{"id":"999473c0-6eda-4432-bb20-961d84162570","arxiv_id":"2505.20805","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Dual-polarized SIM, formed as two independent single-polarization SIMs, supports more interference-free data streams and higher SE/EE than single-polarized SIM in simulated HMIMO links.","lead":"This paper proposes a dual-polarized stacked intelligent metasurface (DPSIM) for holographic MIMO, letting the transmitter and receiver process two orthogonal polarizations separately with stacked phase-shifting layers. The authors report that, in simulation, the design supports more interference-free data streams and higher spectral and energy efficiency than single-polarized SIMs.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (8) forces the polarization-conversion ratio ε out of the channel distribution, so the reported ε-dependent SE/EE gains and the 'approaching upper bound' claim are not supported by the stated model.","rationale":"The paper's central claim is that DPSIM doubles the number of interference-free streams and approaches the SE/EE upper bound under polarization imperfections. The reader's weakest assumption (block-diagonal T/R, polarization unchanged between layers) is a modeling choice the paper explicitly states and that is defensible for high-isolation dual-polarized elements; it is not the point where the argument breaks. The printed channel model in Eq. (8) is internally inconsistent: the repeated correlation matrices make the four polarization channel blocks identical and make the distribution independent of ε. Therefore the reported ε=0.2 vs 0.4 difference in Fig. 6 cannot be produced by the stated model, and the abstract's 'under different polarization imperfection conditions' claim loses its numerical support. The upper bound in Eq. (27) is computed from the same degenerate G, so 'approaching the upper bound' is not a meaningful external benchmark. This is a correctness risk that can be settled by inspecting the linked code. If the code implements the printed Eq. (8), the simulations are invalid for the paper's conclusions; if it implements the likely intended diagonal-correlation form, Eq. (8) must be corrected and all curves regenerated. Either way, the manuscript as written needs a fix, so the CONDITIONAL verdict stands. We do not move to REJECT because the underlying architecture concept (two independent polarization paths, twice as many phase controls) is plausible and the code provides an avenue for verification.","tokens_in":9656,"tokens_out":9500,"duration_ms":98586,"concrete_test":"Run the provided GitHub code with identical random seeds for ε=0.2 and ε=0.4. If the SE/EE curves are statistically identical, Eq. (8) as printed is the simulation model and the claimed polarization-imperfection robustness is an artifact. Alternatively, recompute the distribution of G analytically using the diagonal-correlation form and confirm that the repeated-block form is ε-independent; if the code uses the diagonal form, Eq. (8) is a typo that must still be corrected for the paper's claims to be reproducible.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing problem is not the block-diagonal form of T and R (a stated approximation) but the printed channel model in Eq. (8). With A = R_RX^(1/2) and B = R_TX^(1/2), the pre/post multipliers in Eq. (8) are [A A; A A] and [B B; B B]. Multiplying the block matrix C = [G00 G01; G10 G11] by these rank-one block structures collapses G to four identical blocks A(G00+G10+G01+G11)B. Since G00,G11 have variance (1-ε)PL and G10,G01 have variance εPL, the sum has variance 2PL, independent of ε. Therefore the distribution of G is ε-independent. The distinct EE curves for ε=0.2 and ε=0.4 in Fig. 6 cannot arise from Eq. (8). Moreover, the theoretical upper bound in Eq. (27) is computed from singular values of this same degenerate G, so the claim that DPSIM 'approaches the upper bound' is relative to a model that does not represent dual-polarized propagation. The likely intended form is diag(R_RX^(1/2),R_RX^(1/2)) C diag(R_TX^(1/2),R_TX^(1/2)), which preserves co-/cross-polar structure and makes ε meaningful.