{"id":"5535f827-0ee3-451a-a154-edb6a7ca3222","arxiv_id":"2607.28965","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Scaling 5G MIMO to 256+ ports at 7–8 GHz breaks on common-channel coverage, RF/array power, and CSI overhead; the paper maps a roadmap that decouples radiating elements, RF chains, and acquired channel dimensions.","lead":"This paper explains why simply scaling 5G's many-antenna systems to 256+ ports in the 7–8 GHz band breaks down — coverage for control channels, RF/array power, and channel feedback all fail — and outlines a 6G research roadmap around 'extreme MIMO'. It also contributes a coverage simulation and a 3D-Gaussian-splatting method that renders channel statistics at unvisited locations, both directly relevant to 6G standardization.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"CSI roadmap hinges on untested sublinear rank growth at 256+ ports; the paper's own Sec. II-B admits no data at this scale, so the central decoupling claim is not yet secured.","rationale":"The paper is an invited roadmap, and much of its qualitative diagnosis is well supported by cited measurements and industry prototypes; I do not see a basis for rejection. I focused on the condition that must be true for the central constructive claim—that CSI acquisition and active-chain count can decouple from the radiating aperture. That condition is the low-rank/sublinear-r assumption in Sec. VI. It is also the one the authors explicitly flag as unmeasured at 768–1024 elements, so it is not a manufactured objection. The PDSCH coverage number is a genuine weakness, but the SSB/CSI-RS medians already establish the coverage-asymmetry point, and the paper itself cautions against reading the CDFs as optimized deployments. The 3D-GS result lacks released code and metrics, but it is presented as a preliminary direction, not as a load-bearing part of the central claim. The reader's verdict already conditions on the same rank assumption; my read therefore leaves CONDITIONAL unchanged. The concrete test above would either retire the concern (if r remains small at high element counts) or sharpen it into a central quantitative failure.","tokens_in":34232,"tokens_out":4978,"duration_ms":46954,"concrete_test":"Using the same Herald Square ray-tracing setup as Table II (or the dual-band UMa dataset of [8]/[21]), compute the covariance eigenvalue decay for 7 GHz URAs of 256, 512, and 1024 elements occupying the same physical aperture: r_ε = min{r: Σ_{i≤r} λ_i ≥ 0.9 Σλ_i} per user location and environment. Then check whether r_ε(N_t) is sublinear (e.g., r_1024 < 2×r_256) or roughly linear. If it is linear, repeat with ε=0.01; if even then r_ε≈N_t, Eq. (3) and the payload-saving factor in Eq. (8) lose their basis and Section VI's scaling claim collapses.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central roadmap depends on Eq. (3), h≈U_r g, with r growing sublinearly in N_t so that CSI-RS, feedback, and prediction scale with effective rank rather than port count. This is the least secured premise: the evidence cited in Sec. II-B comes from 32/128-element dual-band measurements and ray tracing, while the paper itself admits that 'the covariance at this scale calls for further measurement and analysis' for 768–1024 elements. Moreover, Sec. II-B states that outdoors the rank 'is maintained or even increases' under equal-aperture scaling, which is compatible with r growing nearly proportionally with N_t if the number of independently resolvable paths scales with aperture. If at 256+ ports the 90% energy dimension r is proportional to N_t, then Eq. (8)'s ε_r floor forces r≈N_t at operating SNR, the claimed N_t/r overhead saving disappears, and the CSI-acquisition half of the paper's decoupling principle is unsupported. The PDSCH -37.7 dB artifact is a real caveat, but it is explicitly labeled an unoptimized illustration; the rank assumption is the condition on which the proposed architecture's feasibility actually rests.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This invited paper argues that a direct extension of the 5G NR architecture to 6G upper-mid-band (FR3, 7–8 GHz) extreme MIMO with 256+ ports runs into four coupled system-level limitations: coverage asymmetry across physical channels and protocol states, wideband/energy-efficient RF and RU implementation, array/beamforming power consumption, and CSI-acquisition overhead. The authors propose a common decoupling principle—N_rad, N_port, N_RF, and the effective channel rank r need not scale together—and illustrate it with protocol-aware beam management, tri-hybrid MIMO architectures, distributed apertures, and AI/scene-based CSI acquisition. The quantitative support consists of a ray-tracing coverage comparison (Fig. 2) and a 3D-GS beamspace-rendering proof-of-concept (Figs. 8–9).","tokens_in":34564,"tokens_out":8627,"duration_ms":85884,"significance":"If the decoupling principle holds, the