{"id":"44d07341-c53b-42df-b5f7-1354f75393fd","arxiv_id":"1908.03133","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A proof that, at the same array location, Massive MIMO always achieves higher SNR than an IRS, despite the IRS's faster N^2 scaling.","lead":"Intelligent reflecting surfaces (IRSs) are often claimed to beat Massive MIMO because their signal-to-noise ratio grows with the square of the number of elements. This paper proves that at the same array location, Massive MIMO always delivers a higher SNR than an IRS, and the gap grows with distance.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; the central inequality is proven within the model's stated scope.","rationale":"The proof of Proposition 1 is transparent and correct: the SNR_IRS expression in (17) follows from the channel model in (9) with optimal phase shifts, and the factorization in (18) shows the extra factor is mu^2 N beta_g, which is at most 1 by energy conservation. The paper explicitly limits the comparison to co-located deployment and free-space propagation with isotropic antennas, and it introduces the exact planar-array model in Section 2.2 to handle large N, where the approximate model would break down. The exact model gives alpha_g <= 1/2, which strengthens rather than weakens the inequality. The numerical results in Fig. 4 are consistent with the analysis. No load-bearing flaw was identified, so the reader's ACCEPT verdict is appropriate.","tokens_in":7615,"tokens_out":18181,"duration_ms":189579,"concrete_test":"Reproduce Fig. 4 using the exact IRS expression (21)-(22) for N up to 10^6 with d_g = 2.5 m and d_h = 25 m, and verify SNR_IRS <= SNR_mMIMO for every N, confirming the bound mu^2 alpha_g <= 1/2.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's Proposition 1 is analytically sound within the co-located, free-space, isotropic-antenna model. The factorization SNR_IRS = (mu^2 N beta_g) SNR_mMIMO in Eq. (18) is algebraically correct, and the energy-conservation bound mu^2 N beta_g <= 1 is valid in the regime where the channel model (9) holds (i.e., N beta_g <= 1). For larger N, the exact planar-array model (4)-(5) gives a total channel gain alpha_g <= 1/2, which preserves the inequality via SNR_IRS = mu^2 N alpha_g beta_h Ptx/sigma^2. The deployment-scope limitation (IRS and mMIMO array at the same location, receiver separated from the IRS) is explicitly stated in Section 3.2 and the conclusion, so it does not undermine the claim as formally presented. No internal inconsistency or hidden assumption was found.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper addresses the claim that intelligent reflecting surfaces (IRSs) enjoy a more favorable power scaling law than Massive MIMO because their SNR scales as N^2 instead of N. Under a single-user line-of-sight free-space model, the authors compare an N-antenna mMIMO receiver with an N-element IRS placed at the same location, with the IRS-aided receiver physically separated from the surface. They derive the SNR expressions, factor the IRS SNR as SNR_IRS = mu^2 N beta_g * SNR_mMIMO, and use the energy-conservation bound mu^2 N beta_g <= 1 to prove Proposition 1: SNR_mMIMO >= SNR_IRS. Numerical results quantify the gap, showing that thousands of IRS elements are needed to match a 64-antenna mMIMO array, and the authors provide an exact planar-array correction for the IRS-to-receiver channel when the far-field approximation breaks down.","tokens_in":7765,"tokens_out":14115,"duration_ms":156211,"significance":"The paper makes a clean, self-contained analytical point that corrects an overinterpretation in the IRS literature: observing an N^2 scaling law does not by itself imply that an IRS is more power-efficient than mMIMO, because the complete SNR expressions include a channel-gain factor that is bounded by energy conservation. The proof is transparent, the assumptions are stated, and the numerical comparison gives concrete, falsifiable orders of magnitude. The paper is honest about its scope, restricting the formal result to the co-located, free-space, LOS setup of Figure 3. This is a useful contribution that should help discipline future claims about IRS power scaling.","major_comments":[],"minor_comments":[{"comment":"The abstract and Proposition 1 state that mMIMO always provides higher SNRs, but the comparison is carried out for the specific co-located deployment in Figure 3, where the IRS and the mMIMO array occupy the same location and