Make your paper interactiveCreate

Text from the interactive explanation

Design and Security Analysis of Quantum Key Distribution Protocol Over Free-Space Optics Using Dual-Threshold Direct-Detection Receiver

Phuc V. Trinh, Thanh V. Pham, Ngoc T. Dang, Hung Viet Nguyen, Soon Xin Ng, Anh T. Pham

Original paper

Trinh, Pham, Dang, Nguyen, Ng & Pham · IEEE Access, 2018

Quantum key distribution normally needs single-photon detectors or a phase-locked laser at the receiver. This paper asks whether a standard laser beam sent through open air between rooftops, read by an ordinary photodiode, could share a secret key instead, and works out with maths and computer simulations when it can.

The paper’s model expects Bob to keep 12.0 % of pulses, with 0.08 % of kept bits wrong; Eve gets 32 % of those bits wrong.

Alice sent 0

Alice sent 1

no bit: discarded

wrong bit

Bob’s two thresholds (drag them)

Simulation of the paper’s model with its parameters (δ = 0.4, ζ = 1.2, weak turbulence, P = 0 dBm), slowed down about 100 million times: the real link sends 10⁹ pulses per second. Pulse colours show Alice’s secret bit, a view nobody on the link has. The beam is drawn in false colour; its 1550 nm infrared light is invisible. The city and the air eddies are illustrative.

01

The key problem

Encryption scrambles a message with a : a long, random string of 0s and 1s. Anyone holding the same key can unscramble it. The difficult part is getting that key to both ends without anyone else learning it. This is the key-distribution problem.

Common key exchange relies on maths puzzles that are hard to solve. The paper points out that such schemes are vulnerable to future advances in computer hardware and algorithms. (QKD) relies on physics instead: Alice and Bob turn random light signals into a shared key that an eavesdropper cannot fully copy.

There are already two main ways to do it. Both work, but both need special hardware. The paper proposes a third.

The setting is a (FSO) link: a laser beamed through open air between buildings. FSO links are cheap and quick to set up, carry fibre-like gigabit rates, and are being considered for 5G backhaul. The paper builds on an earlier fibre version of the dual-threshold idea and adapts it to the open air, where the signal also flickers with turbulence.

Comparison summarised from the paper’s Table 2.

02

Guess the bit

Alice’s laser shines steadily, with a small flicker on top. For a 1 the flicker is nudged one way, and for a 0 it is nudged the mirror-image way. By the time the light reaches Bob, the shimmering air and his receiver’s own electrical noise have blurred that nudge.

Bob’s receiver boils each pulse down to one number . Positive leans towards 1 and negative leans towards 0, but only leans. Try reading them yourself: call each one, or skip it.

Reading 1

Alice has secretly picked a bit. Here is Bob’s reading.

Readings are drawn from the paper’s model: Bob 1 km from Alice, weak turbulence, modulation depth δ = 0.4, 0 dBm laser power. The scale is in units of the receiver’s typical noise.

03

Bob’s two thresholds

Bob’s rule has two cut-off lines, d₀ and d₁. A reading at or below d₀ becomes a 0, a reading at or above d₁ becomes a 1, and anything in between becomes “X”: no bit. This plays the role of BB84’s discarded photons.

Where do the lines go? Start from the average reading for each bit and step outward by ζ (zeta) times the noise level. So ζ is Bob’s pickiness. Bob can place the lines because he knows the current state of the channel. The paper assumes he knows it perfectly: the turbulence changes slowly (over milliseconds), so a few known pilot pulses can track it.

Drag across the curves to move the lines.

Bob keeps 12.0 % of pulses; about 1 in 1,200 kept bits is wrong.

Lines: probability that Bob keeps a pulse (Psift) and the share of kept bits that are wrong (QBER), computed in your browser from the paper’s model (δ = 0.4, ḡ = 15, P = 0 dBm, Bob 1 km away), re-creating the paper’s Fig. 12. Shaded: the paper’s design window, where ≤ 10⁻³ (errors that codes can correct) and ≥ 10⁻² (megabits of key from a gigabit link). That gives 1.2 ≤ ζ ≤ 2.35 in weak turbulence and 1.9 ≤ ζ ≤ 4.35 in strong turbulence.

Thresholds (paper Eq. 24): d₀ = E[i₀] − ζ·σN, d₁ = E[i₁] + ζ·σN, where i₁ = −i₀ = ¼·ℜ·ḡ·P·δ·h is the demodulated signal (Eq. 6), ℜ the photodiode’s responsivity, ḡ its gain, P the transmitted power, δ the modulation depth, and h the channel (loss × turbulence). σN² is the sum of shot noise, background-light noise and thermal noise (Eqs. 7–10).

