S of the rich-scattering environment is estimated with a Virtual VNA procedure, and the load network is characterized by direct VNA measurements of its individual and coupled loads. The end-to-end channel H is then computed for every feasible load configuration using the multiport network theory relation in Eq. (1). The authors exhaustively optimize four KPIs (SISO channel gain, interference-channel sum rate, low-SNR capacity, and high-SNR capacity) over the realizable configuration space, and benchmark the results against open-circuit, diagonal-RIS, ideal-hardware, 1-bit-load, interleaved-BD-RIS, and mutual-coupling-unaware baselines. They report that the coupled loads improve all four KPIs relative to the open-circuit benchmark, that hardware constraints strongly affect the achievable gains, and that mutual-coupling awareness matters little in the realized hardware but matters more under idealized assumptions.","tokens_in":7955,"tokens_out":6945,"duration_ms":65257,"significance":"If the reported results are trusted, this is a useful first hardware demonstration of a reflective BD-RIS architecture and a valuable study of how realistic hardware constraints—finite load sets, tridiagonal connectivity, and switch limitations—change conclusions drawn from idealized BD-RIS models. The strengths of the paper are its direct measurement of load characteristics, its use of exhaustive search over the full realizable configuration space, and its careful definition of several benchmarks that isolate the effect of inter-element connections. However, the central performance claims are entirely computed from the estimated static scattering matrix S and the measured load matrix S_L via Eq. (1); no end-to-end channel measurement for any realized configuration is reported. Because the optimizer selects configurations that maximize the model-predicted KPIs, unquantified modeling error could systematically bias the reported gains. The paper is therefore best viewed as a model-based performance evaluation anchored to component-level measurements, rather than as a direct experimental validation of the prototype's wireless performance.","major_comments":[{"comment":"The end-to-end channel H is never measured for any realized load configuration; it is computed from the estimated static matrix S and the measured load matrix S_L via Eq. (1). All four KPIs and the percentage improvements reported in Sec. IV-D therefore quantify the behavior of the MNT model, not of the physical prototype. Because the exhaustive search in Sec. IV-C specifically selects configurations that maximize these model-predicted KPIs, even a modest bias in Eq. (1) or in its inputs would be amplified into the claimed 24.5%, 50.3%, 13.8%, and 1.7% improvements and into the coupled-load advantage. The paper should report direct measurements of H (e.g., S-parameter measurements between TX and RX ports) for at least the OC, optimized D-RIS, and optimized BD-RIS configurations, and compare them with Eq. (1); without such a comparison, the abstract's statement that the performance evaluations 'validate that inter-element connections are beneficial' is not experimentally established.","section":"Sec. IV"},{"comment":"The Virtual VNA estimate of S is not validated. The gradient-descent procedure is described, but no convergence residual, no comparison with a direct measurement of any portion of S, and no uncertainty estimate is reported; the block-wise sign ambiguity is resolved with a single auxiliary measurement. Since S is a primary input to Eq. (1), any estimation error propagates into every computed H and into the optimized configuration selected by the exhaustive search. The authors should at least validate the estimated S by predicting a held-out set of the 900 measured configurations and reporting the prediction error, or by directly measuring a subset of S entries with the available 8-port VNA.","section":"Appendix"},{"comment":"The assembled eight-port load network is assumed to be exactly characterized by isolated one-port and two-port VNA measurements of the individual and coupled loads. This neglects in-situ effects such as switch leakage, crosstalk between adjacent switch channels, and detuning introduced by the SMA connections and the PCB assembly. Because S_L is the second input to Eq. (1), an unmodeled difference between the assumed and actual S_L would change all computed channels and, consequently, the optimal configurations. The authors should validate S_L in situ, for example by measuring the assembled network directly or by including an S_L sensitivity test in the analysis.","section":"Sec. III"}],"minor_comments":[{"comment":"In the sentence beginning 'In total, there are', the summation symbol appears to be rendered as 'P' in the text; please correct the typography.","section":"Sec. IV-C"},{"comment":"The curves and points are averages over 201 frequency points and all TX/RX selections, but no measure of spread is provided; for the small differences in KPI 4 (1.7% vs. 1.5%), error bars or confidence intervals would help the reader judge whether the