paper provides a useful organizing framework for FR3 E-MIMO and clearly separates the dimensions that must scale from those that should not. Its strengths are the comprehensive synthesis of FR3 spectrum, propagation measurements, and Release 19 channel-model changes; the explicit four-breakpoint formulation; clean mathematical models for tri-hybrid MIMO (Eq. 1), power accounting (Eq. 2), and subspace-limited feedback (Eqs. 3, 6–8); and the candid labeling of the coverage simulation as unoptimized and illustrative. The central risk is that the CSI roadmap rests on an untested scaling law for channel rank at 768–1024 elements, a premise the paper itself flags as open in Sec. II-B.","major_comments":[{"comment":"The entire CSI-acquisition roadmap (beamformed CSI-RS, reduced feedback, 3D-GS rendering) depends on the assumption that the effective rank r in h ≈ U_r g grows sublinearly in the port count N_t, so that the N_t/r overhead saving 'widens along the E-MIMO trajectory.' The evidence cited in Sec. II-B comes from 32- and 128-element dual-band measurements and ray tracing; for 768–1024 elements the text explicitly states that 'the covariance at this scale calls for further measurement and analysis.' Moreover, Sec. II-B also reports that outdoors the rank 'is maintained or even increases' under equal-aperture scaling, which is at least compatible with r growing nearly as fast as N_t. If r ~ N_t at 256+ ports, then Eq. (8)'s floor ε_r cannot be held below 1/ρ at operating SNR without taking r ≈ N_t, and the claimed overhead savings disappear. This is a load-bearing premise for the paper's most","section":"Sec. VI-A and Sec. II-B"},{"comment":"The quantitative coverage study is based on a single ray-tracing environment (Herald Square) with no confidence intervals, and the PDSCH median of -37.7 dB is produced by a deliberately unoptimized baseline (random scheduling, block-diagonalization precoding, imperfect CSI). The text does include a caveat, but Fig. 2(b) and the associated discussion still present the PDSCH number as evidence of channel-dependent coverage loss. Since the specific median is not an equal-aperture bound, the paper should either add error bars and at least a second environment, or clearly separate the illustrative PDSCH value from the qualitative asymmetry claim. The SSB/CSI-RS medians could support the asymmetry conclusion, but they too would benefit from an uncertainty statement.","section":"Sec. III, Fig. 2, Table II"}],"minor_comments":[{"comment":"The sentence 'In the receive and transmit DFT bases, we have' appears incomplete before 'Let p denote...'. Please revise for clarity and define the entries of Ŝ(p) before Eq. (10) is used.","section":"Sec. VI-D, Eq. (9)"},{"comment":"The 3D-GS comparison does not state the number of training locations, Gaussian primitives, hyperparameters, or the baseline sweep policy (all N_r N_t pairs vs. hierarchical). Without these, 'approaches the upper bound' is hard to assess. At minimum, add a comparison to a simple spatial-interpolation baseline such as nearest-neighbor or kriging of beamspace profiles.","section":"Sec. VI-D, Fig. 9"},{"comment":"The spectral- and energy-efficiency tradeoff curves in Fig. 7 have no simulation settings, channel model, number of users, or baseline definitions. As an illustrative reproduction of [49] this is acceptable, but the caption should say so explicitly.","section":"Sec. IV-C, Fig. 7"},{"comment":"The ray-tracing setup would benefit from a statement on ray-tracing configuration (number of reflections, grid resolution, material database) and the number of user drops used for the CDFs. This would support reproducibility of the median values.","section":"Sec. III, Table II"},{"comment":"The claim that 'at N_t = 256 and r = 8 each bit buys 0.43 dB' is correct under the RVQ scaling, but the choice r = 8 is not justified from the cited measurements. Consider adding a short numerical example based on the eigenvalue profiles in [8] or [30].","section":"Sec. VI-C, Eq. (8)"}],"recommendation":"major_revision","confidential_remarks":"This is an invited roadmap, and the paper is appropriate for the journal's format. The main revision should focus on Sec. VI: the rank-scaling premise must be presented as an open hypothesis, with a sensitivity analysis and without overclaiming the N_t/r savings. The coverage simulation is clearly illustrative and can be left as such, but the PDSCH median should not be used as a headline result. I do not think new measurements are strictly required for revision, but the language must be aligned with the evidence the paper itself cites."