arxiv: 2605.02951 · v1 · submitted 2026-05-01 · ⚛️ physics.gen-ph

Recognition: unknown

Binary Classifier Wire-Resistance Attack on KLJN: Impact of Narrowing the Resistor Gap

Mehmet Yildirim , Laszlo B. Kish

Authors on Pith no claims yet

Pith reviewed 2026-05-09 14:24 UTC · model grok-4.3

classification ⚛️ physics.gen-ph

keywords KLJN key exchangewire resistance attackbinary classifierresistor gapthermal noiseinformation leaksecure communication

0 comments

The pith

Narrowing the resistor gap in KLJN key exchange reduces a binary classifier wire-resistance attack's success probability toward the ideal limit of 0.5.

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

The paper establishes that shrinking the difference between the high and low resistor values in a KLJN loop causes the statistical clouds of mean-square noise voltages for HL and LH bits to overlap more in a two-dimensional classifier plane. This overlap lowers the eavesdropper's ability to distinguish the bits using a simple rule based on which end shows higher mean-square voltage. Simulations with realistic cable resistance show that a 10 percent resistor difference yields an attack success rate near 0.7, while the rate would reach 0.5 with equal resistors. The work also reports that greater wire resistance can reduce the leak under this classifier attack, the opposite of its effect in the classical mean-square comparison attack.

Core claim

As the low resistor value approaches the high one at the same cable resistance, the HL and LH point clouds in the classifier plane increasingly overlap, and the measured eavesdropper success probability p drops close to 0.7, approaching the ideal limit p = 0.5 as RL approaches RH.

What carries the argument

The two-dimensional classifier plane whose axes are the mean-square noise voltages measured at Alice's and Bob's ends, with Eve's success quantified by a sign-of-difference decision rule on those voltages.

If this is right

With strongly asymmetric resistors such as 4 kOhm and 10 kOhm the HL and LH clouds remain fully separable, allowing the attack to recover nearly all bits.
Reducing resistor asymmetry improves security specifically against this sign-based classifier attack.
Increasing wire resistance can decrease the information leak to the classifier, unlike its effect in the Bergou-Scheuer-Yariv attack.
The overlap trend implies that equal resistors would eliminate this particular leak even with nonzero cable resistance.

Where Pith is reading between the lines

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

Designers may need to balance resistor closeness against the requirement that thermal noise levels remain distinguishable for legitimate parties.
Other potential attacks on KLJN should be rechecked at small resistor gaps to see whether they also weaken.
The counterintuitive wire-resistance dependence suggests that cable parameters could be tuned as an additional security knob under classifier-style threats.

Load-bearing premise

Time-domain simulations of the non-ideal KLJN loop with finite cable resistance accurately capture the statistical behavior of real thermal noise and wire resistance.

What would settle it

A physical KLJN experiment using RL = 9 kOhm and RH = 10 kOhm that measures whether the actual bit-recovery probability lies near 0.7 or significantly closer to 0.5.

read the original abstract

It is shown that narrowing the difference between the high and low resistor values in the Kirchhoff Law-Johnson Noise (KLJN) key exchange strongly affects security against a recently introduced binary classifier-based wire resistance attack. Using time domain simulations of a non-ideal KLJN loop with finite cable resistance, we generate large ensembles of secure (HL/LH) bits and evaluate the mean-square noise voltages at Alice's and Bob's ends. For each bit, these mean-square values form a point in a two-dimensional classifier plane, where the separation between the HL and LH point clouds characterizes the information available to an eavesdropper (Eve). We quantify Eve's success probability p by a simple decision rule based on the sign of the difference between the measured mean-square voltages. For strongly asymmetric resistors (for example RL = 4 kOhm and RH = 10 kOhm) and realistic wire resistances, the HL and LH clouds are fully separable and Eve's p approaches 1, which confirms that the classifier attack can practically recover all secure bits. As the low resistor value approaches the high one (for example RL = 9 kOhm and RH = 10 kOhm) at the same cable resistance, the HL and LH clouds increasingly overlap, and the measured p drops close to 0.7, approaching the ideal limit p = 0.5 as RL approaches RH. A surprising phenomenon is that, in this classifier-based scenario, increasing the wire resistance can decrease the information leak. This counterintuitive effect is strikingly the opposite of the behavior in the classical Bergou-Scheuer-Yariv wire resistance attack, where the mean-square voltages at the two ends of the wire are simply compared.