coupled-load advantage is meaningful.","section":"Fig. 3"},{"comment":"The claim that this is the first experimental prototype should be reconciled with the fact that the tunable load network PCB was presented in [9]; the paper should state explicitly that the prototype is the integration of that PCB with the antenna array.","section":"Sec. I"},{"comment":"The conclusion that mutual-coupling unawareness barely affects the KPIs in the experimental setup is presented without discussion of whether this is a consequence of the small array size, the specific load impedance values, or the static rich-scattering environment; a sentence of context would improve the paper.","section":"Sec. IV-D"}],"recommendation":"major_revision","confidential_remarks":"The central issue is that all reported performance gains are computed rather than measured; the paper's framing as an experimental validation is not yet supported. The requested end-to-end measurements for at least the optimized configurations are feasible with the existing 8-port VNA setup and would substantially strengthen the contribution. I would encourage the editor to require such validation before acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The headline: this is the first reflective BD-RIS prototype made from actual hardware, but every claimed performance gain is computed from a model, not measured end-to-end. The hardware is genuine: eight antennas terminated by a PCB load network with SP5T switches, three individual loads, and two coupled paths. The authors directly measure the load scattering parameters, which is more than most prototype papers do. They estimate the static environment's scattering matrix via the Virtual VNA method, then exhaustively search all 12,970 realizable configurations and benchmark against diagonal RIS and ideal fully-connected BD-RIS. The systematic study of how hardware constraints affect achievable gains is a useful contribution.\n\nThe soft spot is load-bearing. In Section IV, H is computed for every configuration from Eq. (1) using the estimated S and the measured S_L. No direct measurement of H for any realized configuration is reported. So the abstract's 'validate that inter-element connections are beneficial' and the conclusion's 'our results confirm' are stronger than the evidence supports. What they actually have is a model-based prediction. The exhaustive search makes this more concerning, not less: the optimizer will preferentially select configurations where the model predicts large gains, so even modest model bias can produce the reported 24.5%, 50.3%, 13.8%, and 1.7% improvements. The Virtual VNA estimate of S has no reported residual or uncertainty, and there is a block-wise sign ambiguity resolved by a single auxiliary measurement. With those moving parts, I would want at least a few direct H measurements for optimized configurations before calling the gains validated. The lack of error bars and the absence of released data make independent verification harder.\n\nTo be fair, the paper is transparent about the computation: it says H is computed in software, not measured. The problem is the framing in the abstract and conclusion overclaims relative to that evidence. The prototype and the component characterization are solid; the performance evaluation is careful but remains a simulation layered on measured parameters.\n\nWho is this for? People working on BD-RIS hardware and on model validation for RIS. It deserves a serious referee because the prototype is a first and the analysis is careful, but the referee should push for direct validation or recalibrated language. I would accept it for peer review with a clear directive: either measure H directly for a handful of configurations in a follow-up, or soften every 'validate' and 'confirm' in the abstract and conclusion to 'model-based evaluation suggests.'","headline":"First reflective BD-RIS prototype is real, but every claimed performance gain is computed from an unvalidated model—no end-to-end channel is ever measured.","tokens_in":8507,"tokens_out":2158,"would_cite":true,"duration_ms":20680,"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":"This paper reports the first experimental reflective beyond-diagonal RIS and shows that reconfigurable inter-element connections improve all four tested performance metrics.","keywords":["beyond-diagonal RIS","reconfigurable intelligent surface","mutual coupling","multi-port network theory","Virtual VNA","tridiagonal load network","rich-scattering environment","reverberation chamber"],"falsifier":"Take one of the optimized load configurations reported for a specific frequency and TX/RX pair, measure the actual TX-to-RX scattering parameters with a calibrated VNA, and compare them with the values Eq. (1) predicts from the estimated $S$ and measured $S_L$; disagreement beyond the stated measurement uncertainty would show the reported gains are computational artifacts rather than realized channel improvements.","tokens_in":7496,"feed_emoji":"📡","tokens_out":6919,"duration_ms":59864,"temperature":0.7,"pith_summary":"This