Text from the interactive explanation
Network Coding Aided Cooperative Quantum Key Distribution Over Free-Space Optical Channels
Hung Viet Nguyen, Phuc V. Trinh, Anh T. Pham, Zunaira Babar, Dimitrios Alanis, Panagiotis Botsinis, Daryus Chandra, Soon Xin Ng, Lajos Hanzo
Original paperNguyen et al. · IEEE Access 2017 · interactive explainer
The authors picture rooftop lasers handing out encryption keys whose secrecy rests on the physics of light, not on maths puzzles — protection meant to hold “despite the rapid advances in computational power.” No fibre has to be dug: the light crosses open air, which suits last-mile access, back-haul for mobile networks and disaster recovery.
The catch: every key is finished with an ordinary radio chat between the rooftops, and radio messages get lost. This paper works out what those losses do to the key, and shows how neighbouring rooftop pairs can team up to repair them.
IllustrativeAn animation driven by the paper’s model: photons fade with the computed loss for this distance, and each tile’s colour is drawn from the computed key error rate for this team size, using the paper’s scenario where one radio message in five is lost on a single link.
Switch to a lone pair and watch the ribbons turn red. Why does teamwork fix it? Let’s build it up.
Chapter 1
Light with a direction
A photon is a single particle of light. Each one carries a tilt, called its polarization — picture a tiny arrow pointing sideways, straight up, or along a diagonal.
The paper uses the classic BB84 recipe. The sender — call her Alice — turns each random bit into one photon. For every photon she picks, at random, one of two “alphabets”, called bases :
The receiver — Bob — can’t see which basis Alice chose. For each photon he must pick a filter , straight or diagonal, which steers the photon into one of two detectors: one for 0, one for 1. (In the paper’s receiver a 50:50 beam splitter makes that choice at random.)
Be Bob
Pick a filter to measure the photon. Alice’s choice is revealed after each measurement.
Measure all 8 photons, then compare filters.
A spy in the middle — usually called Eve — faces Bob’s problem: she has to guess a filter for every photon. Whenever she guesses wrong, her measurement re-tilts the photon, so the photon she passes on is sometimes wrong even when Bob’s filter matches Alice’s. When Alice and Bob later compare a small sample of their kept bits openly, those extra errors give her away. Quantum physics also forbids making a perfect copy of an unknown photon, so she can’t keep a spare to measure later.
That comparing step happens over an ordinary radio link . Hold that thought — first, the photons have to cross the sky.
Chapter 2
Across the open air
These photons don’t travel in a glass fibre. They fly through open air between two rooftop telescopes — a link called free-space optics . In the paper’s setup both telescopes have 10 cm openings and the light is infrared, with a wavelength of 1,550 nanometres. Three effects thin out the photons on the way:
Diffraction — every beam spreads. Once it is wider than the 10 cm receiver, much of the light misses.
Turbulence — shimmering air nudges and smears the beam (the paper fixes one turbulence strength).
Extinction — haze absorbs and scatters light, taking a bigger bite with every kilometre.
Detectors also click now and then with no photon from Alice at all — stray light from the surroundings, or the detector’s own electronic “dark counts”. Such a click gives a random bit.
Side view of one link
IllustrativeEach dot is one pulse; the share that reaches the receiver follows the computed photon fraction for this distance. The strip under the receiver is a click log: mint squares are real photons, coral squares are stray-light clicks, mixed in the computed proportion.
ComputedReconstructed from the paper’s equations (photon fraction, Eqs. 28–40, lower bound; sifting and errors, Eqs. 15–17) with its Table 4 parameters. The stray-light level is our estimate — see notes.
As real photons get scarce, stray clicks stop being negligible. Each stray click is a coin toss, so half of them give a wrong bit — the error rate is half the stray share of clicks. That is why errors climb from about 0.01% at 0.5 km to about 30% at 14 km.
The collected fraction is γ = μ · e−αL: μ is the share of the spread-out, turbulence-smeared beam that lands on the receiver, and e−αL is what survives the haze (α = 0.443 per km).
Whether the receiver is “near” or “far” is set by the Fresnel number Df = (π d1 d2 / 4λL)² with 10 cm apertures d1, d2. When Df ≫ 1 (about 26 at 1 km) the receiver sits in the near field and catches almost the whole beam; around Df ≈ 1 (near 5 km) the beam outgrows it; at 14 km Df ≈ 0.13, deep in the far field. The paper derives tight upper and lower bounds on μ for both regimes; the curves here use the lower bound. Right now: Df ≈ 26.
