{"id":"1d0bc3cd-9c79-44f0-a456-9d3d5f008bbb","arxiv_id":"2501.02742","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"A LEO satellite with a transmissive beyond-diagonal RIS and NOMA is optimized via power allocation and phase-shift design, but the derivations contain unit and rank-recovery errors.","lead":"This paper proposes an optimization framework for a low-Earth-orbit satellite that uses a transmissive beyond-diagonal reconfigurable intelligent surface and non-orthogonal multiple access to serve two IoT devices, jointly tuning transmit power and phase shifts to maximize spectral efficiency. The numerical comparison only shows that joint power and phase optimization beats a fixed-power baseline, not that this specific RIS architecture is superior.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"SDR recovery in Section III-C yields a phase matrix that violates the unitary constraint, so the claimed spectral-efficiency gains may be artifacts of the relaxation.","rationale":"The reader's verdict identifies the SDR recovery as the weakest assumption; this stress-test agrees. The abstract and conclusion claim that power allocation and phase shifts are jointly optimized to improve spectral efficiency. For the phase half of that claim to hold, the algorithm must output a feasible Phi_t. The SDR relaxation and the reconstruction are the only bridge from the relaxed variable W back to the system model; without a tightness proof, a rank-recovery procedure, or a randomization/rounding step, the bridge is missing. The concern is not merely about optimality: because any W with Tr(W)=K is admitted, an infeasible W can have a strictly larger objective than any feasible phase matrix, so the numerical gap versus the benchmark can be produced entirely by the relaxation. This is load-bearing regardless of the independent error in Eq. (17), which is a separate issue. We therefore recommend no change to the reader's REJECT verdict.","tokens_in":10091,"tokens_out":12386,"duration_ms":134400,"concrete_test":"Recompute one Monte Carlo point from Fig. 2 (e.g., K=64, P_t=20 W) exactly as described in Section III-C. After CVX returns W*, compute Phi_t = U diag(sqrt(Sigma)) via Eq. (37) and evaluate the feasibility residual ||Phi_t Phi_t^H - I_K||_F and the achievable spectral efficiency at this reconstructed Phi_t. If the residual is not at numerical zero, or if the reconstructed objective is strictly below the relaxed P2.3 objective, the SDR is not tight and the reported curve is not attainable by a feasible T-BD-RIS. A useful variant is to add the constraint W = I_K (or project W* onto the unitary set) and re-run Fig. 2; the gap against the benchmark should then be re-examined.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central numerical claim is carried by the phase-design step in Section III-C. As written, constraint C2 forces W = Phi_t Phi_t^H to be the identity matrix I_K for every feasible Phi_t. The SDR in P2.3 replaces this with all PSD matrices satisfying Tr(W)=K, a strictly larger set whose optimizers generally are not of the form Phi_t Phi_t^H with Phi_t feasible. The recovery step (36)-(37) only factorizes W as U Sigma U^H and sets Phi_t = U diag(sqrt(Sigma)); the reconstructed matrix then satisfies Phi_t Phi_t^H = U Sigma U^H = W, not I_K. Since Sigma is not shown to be I_K and no randomization or projection onto the unitary set is provided, the 'optimal framework' may evaluate a phase matrix that does not exist under the system model. The relaxation gap can only increase the objective, so the improvement over the benchmark in Figs. 2-3 may be an artifact of infeasible W rather than a real gain from joint power and phase optimization.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers a LEO satellite equipped with a transmissive beyond-diagonal RIS (T-BD-RIS) serving two NOMA IoT devices, and it formulates the joint optimization of the NOMA power fractions and the RIS phase-shift matrix to maximize spectral efficiency. The solution strategy is to apply successive convex approximation, split the problem into power allocation and phase-shift design, solve the power subproblem via KKT conditions in closed form, and solve the phase-shift subproblem by semidefinite relaxation. The authors report numerical results showing that the joint optimization outperforms a benchmark with fixed power allocation and optimized phase shifts.","tokens_in":10211,"tokens_out":5551,"duration_ms":60613,"significance":"If the claims were valid, the paper would address a timely scenario—RIS-assisted LEO satellite NOMA—and would