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a dual-polarized stacked intelligent metasurface (DPSIM) architecture for holographic MIMO, in which each metasurface layer independently controls the phase of two orthogonal polarizations. The authors model the end-to-end channel, formulate an optimization problem to shape the effective channel toward the singular-value matrix of the physical channel, and propose a layer-by-layer gradient descent with water-filling (LGD-WF) algorithm. Simulations claim that, for the same number of layers and unit size, DPSIM supports more interference-free streams and achieves higher spectral and energy efficiency than single-polarization SIM and conventional massive MIMO, approaching a theoretical upper bound.","tokens_in":9954,"tokens_out":4383,"duration_ms":48719,"significance":"The core idea—using independent phase control of two polarizations to double the wave-domain processing capability of a stacked metasurface within the same physical footprint—is timely and plausible, and the paper provides a concrete architecture, an iterative algorithm, and a public simulation code link. If the claims were fully supported, this would be a useful extension of the SIM literature. However, the quantitative evidence is currently undermined by a degenerate printed channel model, a circular performance bound, and an energy-efficiency metric that omits all hardware power consumption. The qualitative advantage of DPSIM may survive a corrected analysis, but the present manuscript does not establish it.","major_comments":[{"comment":"As printed, the channel model in Eq. (8) makes the four polarization blocks of G identical and removes all dependence on the polarization conversion ratio ε. With A = R_RX^(1/2) and B = R_TX^(1/2), the pre- and post-multipliers are [A A; A A] and [B B; B B], so each block of G becomes A(G00+G10+G01+G11)B. Since G00,G11 have variance (1−ε)PL and G10,G01 have variance εPL, the sum has variance 2PL, independent of ε. Consequently, the ε-dependent SE/EE curves in Fig. 6 and the claim of robustness to polarization imperfections cannot follow from the stated model. The likely intended form is diag(R_RX^(1/2), R_RX^(1/2)) · [G00 G01; G10 G11] · diag(R_TX^(1/2), R_TX^(1/2)); please correct Eq. (8) and rerun the simulations, or explain explicitly why the printed form is not what was implemented.","section":"§II, Eq. (8)"},{"comment":"The design target Λ in P1 is the singular-value matrix of the physical channel G, and the theoretical upper bound η_SE^ub in Eq. (27) is computed from exactly the same Λ, used also for water-filling power allocation. Therefore the reported 'approach to the theoretical upper bound' measures only how well LGD-WF fits the algorithm's own target, not how close DPSIM comes to any fundamental or architectural limit. This circularity should be acknowledged, and the upper-bound claim should be reframed as, at best, a check of the optimization algorithm's convergence to the chosen target.","section":"§III, P1 and §IV, Eq. (27)"},{"comment":"The energy efficiency definitions in Eq. (28) divide spectral efficiency only by the transmit power P_t, omitting all hardware power consumption: RF chains, DACs/ADCs, metasurface control, and baseband processing. The claim that DPSIM-assisted HMIMO has 'significantly higher EE' than 256×32 and 512×32 massive MIMO systems is therefore not a systems-level energy comparison. Please include a total power model, or relabel the metric as transmit-power-normalized SE and temper the EE conclusions accordingly.","section":"§IV, Eq. (28)"}],"minor_comments":[{"comment":"The stated dimensions are inconsistent: the products in (5) and (6) are built from 2M×2M and 2N×2N block-diagonal matrices, yet T is declared as C^(2M×2S) and R as C^(2S×2N). Please specify how the S active streams are selected from the M or N units (e.g., a truncation or an explicit input/output mapping).","section":"§II, Eqs. (5)–(6)"},{"comment":"There is a typo in the sentence preceding Eqs. (20)–(21): 'Our phase update strategies for TX-DPSIM and TX-DPSIM' should presumably read 'TX-DPSIM and RX-DPSIM'.","section":"§III, Step 3"},{"comment":"The water-filling expression diag(Λ_{1:2S,1:2S}Λ_{1:2S,1:2S}) is confusing; it should be written as the squared singular values, e.g., [Λ]^2_s, to match the SINR terms in Eqs. (26)–(27).","section":"§IV, Eq. (24)"},{"comment":"The GitHub URL in the footnote contains spaces: 'Dual Polarization Stacked Intelligent Metasurfaces for Holographic MIMO.git'. Please provide a URL-encoded link so that the code is actually accessible.","section":"§I, footnote and code link"}],"recommendation":"major_revision","confidential_remarks":"This is a simulation-based extension of the SIM line of work. The dual-polarization concept is plausible, but the printed channel model in Eq. (8) is a load-bearing technical error, and the performance metrics as defined cannot support the paper's central quantitative claims. A major revision that corrects the channel model, reruns the simulations, and revises the upper-bound and EE claims could make the paper acceptable. I do not see a reason to reject