Here η = 0.5 is the detector efficiency, nS = 1 photon per pulse, and nN = 5×10⁻⁵ stray counts per pulse per detector. The paper’s parameter table does not list nN; this value is our estimate, chosen because it reproduces the paper’s curves.
Chapter 3
The chat everyone assumed was perfect
Sifting needs a conversation. Alice tells Bob which basis she used for each photon; Bob tells Alice which bases he used and which photons he actually caught. They talk over a public classical channel — here, an ordinary radio link. Anyone may listen: bases alone reveal nothing about the bits.
Earlier studies of quantum keys sent through the air assumed this radio chat always arrives intact. Real radio links have outages — moments when a message is lost or arrives garbled. The authors model a garbled message as one whose basis is flipped.
The paper’s own example (its Table 3)
Ten photons, already sent and measured. In interval 3 the photon was misread — an error from the light link itself — and in interval 10 nothing was detected. Tap any envelope to garble that radio message.
Now scale it up
200 kept bits on a 0.5 km link. How often is a radio message lost?
This run: 0 of 200 disagree.
With one radio message in five lost, about a third of the key bits disagree — while the light link alone gets only about one bit in 8,000 wrong. The ordinary radio chat, not the quantum link, has become the weakest part.
ComputedA kept bit agrees only if the photon was read correctly and both basis messages arrived: disagree = 1 − (1 − photon errors)(1 − p)². This form reproduces the paper’s key-rate figures; the paper’s own upper bound is photon errors + 2p. The grid draws each bit at random with that probability.
So the fix has to make radio messages harder to lose . The authors’ idea: let neighbours help.
Chapter 4
Neighbours that mix their messages
The fix borrows a trick from wireless networking called network coding . Several rooftop pairs share the radio work. In the broadcast phases each sender (group A) announces its own message — and because radio spreads in every direction, the other senders and all the receivers (group B) can hear it. In the cooperative phases each sender broadcasts a mix : a sum of the messages it heard. A receiver that missed something can often solve for it, like a small algebra puzzle.
Two pairs, as in the paper’s worked example (Section III-D): U1A talks to U1B, and U2A talks to U2B. The messages are stand-in numbers, m1 = 5 and m2 = 3 — the real system mixes bits with finite-field arithmetic, but the logic is the same. Focus on U1B : it needs m1. The schedule:
Broadcast 1: U1A sends m1
Broadcast 2: U2A sends m2
Mix 1: U1A sends m1 + m2
Mix 2: U2A sends m1 + 2·m2 — but only if it heard m1. If the A–A link fails, its mix is just 2·m2 (the paper adjusts the schedule for what each user actually received).
Two pairs already give U1B four chances instead of one. What about bigger teams?
Chapter 5
Bigger teams
More pairs means more mixes in the air — more spare routes for every message. To get a feel for the trend, this toy simulator runs thousands of radio sessions and counts how often U1B ends up unable to recover m1.
Rules: every radio hop fails independently with chance p. Senders hear each other’s broadcasts unless that hop fails; each then broadcasts one mix with random coefficients of every message it heard. U1B collects whatever arrives and checks, by exact elimination, whether it can solve for m1.
The numbers differ from the paper’s, because the paper uses a detailed radio model in which transmit power is shared across the phases. There, a 20% single-link outage drops to 9.1% with 2 pairs and 0.52% with 8 pairs (paper’s Fig. 7). The trend is the lesson.
Chapter 6
The payoff
Now feed the paper’s reported outage values into its key equations. Choose how often a single radio link fails, a team size and a distance. The team sizes carry the paper’s names: G2×4 is the two-pair schedule from Chapter 4; G4×8 and G8×16 are the four- and eight-pair versions.
Radio loss after teamwork: 0.52% (paper, Fig. 7).
Key bits that disagree vs. distance
Extra key vs. a lone pair
ComputedCurves computed from the paper’s equations and the outage values it reports: 20% case — 9.1% (2×4) and 0.52% (8×16) from Fig. 7, 4.9% (4×8) inferred from Fig. 9d; 10% case — 1.9% (2×4) and 0.45% (4×8) from Fig. 7, about 0 for 8×16 (its curve matches error-free radio in Fig. 8a).
Up to about three orders of magnitude fewer key errors. At 0.5 km the ratio Φ = BERlone / BERteam is about 2000 for a 10% single-link outage, about 200 for 1%, 20 for 0.1% and 3 for 0.01% (Fig. 8).
Steady key-rate gains for a 20% outage: about 29% (2×4), 41% (4×8) and 55% (8×16) — close to the 56% of perfectly reliable radio (Fig. 9d).
