{"id":"e7991814-284b-436b-a393-bafec2a16b7f","arxiv_id":"2411.18878","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"RIS elements grouped by Fresnel zone are given equal phase shifts, making the wideband channel a Fourier transform of zone intensities and enabling near-optimal beamforming without true-time-delay hardware.","lead":"A new beamforming method for reconfigurable intelligent surfaces uses the geometry of Fresnel zones to keep wideband signals focused on a receiver, avoiding a frequency-dependent gain loss called near-field beam split. The method needs no extra hardware and its simulated rate comes within a few percent of an ideal upper bound.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (17) silently drops the frequency-dependent path-loss prefactor g0 from Eq. (6), so the Fourier-transform representation and the Parseval-based rate upper bound in Eq. (25) are not proven as stated.","rationale":"The reader's conditional verdict is appropriate. The most load-bearing weakness is not the heuristic phase design—which is plausibly effective and supported by simulation—but the analytical claim that the equivalent wideband channel is exactly a Fourier transform of a fixed reflective intensity. That claim underlies both the upper-bound proof (Eq. 25) and the interpretation of the SP/GS designs as approaching the bound. The 1/f² in g0 is explicit in Eq. (6) and is not addressed anywhere; the paper's amplitude approximation concerns spatial distance variation only. The Jensen bandwidth error in Eq. (25) is also real and must be corrected by inserting 1/B before the integral. Both issues are fixable: the Fourier relation can be restated as an approximation for |f−fc|<<fc with an error bound, and the rate bound can be restated with proper normalization. The concrete numerical test proposed would settle whether the approximation introduces a practically negligible error at the paper's own operating point. Because the central method and simulations appear sound enough to justify conditional acceptance, the reader's verdict should remain CONDITIONAL.","tokens_in":17942,"tokens_out":6198,"duration_ms":53197,"concrete_test":"Evaluate the exact channel g_exact(f) from Eq. (6) without replacing g0(f) by g0(fc), and compare it to g_app(f) from Eq. (17) with v_t computed at fc, using the paper's simulation setup (fc=30 GHz, B=1.5 GHz, D=1 m, TX=(6.4,5,14.4) m, RX=(-4.8,5,6.4) m). Compute the relative change in ∫|g(f)|²df and the achievable-rate difference in Eq. (24) with the corrected Jensen bound (divide the integrated SNR by B). If both changes are below 1%, the frequency dependence is a benign approximation and the upper-bound claim can be restored with a minor caveat; if the rate bound is violated or shifts by more than a few percent, the proof in Sec. V must be revised to include the 1/f² weighting.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central Fourier representation g(f)=∫vt(t)e^{jψt(t)}e^{-j2πft}dt in Eq. (17) requires the Fresnel-zone reflective intensity v(a) in Eq. (15) to be independent of frequency. But v(a)=g0(f)∫J(a,θ)dθ with g0=√N_BS c²/(4π²f²R_BR R_RU d²) from Eq. (6), so v(a) inherits an explicit 1/f² factor. Over the simulated B=1.5 GHz at fc=30 GHz, g0 drops by about 9%, which is small but not negligible. Consequently Eq. (17) is only an approximation evaluated at some fixed frequency; it is not the exact Fourier transform of a fixed intensity, and the Parseval step (c) in Eq. (25), ∫|g(f)|²df=∫v_t²(t)dt, does not follow. The upper bound log2(1+(Sx/Sσ)Eg) is therefore not established by the given proof. A separate error compounds this: inequality (a) applies Jensen without the 1/B normalization of the frequency average, so the bound is dimensionally inconsistent whether or not g0 is treated as constant. The proposed SP and GS phase designs may still work as heuristics, but the theoretical 'upper bound' claim needs either an explicit approximation statement with error bounds or a corrected derivation that accounts for g0(f).","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper addresses near-field wideband beam splitting in RIS-aided links. It models the RIS as a continuous I-RIS, introduces a Fresnel-zone coordinate system in which a constant route length per zone makes the phase response frequency-independent along the zone, and reduces the equivalent channel to a one-dimensional