the receiver is separated from the IRS; please add this qualification to the abstract and to the statement of Proposition 1 so that the claim is not read as applying to other IRS deployment geometries.","section":"Abstract and Proposition 1"},{"comment":"The proof of Proposition 1 should state explicitly that the factorization in Eq. (18) and the bound mu^2 N beta_g <= 1 are used in the far-field regime; for larger N the same conclusion follows by replacing N beta_g with the exact planar-array gain alpha_g <= 1/2 used in Eq. (21), but this replacement is not mentioned in the proof.","section":"§3.2, Eq. (18)"},{"comment":"The sentence after Eq. (20) would be clearer if it distinguished between the norm loss ||v||^2 and the suboptimality of the combining vector v; the IRS SNR equals ||v||^2 times the SNR of an mMIMO receiver using the same v, and it is further below MRC because v = Theta g is not the maximum-ratio combiner.","section":"§3.3, Eq. (20)"},{"comment":"In Figure 4, please distinguish the approximate IRS curve based on Eq. (17) from the exact curve based on Eq. (21) in the legend, and mark the truncation point N = 1/beta_g where the approximate curve is stopped.","section":"Fig. 4"},{"comment":"Reference [19] has a duplicated year in the bibliographic entry, and the rule-of-thumb condition NA/10 < d^2 in §2.2 would benefit from a short justification or a pointer to the error level it corresponds to.","section":"References and §2.2"}],"recommendation":"minor_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Emil and Luca have a short conference paper that should be read as a corrective to the IRS hype. The new thing is the factorization in Eq. (18): SNR_IRS = (mu^2 N beta_g) * SNR_mMIMO, with the first factor provably <= 1 by energy conservation. Earlier papers touted the N^2 scaling without noticing that one of the N's is essentially the fraction of power that escapes between the IRS and a physically separated receiver. That is a genuine and clarifying observation, not just a restatement. The proof is transparent: from the channel models and maximum-ratio combining you get SNR_mMIMO = N beta_h Ptx/sigma^2; with optimal phase shifts the IRS gives mu^2 N^2 beta_g beta_h Ptx/sigma^2; and the energy-conservation bound N beta_g <= 1 closes the inequality. The paper is honest about the model: co-located arrays, free-space, isotropic elements, far-field for g. It shows numerically that even with the receiver 2.5 m from the IRS, thousands of elements are needed to match a 64-antenna mMIMO array.\n\nSoft spots, in proportion. The result is exactly as strong as the stated model. The paper says 'always' in the abstract, but that 'always' is conditional on the IRS and mMIMO array being at the same location and the receiver being physically separated from the IRS. That is not the deployment that motivates most IRS work, where the surface is hung close to the user to shorten the second hop. The authors acknowledge this and partially address it with the d_g = 2.5 m case, but they do not analyze the case where the receiver is effectively attached to the surface (which they mention as 'holographic beamforming' in the conclusion). Also, the comparison is single-user; in a multi-user setting mMIMO's spatial multiplexing only widens the gap, so the conclusion would not flip, but the paper does not make that argument. The far-field model for g limits N below ~1/beta_g, and they note it; the exact planar model (4) preserves the inequality via alpha_g <= 1/2, so the stress-test checks out. The citation pattern is fine: [13] and [18] are their own prior work but they are the right sources for the IRS signal model and the relay intuition.\n\nNet: this is a small but useful paper. It deserves a serious referee; it is the kind of thing that should be in the record so that the 'N^2 beats mMIMO' misreading stops being repeated. I would cite it, and I would bring it to a reading group if anyone in the group works on IRSs. Recommend accept, with a request that the abstract's 'always' be scoped to the assumptions.","headline":"A clean, short proof that under co-located free-space assumptions mMIMO always beats IRS in SNR; the N^2 scaling is real but the second N is just pathloss from the surface to the receiver.","tokens_in":8284,"tokens_out":2014,"would_cite":true,"duration_ms":18993,"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":"A same-location comparison shows that an N-element intelligent reflecting