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

Narrowing the resistor gap weakens this classifier attack on KLJN with a counterintuitive wire-resistance effect, but the results rest only on unvalidated simulations.

read the letter

The key takeaway is that making the resistors closer in value weakens the binary classifier wire-resistance attack on KLJN, dropping Eve's success probability from near 1 down toward 0.7 or lower, and that increasing wire resistance can reduce the information leak in this particular attack. What the paper adds is a systematic look at how the resistor asymmetry controls the separation between the HL and LH point clouds in the mean-square voltage plane. Using large ensembles of time-domain simulations on the non-ideal loop, they plot the clouds and apply the simple sign-difference decision rule to compute p for different RL/RH pairs at fixed cable resistance. The counter-intuitive finding that higher Rw hurts the classifier here, opposite to the Bergou-Scheuer-Yariv case, comes straight from those runs. The simulations are done at scale and produce clean trends that match the described method, so the central claim holds up on its own terms. The soft spots sit in the validation. There is no hardware test, no reported error bars on the p values, and no explicit check that the generated noise matches the expected Johnson-Nyquist spectrum including the wire's own thermal contribution and RC effects. If the sampling or bandwidth in the discrete integration is off, the cloud overlap could be artificial. That matches the stress-test note on simulation fidelity. Readers working on KLJN implementations or physical attacks will get the most from the resistor-tuning results. It is a focused parameter study rather than a broad advance. The work shows clear thinking on the attack mechanics and deserves a serious referee to check the simulation details and perhaps suggest hardware follow-up. I recommend sending it for peer review.

Referee Report

2 major / 2 minor

Summary. The manuscript uses large ensembles of time-domain simulations of a non-ideal KLJN loop with finite cable resistance to generate HL and LH mean-square voltage point clouds. It applies a sign-of-difference decision rule to quantify Eve's success probability p under a binary classifier wire-resistance attack. The central result is that narrowing the RL–RH gap increases cloud overlap and drives p from near 1 (e.g., RL = 4 kΩ, RH = 10 kΩ) toward 0.5 (e.g., p ≈ 0.7 at RL = 9 kΩ, RH = 10 kΩ), while also reporting a counter-intuitive decrease in leakage with increasing wire resistance Rw, opposite to the classical Bergou–Scheuer–Yariv attack.

Significance. If the reported trends are reliable, the work identifies a concrete parameter regime (closer RL and RH values) that can mitigate this particular classifier attack in practical KLJN implementations. The direct, simulation-driven construction of the two-dimensional point clouds and the explicit sign-of-difference rule constitute a clear methodological strength, allowing quantitative p values to be extracted without intermediate fitting.

major comments (2)

[Simulation methodology] Simulation methodology section: the time-domain integration used to produce the voltage traces and mean-square values is not accompanied by any specification of sampling rate relative to the Johnson–Nyquist bandwidth, the generation of correlated thermal noise sources at both ends, or explicit inclusion of the cable's own thermal noise and RC filtering. These omissions are load-bearing because the quantitative claims (p ≈ 0.7 at RL = 9 kΩ / RH = 10 kΩ and the approach to p = 0.5) rest entirely on the fidelity of these statistics.
[Results] Results section (examples with RL = 9 kΩ, RH = 10 kΩ): the reported success probabilities p are given without error bars, bootstrap estimates, or the exact ensemble sizes used to populate the HL/LH clouds. This prevents assessment of whether the observed overlap and the drop from p ≈ 1 to p ≈ 0.7 are statistically robust.

minor comments (2)

[Abstract] Abstract: the statement 'the measured p drops close to 0.7' should specify the wire resistance Rw at which this value was obtained, and the decision rule should be stated explicitly (e.g., sign of which mean-square voltage minus the other).
[Simulation methodology] The manuscript would benefit from a brief statement of how the continuous-time thermal noise spectra are discretized and whether any anti-aliasing or bandwidth limiting was applied before computing the mean-square voltages.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive suggestions. The comments highlight important aspects of reproducibility and statistical rigor that we will address in a revised version of the manuscript.

read point-by-point responses

Referee: [Simulation methodology] Simulation methodology section: the time-domain integration used to produce the voltage traces and mean-square values is not accompanied by any specification of sampling rate relative to the Johnson–Nyquist bandwidth, the generation of correlated thermal noise sources at both ends, or explicit inclusion of the cable's own thermal noise and RC filtering. These omissions are load-bearing because the quantitative claims (p ≈ 0.7 at RL = 9 kΩ / RH = 10 kΩ and the approach to p = 0.5) rest entirely on the fidelity of these statistics.

