{"id":"ec2fcbd2-0508-472f-8870-85e217154492","arxiv_id":"2603.07836","paper_version":2,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":6,"one_line_summary":"Applying a Hadamard transform at the source before modulation is claimed to reduce bit error rates in downlink NOMA by 10–15 dB under imperfect CSI and SIC.","lead":"This paper proposes applying a Hadamard transform to user data before modulation in a NOMA downlink, claiming large bit-error-rate gains over prior NOMA schemes under imperfect channel knowledge. A generalist reader might care because the scheme is simple and claims to fix the main practical weakness of NOMA: sensitivity to imperfect channel state information and error propagation.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Analytical BER for H-NOMA is never derived; the noise-halving argument in Sec. III.A is algebraically invalid, and the four-user results contradict the abstract's universal gain claim.","rationale":"The reader's weakest assumption correctly identifies that the analytical BER expressions are imported from prior T-NOMA work without an H-NOMA-specific derivation. My analysis confirms that the paper's own attempted derivation of a noise-halving effect (Sec. III.A) is algebraically invalid, which strengthens the concern: not only is the BER formula not derived for H-NOMA, but the explanatory mechanism itself is erroneous. The four-user results in Sec. IV.2 additionally contradict the abstract's universal-gain claim, providing direct internal evidence against the headline statement. Because the central quantitative claims are not supported by a correct derivation or credible simulation validation, the paper remains unverified. The reader's UNVERDICTED verdict is appropriate; I do not see reason to move to ACCEPT, and the evidence is too incomplete and internally inconsistent to merit REJECT outright (the underlying idea could still be salvageable with a correct derivation and code). A concrete simulation or independent re-derivation would settle the matter.","tokens_in":11592,"tokens_out":8625,"duration_ms":73340,"concrete_test":"Re-implement the two-user H-NOMA downlink exactly as specified in Sec. II: apply the 2×2 Hadamard transform to the binary data vectors before QPSK modulation, superpose with power allocations α1=0.7 and α2=0.3, use Rayleigh fading with d1=1000 m, d2=400 m, and the SIC receiver described by Eqs. (21)–(25). Simulate >10^6 bits to obtain BER curves for both users over the same Eb/N0 range. Compare these simulated curves with the analytical formulas in Eqs. (38)–(40) and with the T-NOMA baseline. If the simulated H-NOMA BER deviates from the analytical formulas, or if the analytical formulas are numerically identical to the T-NOMA expressions from [15], then the claimed validation is circular and the 15/10 dB gains are unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that source-level Hadamard spreading yields 15 dB/10 dB BER gains in NOMA—rests on the analytical BER expressions in Eqs. (36)–(40). These are quoted from [15] (Assaf et al.) for T-NOMA with QAM; the text provides no derivation showing how the Hadamard transform enters the error probability. The only mechanism offered is the noise-halving argument in Sec. III.A, but that argument is algebraically incorrect. For the near user, Eq. (22) defines \\hat{x}_1 = r/(g_2\\sqrt{P_1}) = x_1 + \\sqrt{P_2/P_1}x_2 + n/(g_2\\sqrt{P_1}), yet Eq. (23) claims \\hat{x}_1 = x_1. The subsequent subtraction of the two equations in Eq. (24) to obtain Eq. (25) is inconsistent and yields signs depending on the order of subtraction. The claimed factor-of-2 noise reduction is thus an artifact of algebra, not a property of the receiver. Furthermore, Sec. IV.1 states that the BER is 'analytically derived and validated using equations [(14), (16)]'—but Eqs. (14) and (16) are channel-generation formulas, not BER expressions. The analytical curves shown in Fig. 4 are therefore indistinguishable from the T-NOMA formulas of [15], making the validation circular. Finally, the four-user experiment in Sec. IV.2 reports a 7 dB degradation for user 1, directly contradicting the abstract's assertion of universal performance improvement. These internal inconsistencies mean the manuscript does not provide a valid argument for its headline dB numbers.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes H-NOMA, a downlink non-orthogonal multiple access scheme in which the Hadamard transform (HT) is applied at the source, before modulation, for each user's data. The authors claim that this source-level spreading mitigates the adverse effects of fading, imperfect channel state information (CSI), and imperfect successive interference cancellation (SIC), yielding large BER gains: 15 dB for the near user at BER 10^-2 and 10 dB for the far user at BER 10^-1 relative to T-NOMA, and 15 dB for the far user relative to Usman-NOMA. The paper presents an analytical SIC error analysis (Section III.A), BER expressions (Section III.B), Monte Carlo simulations for two- and four-user scenarios, and an image transmission example. The central claim is that H-NOMA is more robust than T-NOMA in general multi-user settings.","tokens_in":11985,"tokens_out":2928,"duration_ms":28648,"significance":"If