Fourier transform of a phase-modulated Fresnel-zone intensity. On this basis it derives a sinc approximation and 3-dB bandwidth for narrowband beamforming, proposes an upper bound on achievable rate using Parseval's theorem, and constructs phase profiles via stationary-phase and Gerchberg-Saxton designs. Simulations with a 1 m aperture at 30 GHz and 1.5 GHz bandwidth show flatter in-band gain and higher average rate than narrowband and subarray baselines.","tokens_in":1354,"tokens_out":2518,"duration_ms":116546,"significance":"The conceptual contribution is valuable: the Fresnel-zone decomposition gives a clean structural explanation of near-field beam split and yields phase-only designs that avoid true-time-delay hardware. The derivations are self-contained, the sinc approximation is compared with simulation rather than fitted, and the SP and GS implementations are described with complexity estimates. If the proof issues below are corrected, the paper would be a useful contribution to wideband RIS beamforming. In its current form, however, the two central theoretical claims -- the exact Fourier-transform representation and the Jensen/Parseval upper bound -- are not established as stated.","major_comments":[{"comment":"Equation (17) is presented as an exact Fourier transform of a fixed Fresnel-zone intensity vt(t), but this ignores the explicit frequency dependence in Eq. (6): g0 = sqrt(N_BS)c^2/(4π^2 f^2 R_BR R_RU d^2). Since v(a) in Eq. (15) is defined as the integral of g0 J(a,θ), v(a) and hence vt(t) depend on f. The sentence after Eq. (14) that 'the inner integral along the Fresnel zone in (13) is frequency independent' is therefore inconsistent with the paper's own definitions. As a result, the Parseval step (c) in Eq. (25) does not follow: ∫|g(f)|^2 df = ∫ v_t^2(t) dt requires g(f) to be the Fourier transform of a frequency-independent vt(t)e^{jψt(t)}. The authors should state explicitly that g0 is replaced by its value at a reference frequency and bound the error, or repeat the derivation with a frequency-weighted Parseval identity.","section":"III-C, Eqs. (6), (15), (17)"},{"comment":"The Jensen step in Eq. (25)(a) is dimensionally inconsistent. For the integral in Eq. (24), concavity of log2 gives ∫ log2(1+x(f)) df ≤ B log2(1 + (1/B)∫ x(f) df), not log2(1+∫ x(f) df). The displayed bound omits the bandwidth B and the 1/B normalization, so the claimed upper bound log2(1 + (Sx/Sσ)Eg) is not a consequence of Jensen's inequality and is not an upper bound in general. As a concrete check, for B=10 and flat |g(f)|^2 with x(f)=Sx|g(f)|^2/Sσ=10, the left side of (24) is 10log2(11)≈34.6 normalized units, while the claimed bound is log2(101)≈6.7. With exact Parseval, the correct bound would be B log2(1 + (Sx Eg)/(B Sσ)). This correction affects the theoretical claim in Section V and the rate comparison in Section VII, although the ideal-gain magnitude in Eq. (26) remains correct for a flat in-band gain.","section":"V, Eq. (25), inequality (a)"}],"minor_comments":[{"comment":"The constant Γ0 is given as 0.886 in Eq. (21) but as 0.866 in the verification text; make the two values consistent.","section":"IV, after Eq. (23)"},{"comment":"The text contains the typo 'Frsenel zone-based'; it should read 'Fresnel zone-based'.","section":"VII-B"},{"comment":"The least-squares update requires (A^H A)^{-1}, but no condition is given on K' relative to N_S; if K' < N_S then A^H A is singular, and otherwise it may be ill-conditioned. Please specify that K' ≥ N_S and, if needed, use a regularized pseudo-inverse.","section":"Algorithm 2, Eqs. (35)-(38)"},{"comment":"The paper should state separately that the replacement of the per-element distances by R_BR and R_RU in the amplitude is a spatial approximation and that, for the Fourier representation in (17), g0 is additionally treated as frequency-independent; these are two different approximations and only the first is indicated in Section III-A.","section":"III-A, Eq. (6)"},{"comment":"The symbol V is used for the angular set V(a) in Eq. (15) and for the Fourier transform V(f) in Eq. (18); please rename one of them to avoid ambiguity.","section":"III-C and IV"}],"recommendation":"major_revision","confidential_remarks":"To the editor: the two main proof gaps are correctable and do not, in my judgment, invalidate the design heuristics, which are supported by simulation. However, the paper's claimed theoretical upper bound is currently not a theorem, and the manuscript should not be accepted without the derivations being repaired. The discussion of prior TTD-based work is adequate, and I see no other scope or novelty concerns."