outright, because the core architectural idea is defensible on degrees-of-freedom grounds and the issues appear fixable within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper is an incremental but sensible extension of stacked intelligent metasurfaces: stack dual-polarized RIS layers so each layer controls two orthogonal polarizations independently. The authors model co- and cross-polar channels with four matrices, propose an LGD-WF algorithm to shape the end-to-end channel, and link to a code repository. The qualitative claim that, for a fixed aperture, DPSIM supports more interference-free streams than a single-polarization SIM is defensible: you get twice as many phase controls per layer.\n\nThe soft spots are real, and one is load-bearing. Eq. (8) as printed multiplies the four-block channel matrix by rank-one block-structured pre/post matrices, which collapses all four blocks to the same sum. The polarization conversion ratio epsilon then cancels out of the distribution, so the epsilon-dependent curves in Fig. 6 cannot be produced by the stated model. This looks like a typo — the intended form is almost certainly diag(R_RX^{1/2}, R_RX^{1/2}) C diag(R_TX^{1/2}, R_TX^{1/2}) — but as written it undercuts the paper's central quantitative claims. The authors should fix this and rerun.\n\nTwo additional concerns are more moderate. The EE metric only divides by transmit power, not total hardware power, making 'EE approaching upper bound' less meaningful. And the theoretical upper bound in Eq. (27) is computed from the same singular values that the algorithm targets, so the 'approaching upper bound' language is partly circular. That doesn't kill the architecture, but it means the headline numbers are not independently grounded. The block-diagonal assumption about polarization staying fixed between layers is stated explicitly; it's a simplification, not a hidden flaw.\n\nOverall: the physical idea is worth taking seriously and probably survives the typo correction. The paper deserves a serious referee, but the referee should ask to see the corrected model and a cleaner energy-accounting before the quantitative claims are accepted. I'd send it to review, with heavy revision expected.","headline":"A plausible stacked dual-polarized metasurface idea, undermined as printed by a channel model typo that erases the polarization-mixing parameter and by a partly circular upper bound; still worth refereeing after corrections.","tokens_in":10506,"tokens_out":2136,"would_cite":false,"duration_ms":21261,"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":"This paper claims that independently controlling two orthogonal polarizations on each stacked metasurface layer makes the transceiver behave like two isolated single-polarization SIMs, doubling the interference-free data streams it can…","keywords":["dual-polarized stacked intelligent metasurfaces","holographic MIMO","reconfigurable intelligent surfaces","wave-domain signal processing","spectral efficiency","energy efficiency","polarization cross-interference","LGD-WF algorithm"],"falsifier":"Run a full-wave simulation of two adjacent dual-polarized metasurface layers and compute the complete $2M \\times 2M$ transmission matrix; if the off-diagonal polarization blocks have norm comparable to the diagonal blocks, for example at oblique incidence or with imperfect unit-cell isolation, then the block-diagonal model of Eqs. (5)-(6) fails and the claimed equivalence to two isolated SIMs, together with the per-polarization gradient updates, no longer holds.","tokens_in":9437,"feed_emoji":"📡","tokens_out":7727,"duration_ms":69600,"temperature":0.7,"pith_summary":"This paper claims that a holographic MIMO transceiver can double its wave-domain signal-processing capacity without adding layers or unit cells by letting each metasurface layer control two orthogonal polarizations independently. The proposed dual-polarized stacked intelligent metasurface (DPSIM) is mathematically equivalent to two isolated single-polarization SIMs, so each polarization path can carry its own data streams while polarization cross-interference and inter-stream interference are suppressed. A layer-by-layer gradient-descent with water-filling algorithm (LGD-WF) tunes the phase shifts so the end-to-end channel approaches the target diagonal structure, allowing spatial multiplexing without digital precoding. Simulations show that, for the same number of layers and unit size, DPSIM supports more simultaneous interference-free streams than a conventional SIM and brings spectral and energy efficiency close to the theoretical upper bound even when polarization conversion is