{"id":"40d550e2-2147-464d-a135-b1a354ee949c","arxiv_id":"2508.17149","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":4,"one_line_summary":"Simulation results claim an active multi-layer metasurface on a LEO satellite improves spectral and energy efficiency over single-layer RIS designs, but the supporting equations contain serious internal inconsistencies.","lead":"This paper proposes mounting an active stacked-intelligent-metasurface on a LEO satellite to improve signal delivery to ground users and IoT devices, and compares three optimization algorithms by simulation. The practical interest is that, if the simulation were valid, satellite links could carry more data per hertz and per watt, which matters for 6G satellite IoT.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Per-stream ASIM matrices Φ_c and Φ_l in Eqs. (8)/(11) are undefined and conflict with the single-transfer signal model in Eqs. (3)-(5), making the SINR expressions and all downstream optimization invalid.","rationale":"The reader's weakest_assumption identifies exactly the same core flaw: the SINR expressions in Eqs. (8) and (11) rely on per-stream ASIM matrices Φ_c and Φ_l that are not part of the signal model. I agree this is the most load-bearing concern because it invalidates the mathematical foundation of the entire optimization pipeline and performance evaluation. The central claim—that multi-layer sequential processing in ASIM improves effective channel gains and suppresses interference—is built on these SINR equations. If the correct SINR (with a single T) were used, the gains attributed to per-stream processing might vanish, and the comparison against active RIS/BD-RIS would be unfair because those baselines do not have such extra degrees of freedom. The paper provides no independent verification (e.g., no code, no derivations) to show that the simulations faithfully implement the physical model (Eq. 4) rather than the inconsistent SINR. Thus the rejection verdict is appropriate. I also noticed the optimization over ϑ_sat/ϑ_SIM as a secondary issue, but it does not displace the primary concern; it would only further reinforce rejection. Hence, I maintain the reader's verdict of REJECT without modification.","tokens_in":19434,"tokens_out":4753,"duration_ms":49684,"concrete_test":"Re-derive the SINR for the common stream directly from Eqs. (1)-(5): substitute ROut = T F x + nSIM into y_Ul = g_l^H ROut + n_Ul and compute γ_c,l assuming perfect SIC. Confirm whether any Φ_c or Φ_l appear; the correct SINR contains only the shared T. Then rerun the BCD-SCA optimization from Section IV using this corrected SINR (no Φ_c/Φ_l) with the Table III parameters (M=128, L=3, I=3) and compare the resulting SE in Fig. 4(a) to the reported value. If the corrected SE differs by more than 10%, the headline performance gain is an artifact of the undefined per-stream matrices.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The signal model in Section II-A defines a single ASIM transfer matrix T = ∏_{q=Q}^{1} Φ(q) H(q) (Eq. 3), so the ASIM output is ROut = T F x + nSIM (Eq. 4). For user l, the received signal is y_Ul = g_l^H T F x + g_l^H nSIM + n_Ul. With RSMA, the common-stream SINR should therefore be γ_c,l = σc P_c |g_l^H T F w_c|^2 / (Σ_j σj P_j |g_l^H T F w_j|^2 + ||g_l||^2 σ_SIM^2 + σ_Ul^2), and the private-stream SINR similarly. However, Eqs. (8) and (11) introduce per-stream matrices Φ_c and Φ_l that multiply T F w_c and T F w_l separately, and use |g_l^H Φ_l|^2 σ_SIM^2 as ASIM noise. No such per-stream matrices appear in the system model, are not expressed in terms of Φ(q), and are never optimized or constrained. This implies the ASIM applies different linear transformations to each superposed stream in the same transmitted signal, which is physically impossible for a single transfer matrix. Consequently, the SINR expressions do not follow from the stated signal model. The optimization problem (20) and the three solution algorithms (BCD-SCA, MA-CSAC, MCPPO) are all built on these SINRs, so the objective and constraints do not correspond to the modeled system. The reported SE/EE improvements of ASIM over active RIS and BD-RIS are therefore unsupported by the analysis. A secondary issue is that problem (20) lists ϑ_sat and ϑ_SIM as optimization variables even though they are fixed inverse PA efficiencies (Eq. 16); setting them to zero would artificially minimize P_total, gameing the objective.