the claims were correct, source-level Hadamard spreading would be a simple, low-complexity robustness enhancement for NOMA, and the paper would fill a gap in the literature on transform-domain NOMA. The paper addresses a relevant problem and includes Monte Carlo simulations and an application-oriented image transmission study. However, the analytical core is not self-contained: the BER expressions in Section III.B are imported from prior T-NOMA work without any derivation showing how the Hadamard transform enters the error probability. Moreover, the noise-halving argument in Section III.A is algebraically invalid, and the four-user results in Section IV.2 explicitly report a 7 dB degradation for the farthest user, contradicting the abstract's universal-gain claim. The Monte Carlo validation is also circular because the equations cited for validation are channel-generation formulas, not BER expressions. The paper therefore does not currently provide a valid argument for its headline results. No machine-checked proofs or reproducible code are provided. The significance is conditional on a derivation that is absent.","major_comments":[{"comment":"The noise-halving argument is algebraically invalid. For the near user, Eq. (22) defines \\hat{x}_1 = r/(g_2√P_1) = x_1 + √(P_2/P_1)x_2 + n/(g_2√P_1) unless the SIC step has already removed x_2, yet Eq. (23) claims \\hat{x}_1 = x_1. The later subtraction in Eq. (24) to obtain Eq. (25) is inconsistent; the sign of the noise term depends on which equation is subtracted from which. The claimed factor-of-two noise reduction is therefore an artifact of the algebra, not a property of the H-NOMA receiver. A correct derivation must start from the actual H-NOMA detection rule (e.g., inverse HT after joint symbol estimation) and account for the transformed constellation and SIC errors explicitly.","section":"Section III.A, Eqs. (22)-(25)"},{"comment":"The analytical BER expressions are not derived for H-NOMA. Eq. (34) defines the H-NOMA transmitted signal, but the symbols E_k are undefined and Eq. (35) is a garbled Q-function definition. Eqs. (36)-(40) are stated as 'given in [15] as derived from [5]' and are the T-NOMA BER formulas of Assaf et al. The text never shows how the Hadamard transform modifies the constellation, the effective noise variance, or the decision regions. Consequently, the analytical curves in Fig. 4 are indistinguishable from the T-NOMA formulas relabeled as H-NOMA, and the agreement between 'analytical' and simulation is not confirmatory of the H-NOMA mechanism.","section":"Section III.B, Eqs. (34)-(40)"},{"comment":"The paper states that 'the BER of a downlink H-NOMA system is analytically derived and validated through Monte Carlo simulations using equations [(14), (16)].' Equations (14) and (16) are channel-generation formulas for \\hat{g}_2 and \\tilde{g}_2, not BER expressions. This makes the validation circular: the simulation uses the same channel model and the analytical curve is imported from T-NOMA. The statement must be corrected and the BER derivation must be supplied.","section":"Section IV.1, first paragraph"},{"comment":"The four-user results directly contradict the abstract's universal performance claim. The text says: 'for the user1 (the farest), we obtain a degradation of 7 dB or more at a bit-error-rate (BER) of 10^-3 compared to the state-of-the-art existing T-NOMA.' This is a load-bearing inconsistency: the abstract claims that source-level HT 'mitigates the adverse impact of fading and CSI imperfections' and improves reliability, but the paper's own results show that the farthest user is worse under H-NOMA. Either the claim must be restricted to certain user/channel conditions or the contradiction must be resolved with an explanation (e.g., power allocation, SIC ordering, or constellation effects).","section":"Section IV.2, Fig. 5 and accompanying text"},{"comment":"The receiver specified in Section II.C uses maximum-likelihood detection and inverse Hadamard transform, with SIC performed on the transformed-domain symbols. However, the analytical derivation in Section III.A models the receiver as simple division and subtraction of the superimposed symbols, without any inverse HT or ML detection. The analytical model therefore does not match the described H-NOMA receiver. A correct analysis must either analyze the actual ML/inverse-HT receiver or justify why the simplified model is equivalent.","section":"Section II.C and Section III.A"}],"minor_comments":[{"comment":"Equation (35) is not a valid definition of the Q-function. It appears to be a fragment of a longer expression. Please replace with the standard Q-function definition or remove the equation.","section":"Section III.B, Eq. (35)"},{"comment":"The Sylvester-Hadamard matrix of order 2 is written with 1/√2 normalization, but Eq. (6) then uses the Kronecker product without clarifying whether the normalization is preserved. Define H_N consistently throughout.","section":"Section II.B, Eq. (5) and Eq. (6)"},{"comment":"The text says 'user 1 requires an SNR at least 14 dB lower than user 2 to achieve a BER of 10^-3' but the preceding discussion does not make clear whether this is a comparison