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, the Fresnel-zone coordinate transformation is a genuinely new way to look at near-field RIS beamforming, and the observation that elements on a single Fresnel zone share the same path length, so their phases stay aligned across frequency, is clean and correct. The phase designs — stationary phase and Gerchberg-Saxton — are imported from radar, but the application to RIS is new, and the simulations show they flatten in-band gain without TTD hardware. That part deserves credit.\n\nSecond, the theoretical upper bound is not proven as written. Equation (17) claims g(f) is the Fourier transform of a fixed Fresnel-zone intensity vt(t), but vt(t) inherits a 1/f^2 factor from the path-loss g0 in Eq. (6). So the Parseval step (c) in Eq. (25) doesn't follow. And the Jensen step (a) drops the 1/B normalization, making the bound dimensionally inconsistent; with the missing B the correct bound is B log2(1 + (Sx/Sσ) Eg/B), which is larger than what the paper claims. That is not a minor footnote — the claimed upper bound is lower than the actual upper bound, so it can't hold. The ideal gain in Eq. (26) is unreachable, so the simulations approach something that isn't actually the bound.\n\nThe rest of the paper holds up reasonably. The sinc approximation and the 3dB bandwidth formula are supported by simulation and by the Fourier model, aside from the same frequency-independent-intensity assumption, which is mild over a 1.5 GHz band at 30 GHz (about 9% variation in g0). The complexity claims are plausible. No code or data are shipped, which is a limitation but not disqualifying.\n\nBottom line: this is a promising engineering method with a real explanatory idea, but the paper's central proof needs repair. It should be sent to peer review — a good referee can ask the authors to restate the Fourier relationship as an explicit approximation with error bounds, fix the Jensen normalization, and re-derive or reinterpret the upper bound. If they can do that, it is a solid contribution to the RIS/6G subfield.\n\nI'd bring it to our reading group.","headline":"Fresnel-zone coordinate trick is a genuinely new angle on near-field RIS beamforming, but the paper's advertised rate upper bound has two concrete derivation errors and needs referee attention.","tokens_in":18754,"tokens_out":3688,"would_cite":true,"duration_ms":29082,"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":"The paper claims that grouping RIS elements by Fresnel zone turns the wideband channel into a Fourier transform of zone intensity, allowing phase-only beamforming to approach the rate upper bound despite near-field beam split.","keywords":["reconfigurable intelligent surface","near-field beam split","Fresnel zone","wideband beamforming","stationary phase method","Gerchberg-Saxton algorithm","achievable rate upper bound","mmWave and THz communications"],"falsifier":"Run the exact channel integral in Eq. (6) without dropping the frequency dependence of $g_0(f)\\propto 1/f^2$, and check whether $\\int |g(f)|^2\\,df$ still equals $E_g=\\int v_t^2(t)\\,dt$ and whether the designed flat gain profile survives; if the mismatch grows with fractional bandwidth or with RIS aperture, the Fourier identity and the bound in Eq. (25) are only approximations rather than exact consequences.","tokens_in":17680,"feed_emoji":"📡","tokens_out":8091,"duration_ms":67600,"temperature":0.7,"pith_summary":"The paper argues that the near-field beam split, which makes wideband RIS beams focus at different places for different frequencies, can be overcome by phase shifts alone, without true-time-delay hardware. Its key move is to group RIS elements by Fresnel zone, the ellipsoidal surfaces on which the BS-RIS-UE route length is constant, and to assign the same phase shift to all elements on a zone. Once phases are aligned along each zone, the two-dimensional sum over the RIS plane collapses to a one-dimensional integral, and the frequency response of the equivalent channel becomes the Fourier transform of the phase-modulated reflective intensity across zones. That identity yields a Parseval-based upper