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should read this as a roadmap, not a result. The paper argues that simply scaling 5G NR MIMO to 256+ ports at 7 GHz hits four coupled walls: coverage asymmetry, RF hardware, power, and CSI acquisition. That framing is useful and, as far as I can tell, consistent with the cited measurement work. The genuinely new pieces are the coverage asymmetry study in Fig. 2 and the 3D-GS beamspace rendering in Sec. VI-D. The rendering is a real extension: prior RF NeRF/3D-GS work reconstructs single-link instantaneous channels, while this produces a full N_r x N_t beamspace power profile at unseen locations. That is worth pointing at, even though it is preliminary and ships no code, data, or error metrics.\n\nNow the soft spots. The -37.7 dB PDSCH median in Fig. 2(b) is the eye-catcher, but it comes from a deliberately unoptimized PDSCH baseline — random scheduling, block diagonalization, imperfect CSI — while SSB and CSI-RS use idealized beam sweeps. The paper explicitly calls this an illustration, not an optimized deployment, so I do not read it as a hidden agenda. The bigger problem is the one the stress-test note flags: the entire CSI-acquisition roadmap leans on Eq. (3), i.e., the channel living in an r-dimensional subspace with r growing sublinearly in port count. That is plausible at 32-128 elements, but the paper itself admits the covariance at 768-1024 elements \"calls for further measurement and analysis\". If rank at 256+ ports scales roughly with aperture, the N_t/r overhead saving in Eq. (8) evaporates and the decoupling principle loses its quantitative footing. This is an acknowledged gap, not a sleight of hand, but it is load-bearing.\n\nCredit where due: the paper is carefully hedged, the four-breakpoint structure is clear, and the authors do not oversell the 3D-GS step as more than a promising direction. It is an invited paper, and it reads like one written by people who know the area.\n\nWho benefits: anyone working on FR3 channel modeling, 6G standardization, or massive MIMO architectures. It is not a breakthrough paper, but it is a competent synthesis with two small new results and a clear research agenda.\n\nRecommendation: send it to peer review. A good referee should push the authors to either provide more evidence on rank scaling at large arrays or explicitly scope the claims to the measured regime, and to add confidence intervals or alternative environments to Fig. 2. But the paper deserves referee time; desk rejection would be wrong.","headline":"A solid, honest 6G roadmap whose central decoupling thesis is plausible but still hangs on an unmeasured rank-scaling assumption; worth refereeing, not because the claims are proven, but because the field needs this synthesis.","tokens_in":35133,"tokens_out":1907,"would_cite":true,"duration_ms":21112,"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":"Scaling 5G MIMO from tens to hundreds of antenna ports breaks at four coupled system bottlenecks, so 6G must decouple aperture size from active chains and channel-sounding cost.","keywords":["FR3","upper-mid band","extreme MIMO","6G","massive MIMO scaling","CSI acquisition","tri-hybrid MIMO","coverage asymmetry"],"falsifier":"Measure the spatial covariance rank at a 256-, 512-, and 1,024-element equal-aperture FR3 array in outdoor urban NLOS. If the 95% energy rank grows roughly proportionally with port count (e.g., r ≈ N_t/4 or worse) rather than sublinearly, the overhead savings, the 3/(r-1) bit-gain figure, and the 3D-GS rendering roadmap collapse.","tokens_in":34120,"feed_emoji":"📡","tokens_out":4418,"duration_ms":36997,"temperature":0.7,"pith_summary":"This paper argues that extreme MIMO — packing 256 or more antenna ports into the same base-station panel used at 3.5 GHz — cannot be achieved by simply enlarging the 5G New Radio design. Four coupled limits break it: wide-beam common and control channels get no aperture gain; RF devices and power amplifiers fall behind at 7 GHz over fourfold wider bandwidths; powering hundreds of always-on digital chains becomes unsustainable; and channel-state acquisition overhead grows with port count. The central constructive claim is that these four quantities — radiating elements, antenna ports, active RF chains, and the dimension of acquired channel information — can and should scale at different rates. The paper lays out a roadmap of protocol-aware coverage, low-power tri-hybrid array architectures, distributed apertures, and rank-based, scene-rendered CSI acquisition, with simulation evidence that equal-aperture scaling alone leaves common, control, and data channels 16–38 dB below 5G targets in the worst-case environment.","feed_headline":"Scaling 5G MIMO to 256 ports breaks in four ways","feed_subtitle":"Equal-aperture 7 GHz arrays leave common and control channels 16–38 dB short of 5G; 6G must decouple aperture from active chains.","key_machinery":"Two formal devices carry the argument. First, the tri-hybrid signal model x_ant = F_ant F_ana F_dig s, which separates the number of physical radiating elements Nrad from actively driven ports Nport and RF chains NRF; dynamic-metasurface and fluid-antenna realizations put spatial