of H-NOMA curves or a property of the power allocation. Please clarify.","section":"Section IV.1"},{"comment":"Several typos: 'bleu' should be 'blue', 'depected' should be 'depicted', 'are are' should be 'are'. These should be corrected.","section":"Section IV.2"},{"comment":"The table caption is 'ABBREVIATION' but the table contains mathematical definitions. Rename the table and ensure all symbols used in Eqs. (36)-(40) are defined consistently.","section":"Table I"},{"comment":"The notation for channel coefficients alternates between g_k, h_k, and \\mathcal{h}_k. Unify the notation and define all symbols before first use.","section":"General"}],"recommendation":"reject","confidential_remarks":"The manuscript's central analytical claim is unsupported: the BER expressions are taken from prior T-NOMA work without derivation, the noise-halving argument is algebraically invalid, and the four-user results contradict the abstract. The Monte Carlo validation is circular. These are load-bearing issues that cannot be fixed within the manuscript's current scope. I see no basis for acceptance even after minor revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the source-level Hadamard idea is plausible, but the paper's analytical core is not derived for H-NOMA, the noise-halving argument is algebraically wrong, and the four-user results contradict the abstract. The headline dB gains are unsupported as written.\n\nWhat is genuinely new is placing the Hadamard transform before IQ modulation; prior work applied it post-modulation. The two-user Monte Carlo results are directionally consistent with the idea, and the image-transmission PSNR table shows a real effect. The paper also cites the relevant prior work fairly.\n\nBut the soft spots are load-bearing. First, Eqs. (36)–(40) are imported from [15] and labeled as the analytical BER for H-NOMA, with no derivation showing how the transform enters. The text says these expressions come from [15] as derived from [5]; it simply relabels T-NOMA formulas. The simulation section says the BER is 'analytically derived and validated using equations [(14), (16)]', but those are channel-generation formulas, not error probabilities. That makes the validation circular. Second, the mechanism argument in Sec. III.A is algebraically inconsistent. Eq. (22) gives \\hat{x}_1 = x_1 + \\sqrt{P_2/P_1}x_2 + noise, but Eq. (23) silently drops the interference. Subtracting the two equations to get a factor-of-two noise reduction is an artifact of the algebra, not a receiver property. Third, the four-user experiment in Sec. IV.2 reports a 7 dB degradation for the far user, directly contradicting the abstract's claim of universal improvement. The text acknowledges it but does not reconcile it with the headline numbers. Fourth, the figures from which the gains are taken (Figs. 4, 6, 7) are absent from the submission, and the imperfect-CSI variance \\sigma_E^2 is never stated. There is also a mismatch between the metadata author list and the three authors in the full text.\n\nIn short, the central analytical contribution does not exist in this manuscript. The simulations might be fine, but I cannot verify them without the missing figures and parameters. This paper is not ready for peer review. I would desk-reject it in its current form. If the authors provide a real derivation or re-cast the paper as purely empirical, fix the four-user inconsistency, and clean up the author list, it could be resubmitted. As it stands, it doesn't deserve referee time.","headline":"The source-level Hadamard idea is plausible, but the paper's analytical core is not derived for H-NOMA, the noise-halving argument is algebraically wrong, and the four-user results contradict the abstract—so the headline dB gains are unsupported.","tokens_in":12578,"tokens_out":5045,"would_cite":false,"duration_ms":41900,"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":"This paper claims that applying the Hadamard transform to user data before modulation, rather than after it, makes downlink NOMA markedly more robust to fading, imperfect channel knowledge, and imperfect interference cancellation, with repo","keywords":["Hadamard transform","NOMA","bit error rate","imperfect CSI","successive interference cancellation","source-level spreading","QAM","downlink"],"falsifier":"Recompute the two-user BER with a direct Monte Carlo simulation of the full H-NOMA encoder/decoder (Hadamard spreading, QAM, SIC, inverse Hadamard) and compare the near-user error rate against the factor-of-two noise prediction; if the BER does not improve by approximately 3 dB relative to T-NOMA at the operating point, the claimed cancellation mechanism is not driving the reported gains.","tokens_in":11353,"feed_emoji":"📡","tokens_out":3120,"duration_ms":30264,"temperature":0.7,"pith_summary":"The paper proposes H-NOMA, in which each user's bit vector is spread with the Hadamard transform before QAM modulation and despread at the receiver. It argues that this source-level spreading partially cancels post-SIC noise, making detection less sensitive to CSI errors and SIC error propagation. Simulations are reported showing a 15 dB gain for the near user at a bit error rate of 10^-2 and a 10 dB