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies a LEO satellite downlink in which an active stacked intelligent metasurface (ASIM), integrated on the satellite's solar-panel backplate, serves L RSMA users and I passive IoT backscatter devices. The system model defines a single ASIM transfer matrix T (Eq. 3) and a single output ROut (Eq. 4); the paper then formulates SINR/rate expressions, a joint optimization problem (20), and three solution algorithms (BCD-SCA, MA-CSAC, MCPPO) to maximize weighted spectral/energy efficiency. Simulations compare these algorithms and claim superiority of ASIM over active RIS and active BD-RIS.","tokens_in":20019,"tokens_out":7345,"duration_ms":88362,"significance":"If the modeling were consistent, the paper would address a relevant 6G IoT/satellite integration problem and provide a useful multi-algorithm benchmark. The ASIM concept is interesting and the paper covers relevant prior work; the inclusion of energy harvesting and power-amplifier modelling is a strength, and the comparative evaluation of three optimization approaches is ambitious. However, the central SINR derivation and the optimization objective are not tied to the signal model, so the reported SE/EE improvements are not supported. I cannot recommend publication in its current form.","major_comments":[{"comment":"The SINR expressions introduce per-stream ASIM matrices Φ_c and Φ_l that never appear in the system model. From Eqs. (3)-(5), y_Ul = g_l^H T F x + g_l^H nSIM + n_Ul, so the SINR must be computed with one common transfer matrix T, and the ASIM noise contribution is ||g_l||^2 σ_SIM^2. Instead, Eq. (8) uses U_c = g_l^H Φ_c T F w_c and U_l = g_l^H Φ_l T F w_l, and Eq. (11) uses a further Φ_ℓ. These matrices are undefined, are not expressed in terms of Φ(q), and are never constrained. Since R_c,l and R_p,l feed constraints (20g)-(20h) and all numerical results, the objective and algorithms optimize a different system from the one modeled. This is a load-bearing modeling error.","section":"II-C, Eqs. (8) and (11)"},{"comment":"The optimization variables in (20) include ϑ_sat and ϑ_SIM, but Eq. (16) defines these as fixed inverse PA efficiencies (1/η_PA). Treating them as decision variables allows the optimizer to set them to zero (or arbitrarily small) to remove the ϑ_sat P_sat + ϑ_SIM P_SIM terms from P_total. No lower bound or feasibility constraint is imposed. The same variables appear in the DRL action space (28). This can artificially deflate P_total and inflate EE, so the reported EE comparisons are not meaningful.","section":"Section III, Problem (20) and Eq. (28)"},{"comment":"The comparison of ASIM versus active RIS/active BD-RIS is not controlled. The text states that the comparison uses 128 elements per surface and 4 metasurfaces for ASIM, so ASIM has 4×128 = 512 active elements, while each baseline uses a single 128-element surface. The horizontal axis is surface transmit power Pmax, but it is not specified whether this is total power or per-element/per-layer power. With 4× the number of active elements, the ASIM has substantially more hardware resources; the observed gain cannot be attributed solely to multi-layer processing or to the proposed algorithm. The claim that ASIM outperforms active RIS/BD-RIS is therefore unsupported. A fair comparison should fix total number of active elements or total power and report per-surface transmit power consistently.","section":"Section VI-D, Fig. 6"},{"comment":"The rate expressions use B log2(1 + γ/B). This is not the Shannon capacity formula, which is B log2(1 + γ); γ is dimensionless, so γ/B has units of 1/Hz and the expression is dimensionally inconsistent. The paper reports rates in bps/Hz, but multiplying by B yields bps. This affects all SE/EE results, including comparisons in Figs. 4-7. Unless the authors define a specific finite-blocklength or bandwidth-normalization convention, the rate model must be corrected.","section":"Section II-C, Eqs. (9), (12), and (15)"}],"minor_comments":[{"comment":"'BSD-SCA' should be 'BCD-SCA'.","section":"Section IV, first paragraph"},{"comment":"The notation is confusing: the constraint is imposed ∀q, but the second term sums over q=1..Q inside each constraint. If the intent is a per-layer power budget, the sum should not be repeated, and the first term should be the power through the complete ASIM transfer T rather than Φ(q) F W.","section":"Eq. (20f)"},{"comment":"The text says 'k-th metasurface layer' but the index is q; please fix the inconsistency.","section":"Eq. (2)"},{"comment":"The BCD-SCA convergence plot labels an 'Asymptotic Optimum' without defining how it is computed; this should be clarified.","section":"Fig. 3(b)"},{"comment":"Propositions 1-2 are stated for general MDPs; the paper does not verify the Lipschitz and bounded-variance assumptions for the specific reward/cost functions, so the guarantees cannot be directly applied to this system.","section":"Section V-D"},{"comment":"The table mixes dBm and mW