surface can never exceed the SNR of an N-antenna Massive MIMO receiver, because the quadratic scaling contains an energy-conservation factor that is at most one.","keywords":["intelligent reflecting surface","massive MIMO","power scaling law","signal-to-noise ratio","energy conservation","wireless propagation","reconfigurable intelligent surface","passive beamforming"],"falsifier":"Under the paper's assumptions, compute or measure the SNR of an $N$-element IRS and an $N$-antenna mMIMO receiver with optimal phases and perfect channel knowledge. Any instance with $\\mathrm{SNR}_{\\mathrm{IRS}} > \\mathrm{SNR}_{\\mathrm{mMIMO}}$ refutes Proposition 1; the analytic check is whether the prefactor $\\mu^2 N \\beta_g$ can exceed one while respecting energy conservation.","tokens_in":7425,"feed_emoji":"📡","tokens_out":12062,"duration_ms":101805,"temperature":0.7,"pith_summary":"This paper challenges the widespread interpretation that intelligent reflecting surfaces (IRSs) have a better power scaling law than Massive MIMO because their SNR grows as $N^2$ rather than as $N$. It claims that, when an IRS and an $N$-antenna Massive MIMO receiver occupy the same location and are compared under free-space line-of-sight propagation, the quadratic scaling is the artifact of an incomplete comparison. The paper proves that Massive MIMO always achieves at least as high an SNR, and its numerics show that an IRS needs thousands of reflecting elements to match a 64-antenna array. The practical message is that the advertised $N^2$ scaling does not translate into a power advantage for IRSs.","feed_headline":"Massive MIMO always outperforms an intelligent reflecting surface","feed_subtitle":"Quadratic IRS SNR scaling hides a loss that same-size Massive MIMO avoids.","key_machinery":"The load-bearing identity is the factorization in Eq. (18): $\\mathrm{SNR}_{\\mathrm{IRS}} = (\\mu^2 N \\beta_g)\\, (N \\beta_h P_{\\mathrm{tx}}/\\sigma^2)$, where the second factor is exactly $\\mathrm{SNR}_{\\mathrm{mMIMO}}$. The apparent quadratic gain is isolated in the prefactor $\\mu^2 N \\beta_g$: $\\beta_g$ is the free-space channel gain between the IRS and the receiver, $N\\beta_g$ is the total channel gain of that hop, and energy conservation bounds it above by one, while $\\mu^2$ is the fraction of incident power the surface reflects. Proposition 1 follows because that prefactor cannot exceed unity; equivalently, the IRS behaves as an $N$-element receiver using a fixed combiner $v = \\Theta g$ instead of maximum-ratio combining.","core_discovery":"The central discovery is Proposition 1: under optimal phase-shift configuration and perfect channel knowledge, the SNR of an IRS-aided link factors as $\\mathrm{SNR}_{\\mathrm{IRS}} = (\\mu^2 N \\beta_g)\\, \\mathrm{SNR}_{\\mathrm{mMIMO}}$, where $\\mu^2 \\le 1$ is the reflection efficiency, $\\beta_g$ is the channel gain from the IRS to the single-antenna receiver, and $N \\beta_g \\le 1$ by energy conservation. Since the prefactor is at most one, $\\mathrm{SNR}_{\\mathrm{mMIMO}} \\ge \\mathrm{SNR}_{\\mathrm{IRS}}$ for every $N$. The apparent $N^2$ growth is really one factor of $N$ from the same aperture gain that Massive MIMO already enjoys, multiplied by the bounded fraction of reflected power that reaches the receiver.","pith_inferences":["Beyond the paper, the same factorization argument suggests that any passive, power-conserving reradiating structure pays a comparable two-hop loss, so quadratic scaling is not a general route to beating a same-aperture active array.","The paper's comparison fixes the IRS and the mMIMO array at one site; an undeveloped regime is the common deployment where the surface sits near the user while the base station cannot host a large active array there, in which case the relevant baseline is a remote mMIMO array with different pathloss.","A testable extension would allow directive elements or multipath propagation; under those conditions the bound $N \\beta_g \\le 1$ may loosen or tighten, so the inequality should be re-checked before being used as a design rule."],"forward_implications":["An IRS-aided link cannot achieve a higher SNR or information rate than an $N$-antenna Massive MIMO receiver in the modeled same-location, line-of-sight setup, regardless of $N$.","The advertised $N^2$ scaling of IRS SNR should not be read as a power advantage; the extra factor is the bounded fraction of reflected power that reaches the receiver.","Numerically, matching