arxiv: 2604.24918 · v1 · submitted 2026-04-27 · 💻 cs.IT · eess.SP· math.IT

Recognition: unknown

Covariance-Aware Demapping on Fourier-Curve Constellations

Bin Han , Muxia Sun , H. Vincent Poor , Hans D. Schotten

Authors on Pith no claims yet

Pith reviewed 2026-05-07 17:48 UTC · model grok-4.3

classification 💻 cs.IT eess.SPmath.IT

keywords covariance-aware demappingFourier-curve constellationstangent artificial noisematched maximum-likelihood decoderrank-one correctionLDPC codingphysical-layer securityBICM achievable rate

0 comments

The pith

A rank-one covariance correction in the maximum-likelihood decoder improves BLER performance by 5 dB for tangent-space artificial noise on Fourier-curve constellations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that tangent-space artificial noise on a curved constellation produces observations whose covariance matrix is rank-one and varies with the transmitted symbol. This structure allows a matched decoder to differ from the ordinary Euclidean nearest-neighbor rule by one inexpensive rank-one update per candidate symbol. For a Fourier-curve constellation used with regular LDPC coding, the correction widens the operating range at 10^{-1} block error rate by about 5 dB relative to the mismatched Euclidean decoder on the identical transmitter. The advantage is corroborated by both coded block-error-rate curves and bit-interleaved achievable-rate calculations. The same correction generalizes to per-tone Ricean fading through the Woodbury identity and remains effective after 6-bit quantization of the required lookup table.

Core claim

Injecting artificial noise along the tangent space of a curved constellation makes each transmitted symbol induce a Gaussian observation with a symbol-dependent rank-one covariance, so the matched maximum-likelihood decoder differs from the Euclidean nearest-neighbor decoder by a single rank-one correction per candidate. A baseband-demapper realization of this correction is developed for the Fourier-curve constellation and tested in a regular (3,6) LDPC-coded link at (k,M)=(20,64).

What carries the argument

The symbol-dependent rank-one correction term applied to the Euclidean distance metric, arising directly from the tangent-space artificial-noise covariance.

If this is right

The matched decoder widens the BLER=10^{-1} operating range by approximately 5 dB over the Euclidean-mismatched decoder on the same tangent-AN transmitter.
A bit-interleaved coded-modulation achievable-rate calculation reproduces the same matched-metric advantage under the tested labeling and max-log demapper.
The Woodbury identity extends the rank-one correction to per-tone Ricean fading without changing the core demapper structure.
Quantizing the constellation-tangent lookup table to 6 bits produces no measurable degradation in coded block error rate at the tested points.
A design-aware bounded-search eavesdropper lacking the phase key fails to decode any tested LDPC codeword within a 10^3 non-code-aided search budget.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same covariance structure could be exploited for other curved or manifold constellations where tangent artificial noise is feasible.
The modest 50-100 percent increase in multiply-accumulate operations per symbol suggests the technique is compatible with existing baseband hardware pipelines.
Physical-layer security arguments would be strengthened by quantifying the eavesdropper degradation under code-aided and known-preamble attacks.
The approach may reduce the power penalty usually paid for artificial-noise security by allowing reliable operation closer to the Shannon limit of the legitimate link.

Load-bearing premise

Each transmitted symbol produces an observation that is exactly Gaussian with a covariance matrix that is rank one and different for every symbol.

What would settle it

Monte-Carlo simulation of the LDPC-coded block error rate curve for the matched decoder versus the Euclidean decoder on the identical tangent-AN transmitter at the tested operating point; the 5 dB horizontal shift at BLER=10^{-1} either appears or does not.

Figures

Figures reproduced from arXiv: 2604.24918 by Bin Han, Hans D. Schotten, H. Vincent Poor, Muxia Sun.

**Figure 1.** Figure 1: Tangent-AN geometry at M=8, β=0.3, σc=0.14. Top: received-signal densities with tˆi. Bottom: matched (red) vs. Euclidean (dashed) decision regions. Columns (a,d) k=1 ambient (8-PSK on the unit circle); (b,e) k=2 ambient projected to (cos θ, sin 2θ)/ √ 2 with regions lifted to R4 ; (c,f) intrinsic (θ, r) at k=2. Decision regions in (e) are projection artefacts, not rigorous R4 regions. not restore the AWGN … view at source ↗

**Figure 2.** Figure 2: End-to-end Tx–channel–Rx chain. Red blocks: tangent-AN injector (Tx) and rank-one correction MF bank (Rx). The AWGN cloud is the post-sync, view at source ↗