values without consistent conversion; e.g., Pphs is listed as 7 dBm but the text says 1.5-7.8 mW. Please clarify units.","section":"Table III"}],"recommendation":"reject","confidential_remarks":"The manuscript has a fundamental model mismatch: the SINR expressions in Eqs. (8)/(11) are not derived from the single-transfer-matrix model of Eqs. (3)-(5), and the optimization problem includes fixed hardware efficiencies as decision variables. These are not presentation issues; they invalidate the central claims and would require a full re-derivation and re-simulation. I therefore recommend rejection. If the authors rework the system model and perform fair comparisons, the underlying idea may be worth revisiting."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nQuick take: this paper combines active stacked intelligent metasurfaces, RSMA, and symbiotic radio for a LEO satellite downlink, then runs three optimization algorithms. The components are all known, but the specific combination appears new. The paper is clearly written and the simulation setup is thorough. But there is a load-bearing problem in the SINR derivation that invalidates the reported gains.\n\nWhat it does well: the system model is detailed, the power consumption model is reasonable, and the three-way comparison between BCD-SCA, MA-CSAC, and MCPPO is a useful engineering exercise. I also appreciate the concrete hardware story of mounting the ASIM on the solar panel backplate.\n\nThe problem: Eqs. (3)–(5) define a single ASIM transfer matrix T. The received signal at user l is g_l^H T F x + g_l^H n_SIM + noise. But Eqs. (8) and (11) introduce per-stream matrices Φ_c and Φ_l that multiply T F w_c and T F w_l separately, and they also use |g_l^H Φ_l|^2 for the ASIM noise power. Those matrices appear nowhere in the system model, are never expressed in terms of the layer matrices Φ(q), and are not constrained. Since the ASIM applies one linear transformation to the whole incident signal, per-stream processing is physically impossible. The SINR expressions therefore do not follow from the model, and the rates, the objective, and all three algorithms optimize a system that is not the one described.\n\nSecondary issues: the objective in (20) lists ϑ_sat and ϑ_SIM as optimization variables, but they are fixed inverse PA efficiency constants defined in (16). Optimizing them would let the algorithm game the energy term by pretending the amplifiers are arbitrarily efficient. Also, the head-to-head comparison with active RIS and BD-RIS uses a single 128-element surface for the baselines while the ASIM uses 4×128 = 512 elements. That is an unequal hardware budget, so the superiority claim is built into the setup.\n\nOn the positive side, the paper does not hide these choices; they are in plain sight. If the SINRs are rederived from a single transfer matrix and the baselines are given equal element counts, the framework could support a fair comparison. As it stands, the central claims are unsupported.\n\nI would not cite the results as given, and I would not put it in a reading group as a model of derivation. But it is a real attempt with enough substance that a serious referee could help the authors fix it. I would send it to peer review, expecting major revision. The reviewer's main job should be to force a clean signal model and equal-resource baselines.","headline":"Solid system concept, but the SINR derivation contradicts the signal model and the hardware comparison is uneven—performance gains as reported don't hold up.","tokens_in":20452,"tokens_out":3260,"would_cite":false,"duration_ms":31383,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"An active stacked intelligent metasurface (ASIM) mounted on a LEO satellite's solar-panel backplate can lift both spectral and energy efficiency above what active RIS or beyond-diagonal RIS achieve, by processing the wavefront in several tu","keywords":["active stacked intelligent metasurface","LEO satellite","rate-splitting multiple access","symbiotic radio","IoT backscatter","spectral efficiency","energy efficiency","deep reinforcement learning"],"falsifier":"The decisive check is to rerun the simulations using only the single shared transfer matrix T described in the signal model, with no separate matrices for the common and private streams; if ASIM no longer beats a single-layer active RIS at equal power, the multi-layer advantage is an artifact of that unmodeled extra freedom. A hardware or full-wave test at 20 GHz with four stacked active layers would settle it directly.","tokens_in":19406,"feed_emoji":"🛰️","tokens_out":6397,"duration_ms":70261,"temperature":0.7,"pith_summary":"The paper argues that equipping a low Earth orbit satellite with an