paper reports the first hardware prototype of a reflective beyond-diagonal reconfigurable intelligent surface (BD-RIS), a surface whose antenna ports are terminated by a tunable network that can either apply one of three individual loads or connect a port to an adjacent one. Using multi-port network theory and a Virtual VNA estimate of the static scattering matrix, the authors compute the end-to-end wireless channels for every one of the 12,970 load configurations of their eight-element prototype inside a reverberation chamber. They then exhaustively optimize four KPIs: one SISO channel gain and three MIMO capacities. Averaged over frequency and transmitter/receiver choices, the full prototype improves all four KPIs relative to open-circuit terminations by 24.5%, 50.3%, 13.8%, and 1.7%, and the coupled loads add gains over a diagonal-only RIS. The paper argues that realistic hardware constraints, not only the number of loads but their exact scattering properties, substantially change the conclusions drawn from idealized BD-RIS theory.","feed_headline":"First beyond-diagonal RIS prototype shows coupling helps","feed_subtitle":"A switch network linking adjacent antenna ports outperforms a conventional RIS on all four tested metrics.","key_machinery":"The load-bearing object is the multi-port network theory relation $$H = S_{RT} + S_{RS}($S_L^{{-1}}$ - S_{SS})^{-1}S_{ST},$$ which turns the static scattering matrix $S$ of the environment, together with the tunable load network scattering matrix $S_L$, into the end-to-end channel $H$. $S$ is estimated without a 15-port VNA using the Virtual VNA technique, which reconstructs the full matrix from measurements made through a subset of ports while the remaining ports are terminated by tunable and coupled loads. $S_L$ is obtained from direct one- and two-port VNA measurements of the PCB's loads; hardware constraints make $S_L$ tridiagonal and prevent simultaneous use of both adjacent couplings on one element. Because $S$ and $S_L$ are known, the paper enumerates all 12,970 realizable configurations and exhaustively optimizes each KPI.","core_discovery":"The central claim is that a reflective BD-RIS with reconfigurable inter-element connections is practically realizable and that the connections are worth having. With the experimentally realized tridiagonal load network, all four considered KPIs improve over an open-circuit baseline, and the availability of coupled loads yields additional improvement over the same prototype used as a conventional diagonal RIS. The paper also establishes that hardware constraints shape the conclusions: reducing the three individual loads to two can either barely matter or cost notably, depending on which load is dropped, and the gap between BD-RIS and diagonal RIS widens sharply under the idealized continuously tunable, fully connected load matrices assumed in much theory. In this constrained prototype, ignoring mutual coupling during optimization barely changes the achieved KPIs, whereas under ideal hardware it can cut one KPI's gain nearly in half.","pith_inferences":["Because no end-to-end channel is directly measured, the reported gains are model-based; a direct VNA measurement of $H$ for optimized configurations would test whether the modeled gains appear over the air.","The same PCB that enables Virtual VNA characterization also acts as the BD-RIS itself, suggesting that tunable load networks can serve dual roles: measurement hardware in the lab and operational RIS hardware in deployment.","The hardware constraint that a switch cannot simultaneously connect to both neighbors may be a central reason the coupled-load gains are modest; alternative topologies that allow simultaneous couplings could show larger gains.","The strong KPI dependence suggests BD-RIS benefits should be reported as a distribution over KPIs and environments rather than as a single number."],"forward_implications":["Inter-element connections help in practice, but the size of the benefit depends strongly on the KPI: 50.3% improvement for one MIMO metric versus 1.7% for another.","Hardware constraints substantially shrink the theoretical advantage of BD-RIS over diagonal RIS; ideal fully connected tunable loads show much larger gains and a larger BD-RIS versus D-RIS gap.","The exact scattering properties of the available individual loads matter, not just their count, since dropping one reflective load hurts performance much more than dropping the absorptive load.","Mutual-coupling-unaware optimization is nearly lossless for this constrained hardware but harmful under idealized hardware, so MC awareness matters most when the load network is nearly ideal.","No notable BD-RIS versus interleaved BD-RIS difference appears in this eight-element prototype, leaving the predicted advantage of interleaving untested at this scale."],"supporting_citations":[{"why":"Supplies the scattering-parameter network analysis that yields the end-to-end channel formula in Eq. (1).","marker":"[1]"},{"why":"Provides the mutual-coupling-aware BD-RIS model and the theoretical claim that