**Figure 3.** Figure 3: Coded-BLER, (k, M)=(20, 64), regular (3, 6) LDPC, 95% CP bands. (a) Matched extends BLER=10−1 by ∼5.1 dB over Euclidean (B1); B2’s isotropic covariance carries no candidate-dependent label, so it tracks the proposed curve; B4, a 20×-energy repetition sanity check rather than a same-energy baseline, stays at BLER=1. (b) Matched is robust across the tested β range; Euclidean degrades sharply once β ≳ 0.25. 6… view at source ↗

**Figure 4.** Figure 4: BICM-AIR (bits/complex-slot) vs. ρslot at (k, M)=(20, 64), β=0.3. Solid: s⋆-optimised; dashed: s=1 for matched/Euclidean. Horizontal grey: operating SE. 4 6 8 16 32 LUT bit-width 0.0 0.2 0.4 0.6 0.8 1.0 BLER Matched, c=0.20 Eucl., c=0.20 Matched, c=0.24 Eucl., c=0.24 Matched, c=0.28 Eucl., c=0.28 view at source ↗

**Figure 5.** Figure 5: BLER vs. LUT bit-width at β=0.3 and σc ∈ {0.20, 0.24, 0.28} (pre-/mid-/post-waterfall). Quantizer: uniform symmetric fixed-point clipped at ±1, round-to-nearest, applied to the LUT entries only (both x¯i and tˆi); branch-metric and LDPC-LLR accumulation remain double-precision. The tangents are not renormalized after quantization, so the receiver evaluates an approximate matched metric whose Sherman–Morris… view at source ↗

read the original abstract

Injecting artificial noise (AN) along the tangent space of a curved constellation makes each transmitted symbol induce a Gaussian observation with a symbol-dependent rank-one covariance, so the matched maximum-likelihood (ML) decoder differs from the Euclidean nearest-neighbor decoder by a single rank-one correction per candidate. We develop a baseband-demapper realization of this correction for the Fourier-curve constellation and instantiate a regular $(3,6)$ low-density parity-check (LDPC)-coded link at $(k,M){=}(20,64)$. Against four baselines (Euclidean-mismatched, flat-constellation isotropic-AN, no-AN, and same-spectral-efficiency narrowband), the matched decoder extends the BLER${=}10^{-1}$ operating range by approximately $5$\,dB over the Euclidean-mismatched counterpart on the same tangent-AN transmitter, at a cost of $2kM$ additional multiply-accumulate operations per symbol ($+50\%/+100\%$ under residual/template-correlation accounting) and a $20$\,KB constellation--tangent lookup table ($10$\,KB incremental over a Euclidean template-only LUT). A bit-interleaved coded-modulation achievable-rate (BICM-AIR) computation supports the same matched-metric advantage at the tested labeling and max-log demapper, indicating that the BLER gain is not merely an artifact of this particular LDPC simulation, and a Woodbury extension generalizes the rank-one correction to per-tone Ricean fading. In the tested Monte-Carlo runs, a design-aware bounded-search eavesdropper without the phase-key shows no successful LDPC decoding at any tested $k\in\{2,8,20\}$ within a $B{=}10^{3}$ non-code-aided search budget; code-aided, multi-frame, and known-preamble attacks are left to follow-up work. LUT quantization down to $6$ bits yields no measurable coded-BLER degradation at the tested operating points.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper shows how to turn tangent-space AN on Fourier-curve constellations into a simple rank-one correction on top of Euclidean demapping, and the LDPC simulations back a clean 5 dB BLER gain at the tested point.

read the letter

The main thing to know is that the matched decoder for this setup reduces to the usual nearest-neighbor metric plus one rank-one term per candidate, and they have a baseband implementation that works with a 20 KB LUT and only 50-100% extra MACs. They pair it with a regular (3,6) LDPC code at (k,M)=(20,64) and report the 5 dB extension of the BLER=10^{-1} range over the Euclidean-mismatched decoder on the same tangent-AN transmitter. BICM-AIR calculations line up with the BLER curves, and the Woodbury version handles per-tone Ricean fading without extra trouble. 6-bit quantization of the LUT shows no measurable loss in the runs they did. The eavesdropper note is that a design-aware bounded search without the phase key fails to decode within the tested budget, though they flag code-aided and multi-frame attacks as open. The model itself is direct: the tangent AN produces exactly the symbol-dependent rank-one covariance update, so the correction follows from the matrix inversion lemma with no extra fitting. That keeps the circularity burden low. The comparisons to flat isotropic AN, no-AN, and narrowband baselines are useful for context. One soft spot is that the headline gain sits at BLER=10^{-1} rather than the lower rates where coded systems usually operate, so it would help to see the curves at 10^{-3} or below. The complexity numbers are honest but the overhead is real. This is aimed at people doing physical-layer security or waveform design for 5G/6G links. A reader who needs concrete demapper details and simulation numbers will get something usable. The work is grounded enough and the claims are checkable, so it deserves a serious referee.