Averaged over the turbulence, the sift probability is Psift = P(0,0) + P(0,1) + P(1,0) + P(1,1) and QBER = Perror / Psift (Eqs. 18–20), with each joint probability a Gaussian tail Q(·) averaged over the fading (Eqs. 22–23). The paper evaluates these in closed form (Gauss–Hermite and Gauss–Laguerre series) and checks them with Monte-Carlo simulation; this page averages over 3,000 simulated fading samples.

04

Saying it out loud

Next comes a step that happens in public. Over an ordinary channel that anyone can overhear, Bob announces which time slots gave him a bit, but never the values. Alice drops all the other slots. Now they both hold the same sifted key. A final step, error correction plus privacy amplification (hashing the key into a shorter one), removes the rare errors and squeezes out anything an eavesdropper might know.

A batch of ten time slots with at least two kept. At ζ = 1.2 Bob keeps about 1 slot in 8.

Your prediction

Eve has parked her own receiver in the beam’s spill, 500 m behind Bob. She hears the announcement too, so she knows exactly which slots to keep. In those slots, will her bits be as reliable as Bob’s?

Choose an answer to send 20,000 simulated pulses and find out.

05

Whisper, don’t shout

Alice’s main defence is how hard she nudges the light: the δ. A big nudge makes 0s and 1s easy to tell apart, for Bob but also for Eve. A small nudge keeps Eve’s two curves piled on top of each other.

The paper asks that Eve guess wrong more than 10 % of the time (e > 0.1), even with her receiver right beside Bob’s and her photodiode’s gain tuned to its best value. It finds that δ ≤ 0.4 achieves this under weak, moderate and strong turbulence. A nudge that is too small hurts Bob too, so the paper picks δ = 0.4.

Eve guesses wrong 10.3 % of the time.

Eve 1 km from Alice (right beside Bob), P = 0 dBm, computed from the paper’s model in your browser, re-creating the paper’s Figs. 9 and 11. The bell curves use weak turbulence. Readings are scaled by Eve’s typical noise. Bob’s line uses his own gain of 15 and ζ = 1.2.

Eve uses one threshold at zero, the best single cut for two mirror-image signals. Her error is e = P(0,1) + P(1,0) (Eq. 41), each a Gaussian tail Q(i₁/σN) averaged over her fading. Her information is I(A;E) = 1 + e·log₂e + (1−e)·log₂(1−e) (Eq. 40).

Noise (Eqs. 7–10): σN² = 2qḡ²ℜFA(¼Pδh)Δf + 2qḡ²ℜFAPbΔf + 4kBTFnΔf/RL, with excess-noise factor FA = kAḡ + (2 − 1/ḡ)(1 − kA). Raising the gain lifts the signal above thermal noise, but past a point the photodiode’s own multiplication noise wins. That is why Eve’s error is lowest near ḡ ≈ 15.

06

Where can Eve hide?

The paper argues that Eve’s most practical attack is to put an unauthorised receiver in the beam near or behind Bob. The beam is narrow and invisible, so cutting in mid-path is very hard, and cameras at Alice’s end can deter anyone from splitting the beam there. But the beam spreads as it travels, to about a 0.5 m radius at 1 km, so most of the light flows on past Bob’s 2 cm lens.

The security score is the secret-key rate S = I(A;B) − I(A;E) : how much Bob learns about Alice’s bits, minus how much Eve learns, per pulse. If S is positive, error correction and privacy amplification can distil a key that Eve knows essentially nothing about.

Drag Eve along the beam, or use the slider.

Computed in your browser from the paper’s model (δ = 0.4, ḡ = 15, P = 0 dBm, Bob 1 km from Alice), re-creating the paper’s Figs. 13–14. Turbulence toggle: C²n = 10⁻¹⁵ (log-normal) or 10⁻¹⁴ (Gamma-Gamma). The paper reports S > 0 once Eve is at least 1.5 km from Alice in weak turbulence with ζ = 1.2, and at least 1.7 km for C²n = 10⁻¹⁴ with ζ = 1.9. A larger ζ throws away more of Bob’s bits, so the paper recommends the smallest ζ in the design window . Beam side view is not to scale.

Channel loss (Eq. 11): hl = (π(D/2)²) / (π(θL/2)²) · exp(−βlL), with a 2 cm aperture D, divergence θ = 1 mrad and attenuation βl = 0.43 per km. Turbulence strength is set by the Rytov variance σR² = 1.23·(2π/λ)7/6·L11/6·C²n (Eq. 17): log-normal fading for weak turbulence, Gamma-Gamma for moderate to strong.

Bob’s link is an erasure channel with errors (Eq. 39): with p = P(Bob gets it right) and ε = P(Bob gets it wrong), I(A;B) = p·log₂p + ε·log₂ε − (p+ε)·log₂((p+ε)/2). The secret-key rate is S = I(A;B) − I(A;E) (Eq. 37). Eve’s fading is drawn independently of Bob’s, at her own distance.