ignoring mutual coupling degrades performance.","marker":"[2]"},{"why":"Supplies the physics-compliant diagonal representation argument used to connect the prototype's architecture to existing diagonal-RIS algorithms.","marker":"[4]"},{"why":"Provides the Virtual VNA estimation method that reconstructs $S$ from measurements through a subset of ports.","marker":"[7]"},{"why":"Presents the PCB tunable load network that the paper repurposes as the BD-RIS hardware.","marker":"[9]"},{"why":"Defines the tridiagonal RIS architecture and the graph-theoretic argument for its efficiency, which the prototype instantiates.","marker":"[10]"},{"why":"Motivates the interleaved BD-RIS benchmark by predicting that interleaving can outperform standard BD-RIS.","marker":"[13]"},{"why":"Quantifies performance loss from mutual-coupling-unaware RIS optimization, the effect the paper tests experimentally.","marker":"[14]"}],"fun_headline_variants":["First beyond-diagonal RIS prototype proves coupling pays","Experimental BD-RIS: inter-element connections boost performance","Beyond-diagonal RIS prototype shows links outweigh isolation","First BD-RIS prototype: coupling helps, but hardware limits matter"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire performance evaluation computes the end-to-end channel $H$ in software from an estimated static scattering matrix $S$ and a measured load scattering matrix $S_L$, so the reported gains depend on the accuracy of the multi-port network model and the Virtual VNA estimate of $S$.","fun_headline_variants_meta":{"raw":{"variants":["First beyond-diagonal RIS prototype proves coupling pays","Experimental BD-RIS: inter-element connections boost performance","Beyond-diagonal RIS prototype shows links outweigh isolation","First BD-RIS prototype: coupling helps, but hardware limits matter"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00096,"raw_usage":{"total_tokens":4019,"prompt_tokens":807,"completion_tokens":3212,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":423,"completion_tokens_details":{"reasoning_tokens":3148}},"tokens_in":423,"tokens_out":3212,"duration_ms":20117,"temperature":1.0,"reasoning_tokens":3148,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:13:57.041077+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take one of the optimized load configurations reported for a specific frequency and TX/RX pair, measure the actual TX-to-RX scattering parameters with a calibrated VNA, and compare them with the values Eq. (1) predicts from the estimated $S$ and measured $S_L$; disagreement beyond the stated measurement uncertainty would show the reported gains are computational artifacts rather than realized channel improvements.","supporting_citations":[{"cited_title":"Modeling and architecture design of reconfigurable intelligent surfaces using scattering parameter network analysis,","cited_arxiv_id":null,"evidence_quote":"Supplies the scattering-parameter network analysis that yields the end-to-end channel formula in Eq. (1)."},{"cited_title":"Beyond diagonal reconfigurable intelligent surfaces with mutual coupling: Modeling and optimization,","cited_arxiv_id":null,"evidence_quote":"Provides the mutual-coupling-aware BD-RIS model and the theoretical claim that ignoring mutual coupling degrades performance."},{"cited_title":"A physics-compliant diagonal representation for wireless channels parametrized by beyond-diagonal reconfigurable intelligent surfaces,","cited_arxiv_id":null,"evidence_quote":"Supplies the physics-compliant diagonal representation argument used to connect the prototype's architecture to existing diagonal-RIS algorithms."},{"cited_title":"Virtual VNA 2.0: Ambiguity-free scattering matrix estimation by terminating not-directly-accessible ports with tunable and coupled loads,","cited_arxiv_id":null,"evidence_quote":"Provides the Virtual VNA estimation method that reconstructs $S$ from measurements through a subset of ports."},{"cited_title":"Scalable Multiport Antenna Array Characterization with PCB-Realized Tunable Load Network Providing Additional \"Virtual\" VNA Ports","cited_arxiv_id":"2503.16256","evidence_quote":"Presents the PCB tunable load network that the paper repurposes as the BD-RIS hardware."},{"cited_title":"Beyond diagonal reconfigurable intelligent surfaces utilizing graph theory: Modeling, architecture design, and optimization,","cited_arxiv_id":null,"evidence_quote":"Defines the tridiagonal RIS architecture and the graph-theoretic argument for its efficiency, which the prototype instantiates."},{"cited_title":"Static grouping strategy design for beyond diagonal reconfigurable intelligent surfaces,","cited_arxiv_id":null,"evidence_quote":"Motivates the interleaved BD-RIS benchmark by predicting that interleaving can outperform standard BD-RIS."},{"cited_title":"Mutual coupling and unit cell aware optimization for reconfigurable intelligent surfaces,","cited_arxiv_id":null,"evidence_quote":"Quantifies performance loss from mutual-coupling-unaware RIS optimization, the effect the paper tests experimentally."}],"review_version":1}