Referee Report

2 major / 3 minor

Summary. The paper claims that tangent-space artificial noise injection on Fourier-curve constellations produces symbol-dependent rank-one covariance updates to the observation, so the matched ML demapper reduces to Euclidean distance plus a single rank-one correction (via the matrix inversion lemma). For a regular (3,6) LDPC-coded system at (k,M)=(20,64), Monte-Carlo simulations show this matched decoder extends the BLER=10^{-1} operating range by approximately 5 dB relative to Euclidean demapping on the identical tangent-AN transmitter. The advantage is corroborated by BICM-AIR computations at the tested labeling and max-log demapper; a Woodbury extension handles per-tone Ricean fading; a 20 kB LUT implementation (6-bit quantization) incurs no measurable coded-BLER loss; and a design-aware bounded-search eavesdropper without the phase key fails to decode within the tested budget.

Significance. If the Gaussian rank-one model holds, the result is significant: it converts the structural side-effect of tangent AN into a concrete decoding gain of 5 dB at a practical operating point, with only +50-100% MAC overhead and a modest incremental LUT. The BICM-AIR cross-check, Woodbury generalization, and quantization robustness are concrete strengths that make the contribution reproducible and extensible. The work directly addresses the gap between curved-constellation security constructions and practical demapping.

major comments (2)

[§3] §3 (Demapper Realization): the reduction to Euclidean distance plus rank-one correction is presented as following directly from the AN construction, but the manuscript should include an explicit verification (perhaps via the tangent-space definition) that no additional cross-symbol or higher-rank terms arise under the stated baseband model.
[Simulation section] Simulation section (results on BLER curves): the central 5 dB claim at BLER=10^{-1} is load-bearing; the number of Monte-Carlo trials, error-bar handling, and data-exclusion rules must be stated so that the reported gain can be assessed for statistical reliability.

minor comments (3)

[Abstract] The abstract is information-dense; early introduction of the precise definition of 'tangent-AN transmitter' and the four baselines would improve readability.
[Complexity discussion] Complexity accounting (+50 % / +100 % under residual/template correlation) would benefit from an explicit table breaking down the 2kM additional MACs per symbol.
[Figures] Figure captions for the BLER and AIR plots should list all compared schemes (Euclidean-mismatched, flat isotropic-AN, no-AN, narrowband) rather than referring only to 'baselines'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and the recommendation for minor revision. The suggestions will improve the clarity and reproducibility of the manuscript, and we address each major comment below.

read point-by-point responses

Referee: [§3] §3 (Demapper Realization): the reduction to Euclidean distance plus rank-one correction is presented as following directly from the AN construction, but the manuscript should include an explicit verification (perhaps via the tangent-space definition) that no additional cross-symbol or higher-rank terms arise under the stated baseband model.

Authors: We agree that an explicit verification strengthens the presentation. In the revised §3 we will insert a short derivation that begins from the tangent-space definition of the injected AN, substitutes into the baseband observation model, obtains the symbol-dependent rank-one covariance, and applies the matrix inversion lemma to show that the ML metric reduces precisely to Euclidean distance plus one rank-one correction, with no cross-symbol or higher-rank terms introduced by the model. revision: yes
Referee: [Simulation section] Simulation section (results on BLER curves): the central 5 dB claim at BLER=10^{-1} is load-bearing; the number of Monte-Carlo trials, error-bar handling, and data-exclusion rules must be stated so that the reported gain can be assessed for statistical reliability.

