Make your paper interactiveCreate

Text from the interactive explanation

The Road to Near-Capacity CV-QKD Reconciliation: An FEC-Agnostic Design

Xin Liu, Chao Xu, Yasir J. Noori, Soon Xin Ng, Lajos Hanzo

Original paper

Quantum key distribution · error correction

Faint pulses of laser light can give two people a shared secret key that no eavesdropper can copy unnoticed. But the light arrives noisy, and the two copies of the key must be repaired by talking in public without giving the secret away. This paper redesigns that repair step so that any error-correcting code can do it, including the very best.

Predict first Which code keeps the secret alive at 30 km? Pick one, then drag the distance.

light pulses in the fibre

Bob's codeword (quantum channel, ←)

Alice's codeword (public radio, →)

bit flipped by noise

recent keys: part kept after privacy amplification (frame = full key)

Eve

Illustration of the paper's proposed codeword swap, drawn with a 7-bit toy code. Key rates, reach and the share of each key that survives are computed from the paper's security model with its simulated code efficiencies (Table 5).

01

A key that physics guards

The only encryption proven unbreakable is the one-time pad : scramble a message with a random key as long as the message, use the key once, then destroy it. Its weak point is delivery. Alice and Bob need fresh, identical, secret keys all the time.

Today's internet hands out keys using mathematical puzzles, like factoring huge numbers, that ordinary computers can't solve in any reasonable time. A large quantum computer running Shor's algorithm could. Quantum key distribution (QKD) skips the puzzle. It grows keys from light, and quantum physics makes it impossible to copy that light without leaving traces.

Once the light has been measured, four clean-up steps turn raw readings into a key. Tap each one:

Reconciliation. They fix every mismatch by talking over a public channel, so both end with identical bits without revealing them. This paper redesigns this step.

02

Bob hears a noisy echo

This paper is about continuous-variable QKD . Alice doesn't send single photons. She sends faint laser pulses whose strengths, measured along two directions (physicists call them quadratures), are random numbers drawn from a bell curve. Bob measures them with a homodyne detector, the kind of receiver already used in optical telecoms.

What Bob reads is a weakened, noisy echo of what Alice sent. Each dot is one pulse: Alice's value across, Bob's reading up. Dots on the dashed line would be perfect echoes.

Slide the signal down: the cloud spreads and each pulse carries less than a bit. Illustrative simulation: Bob's reading is Alice's value plus random noise.

Notice the ceiling. No method can extract more than ½·log₂(1 + SNR) bits per pulse, Claude Shannon's capacity. Over long fibre the signal is weak, so each pulse carries only a fraction of a bit. Any error correction that falls short of this ceiling wastes secret key the link could have delivered. This shortfall will come back later as the code's efficiency .

03

Turning a cloud of numbers into bits

Error-correcting codes work on bits, but Alice and Bob hold messy real numbers. The paper builds on a trick called multidimensional reconciliation . Chop the numbers into small groups and scale each group onto a sphere. Bob privately picks random bits, and every bit pattern is a fixed point on that sphere.

Bob announces just one thing in public: the turn that carries his measured point onto his chosen bit point. Alice applies the same turn to her own point. Since her point sits near his, she lands near his bits. Eve hears the turn too, but without Bob's measurement she can't tell where it started.

Faint dots show where Alice landed in recent rounds: amber when she read Bob's bits correctly, red when noise pushed her into the wrong quadrant.

Here's the surprise: the quantum link now behaves like an ordinary noisy channel carrying Bob's bits to Alice (the paper calls it the virtual equivalent channel). The real scheme uses groups of eight numbers on an eight-dimensional sphere; we draw two so you can see it, as the paper's Fig. 4 does. Alice's copy still has wrong bits, so an error-correcting code has to finish the job.

04

Fixing errors in public: the classic recipe

Take the paper's own seven-bit example. Bob's key is 1111010; Alice's copy arrived as 1111011. Bob can't read his key aloud. Instead he announces a syndrome : three parity bits, each saying whether the bits inside one circle add up to odd (1) or even (0). That reveals little about the key, but it lets Alice hunt down her mistake.