present.","feed_headline":"Dual-polarized metasurface stack doubles interference-free streams","feed_subtitle":"Stacking dual-polarized surfaces lets holographic MIMO suppress interference and approach the spectral-efficiency ceiling.","key_machinery":"The central object is the DPSIM transfer-matrix pair in Eqs. (5)-(6): interleaved per-layer diagonal phase matrices $\\Phi_p^l$ and $\\Psi_p^k$ for the two polarizations $p\\in\\{0,1\\}$ with Rayleigh-Sommerfeld free-space propagation matrices $V^l$ and $U^k$. Because both kinds of matrices are block-diagonal across polarizations, the whole stack factorizes into two isolated SIMs. The LGD-WF algorithm differentiates the fitting objective $\\Gamma=\\|\\alpha RGT-\\Lambda_{1:2S,1:2S}\\|_F^2$ with respect to each per-polarization phase, normalizes the gradients to avoid explosion or vanishing, updates the phases layer by layer with a decaying learning rate, and finishes with water-filling power allocation across the $2S$ streams.","core_discovery":"The central claim is that a stacked intelligent metasurface made of dual-polarized units, with independent phase control per polarization, splits into two parallel, isolated single-polarization SIMs. Because both the per-layer phase matrices and the inter-layer propagation matrices are block-diagonal in the polarization index, the overall transmitter and receiver transfer matrices factor into two identical single-polarization paths; the end-to-end channel can then be driven toward the truncated singular-value matrix of the physical channel by per-polarization gradient updates. At fixed layer count and unit-cell count this yields up to twice as many interference-free data streams as a single-polarization SIM, and the spectral and energy efficiency follow the theoretical upper bound closely across the simulated polarization-imperfection levels.","pith_inferences":["Because the stack factorizes by polarization, one could run the LGD-WF updates independently on each polarization block, roughly halving the optimization dimension; the paper does not explicitly use this computational saving.","If measurable polarization conversion occurs between layers, the block-diagonal model would need to be replaced by full $2M \\times 2M$ inter-layer blocks; the same gradient framework could still be applied to all four polarization blocks, trading the clean doubling for robustness.","The DPSIM substitution is a drop-in structural upgrade, so the same idea may carry over to other SIM-based systems such as multiuser beamforming, integrated sensing and communication, and semantic communication, potentially doubling their wave-domain degrees of freedom.","A direct testable consequence of the equivalence claim is that a DPSIM with $M$ dual-polarized units per layer should match the channel-fitting quality of a single-polarization SIM with $2M$ units per layer under the same optimization budget."],"forward_implications":["For a fixed physical footprint, DPSIM supports more simultaneous data streams with negligible inter-stream interference than a single-polarization SIM, because each of the two polarization paths provides an independent parallel channel.","The LGD-WF algorithm converges quickly, within about 20 iterations in the simulations, and produces end-to-end channel matrices close to the target diagonal form for both SIM and DPSIM.","Increasing the number of metasurface layers improves channel fitting and interference suppression, but DPSIM reaches the same or better fitting at the same layer count, leaving integration space for other functions.","Spectral and energy efficiency of DPSIM-assisted HMIMO approach the theoretical upper bound for polarization conversion power ratios of 0.2 and 0.4, indicating robustness to polarization cross-interference.","DPSIM-assisted HMIMO achieves higher energy efficiency than conventional massive MIMO because it avoids digital precoding and uses low-resolution data converters."],"supporting_citations":[{"why":"introduces the SIM architecture and the truncated-SVD channel-fitting objective that DPSIM extends and uses as baseline.","marker":"[4]"},{"why":"supplies the four-block polarization channel model (00, 11, 10, 01) and the claim that dual-polarized RIS suppresses polarization cross-interference and inter-stream interference.","marker":"[12]"},{"why":"provides the Rayleigh-Sommerfeld diffraction formula used for the inter-layer transmission coefficients in the transfer matrices.","marker":"[14]"},{"why":"is the basis for assuming the polarization state is unchanged between DPSIM layers, since scatterers are the main polarization-changing mechanism.","marker":"[15]"},{"why":"gives