bound on achievable rate, and the paper designs phases by stationary-phase and Gerchberg-Saxton methods to approach flat in-band gain. If correct, this gives a practical, hardware-free cure for near-field beam split in large-aperture mmWave and THz RIS systems.","feed_headline":"Fresnel zones cure near-field beam split in RIS","feed_subtitle":"Equal phases per zone make the wideband channel a Fourier transform, letting phase-only beamforming approach the rate limit.","key_machinery":"The central object is the Fresnel-zone coordinate system for the RIS plane. Fresnel zones are a family of concentric ellipsoids whose foci are the BS and the UE; their intersections with the RIS plane are nested ellipses indexed by the semi-major axis $a$, which is half the reflection route length. Transforming the RIS plane from $(x,y)$ to $(a,\\theta)$ makes the phase of the cascaded channel linear in $a$, and locking the phase on each zone $\\phi(a,\\theta)=\\psi(a)$ removes frequency dependence along the zone. The surviving quantity is the reflective intensity $v(a)=\\int g_0 J(a,\\theta)\\,d\\theta$, and the substitution $t=2a/c$ turns the channel into the Fourier transform $g(f)=\\int v_t(t)e^{j\\psi_t(t)}e^{-j2\\pi f t}\\,dt$. That Fourier identity, together with Parseval's theorem and the stationary-phase / Gerchberg-Saxton spectrum-shaping tools, carries the entire argument.","core_discovery":"The paper establishes, under its amplitude-approximation model, that the equivalent wideband channel is $g(f)=\\int v_t(t)e^{j\\psi_t(t)}e^{-j2\\pi f t}\\,dt$, the Fourier transform of the Fresnel-zone reflective intensity $v_t(t)$ multiplied by the designed phase modulation $e^{j\\psi_t(t)}$. It follows from Parseval's theorem that $\\int |g(f)|^2 df = E_g = \\int v_t^2(t)\\,dt$ is fixed for every phase configuration, so the achievable rate is bounded by $\\log_2\\left(1+\\frac{S_x}{S_\\sigma}E_g\\right)$, with equality approached only by an ideal gain that is flat across the band and zero outside it. The paper then treats the phase design as a one-dimensional spectrum-shaping problem: the stationary-phase method gives a nearly closed-form phase profile $\\psi(a)$ across Fresnel zones, and the Gerchberg-Saxton algorithm iterates between the zone domain and frequency domain to flatten the gain. Simulations show the proposed methods keep gain approximately flat over the band, reduce out-of-band leakage, and achieve rates within a few percent of the upper bound while outperforming narrowband and virtual-subarray baselines.","pith_inferences":["The Fourier-transformed channel makes the RIS phase design mathematically identical to radar waveform design; techniques from pulse compression and spectral shaping could be imported directly, and the one-dimensional zone-domain formulation suggests delay-domain optimization as a scalable route for extremely large surfaces.","The Parseval bound implies a phase-only RIS cannot create more in-band energy than the fixed intensity supports; any attempt to boost edge-subcarrier gain must steal out-of-band energy, so flat in-band gain is the best achievable spectrum shape.","If the reflective intensity $v_t(t)$ is measurably frequency-dependent through the $1/f^2$ path-loss factor at very wide fractional bandwidths, the exact Fourier identity would fail; the phase designs would need an iterative frequency-domain correction, with the stationary-phase profile as initialization.","The zone grouping principle might extend to multi-user or cell-free settings by treating each user's Fresnel-zone intensity as a separate waveform constraint, and to holographic or continuous-aperture metasurfaces where the per-zone phase profile is sampled directly."],"forward_implications":["Near-field beam split can be mitigated with phase-only RIS control; no true-time-delay or delay-adjustable metasurface hardware is needed, avoiding the insertion loss and cost of TTD modules.","For any phase-only design, total gain energy over all frequencies is fixed at $E_g=\\int v_t^2(t)\\,dt$; the rate bound $\\log_2(1+S_xE_g/S_\\sigma)$ is therefore a fundamental limit of phase-shift-based wideband RIS, tighter than bounds that assume TTD hardware.","Under narrowband beamforming, the 3 dB bandwidth shrinks roughly as $B_{3\\mathrm{dB}}\\approx c\\Gamma_0/(\\iota D_0)$, so larger apertures and BS/UE directions pointing similarly make beam split