processing into the aperture itself, letting the accessible aperture grow without always-on chains. Second, the covariance-subspace channel model h ≈ U_r g, where U_r holds the r dominant eigenvectors of the spatial covariance R; when the FR3 channel's angular spread narrows, r grows sublinearly in port count, and sounding, feedback, codebook quantization, prediction, and scheduling can all be confined to r dimensio","core_discovery":"The paper's central discovery is that the equal-aperture argument for upper-mid-band 6G is a statement about the user-data channel only. The PDSCH uses UE-specific narrow beams and exploits the full aperture gain, while SSB, CSI-RS, SRS, and uplink channels do not; in a representative equal-aperture 7 GHz simulation the median relative gains of SSB, CSI-RS, and PDSCH fall 15.9, 17.2, and 37.7 dB below the corresponding 3.5 GHz link in the worst combined deep-NLOS and outdoor-to-indoor environment. From this coverage asymmetry together with RF, power, and CSI scaling mismatches, the paper concludes that a uniform extension of 5G NR is infeasible at the E-MIMO scale and that the system must de","pith_inferences":["The 37.7 dB PDSCH deficit is an artifact of the paper's deliberately unoptimized scheduling and precoding; a realistic optimizer would shift that curve substantially, so the true coverage asymmetry between data and control channels is likely smaller than the headline number — but the SSB and CSI-RS deficits are robust.","If the low-rank premise holds, the same rendering idea could be extended from beamspace power profiles to full covariance rendering, eliminating per-UE sounding entirely; the paper leaves this as open.","The decoupling principle generalizes beyond FR3: any array where aperture can grow without proportional RF chains, such as lens-based or metasurface arrays at mmWave, inherits the same rank-scaling CSI argument.","The paper's coverage study treats only co-located RUs; an equal-aperture distributed deployment would close part of the indoor gap, suggesting the roadmap's DMIMO section and coverage section are best read as one design."],"forward_implications":["6G base stations can host 512–1,024 radiating elements while keeping only tens of active RF chains, with beamforming gain retained through passive beamforming networks and reconfigurable apertures.","CSI acquisition overhead, which scales with port count in 5G, can instead scale with effective rank; at 256 ports and r=8, each feedback bit's accuracy improves by a factor of about 36.","Channel statistics can be treated as properties of locations rather than of terminals, so a scene learned from past observations can render beamspace profiles at unseen positions and cut sweeping overhead.","Coverage must be designed channel-by-channel: closing wide-beam and indoor-penetration gaps may require hierarchical beam management and in-building solutions rather than higher transmit power.","Distributed apertures convert installed spatial diversity into proximity gain, but only if RU activation, synchronization, and fronthaul costs are jointly controlled."],"fun_headline_variants":["Why 5G MIMO breaks at 256 ports for 6G upper-mid-band","6G extreme MIMO: 5G scaling fails on coverage, RF, CSI","Equal-aperture myth: 7 GHz MIMO leaves controls 38 dB short","5G NR can't scale to 256 ports—6G needs a new architecture","Upper-mid-band 6G: the four cracks in scaling 5G MIMO"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The whole CSI-acquisition roadmap assumes the FR3 channel at 256+ elements lives in a low-dimensional subspace whose rank grows much more slowly than the number of antenna ports; the paper's own Section II-B admits that covariance behavior at 768–1,024 elements awaits measurement.","fun_headline_variants_meta":{"raw":{"variants":["Why 5G MIMO breaks at 256 ports for 6G upper-mid-band","6G extreme MIMO: 5G scaling fails on coverage, RF, CSI","Equal-aperture myth: 7 GHz MIMO leaves controls 38 dB short","5G NR can't scale to 256 ports—6G needs a new architecture","Upper-mid-band 6G: the four cracks in scaling 5G MIMO"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000234,"raw_usage":{"total_tokens":1411,"prompt_tokens":899,"completion_tokens":512,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":643,"completion_tokens_details":{"reasoning_tokens":398}},"tokens_in":643,"tokens_out":512,"duration_ms":4825,"temperature":1.0,"reasoning_tokens":398,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T16:21:48.333434+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the spatial covariance rank at a 256-, 512-, and 1,024-element equal-aperture FR3 array in outdoor urban NLOS. If the 95% energy rank grows roughly proportionally with port count (e.g., r ≈ N_t/4 or worse) rather than sublinearly, the overhead savings, the 3/(r-1) bit-gain figure, and the 3D-GS rendering roadmap collapse.","supporting_citations":[],"review_version":1}