active stacked intelligent metasurface—a cascade of programmable surfaces that amplifies and shapes the signal as it passes through—can serve both cellular users and passive IoT devices in one spectrum. The claim is that the multi-layer sequence gives the satellite finer beam control, improving effective channel gains and suppressing inter-user interference, so it outperforms single-layer active RIS and beyond-diagonal RIS alternatives. To make the design work, the authors formulate a non-convex resource-allocation problem and test three solvers: a classic block-coordinate convex-approximation method and two deep reinforcement learning policies. The simulations say BCD-SCA converges fastest and most stably, MA-CSAC achieves the highest long-term spectral and energy efficiency at scale, and MCPPO is a middle ground. If right, this points to a scalable, energy-efficient route to next-generation LEO satellite communication for IoT and cellular users.","feed_headline":"Multi-layer active SIM edges out RIS in LEO satellite links","feed_subtitle":"Stacked metasurface boosts channel gains and cuts interference for satellite IoT and cellular links.","key_machinery":"The central object is the ASIM's overall transfer matrix T = ∏_{q=Q}^{1} Φ(q) H(q), where Φ(q) is the diagonal amplitude-and-phase tuning of layer q and H(q) is the inter-layer coupling between surfaces. Because the signal passes through several active layers, each layer's tuning pattern can be adjusted sequentially, which is what produces stronger effective channels and lower inter-user interference. The optimization then tunes the satellite precoder W, the ASIM layers {Φ(q)}, and the backscatter parameters to maximize a weighted sum of throughputs minus power consumption.","core_discovery":"The central discovery is that multi-layer sequential processing inside an ASIM is a better use of satellite surface power than single-layer active RIS or block-diagonal active RIS. The transfer matrix T = product over layers of diagonal tuning matrices Φ(q) and inter-layer coupling matrices H(q) lets each layer reshape the wavefront, so the composite can strengthen weak channels and cancel interference across users. In the paper's simulations, ASIM achieves higher spectral efficiency than Active RIS and Active BD-RIS at equal surface transmit power, and the best optimization strategy—MA-CSAC—balances energy and spectral efficiency in larger networks. The paper embeds this in a system serving","pith_inferences":["A reader should not assume the per-stream processing is physically available: the SINR formulas use separate ASIM matrices for the common and each private stream (Φc, Φl), while the signal model and hardware description contain only one shared transfer matrix T. If the separate matrices are not realizable, the reported gains are optimistic.","The paper's SE–EE plots assume static channel snapshots per optimization step; LEO satellites sweep across the sky, so a natural extension is to test how fast the ASIM layers and beamformer can be retuned against Doppler and elevation changes.","Because the whole advantage over active RIS rests on inter-layer coupling matrices H(q), a direct measurement of those near-field couplings in a prototype would turn the architecture's benefit into an engineering quantity rather than a simulation parameter."],"forward_implications":["If the central claim is correct, ASIM-equipped LEO satellites can offer higher spectral efficiency than active RIS or beyond-diagonal RIS at the same surface transmit power.","Splitting power amplification between the satellite's main amplifier and the ASIM can reduce the power amplifier burden enough that solar harvesting remains a viable energy source.","RSMA combined with ASIM is more energy-efficient than NOMA combined with ASIM, because common-stream decoding handles interference better in the multi-user setting.","The choice of optimizer matters: BCD-SCA fits convex, energy-constrained regimes, MA-CSAC fits large dynamic networks, and MCPPO suits fast-deployment scenarios."],"supporting_citations":[{"why":"Introduces stacked intelligent metasurfaces as layered holographic MIMO surfaces, which the ASIM design extends.","marker":"[15]"},{"why":"Applies stacked intelligent metasurfaces to LEO satellite links with statistical CSI, grounding the satellite-specific use of SIM.","marker":"[23]"},{"why":"Analyzes active-RIS-assisted rate-splitting multiple access spectral and energy tradeoffs, supplying the active-element power model and RSMA framing.","marker":"[32]"},{"why":"Provides the UAV-mounted BD-active-RIS RSMA baseline and DRL optimization approach that the paper compares against and extends.","marker":"[22]"},{"why":"Shows RIS-assisted