a 64-antenna mMIMO array requires more than $10^4$ IRS elements when the receiver is 25 m away and roughly $3\\times10^3$ elements when it is 2.5 m away.","For very large $N$, the far-field linear model must give way to the planar-array formula, whose received power saturates at half the transmit power; the IRS comparison inherits that saturation.","Placing the receiver directly behind the surface (holographic beamforming) is the configuration that can theoretically match mMIMO performance."],"supporting_citations":[{"why":"Supplies the Massive MIMO signal model and maximum-ratio combining SNR expression used as the baseline.","marker":"[3]"},{"why":"Introduced the IRS beamforming model and the $N^2$ SNR scaling that this paper reinterprets.","marker":"[6]"},{"why":"Establishes that the IRS forwards the signal without amplification, the relay interpretation used to explain the SNR gap.","marker":"[13]"},{"why":"Supplies the conventional mMIMO power scaling law that the paper contrasts with the IRS scaling.","marker":"[15]"},{"why":"Provides the exact planar-array received-power expression used to check the far-field approximation and the near-field IRS SNR.","marker":"[17]"},{"why":"Derives the signal model $y = g^T \\Theta h \\sqrt{P_{\\mathrm{tx}}}s + n$ that underlies the IRS SNR calculation.","marker":"[18]"}],"fun_headline_variants":["IRS's N^2 gain is capped by energy conservation; Massive MIMO wins","Massive MIMO always outperforms IRS even when IRS scales quadratically","Quadratic IRS scaling still loses to linear same-size Massive MIMO","Energy conservation breaks IRS's scaling illusion; Massive MIMO is superior"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The result assumes the IRS and the Massive MIMO array are compared at the same location, with the receiver physically separated from the IRS, under free-space line-of-sight propagation and isotropic antennas; outside that deployment the ranking is not proved.","fun_headline_variants_meta":{"raw":{"variants":["IRS's N^2 gain is capped by energy conservation; Massive MIMO wins","Massive MIMO always outperforms IRS even when IRS scales quadratically","Quadratic IRS scaling still loses to linear same-size Massive MIMO","Energy conservation breaks IRS's scaling illusion; Massive MIMO is superior"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000433,"raw_usage":{"total_tokens":2160,"prompt_tokens":854,"completion_tokens":1306,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":470,"completion_tokens_details":{"reasoning_tokens":1226}},"tokens_in":470,"tokens_out":1306,"duration_ms":10616,"temperature":1.0,"reasoning_tokens":1226,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:21:54.570028+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Under the paper's assumptions, compute or measure the SNR of an $N$-element IRS and an $N$-antenna mMIMO receiver with optimal phases and perfect channel knowledge. Any instance with $\\mathrm{SNR}_{\\mathrm{IRS}} > \\mathrm{SNR}_{\\mathrm{mMIMO}}$ refutes Proposition 1; the analytic check is whether the prefactor $\\mu^2 N \\beta_g$ can exceed one while respecting energy conservation.","supporting_citations":[{"cited_title":"more efficient than conventional mMIMO","cited_arxiv_id":null,"evidence_quote":"Supplies the Massive MIMO signal model and maximum-ratio combining SNR expression used as the baseline."},{"cited_title":"Noncooperative cellular wireless with unlim- ited numbers of base station antennas,","cited_arxiv_id":null,"evidence_quote":"Introduced the IRS beamforming model and the $N^2$ SNR scaling that this paper reinterprets."},{"cited_title":"Smart radio environments empowered by re- configurable AI meta-surfaces: an idea whose time has come,","cited_arxiv_id":null,"evidence_quote":"Establishes that the IRS forwards the signal without amplification, the relay interpretation used to explain the SNR gap."},{"cited_title":"Anomalous terahertz re- flection and scattering by flexible and conformal coding meta- materials,","cited_arxiv_id":null,"evidence_quote":"Supplies the conventional mMIMO power scaling law that the paper contrasts with the IRS scaling."},{"cited_title":"Intelligent Reflecting Surface vs. Decode-and-Forward: How Large Surfaces Are Needed to Beat Relaying?","cited_arxiv_id":"1906.03949","evidence_quote":"Provides the exact planar-array received-power expression used to check the far-field approximation and the near-field IRS SNR."}],"review_version":1}