offer a tractable two-step optimization procedure for a nonconvex problem. The identification of the scenario and the explicit formulation of the coupled power-and-phase problem are useful starting points. However, the central numerical claim is not supported by the derivations as written: the power-splitting formula is dimensionally inconsistent, and the semidefinite-relaxation recovery step returns phase matrices that violate the imposed unitary constraint. Because the reported gains in Figures 2 and 3 rest on these two steps, the paper's main conclusion currently has no sound basis. No code, proofs of SDR tightness, or convergence analysis are provided, so the numerical experiments cannot be independently checked.","major_comments":[{"comment":"The SDR step is invalid and this directly affects the central claim. Constraint C2 in (7) requires Phi_t Phi_t^H = I_K, so every feasible Phi_t gives W = Phi_t Phi_t^H = I_K. The relaxed problem P2.3, however, only imposes W ⪰ 0 and Tr(W) = K, which is a strictly larger feasible set. The reconstruction in (36)–(37) produces Phi_t with Phi_t Phi_t^H = U Sigma U^H = W, and since W is not shown to equal I_K, the reconstructed matrix generally violates C2. No tightness proof, randomization procedure, or projection onto the unitary set is provided. Consequently, the spectral-efficiency improvements attributed to the proposed framework in Figures 2 and 3 may be artifacts of evaluating infeasible matrices rather than genuine gains from joint power-and-phase optimization.","section":"§III-C, Eqs. (34)–(37)"},{"comment":"Equation (17), p_j^* = (1 - p_i^*) P_t, is dimensionally inconsistent and violates constraint C3 in (9). The variables p_i and p_j are dimensionless power coefficients satisfying p_i + p_j <= 1, as stated in Section II. Substituting the simulated value P_t = 20 W, the formula gives p_j^* = 20 (1 - p_i^*) W, which is not a power fraction, can greatly exceed 1, and does not respect C3. The correct update is p_j^* = 1 - p_i^*. Since this formula determines the NOMA power allocation for all subsequent results, the KKT-based power solution, and hence the numerical comparison, is not valid as written.","section":"§III-B, Eq. (17)"},{"comment":"The dimensions of the channel vectors and the phase-shift matrix are inconsistent. The paper defines h_eta in C^{K×1} but then writes y_i = h_i Phi_t x, which requires h_i to be a row vector of length compatible with Phi_t; simultaneously Phi_t is defined in C^{K_x × K_y} and constraint C2 requires Phi_t Phi_t^H = I_K. These conditions are mutually incompatible unless K_x = K_y = K, while the Kronecker form in (4) implies a channel vector of length K_x K_y. The system model needs one consistent convention for the sizes of h_eta and Phi_t before the objective and constraints can be evaluated.","section":"§II and §III-A, Eqs. (1)–(7)"},{"comment":"The convexity classification used to justify the Taylor approximation is stated incorrectly. The term g(W) in (28) is a concave function of W because it is the logarithm of an affine positive function, yet it is described as the convex part. This misstatement does not by itself invalidate the resulting surrogate, but it indicates that the SCA derivation needs revision, and the convergence properties of the alternating procedure are not discussed.","section":"§III-C, Eq. (27)"}],"minor_comments":[{"comment":"The notation h_eta is used inconsistently: it is declared as a column vector in C^{K×1}, but later expressions such as h_i Phi_t x require a row-vector convention. The authors should write h_eta^T or define h_eta as a row vector consistently throughout.","section":"§II, Eq. (4)"},{"comment":"Constraint C4, P_t (p_i + p_j) <= P_max, is redundant with C3 when P_max >= P_t, and no value of P_max is given in the numerical setup. The distinction between the two constraints should be clarified.","section":"§III-A, P0"},{"comment":"The benchmark framework is only described as 'fixed NOMA power allocation,' but the specific fixed power split used to generate Figures 2 and 3 is not stated, which makes the comparison difficult to reproduce.","section":"§IV"},{"comment":"The phrase 'removing the rank-1 constraint' is imprecise: the rank of W = Phi_t Phi_t^H is governed by the dimensions of Phi_t and is not necessarily 1, and the more important missing condition is the unitary structure C2 itself.","section":"§III-C"}],"recommendation":"reject","confidential_remarks":"The paper is not ready for publication in its current form. The SDR recovery issue in Section III-C is load-bearing: the proposed phase-shift design can return matrices that are infeasible under the system model, so the numerical gains over the benchmark may be spurious. The power-allocation formula in Eq. (17) is also dimensionally wrong. These are not presentation issues but fundamental correctness concerns that would require reworking the optimization methodology and the numerical comparisons. I would not encourage acceptance even with minor revision; a substantial revision that proves the tightness of the relaxation or replaces it with a valid unitary-recovery step, corrects the power split, and restates the system model consistently could be reconsidered as a new submission."