worse; the zone-based designs restore flat gain across the designed band.","Achievable rate grows with RIS size and bandwidth under the proposed designs, while narrowband beamforming plateaus; 2-bit phase quantization is enough to nearly match continuous-phase performance.","The same Fresnel-zone construction works in far-field scenarios, since far-field is a special case of the near-field model used here."],"supporting_citations":[{"why":"Supplies the constant-reflection-amplitude RIS model and the frequency-independent phase-shift mismatch that causes beam split.","marker":"[12]"},{"why":"Establishes the near-field wideband beamforming problem for extremely large arrays that this paper targets.","marker":"[15]"},{"why":"Provides the near-field spherical-wave channel model used for RIS-UE and BS-RIS links.","marker":"[16]"},{"why":"Documents that near-field beam split degrades power even in the boresight direction, motivating the work.","marker":"[17]"},{"why":"Gives the prior TTD-based upper bound and system model that the paper compares with and tightens.","marker":"[20]"},{"why":"Defines the virtual-subarray baseline whose separate beamforming suffers $1/N_{\\text{sub}}^2$ gain loss.","marker":"[27]"},{"why":"Justifies the amplitude approximations in the continuous RIS model and the near-field to far-field reduction.","marker":"[30]"},{"why":"Supplies Parseval's theorem and the sinc-function 3 dB bandwidth constant used in the rate bound and gain analysis.","marker":"[33]"},{"why":"Provides the stationary-phase method for designing phase-modulated waveforms with a desired spectrum.","marker":"[34]"},{"why":"The reference cited for the Gerchberg-Saxton alternating-projection procedure used in Algorithm 2.","marker":"[37]"}],"fun_headline_variants":["Fresnel zones flatten wideband RIS beams","Phase-only RIS beamforming via Fresnel zones","Near-field beam split solved with Fresnel zones","Fourier shaping of RIS beams using Fresnel zones","RIS wideband gain bound achieved via Fresnel zones"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The Fourier-transform picture and the Parseval-based rate bound require the reflective intensity of each Fresnel zone to be independent of frequency, although the path-loss amplitude in the model carries an explicit $1/f^2$ factor and the paper only justifies amplitude approximations in space, not in frequency.","fun_headline_variants_meta":{"raw":{"variants":["Fresnel zones flatten wideband RIS beams","Phase-only RIS beamforming via Fresnel zones","Near-field beam split solved with Fresnel zones","Fourier shaping of RIS beams using Fresnel zones","RIS wideband gain bound achieved via Fresnel zones"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000657,"raw_usage":{"total_tokens":3061,"prompt_tokens":1051,"completion_tokens":2010,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":667,"completion_tokens_details":{"reasoning_tokens":1936}},"tokens_in":667,"tokens_out":2010,"duration_ms":12716,"temperature":1.0,"reasoning_tokens":1936,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T10:48:16.931067+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the exact channel integral in Eq. (6) without dropping the frequency dependence of $g_0(f)\\propto 1/f^2$, and check whether $\\int |g(f)|^2\\,df$ still equals $E_g=\\int v_t^2(t)\\,dt$ and whether the designed flat gain profile survives; if the mismatch grows with fractional bandwidth or with RIS aperture, the Fourier identity and the bound in Eq. (25) are only approximations rather than exact consequences.","supporting_citations":[{"cited_title":"Reconfigurable intelligent surface-based wireless communications: Antenna design, pro- totyping, and experimental results,","cited_arxiv_id":null,"evidence_quote":"Supplies the constant-reflection-amplitude RIS model and the frequency-independent phase-shift mismatch that causes beam split."},{"cited_title":"Near-field wideband beamforming for extremely large antenna arrays,","cited_arxiv_id":null,"evidence_quote":"Establishes the near-field wideband beamforming problem for extremely large arrays that this paper targets."},{"cited_title":"Hierarchical beam training for extremely large-scale mimo: From far-field to near-field,","cited_arxiv_id":null,"evidence_quote":"The reference cited for the Gerchberg-Saxton alternating-projection procedure used in Algorithm 2."}],"review_version":1}