the spatially-stationary sinc correlation model used for the spatial correlation matrices at the metasurface.","marker":"[17]"},{"why":"supports the truncated singular-value decomposition strategy for realizing direct spatial multiplexing without precoding.","marker":"[18]"},{"why":"supplies the water-filling algorithm used for optimal power allocation across data streams.","marker":"[19]"},{"why":"provides the millimeter-wave path loss and shadowing model used in the simulations.","marker":"[22]"}],"fun_headline_variants":["Dual-polarized SIM doubles interference-free streams","Polarization split yields twice the MIMO streams","Dual-pol metasurface stack nears spectral efficiency ceiling","Stacked dual-polarized surfaces double MIMO capacity","Holographic MIMO with dual-polarized SIMs: 2x streams"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes that a wave keeps its polarization state while travelling between metasurface layers, so the inter-layer propagation matrices are block-diagonal and the two polarizations never mix inside the stack; if polarization conversion between layers is significant, the two paths couple and the system is no longer equivalent to two isolated SIMs.","fun_headline_variants_meta":{"raw":{"variants":["Dual-polarized SIM doubles interference-free streams","Polarization split yields twice the MIMO streams","Dual-pol metasurface stack nears spectral efficiency ceiling","Stacked dual-polarized surfaces double MIMO capacity","Holographic MIMO with dual-polarized SIMs: 2x streams"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000551,"raw_usage":{"total_tokens":2596,"prompt_tokens":880,"completion_tokens":1716,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":496,"completion_tokens_details":{"reasoning_tokens":1634}},"tokens_in":496,"tokens_out":1716,"duration_ms":12552,"temperature":1.0,"reasoning_tokens":1634,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T13:47:18.145930+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a full-wave simulation of two adjacent dual-polarized metasurface layers and compute the complete $2M \\times 2M$ transmission matrix; if the off-diagonal polarization blocks have norm comparable to the diagonal blocks, for example at oblique incidence or with imperfect unit-cell isolation, then the block-diagonal model of Eqs. (5)-(6) fails and the claimed equivalence to two isolated SIMs, together with the per-polarization gradient updates, no longer holds.","supporting_citations":[{"cited_title":"Stacked intelligent metasurfaces for efficient holographic MIMO communications in 6G,","cited_arxiv_id":null,"evidence_quote":"introduces the SIM architecture and the truncated-SVD channel-fitting objective that DPSIM extends and uses as baseline."},{"cited_title":"Dual-polarized RIS-assisted mobile communications,","cited_arxiv_id":null,"evidence_quote":"supplies the four-block polarization channel model (00, 11, 10, 01) and the claim that dual-polarized RIS suppresses polarization cross-interference and inter-stream interference."},{"cited_title":"All-optical machine learning using diffractive deep neural networks,","cited_arxiv_id":null,"evidence_quote":"provides the Rayleigh-Sommerfeld diffraction formula used for the inter-layer transmission coefficients in the transfer matrices."},{"cited_title":"Analysis and modeling on co- and cross-polarized urban radio propaga- tion for dual-polarized mimo wireless systems,","cited_arxiv_id":null,"evidence_quote":"is the basis for assuming the polarization state is unchanged between DPSIM layers, since scatterers are the main polarization-changing mechanism."},{"cited_title":"Spatially-stationary model for holographic MIMO small-scale fading,","cited_arxiv_id":null,"evidence_quote":"gives the spatially-stationary sinc correlation model used for the spatial correlation matrices at the metasurface."},{"cited_title":"Low-complexity channel estima- tion and passive beamforming for RIS-assisted MIMO systems relying on discrete phase shifts,","cited_arxiv_id":null,"evidence_quote":"supports the truncated singular-value decomposition strategy for realizing direct spatial multiplexing without precoding."},{"cited_title":"A mathematical theory of communication,","cited_arxiv_id":null,"evidence_quote":"supplies the water-filling algorithm used for optimal power allocation across data streams."},{"cited_title":"Wide- band millimeter-wave propagation measurements and channel models for future wireless communication system design,","cited_arxiv_id":null,"evidence_quote":"provides the millimeter-wave path loss and shadowing model used in the simulations."}],"review_version":1}