Match the code to the conditions: the powerful 8×16 code is needed when the outage is 10%; the simpler 2×4 code suffices at 0.01%.
Theory checked by simulation: the analytical bounds closely match Monte-Carlo simulations (Fig. 7). All results are calculations and simulations; no hardware link was built.
Chapter 7
Back on the rooftops
Teamwork can be spent in two ways: more key at the same distance, or more reach at the same key rate. Try both with the paper’s 20% scenario.
More key, or more reach?
ComputedFrom the paper’s equations with the reported 20% → 0.52% outage for 8×16.
The authors suggest switching cooperation on exactly when it pays: when a user needs a higher key rate, or moves out of good range. Their case rests on analytical bounds checked against Monte-Carlo simulations, not on a field trial.
Back to the opening picture: rooftop lasers handing out keys that physics protects. This paper’s contribution is to take the humble radio chat seriously — and to show that neighbours who mix their messages can make it almost as dependable as a perfect link.
Sources & notes
Hung Viet Nguyen, Phuc V. Trinh, Anh T. Pham, Zunaira Babar, Dimitrios Alanis, Panagiotis Botsinis, Daryus Chandra, Soon Xin Ng and Lajos Hanzo, “Network Coding Aided Cooperative Quantum Key Distribution Over Free-Space Optical Channels,” IEEE Access, vol. 5, pp. 12301–12317, 2017. DOI: 10.1109/ACCESS.2017.2712288.
The article is open access under the Creative Commons Attribution 3.0 licence (CC BY 3.0). The authors state that the data supporting the study is available from the University of Southampton repository: doi.org/10.5258/SOTON/D0151.
What is what on this page
Reported by the paper: the evidence card in Chapter 6, the outage values used for each team, the Table 3 example in Chapter 3, and the network-coding schedule and matrix in Chapter 4.
Computed from the paper’s equations: the curves and numbers in Chapters 2, 3 (rates), 6 and 7, using its Table 4 parameters (1,550 nm, 10 cm apertures, α = 0.443 per km, Cn² = 10⁻¹⁵, detector efficiency 0.5, one photon per pulse) and the tight lower bound on the collected photon fraction.
Illustrative: the rooftop animations (their randomness is driven by the computed probabilities, but the choreography of packets is a picture, not a simulation of the paper’s radio model), the side-view link, the puzzle numbers m1 = 5 and m2 = 3, and the Chapter 5 toy simulator.
Modelling notes
The stray-light and dark-count level nN = 5×10⁻⁵ per pulse per detector is not printed in the paper’s parameter table; it is our estimate, chosen because it reproduces the paper’s Fig. 5 curves.
A kept bit is counted as correct only if the photon was read correctly and both basis messages arrived: BER ≈ 1 − (1 − BERphotons)(1 − P)². This reproduces the paper’s Fig. 9d key-rate gains; the paper’s analytical upper bound is BERphotons + 2P (Eqs. 23 and 25).
The 4.9% outage for the 4×8 code in the 20% scenario is inferred from the 41% gain in Fig. 9d. Values read from figures make our 10%-scenario error reduction at 0.5 km come out near 1,500× rather than the paper’s reported ≈2000×.
Key rates are shown per 1,000 pulses (Eq. 26 with ψ = 1); absolute bit rates depend on hardware assumptions and are not shown.
All visuals are original code reconstructions and illustrations based on the paper’s equations and reported values; no figures are reproduced.
Topics in this iPaper
Engineering
we conceive network coding aided cooperative QKD over free space optical systems for improving the bit error ratio and either the key rate or the reliable operational distance
Abstract, page 12301
Physics
Since the FSO channel imposes deleterious effects, such as diffraction, atmospheric turbulence and extinction
Section II-B Photon detection in QKD systems, page 12303
Computer science
the transmission session can be summarised by a transfer matrix having two rows and 4 columns
Section III-D, page 12312
quantum key distribution
INDEX TERMS Quantum key distribution, network coding, free space optical.
Index terms, page 12301
network coding
each user in the system is capable of supporting the others with the aid of Network Coding (NC)
Section III-A, page 12307
free space optical communication
We have considered realistic error-prone public wireless channels in the context of quantum key distribution (QKD) systems relying on an FSO link.
Section V Conclusion, page 12315
bb84 protocol
the key-sharing process is based on the popular QKD protocol BB84
Section III-A, page 12306
cooperative communication
there are two sets of phases, namely the broadcast phase (BP) and the cooperative phase (CP)
Section III-D, page 12312
monte carlo simulation
which were obtained by Monte-Carlo simulations based on the parameters listed in Table 4
Section IV-A, page 12314