07

The sweet spot of power

More laser power helps Bob, but it helps Eve even faster. The paper’s final key-creation rate, Rf = Psift × Rb × S (secret bits per second from a 1 Gbps link, with Eve 500 m behind Bob), climbs with power, peaks, then falls.

Weak (C²n = 10⁻¹⁵, ζ = 1.2): peak ≈ 10 Mbps at −2.5 dBm

Moderate (C²n = 5×10⁻¹⁴, ζ = 1.9): ≈ 3.6 Mbps at −2 dBm

Strong (C²n = 10⁻¹³, ζ = 1.9): ≈ 5.2 Mbps at −1.5 dBm

Why the rate falls: weak turbulence at this power

Curves: values traced from the paper’s Fig. 15 (approximate), δ = 0.4, Bob at 1 km, Eve at 1.5 km, ζ = 1.2 (weak) or 1.9 (moderate and strong). Beyond each curve’s end the paper’s S turns negative and no key can be made. Bars: simplified in-browser re-creation from the same model, in bits per pulse. Power in .

A surprise: strong turbulence (peak about 5.2 Mbps) beats moderate turbulence (about 3.6 Mbps). With the same ζ = 1.9, stronger turbulence scrambles Eve’s readings more, which lowers what she learns. That is the paper’s explanation.

Final key-creation rate (Eq. 38): Rf = Rs·(I(A;B) − I(A;E)), with sifted-key rate Rs = Psift·Rb and system bit rate Rb = 1 Gbps. The paper assumes 100 % efficient error correction, so these rates are an upper bound.

08

What this brings closer

Back to the opening question. According to the paper’s analysis and simulations, the key-sharing function of QKD can ride on the ordinary signal level of a laser beam, as used in standard optical systems. The hardware is the commercially available intensity-modulation/direct-detection kind that already runs at gigabits per second, and in the simulated scenarios it produces secret keys at megabits per second.

If that holds up in practice, secure networks could be built from standard telecom devices instead of costly single-photon equipment, and the large toolbox already developed for free-space laser links could be reused directly.

Next steps the authors name

Build it: experiments on real free-space links, both horizontal (between buildings) and space links.

Add pointing errors and weather effects to the analysis.

Try diversity and relaying techniques.

Try other error-correction (reconciliation) schemes, such as bidirectional or reverse reconciliation.

The setting of the analysis

A proposed protocol, analysed with closed-form maths and checked with Monte-Carlo computer simulations. No hardware experiment.

Alice’s and Bob’s devices are trusted.

Bob knows the channel state perfectly, since the fading varies over milliseconds and can be estimated with pilot symbols.

Error correction is assumed 100 % efficient, so key rates are an upper bound.

Eve performs an unauthorised-receiver attack inside the beam near or behind Bob.

Words used on this page

Source

Phuc V. Trinh, Thanh V. Pham, Ngoc T. Dang, Hung Viet Nguyen, Soon Xin Ng and Anh T. Pham, “Design and Security Analysis of Quantum Key Distribution Protocol Over Free-Space Optics Using Dual-Threshold Direct-Detection Receiver,” IEEE Access, vol. 6, pp. 4159–4175, 2018. doi:10.1109/ACCESS.2018.2800291

All visuals are original illustrations and simulations built for this page from the equations and parameters in the paper. The curves in the power section are values traced from the paper’s Fig. 15. The live scenes are simulations of the paper’s model, slowed down; the city skyline, the air eddies and the beam’s colour are illustrative.

Topics in this iPaper

Engineering
the proposed QKD protocol can be implemented on standard FSO systems using subcarrier intensity modulation binary phase shift keying and direct detection with a dual-threshold receiver

Abstract, page 4159

Computer science
we define the ergodic secret-key rate as the maximum transmission rate at which the eavesdropper is unable to decode any information

Section V-B, page 4167

quantum key distribution
This paper proposes a novel design and analyzes security performance of quantum key distribution (QKD) protocol over free-space optics (FSO).

Abstract, page 4159

free space optical communication
Figure 4 presents a block diagram of the proposed free-space CV-QKD system using SIM/BPSK and D-T/DD receiver with an APD.

Section III-B, page 4163

atmospheric turbulence
log-normal (LN) and Gamma-Gamma (GG) distribution models have been widely used for characterizing weak and moderate-to-strong turbulence regimes, respectively.

Section IV-B, page 4165

binary phase shift keying
the source data d(t) is first modulated onto a radio frequency (RF) subcarrier signal using BPSK scheme

Section III-B, page 4163

monte carlo method
Monte-Carlo simulations are performed to verify analytical results and the feasibility of the proposed QKD protocol.

Abstract, page 4159

Design and Security Analysis of Quantum Key Distribution Protocol Over Free-Space Optics Using Dual-Threshold Direct-Detection Receiver — Interactive explanation | iPaper.ai