A circle glows red when Alice's bits inside it disagree with Bob's parity bit. Tap the bit that sits inside exactly the red circles.

Problem 1 · The public channel isn't perfect

Every earlier syndrome scheme assumed the public channel never makes mistakes. Real links, especially wireless ones, add noise and fading. As you just saw, one garbled parity bit sends Alice to the wrong bit. You can protect the syndrome with its own error-correcting code (the paper's System B), but then Alice runs two heavy decoders while Bob does almost no work.

Problem 2 · Locked to one family of codes

Syndromes only exist for some codes, such as the LDPC codes used throughout earlier work. Convolutional codes, common in communication standards, can't be plugged in directly. Neither can polar codes, nor the near-capacity irregular convolutional codes.

LDPC ✓Convolutional ✗Polar ✗IRCC ✗

05

The swap: two codewords, one key

The paper's answer (its System D) stops sending syndromes. Alice and Bob each send the other a codeword : a bit string carrying the code's built-in redundancy, so a single error can be repaired. Each side repairs what arrives, then combines it with its own codeword using XOR (bit by bit: same → 0, different → 1). The first round below is the paper's worked example.

Why the keys agree. Alice computes b̂ ⊕ c and Bob computes b ⊕ ĉ. When both repairs succeed, b̂ = b and ĉ = c, so the two keys are identical. Eve can copy Alice's codeword off the radio, but the key also depends on Bob's codeword, which travelled through the quantum channel. The paper argues the security is the same as the syndrome recipe's. A block whose repair fails is detected and discarded, as in earlier schemes.

Plug in any code

Nothing in the swap asks the code for a syndrome. All it needs is an encoder and a decoder. Pick a code:

Syndrome recipe, protected (System B)

Codeword swap (System D)

Balanced work suits two similar devices talking to each other; the paper points to device-to-device wireless links.

06

Why the best code reaches farthest

Codes differ in how weak a signal they can still repair. The paper simulated three codes at the same rate: every 2 bits sent carry 1 bit of key (rate ½). The chart shows how often a 10,000-bit block fails to decode as the quantum channel's signal-to-noise ratio rises. Each code falls off a cliff at its own threshold.

Redrawn from the paper's Fig. 20 (simulated, noisy public channel at 6 dB). The 10%-failure thresholds are the paper's Table 5 values; curve shapes are approximate.

The paper scores each code by its reconciliation efficiency β : the code's rate divided by Shannon's ceiling at the signal level where the code starts working. A perfect code would work right at 0 dB, β = 100%.

Paper's Table 5 (simulated, rate-½ codes, 10,000-bit blocks unless noted).

Why efficiency buys distance: Alice can whisper

In the paper's model, Alice tunes her pulse strength to exactly the signal her code needs. Every code then extracts the same ½ bit of corrected key per pulse, but a code that works with a fainter signal lets Alice whisper , and a whisper leaks less to an eavesdropper. The secret is what's left after subtracting the most Eve could know.

07

The road, measured

Now the payoff. The paper's security model uses its fibre loss (0.2 dB per km), detector and noise values, and the simulated efficiencies above. From it we can compute the secret key rate at every distance. Eve's possible share grows with every kilometre until it swallows the key. Where a curve plunges is that code's limit.

Maximum secure distance reported in the paper's Fig. 21 (simulated, 10% block failures). Our recomputation from its equations gives 8.0, 28.4, 34.4 and 38.0 km.

Does a noisy public channel spoil it?

The paper also simulated what happens when the public channel adds noise (AWGN) or fades (Rayleigh). Choose a case it tested:

The lesson: give the public channel a code that suits its quality and the system performs as if it were perfect. Make it too weak and an error floor appears that no amount of quantum signal can remove.

The model assumes the strongest known attack on this protocol, a Gaussian collective attack in which Eve controls the whole fibre. Her maximum knowledge is the Holevo information χ, computed from the covariance matrix of the light. The secret key rate is K = (1 − BLER)·(β·I − χ − Δ). Here I is the Shannon information Alice and Bob share, BLER the block failure rate (0.1) and Δ a small finite-size penalty. Parameters: excess noise 0.002, detector efficiency 0.98, electronic noise 0.01, 10¹² pulses, security parameter 10⁻¹⁰. The PLOB bound, −log₂(1 − T), is the highest key rate any protocol can reach over a direct fibre link with transmission T.