Authors: We will augment the Simulation section with the requested details: the total number of Monte-Carlo trials executed per SNR point, the method used to compute error bars, and the precise data-exclusion rules applied when generating the BLER curves. These additions will allow readers to evaluate the statistical reliability of the reported 5 dB gain. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper derives the covariance-aware ML demapper directly from the standard maximum-likelihood metric applied to the Gaussian observation model with symbol-dependent rank-one covariance induced by tangent-space AN. This is a straightforward application of the matrix inversion lemma (Woodbury identity) to obtain the rank-one correction term, without any self-definitional loops, fitted parameters renamed as predictions, or load-bearing self-citations. The reported 5 dB BLER gain at 10^{-1} is corroborated by independent LDPC simulations and BICM-AIR computations, and the LUT/Woodbury extensions are presented as implementation details. The derivation chain is self-contained against the stated model assumptions and external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the standard additive Gaussian noise model with the additional structure that tangent AN produces a symbol-dependent rank-one covariance perturbation; no free parameters are explicitly fitted in the abstract, and no new entities are postulated beyond the Fourier-curve geometry (likely defined in prior work).

axioms (1)

domain assumption The received signal is a Gaussian random vector whose covariance is the identity plus a rank-one term depending on the transmitted symbol.
Directly stated in the abstract as the effect of injecting AN along the tangent space.

pith-pipeline@v0.9.0 · 5670 in / 1488 out tokens · 94984 ms · 2026-05-07T17:48:18.116565+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

13 extracted references · 3 canonical work pages · 1 internal anchor

[1]

Matched and Euclidean-Mismatched Decoding on Fourier-Curve Constellations with Tangent Noise

B. Han, H. Chen, M. Sun, H. V . Poor, and H. D. Schotten, “Matched and Euclidean-mismatched decoding on Fourier-curve constellations with tangent noise,” 2026,arXiv:2604.14844 [cs.IT]

work page internal anchor Pith review Pith/arXiv arXiv 2026
[2]

The wire-tap channel,

A. D. Wyner, “The wire-tap channel,”Bell Syst. Tech. J., vol. 54, pp. 1355–1387, 1975

1975
[3]

Broadcast channels with confidential mes- sages,

I. Csisz ´ar and J. K ¨orner, “Broadcast channels with confidential mes- sages,”IEEE Trans. Inf. Theory, vol. 24, pp. 339–348, 1978

1978
[4]

Guaranteeing secrecy using artificial noise,

S. Goel and R. Negi, “Guaranteeing secrecy using artificial noise,” IEEE Trans. Wireless Commun., vol. 7, pp. 2180–2189, 2008

2008
[5]

Secure transmission with multiple antennas—II: The MIMOME wiretap channel,

A. Khisti and G. W. Wornell, “Secure transmission with multiple antennas—II: The MIMOME wiretap channel,”IEEE Trans. Inf. Theory, vol. 56, pp. 5515–5532, 2010

2010
[6]

Principles of physical layer security in multiuser wireless networks: A survey,

A. Mukherjee, S. A. A. Fakoorian, J. Huang, and A. L. Swindlehurst, “Principles of physical layer security in multiuser wireless networks: A survey,”IEEE Commun. Surveys Tuts., vol. 16, pp. 1550–1573, 2014

2014
[7]

A survey on artificial noise for physical layer security,

H. Niuet al., “A survey on artificial noise for physical layer security,” 2025,arXiv:2507.06500

work page arXiv 2025
[8]

Mismatched decoding: Error exponents, second-order rates and saddlepoint approx- imations,

J. Scarlett, A. Martinez, and A. Guill ´en i F `abregas, “Mismatched decoding: Error exponents, second-order rates and saddlepoint approx- imations,”Found. Trends Commun. Inf. Theory, vol. 17, pp. 93–300, 2020

2020
[9]

Bit-interleaved coded modulation revisited: A mismatched decoding perspective,

A. Martinez, A. Guill ´en i F `abregas, G. Caire, and F. M. J. Willems, “Bit-interleaved coded modulation revisited: A mismatched decoding perspective,”IEEE Trans. Inf. Theory, vol. 55, pp. 2756–2765, 2009

2009
[10]

Directional modulation technique for phased arrays,

M. P. Daly and J. T. Bernhard, “Directional modulation technique for phased arrays,”IEEE Trans. Antennas Propag., vol. 57, pp. 2633–2640, 2009

2009
[11]

Decomposed and distributed directional modulation for secure wireless communications,

B. Qiu, J. Cheng, and W. Zhang, “Decomposed and distributed directional modulation for secure wireless communications,” 2024, arXiv:2407.13289

work page arXiv 2024
[12]

H. V . Poor,An Introduction to Signal Detection and Estimation, 2nd ed. Springer, 1994

1994
[13]

Bit-interleaved coded modula- tion,

G. Caire, G. Taricco, and E. Biglieri, “Bit-interleaved coded modula- tion,”IEEE Trans. Inf. Theory, vol. 44, pp. 927–946, 1998

1998