Authors: We agree that these methodological details are essential for full reproducibility and for confirming that the reported mean-square voltage statistics are physically faithful. In the revised manuscript we will add an explicit subsection describing: (i) the sampling rate (set to 20× the Johnson–Nyquist bandwidth of the highest-frequency component), (ii) the generation of independent thermal noise sources at Alice’s and Bob’s resistors together with the appropriate cross-correlation induced by the finite wire resistance, and (iii) the inclusion of the cable’s own Johnson noise and first-order RC filtering. These additions will be accompanied by a brief validation that the simulated power spectral densities match the analytic expectations for the non-ideal KLJN loop. revision: yes
Referee: [Results] Results section (examples with RL = 9 kΩ, RH = 10 kΩ): the reported success probabilities p are given without error bars, bootstrap estimates, or the exact ensemble sizes used to populate the HL/LH clouds. This prevents assessment of whether the observed overlap and the drop from p ≈ 1 to p ≈ 0.7 are statistically robust.

Authors: We acknowledge that quantitative error estimates are required to substantiate the claimed trends. The original simulations used ensembles of 10^5 independent HL and LH realizations for each resistor pair; we will state this number explicitly. In the revision we will also report bootstrap-derived 95 % confidence intervals on all p values (obtained from 2000 resamples) and will confirm that the reduction from p ≈ 1 to p ≈ 0.7 remains statistically significant (standard error < 0.01) even after accounting for finite-sample fluctuations. revision: yes

Circularity Check

0 steps flagged

No significant circularity; results are direct simulation outputs

full rationale

The paper's central results on HL/LH cloud overlap and Eve's success probability p are obtained by generating voltage traces via time-domain simulations of the non-ideal KLJN loop, computing mean-square values at each end, and applying the sign-of-difference decision rule to the resulting point clouds. These p values (e.g., p approaching 1 for RL=4kΩ/RH=10kΩ and p≈0.7 for RL=9kΩ/RH=10kΩ) are measured outcomes of the simulated ensembles rather than fitted parameters, self-defined quantities, or predictions derived from prior self-citations. No load-bearing step in the derivation chain reduces by construction to its own inputs; the work is self-contained against the stated simulation model.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard Johnson-Nyquist noise modeling and Kirchhoff's laws applied to a lumped-element circuit with added series wire resistance; no new physical entities are postulated and the only free parameters are the chosen resistor and wire values that are varied explicitly.

free parameters (2)

RL and RH resistor pair
Specific numerical values (e.g., 4 kΩ/10 kΩ, 9 kΩ/10 kΩ) are chosen to illustrate the narrowing effect; these are input parameters, not fitted constants.
wire resistance Rw
Finite cable resistance is introduced as a realistic non-ideality and varied to observe the counter-intuitive security effect.

axioms (2)

standard math Thermal noise voltages obey Johnson-Nyquist statistics with mean-square voltage proportional to resistance and temperature.
Invoked implicitly in the generation of the simulated voltage traces.
domain assumption The KLJN loop can be modeled as a lumped circuit with series wire resistance.
The non-ideal loop model used for all time-domain simulations.

pith-pipeline@v0.9.0 · 5620 in / 1492 out tokens · 36116 ms · 2026-05-09T14:24:37.889439+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages

[1]

Resilience of gated avalanche photodiodes against bright illumination attacks in quantum cryptography

Simulations of narrowing the resistance gap In preliminary simulations, generating the noise at Alice and Bob from disjoint segments of the same random-number stream led to a small but systematic non‑zero average power flow between the two sides in the secure states. This behavior contradicts the zero‑power‑flow condition in ideal KLJN and indicates that ...

work page doi:10.1142/s0219477525500324 1949
[2]

Generalized

L.B. Kish, Enhanced secure key exchange systems based on the Johnson-noise scheme, Metrol. Meas. Syst. 20 (2013) 191-204. [51] L.J. Gunn, A. Allison and D. Abbott, A new transient attack on the Kish key distribution system, IEEE Access 3 (2015) 1640-1648. [52] G. Vadai, Z. Gingl and R. Mingesz, Generalized attack protection in the Kirchhoff-law-Johnson-no...

work page arXiv 2013
[3]

Kish's key exchange scheme is insecure

L.B. Kish and O. Saidi, Unconditionally secure computers, algorithms and hardware, such as memories, processors, keyboards, flash and hard drives, Fluct. Noise Lett. 8 (2008) L95–L98. [77] L.B. Kish, K. Entesari, C.-G. Granqvist and C. Kwan, Unconditionally secure credit/debit card chip scheme and physical unclonable function, Fluct. Noise Lett. 16 (2017)...

work page doi:10.1142/s0219477521500504 2008