An irregular convolutional code mixes 17 convolutional codes with rates from 0.1 to 0.9, giving each a tuned share of the bits (for example, the rate-0.35 code handles about 61% of them). Its decoder works in two halves that pass beliefs back and forth. The mix is chosen so their exchange keeps a narrow but open "tunnel" to perfect decoding; the narrower the tunnel, the closer the code works to capacity (paper Figs. 18–19).

Under the old error-free-channel assumption, the paper first compared two ways of decoding the same LDPC code. Simple bit-flipping reached β = 33.18% and a secure distance of about 1 km. Belief propagation, which passes soft probabilities and also uses the syndrome, reached β = 81.80% and about 30 km (Figs. 12 and 15).

08

Back to the crossing

So how far can a secret travel on light? As far as Alice and Bob can repair their keys in public without wasting what the light carried. This paper turns that repair into a swap of codewords. Both channels get error correction, noise and fading on the public channel are accounted for, both ends do equal work, and the best codes can finally be plugged in. In the authors' simulations, irregular convolutional codes stretched the secure fibre distance from about 28 km (LDPC) to about 35 km, and about 37 km with longer blocks. That inches toward the PLOB bound, the fundamental limit for a direct link.

The authors place QKD in tomorrow's beyond-5G and 6G networks. There, links may be wireless, may run at microwave or terahertz frequencies, and may connect similar devices that should share the work fairly. A reconciliation step that accepts any code and tolerates an imperfect public channel is one stretch of that road.

Source and credits

Xin Liu, Chao Xu, Yasir Noori, Soon Xin Ng and Lajos Hanzo, “The Road to Near-Capacity CV-QKD Reconciliation: An FEC-Agnostic Design,” IEEE Open Journal of the Communications Society, 2024. doi:10.1109/OJCOMS.2024.3382225

This page is based on the arXiv preprint version of that article, arXiv:2403.16180v1 (24 March 2024), distributed under a Creative Commons Attribution 4.0 licence.

All results quoted come from the paper's computer simulations and analysis, not a physical experiment.

Key-rate curves and the secret budget are recomputed from the paper's equations (7)–(25), using its stated parameters and the efficiencies in its Table 5. Our maximum distances (8.0, 28.4, 34.4 and 38.0 km) agree with its Fig. 21.

The block-failure chart is an approximate redraw of Fig. 20: the 10%-failure thresholds are the paper's values; curve shapes are approximate. Public-channel outcomes summarise Figs. 13, 14, 16 and 17.

Animations illustrate the protocols described in the paper. The seven-bit toy code's parity rules were chosen to reproduce the paper's worked examples (Sections III-A and III-D). The circle uses the worked example of Section II-B. The noisy-echo and circle widgets are illustrative simulations.

No figures from the paper are reproduced; all visuals are original drawings made for this page. Heading typeface: Instrument Serif (SIL Open Font License).

Topics in this iPaper

Engineering
both the authenticated classical channel (ClC) and the quantum channel (QuC) for QKD are protected by separate forward error correction (FEC) coding schemes

Abstract, page 1

Physics
the CV-QKD information is modulated or mapped onto the quadrature components of electromagnetic fields

Section I (Introduction), page 3

Computer science
Hence, a quantum-safe cryptosystem is needed to tackle this threat.

Section I (Introduction), page 3

quantum key distribution
New near-capacity continuous-variable quantum key distribution (CV-QKD) reconciliation schemes are proposed

Abstract, page 1

forward error correction
Alice sends an FEC-protected codeword to Bob through a ClC, while Bob sends a separate FEC protected codeword to Alice through a QuC

Abstract, page 1

low density parity check codes
LDPC codes constitute a class of linear block codes defined by a sparse parity-check matrix (PCM)

Section II-C, page 9

convolutional codes
using a 1/2-rate CC of constraint-length 7 under AWGN and Rayleigh channels

Section V-C, page 19

The Road to Near-Capacity CV-QKD Reconciliation: An FEC-Agnostic Design — Interactive explanation | iPaper.ai