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the system-level idea—putting a transmissive beyond-diagonal RIS on a LEO satellite and serving two NOMA IoT users—is a reasonable niche to explore, and the paper is the first to put that exact label together. The literature review is fine, and the decomposition into power control and phase design is the standard SCA/KKT/SDR pipeline. But as written, the math does not hang together. I agree with the reader's rejection.\n\nThe main problem is in Section III-C. The constraint is Phi_t Phi_t^H = I_K. That makes W = Phi_t Phi_t^H identically I_K, so the objective in P2.1 is constant in Phi_t. The SDR then relaxes W to any PSD matrix with trace K, solves a different problem, and reconstructs Phi_t via (36)-(37). The reconstructed Phi_t satisfies Phi_t Phi_t^H = W, not I_K, so the \"optimal\" phase matrix violates the system model unless W happens to be I_K, which is not shown and generally false. No randomization or projection onto the unitary set is provided, so the gains in Figs. 2-3 may be artifacts of an infeasible relaxation. This is load-bearing, not a typo.\n\nThere are also smaller but real issues. Eq. (17) sets p_j^* = (1-p_i^*) P_t, mixing a dimensionless power-split coefficient with total power; it can violate the budget and the units. The KKT closed form in (13) leaves all multipliers unspecified and has a ± root with no selection rule. And the dimensions of the channel vectors and the phase matrix are never reconciled: h_i is K×1, Phi_t is Kx×Ky, and the product h_i^H Phi_t only makes sense if K=Kx, while the Kronecker model in (4) suggests K=Kx*Ky. The paper also compares only joint optimization against fixed-power phase optimization, so it does not isolate the claimed T-BD-RIS benefit.\n\nCredit where due: the topic is timely, the references are relevant, and the algorithmic skeleton (SCA lower bound, KKT power, SDR phase) is the right toolkit. None of that rescues the central step. I would not send this to peer review in its current form; a desk reject with a clear list of these issues is fair. If the authors fix the phase recovery, correct the power formula, and add diagonal-RIS/no-RIS baselines, it could be worth a second look.","headline":"The T-BD-RIS + NOMA LEO combination is new, but the phase-design optimization is mathematically broken as written, so the reported gains are not credible.","tokens_in":10835,"tokens_out":7066,"would_cite":false,"duration_ms":70694,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that in a two-user NOMA downlink from a LEO satellite equipped with a transmissive beyond-diagonal RIS, jointly optimizing the NOMA power split and the RIS phase-shift matrix improves spectral efficiency over optimizing…","keywords":["LEO satellite communications","transmissive beyond diagonal RIS","NOMA power allocation","spectral efficiency optimization","semidefinite relaxation","Internet of Things","phase shift design"],"falsifier":"Take a small instance, say $K=4$ and two users; solve the relaxed problem P2.3 and inspect the eigenvalues of the optimal $\\mathbf{W}$. If in a nontrivial fraction of random Rayleigh-fading channel draws the second-largest eigenvalue is not negligible, the reconstruction $\\boldsymbol{\\Phi}_t = \\mathbf{U} \\mathrm{diag}(\\sqrt{\\boldsymbol{\\Sigma}})$ does not recover the relaxed objective, and the claimed optimality of the phase design fails at that instance.","tokens_in":9779,"feed_emoji":"🛰️","tokens_out":6638,"duration_ms":61183,"temperature":0.7,"pith_summary":"This paper aims to show that a LEO satellite serving two Internet-of-Things devices under downlink non-orthogonal multiple access (NOMA) can achieve higher spectral efficiency when both the transmit-power split and the phase-shift matrix of a transmissive beyond-diagonal reconfigurable intelligent surface (T-BD-RIS) are optimized together. The authors construct a non-convex sum-rate optimization, approximate