gain for the far user at 10^-1 relative to traditional NOMA, with a further 15 dB far-user gain over a post-modulation Hadamard scheme. If the claim holds, a low-complexity transform at the transmitter gives a large reliability boost without changing the spectral efficiency of NOMA.","feed_headline":"Hadamard transform cuts NOMA bit errors by up to 15 dB","feed_subtitle":"Source-level Hadamard spreading makes NOMA resilient to fading and channel-estimation errors, with reported gains of 10-15 dB.","key_machinery":"The Sylvester-Hadamard transform (H_N), a unitary matrix built recursively as H_N = H_{N-1} ⊗ H_2 with H_2 = (1/sqrt(2))[[1,1],[1,-1]], applied at the source before IQ modulation and inverted at the receiver. Its role is to spread each user's data across the transform domain so that the receiver's SIC estimates, when combined, cancel part of the noise and reduce residual interference, which the paper proposes as the mechanism behind H-NOMA's robustness.","core_discovery":"The central claim is that spreading user data with the Sylvester-Hadamard matrix before modulation cancels part of the noise that survives successive interference cancellation. In the two-user derivation, the near-user symbol estimate in H-NOMA has noise n/(2*g2*sqrt(P2)), half the noise of the traditional estimate n/(g2*sqrt(P2)); for the far user, the estimate also contains a partially cancelled near-user interference term rather than the full interference of T-NOMA. The paper attributes the reported bit-error-rate improvements to this factor-of-two noise reduction and to the fact that all transformed components participate in each user's data estimate. Analytical BER expressions and Monte","pith_inferences":["The analytical BER expressions are quoted from prior non-Hadamard NOMA work, and no derivation shows how the factor-of-two noise reduction enters those Q-function formulas; if the analytical curve is just the T-NOMA formula relabeled, the reported simulation \"validation\" is circular rather than confirmatory.","A natural test is to simulate the exact H-NOMA encoder/decoder with the inverse Hadamard at the receiver and check whether the near-user BER shifts by the expected 3 dB from the noise factor of two; if it does not, the claimed cancellation mechanism is not the source of the gains.","Because HT spreading changes the effective symbol constellation seen by SIC, the optimal power-allocation coefficients for H-NOMA likely differ from T-NOMA; re-optimizing them could shift or shrink the reported gains.","The factor-of-two noise reduction is derived only for two users; whether it grows with transform order for larger N is an open, testable extension."],"forward_implications":["If correct, H-NOMA gives a low-complexity robustness fix that can be added to existing downlink NOMA without changing power allocation or spectrum use.","The reported 10-15 dB BER gains imply users can meet target error rates at substantially lower SNR, which could extend cell coverage or reduce transmit power.","In the four-user results, near users gain 1-5 dB at BER 10^-3 while the farthest user loses about 7 dB, indicating that HT spreading improves most users but can trade away the worst user's performance.","The image-transmission results suggest the scheme also improves perceptual quality for multimedia NOMA, not just raw bit errors."],"fun_headline_variants":["Hadamard spreading halves NOMA noise, cuts errors by 15 dB","Hadamard-NOMA: 15 dB BER gain via source-level spreading","Source Hadamard transform boosts NOMA by up to 15 dB","Hadamard-NOMA: 14-15 dB gains from noise-cancelling spreading","Hadamard transform makes NOMA robust to imperfect CSI"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The paper assumes, without deriving, that the BER expressions taken from prior T-NOMA analysis describe the H-NOMA receiver; if those analytical curves are just the T-NOMA formulas relabeled, the simulation \"validation\" does not confirm the claimed Hadamard mechanism.","fun_headline_variants_meta":{"raw":{"variants":["Hadamard spreading halves NOMA noise, cuts errors by 15 dB","Hadamard-NOMA: 15 dB BER gain via source-level spreading","Source Hadamard transform boosts NOMA by up to 15 dB","Hadamard-NOMA: 14-15 dB gains from noise-cancelling spreading","Hadamard transform makes NOMA robust to imperfect CSI"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000171,"raw_usage":{"total_tokens":1165,"prompt_tokens":856,"completion_tokens":309,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":600,"completion_tokens_details":{"reasoning_tokens":210}},"tokens_in":600,"tokens_out":309,"duration_ms":3202,"temperature":1.0,"reasoning_tokens":210,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T18:33:29.020897+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the two-user BER with a direct Monte Carlo simulation of the full H-NOMA encoder/decoder (Hadamard spreading, QAM, SIC, inverse Hadamard) and compare the near-user error rate against the factor-of-two noise prediction; if the BER does not improve by approximately 3 dB relative to T-NOMA at the operating point, the claimed cancellation mechanism is not driving the reported gains.","supporting_citations":[],"review_version":1}