Make your paper interactiveCreate

Text from the interactive explanation

Short-Block Polar-Coded Reverse and Direct Reconciliation in CV-QKD

Dingzhao Wang, Xin Liu, Chao Xu, Soon Xin Ng, Lajos Hanzo

Original paper

IEEE Open Journal of Vehicular Technology · 2025

Future 6G wireless networks may lock their control messages with keys carried by light: keys whose secrecy rests on quantum physics rather than on hard maths puzzles. But light arrives noisy, so the two ends of the link hold keys that are almost identical. This paper asks how to make them exactly identical, quickly, for the short blocks that fast networks need.

The scene is running for real: your browser is encoding random keys with a polar code , sending them through quantum-channel noise, and trying to repair them, block after block.

Simulated in your browser: rate ½ polar code, Gaussian-approximation construction, list decoder keeping 16 candidates (as in the paper), equivalent binary-input Gaussian-noise channel.

1 · A key made of light

Why Alice's and Bob's copies differ

Quantum key distribution (QKD) lets two parties, traditionally Alice and Bob, grow a shared secret key while any eavesdropper (Eve) leaves detectable traces. The first scheme, BB84 in 1984, used single photons; that family needs costly single-photon detectors, often cooled to low temperatures.

Continuous-variable QKD instead writes random numbers onto the amplitude and phase (the two “quadratures”) of ordinary laser pulses, and reads them with the room-temperature detectors already used in telecoms, even on silicon chips. The catch: every pulse carries an irreducible quantum fuzz called shot noise, and the fiber loses light along the way.

Illustrative model. Gold dot: the value Alice encoded; cyan halo: shot-noise fuzz (one shot-noise unit). Readouts use the paper's channel equations (13)–(14) with its detector settings (efficiency 0.98, electronic noise 0.01, excess noise 0.002 shot-noise units, 0.2 dB/km fiber loss) and an illustrative modulation strength VS = 4, which the paper does not specify.

After measuring, Alice and Bob normalise their data and apply multidimensional reconciliation (the paper uses 8 dimensions): Bob's random bits are mapped onto points of a unit sphere and a public rotation is shared. The result behaves like an ordinary binary channel with Gaussian noise (a “BI-AWGN” channel). The paper simulates its codes on this equivalent channel; the hero scene does the same.

2 · Repairing a key out loud

The paper's own 8-bit example

Alice and Bob may only talk over a public line that Eve hears. So Bob never sends his key. He builds a longer codeword from it, padding it with “frozen” placeholder bits that everyone agrees are 0, and mixes everything through a butterfly of exclusive-or (XOR) operations. The only public hint is which positions are frozen.

In the paper's example Bob's key is 1 1 1 1, the code is 8 bits long, and the frozen positions are {1, 2, 3, 5}. Change the key, drag the received values, and watch Alice's decoder.

Drag a bar (or focus it and use ↑/↓). Up = “looks like 0”, down = “looks like 1”; length = confidence. Coral outline = the bar points the wrong way.

Alice's decoder: which legal codeword explains what arrived?

Legal codewords use frozen positions {1, 2, 3, 5}. There are 16 of them, one per possible key.

The polar encoder computes x = u·GN with GN = BNF2⊗n, where F2 = [[1,0],[1,1]] and BN is a bit-reversal permutation (paper eqs. 1–4). Here u holds the key bits in the reliable positions and zeros in the frozen ones. The frozen set is chosen by estimating every position's reliability at the channel's signal-to-noise ratio with a Gaussian approximation. With key 1111 the encoder gives codeword 0 1 1 0 1 0 0 1. The received word in the paper, 0 1 1 0 1 0 0 0, has one error in the last position, and decoding still returns 1111.

3 · Sorting noise into perfect and useless

Channel polarization

Why does a butterfly of XORs help? Each stage pairs up channel uses: one partner gets a slightly worse view of its bit, the other a slightly better one. Repeat the pairing log₂N times and N equally noisy channel uses drift apart into “virtual channels” that are nearly perfect or nearly useless. Put the key on the good ones; park the known placeholders on the bad ones.

Polarization doesn't create information, it sorts it: the average stays exactly where it started. Scrub through the stages, then make the block longer.

Computed in your browser with the Gaussian approximation the paper uses to pick frozen positions. Bar height = capacity of each virtual channel (0 = useless, 1 = perfect). Gold = the N/2 most reliable (carry key bits); indigo = frozen. Dashed line = the raw channel's capacity.

4 · The short-block race

Polar versus LDPC, block by block

The usual choice for CV-QKD reconciliation is a low-density parity-check (LDPC) code: many sparse parity checks, decoded by passing beliefs back and forth along a graph for many rounds (the paper assumes 50). Its public hint is the syndrome, the pattern of failed checks. LDPC codes shine for long blocks, but designing good short ones is hard. Polar codes are simple to construct and are already used for short control messages in 5G.

Curves redrawn from Fig. 8 of the paper (rate ½, ideal public channel). Gold solid = polar (list decoding, L = 16); lilac dashed = LDPC (belief propagation). Drag on the chart to move the cursor.

Coding gain at bit error rate 10⁻³

Redrawn from Fig. 9. Higher = more noise tolerated.