it with successive convex approximation, split it into a power-allocation problem and a phase-design problem, and solve the former in closed form from KKT conditions and the latter by semidefinite relaxation. Monte Carlo simulations compare this joint design against a benchmark that keeps NOMA power fixed and optimizes only the phases, and report a consistent gain that grows with transmit power and with the number of RIS elements. The significance of this claim is that satellite downlinks to small battery-powered IoT devices are link-budget-limited, and a transmission-mode RIS that avoids reflective self-interference may be a practical way to improve spectrum use without ground infrastructure.","feed_headline":"Joint NOMA power and RIS phase tuning lifts LEO spectral efficiency","feed_subtitle":"Jointly optimizing satellite transmit power and RIS phase shifts outperforms phase-only optimization in Monte Carlo trials.","key_machinery":"The central object is the T-BD-RIS phase-shift matrix $\\boldsymbol{\\Phi}_t \\in \\mathbb{C}^{K_x \\times K_y}$ satisfying the semi-unitary constraint $\\boldsymbol{\\Phi}_t \\boldsymbol{\\Phi}_t^{H} = \\mathbf{I}$, with $K = K_x K_y$. For optimization, the paper defines $\\mathbf{W} = \\boldsymbol{\\Phi}_t \\boldsymbol{\\Phi}_t^{H}$ and relaxes the rank constraint, turning the phase design into the semidefinite program P2.3 (maximize a concave objective in $\\mathbf{W}$ subject to $\\mathbf{W} \\succeq 0$, $\\mathrm{Tr}(\\mathbf{W}) = K$). The reconstruction step $\\mathbf{W} = \\mathbf{U} \\boldsymbol{\\Sigma} \\mathbf{U}^{H}$ and $\\boldsymbol{\\Phi}_t = \\mathbf{U} \\mathrm{diag}(\\sqrt{\\boldsymbol{\\Sigma}})$ carries the argument, because it must return a feasible phase-shift matrix from the relaxed solution. The transmit-power side is carried by the KKT conditions applied to a Lagrangian of the power-allocation subproblem, yielding a closed form for $p_i^*$ and then $p_j^* = (1 - p_i^*) P_t$.","core_discovery":"On its own terms, the paper's finding is that a joint design, using a closed-form NOMA power allocation and a semidefinite-relaxation phase-shift design for a T-BD-RIS, yields the highest spectral efficiency among the compared schemes for a LEO satellite transmitting to two NOMA users. The effective channel gains are written as $\\mathrm{Tr}(\\mathbf{W} \\mathbf{F}_\\eta)$ with $\\mathbf{W} = \\boldsymbol{\\Phi}_t \\boldsymbol{\\Phi}_t^{H}$, so the phase design becomes a semidefinite program with $\\mathbf{W} \\succeq 0$ and $\\mathrm{Tr}(\\mathbf{W}) = K$, and $\\boldsymbol{\\Phi}_t$ is reconstructed by eigenvalue decomposition. Because the benchmark fixes the NOMA power allocation and only optimizes phases, the reported gap is evidence that power allocation and phase design should be solved jointly rather than sequentially with ad hoc power splitting.","pith_inferences":["A straightforward extension, which the paper lists as future work, is multi-carrier NOMA; since the two-user per-carrier structure repeats, the same power/phase decomposition is likely to carry over with per-carrier power splits.","The size of the reported gain is tied to the benchmark's fixed power allocation; a benchmark that re-optimized power by exhaustive search would give a sharper measure of how much the joint design adds.","If the SDR tightness assumption holds empirically, the same rank-one reconstruction trick could be applied to other semi-unitary RIS designs, including reflective and hybrid beyond-diagonal RIS, where the constraint set has the same algebraic form.","An immediate validation step is to test the SDR reconstruction against a grid search or local search over feasible phase matrices on small $K$; any systematic gap would indicate that the proposed algorithm underdelivers relative to the relaxed bound."],"forward_implications":["If the algorithm works as claimed, LEO IoT downlinks can use a transmission-mode RIS and NOMA to improve spectral efficiency without additional ground hardware.","The reported gain over the benchmark grows with both LEO transmit power (5 to 30 W) and RIS element count (16 to 144), so the joint design matters most in the regimes where link budgets are tight.","The QoS constraint $R_\\eta \\ge R_{\\min}$ is carried through the SCA reformulation, so the optimized operating point respects minimum-rate requirements for both users.","Because the phase design reduces to a convex semidefinite program at each SCA iteration, the method is implementable with standard convex solvers and has bounded per-iteration complexity."],"supporting_citations":[{"why":"Shows