satellites for IoT networks, motivating the energy-efficiency problem for satellite IoT.","marker":"[16]"},{"why":"Optimizes energy efficiency in RIS-assisted LEO satellite NOMA, providing a baseline approach for satellite RIS optimization.","marker":"[17]"},{"why":"Defines symbiotic radio as the communication paradigm for passive IoT devices used in the IoT sub-network.","marker":"[8]"},{"why":"Supplies proximal policy optimization, the base algorithm extended into MCPPO for constrained optimization.","marker":"[36]"},{"why":"Supplies constrained policy optimization for guaranteeing long-term constraint satisfaction, used in the DRL theoretical guarantees.","marker":"[37]"}],"fun_headline_variants":["ASIM beats RIS in LEO satellite spectral efficiency","Stacked metasurface outperforms active RIS for satellite IoT","MA-CSAC maximizes LEO satellite energy-spectral efficiency","Multi-layer active SIM improves LEO satellite links","Active stacked metasurface edges out RIS in LEO"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The optimization depends on the ASIM being able to apply different processing to the common stream and to each private stream simultaneously, but the signal model uses one shared ASIM transfer matrix and the hardware description does not show how the separate per-stream processing would be realized.","fun_headline_variants_meta":{"raw":{"variants":["ASIM beats RIS in LEO satellite spectral efficiency","Stacked metasurface outperforms active RIS for satellite IoT","MA-CSAC maximizes LEO satellite energy-spectral efficiency","Multi-layer active SIM improves LEO satellite links","Active stacked metasurface edges out RIS in LEO"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000234,"raw_usage":{"total_tokens":1335,"prompt_tokens":751,"completion_tokens":584,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":495,"completion_tokens_details":{"reasoning_tokens":506}},"tokens_in":495,"tokens_out":584,"duration_ms":6834,"temperature":1.0,"reasoning_tokens":506,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T16:59:30.095884+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"The decisive check is to rerun the simulations using only the single shared transfer matrix T described in the signal model, with no separate matrices for the common and private streams; if ASIM no longer beats a single-layer active RIS at equal power, the multi-layer advantage is an artifact of that unmodeled extra freedom. A hardware or full-wave test at 20 GHz with four stacked active layers would settle it directly.","supporting_citations":[{"cited_title":"Stacked Intelligent Metasurfaces fo r efficient holographic MIMO communications in 6G,","cited_arxiv_id":null,"evidence_quote":"Introduces stacked intelligent metasurfaces as layered holographic MIMO surfaces, which the ASIM design extends."},{"cited_title":"Stacked Int elligent Metasurface enabled LEO satellite communications relying on statistical CSI,","cited_arxiv_id":null,"evidence_quote":"Applies stacked intelligent metasurfaces to LEO satellite links with statistical CSI, grounding the satellite-specific use of SIM."},{"cited_title":"Active RIS assisted Rate-Splitting Multiple Access network: Spectral and energy efficiency tradeoff,","cited_arxiv_id":null,"evidence_quote":"Analyzes active-RIS-assisted rate-splitting multiple access spectral and energy tradeoffs, supplying the active-element power model and RSMA framing."},{"cited_title":"Energy Efficient RSMA-Based LEO Satellite Communications Assisted by UAV-Mounted BD-Active RIS: A DRL Approach","cited_arxiv_id":"2505.04148","evidence_quote":"Provides the UAV-mounted BD-active-RIS RSMA baseline and DRL optimization approach that the paper compares against and extends."},{"cited_title":"Energy- efficient RIS- assisted satellites for IoT networks,","cited_arxiv_id":null,"evidence_quote":"Shows RIS-assisted satellites for IoT networks, motivating the energy-efficiency problem for satellite IoT."},{"cited_title":"RIS-assisted energy-efficient LEO satellite commun ications with NOMA,","cited_arxiv_id":null,"evidence_quote":"Optimizes energy efficiency in RIS-assisted LEO satellite NOMA, providing a baseline approach for satellite RIS optimization."},{"cited_title":"Symbi otic Radio: A new communication paradigm for passive Internet of Things,","cited_arxiv_id":null,"evidence_quote":"Defines symbiotic radio as the communication paradigm for passive IoT devices used in the IoT sub-network."},{"cited_title":"Constraine d Policy Op- timization,","cited_arxiv_id":null,"evidence_quote":"Supplies constrained policy optimization for guaranteeing long-term constraint satisfaction, used in the DRL theoretical guarantees."}],"review_version":1}