Decoding effort for this block length

Computed from the paper's operation counts (eq. 35; LDPC belief propagation with 3 ones per column, 50 rounds), as in Fig. 10.

5 · The public line is noisy too

Double protection, and sharing the work

Earlier schemes assumed that the public conversation arrives error-free. Real wireless links fade and hiss. For polar reconciliation, the public message is the frozen-position list, and you saw in chapter 2 what one garbled index does. Sent naively it is also bulky: at N = 1,024 each position needs 10 bits, so a half-rate code's 512 frozen positions cost 5,120 bits.

The paper's answer is double protection : a polar code guards the quantum link, and a separate LDPC code guards the public link.

Radio animation: illustrative (fading strength and corrupted bits drawn at random). Chart redrawn from Fig. 11: System B, polar N = 512, rate ½, list size 16, public channel with Rayleigh fading at SNRc.

Who does the heavy lifting?

The paper compares four arrangements. Each decoder is a big job, so where it runs matters.

In reverse reconciliation Bob's measurements are the reference, so the error correction is modelled as if Bob's codeword travelled to Alice through an equivalent noisy channel (cyan).

6 · Whose copy is the reference?

Direct versus reverse reconciliation

In direct reconciliation (DR) , Alice's data is the reference and Bob corrects toward it. In reverse reconciliation (RR) , Bob's measurement results are the reference and Alice corrects toward them. Why should that matter? Because any light lost in the fiber could be light Eve collected with a beam splitter.

Idealized illustration: all loss treated as Eve's beam splitter, 0.2 dB/km. Once more than half the light is lost (3 dB, about 15 km here), Eve's copy of Alice's signal beats Bob's, so DR cannot be secure. In RR the reference is Bob's own measurement, which includes noise Eve never sees, so a key remains possible.

How far can a secret key reach?

Redrawn from Figs. 14–16 of the paper. Secret key rate in bits per channel use (log scale) versus fiber distance. Assumes detector efficiency 0.98 and electronic noise 0.01 shot-noise units; reconciliation efficiency β is evaluated where the block error rate is 0.1.

7 · What this brings closer

Short keys, repaired well

Back to the opening idea: control messages in fast wireless networks are short, and quantum keys for them would be reconciled in short blocks. Across simulations of four system designs, the paper finds:

Polar codes reach a lower block error rate than LDPC codes for every block length up to 512 bits; LDPC is better at 1,024.

Polar list decoding needs about a tenth of the arithmetic of 50-round LDPC decoding (e.g. 12,272 vs 121,600 operations for 32 key bits).

A fading public line causes an error floor; LDPC-protected side information brings performance back to the ideal case once the public link reaches about 6.5 dB.

With reverse reconciliation, short polar blocks reach about 24–29 km versus 16–26 km for LDPC at the same lengths. Direct reconciliation reaches only about 2–3 km but gives the higher rate at very short range.

Together with splitting the decoding load between the two ends, these results point toward practical, quantum-safe key exchange for next-generation wireless networks. The authors name the next steps: codes whose rate and block length can adapt (designed with EXIT charts), and turbo codes with automatic repeat requests.

All findings above come from the paper's computer simulations: an equivalent binary-input Gaussian-noise quantum channel, a Rayleigh-fading public channel, and a secret-key-rate analysis under collective attacks.

Source paper. Dingzhao Wang, Xin Liu, Chao Xu, Soon Xin Ng and Lajos Hanzo, “Short-Block Polar-Coded Reverse and Direct Reconciliation in CV-QKD,” IEEE Open Journal of Vehicular Technology, vol. 6, pp. 2195–2209, 2025. doi:10.1109/OJVT.2025.3591417

Charts are approximate redraws, restyled and traced from Figs. 8, 9, 11, 14, 15 and 16 of the paper, © 2025 The Authors, licensed under CC BY 4.0. The paper's 8-bit example is reproduced as an interactive. Live simulations, illustrative animations and the phase-space model are new to this page and labelled as such.

Topics in this iPaper

Engineering
quantum-safe wireless networks in the emerging 6 G era, mapping quantum information onto the amplitude or phase of electromagnetic waves

Abstract, p. 2195

Physics
Alice firstly prepares Gaussian-modulated coherent states and then transmits them to Bob over the quantum channel.

Section III, p. 2198

Computer science
Polar codes, invented by Arikan [35] have been shown to approach the Shannon limit in the Binary Discrete Memoryless Channel (BDMC).

Section II, p. 2197

quantum key distribution
Continuous-variable quantum key distribution (CV-QKD) is a promising technique

Abstract, p. 2195

polar codes
Polar codes safeguard quantum transmissions due to their superior performance for short block lengths

Abstract, p. 2195

low density parity check codes
LDPC codes robustly protect the classical side information exchanged over auxiliary classical channels

Abstract, p. 2195

information reconciliation
there are two types of reconciliation schemes, which are DR and RR

Section III-A, p. 2198

forward error correction
both Alice and Bob try to obtain the reconciled keys with the aid of FEC codes

Section III, p. 2198

secret key rate
the SKR of a CV QKD system using RR is defined as [27]

Section V-A, p. 2202

Short-Block Polar-Coded Reverse and Direct Reconciliation in CV-QKD — Interactive explanation | iPaper.ai