BD-RIS outperforms diagonal RIS in beamforming and capacity, motivating the phase-structure choice in this paper.","marker":"[16]"},{"why":"Identifies the self-interference problem in reflective BD-RIS that transmissive operation avoids.","marker":"[20]"},{"why":"Uses RIS-assisted satellite links to overcome path loss, establishing the baseline scenario this paper extends.","marker":"[22]"},{"why":"Optimizes transmit power and passive beamforming jointly in RIS satellite networks, the algorithmic template for joint power and phase design.","marker":"[23]"},{"why":"Maximizes energy efficiency in RIS-assisted NOMA satellite networks by alternating power and phase optimization, a direct precursor.","marker":"[24]"},{"why":"Proposes transmissive RIS mounted LEO satellite communication maximizing sum rate, the closest prior architecture.","marker":"[28]"},{"why":"Applies NOMA and transmissive RIS to a cognitive LEO satellite network, motivating NOMA with transmissive RIS.","marker":"[29]"}],"fun_headline_variants":["Joint NOMA power and RIS phase design boosts LEO spectral efficiency","T-BD-RIS: Joint power-phase tuning beats phase-only in LEO NOMA","LEO NOMA gets spectral edge from joint power-RIS phase optimization","Beyond diagonal RIS: joint power-phase tuning lifts LEO NOMA","Closed-form NOMA power plus RIS phase beats phase-only in LEO"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method depends on the assumption that the relaxed mathematical version of the phase-shift problem has a solution that can be factored back into a phase-shift matrix satisfying the original constraint without losing performance.","fun_headline_variants_meta":{"raw":{"variants":["Joint NOMA power and RIS phase design boosts LEO spectral efficiency","T-BD-RIS: Joint power-phase tuning beats phase-only in LEO NOMA","LEO NOMA gets spectral edge from joint power-RIS phase optimization","Beyond diagonal RIS: joint power-phase tuning lifts LEO NOMA","Closed-form NOMA power plus RIS phase beats phase-only in LEO"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001021,"raw_usage":{"total_tokens":4296,"prompt_tokens":923,"completion_tokens":3373,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":539,"completion_tokens_details":{"reasoning_tokens":3275}},"tokens_in":539,"tokens_out":3373,"duration_ms":23064,"temperature":1.0,"reasoning_tokens":3275,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:06:29.746825+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a small instance, say $K=4$ and two users; solve the relaxed problem P2.3 and inspect the eigenvalues of the optimal $\\mathbf{W}$. If in a nontrivial fraction of random Rayleigh-fading channel draws the second-largest eigenvalue is not negligible, the reconstruction $\\boldsymbol{\\Phi}_t = \\mathbf{U} \\mathrm{diag}(\\sqrt{\\boldsymbol{\\Sigma}})$ does not recover the relaxed objective, and the claimed optimality of the phase design fails at that instance.","supporting_citations":[{"cited_title":"Beyond diagonal RIS for 6G non-terrestrial networks: Potentials and challenges,","cited_arxiv_id":null,"evidence_quote":"Shows BD-RIS outperforms diagonal RIS in beamforming and capacity, motivating the phase-structure choice in this paper."},{"cited_title":"MIMO transmission through reconfigurable intelligent surface: System design, analysis, and implementation,","cited_arxiv_id":null,"evidence_quote":"Identifies the self-interference problem in reflective BD-RIS that transmissive operation avoids."},{"cited_title":"Energy-efficient RIS-assisted satellites for IoT networks,","cited_arxiv_id":null,"evidence_quote":"Uses RIS-assisted satellite links to overcome path loss, establishing the baseline scenario this paper extends."},{"cited_title":"RIS-assisted energy-efficient LEO satellite commu- nications with NOMA,","cited_arxiv_id":null,"evidence_quote":"Optimizes transmit power and passive beamforming jointly in RIS satellite networks, the algorithmic template for joint power and phase design."},{"cited_title":"Energy effi- ciency design in RIS-assisted satellite–terrestrial integrated networks with NOMA,","cited_arxiv_id":null,"evidence_quote":"Maximizes energy efficiency in RIS-assisted NOMA satellite networks by alternating power and phase optimization, a direct precursor."},{"cited_title":"Transmissive RIS-empowered LEO-satellite communica- tions with hybrid-NOMA under residual hardware impairments,","cited_arxiv_id":null,"evidence_quote":"Proposes transmissive RIS mounted LEO satellite communication maximizing sum rate, the closest prior architecture."}],"review_version":1}