Covert Optical Communication

• Reliability requires codewords to remain distinguishable after channel corruption, keeping Bob’s decoding error probability arbitrarily small for sufficiently large n.Error-correction codes are used in practice to enable reliability.Source: INFORMATION-THEORETICALLY COVERT… · p. 2
• Willie tests whether Alice transmitted using a binary hypothesis test, with error probability equal to the mean of false-alarm and missed-detection probabilities.Unequal prior probabilities do not affect the stated asymptotics.Source: INFORMATION-THEORETICALLY COVERT… · p. 2
• Covert communication requires Willie’s detection error probability to remain at least 1/2 − ǫ, the error rate of a random guess minus an arbitrarily small margin.Alice can reduce Willie’s test effectiveness by lowering transmission power, at the expense of Bob’s decoding reliability.Source: INFORMATION-THEORETICALLY COVERT… · p. 2
• The scheme assumes Alice and Bob pre-share a secret, whose cost is taken to be substantially less than the cost of being detected.The passage notes that some scenarios may permit secretless optical covert communication.Source: INFORMATION-THEORETICALLY COVERT… · p. 2

• The channel model assigns fraction ηb of Alice’s photons to Bob and the remaining ηw = 1 − ηb to Willie, who collects them while remaining passive.It models single-spatial-mode free-space and single-mode fiber channels.Source: ANALYSIS OF COVERT OPTICAL COMMUNICATION · p. 2, Figure 1 · p. 3
• On a pure-loss channel, reliable covert communication is impossible because Willie’s single-photon detector reveals any non-vacuum codeword with a click.The impossibility holds even with an infinitely large pre-shared secret; non-adversarial excess noise enables covert communication.Source: ANALYSIS OF COVERT OPTICAL COMMUNICATION · p. 3
• With thermal noise above zero, Alice can reliably send O(√n) covert bits in n optical modes to Bob using conventional homodyne detection.The stated mean photon number per mode is O(1/√n).Source: ANALYSIS OF COVERT OPTICAL COMMUNICATION · p. 3
• Exceeding the O(√n) limit leaves transmission either detectable or unreliable.The analysis states that the square-root limit cannot be outperformed.Source: ANALYSIS OF COVERT OPTICAL COMMUNICATION · p. 3
• Detector excess noise, including dark counts, can enable covert communication on a pure-loss channel using OOK with an on symbol sent at probability q = O(1/√n).The cited result gives O(√n) covert bits using n OOK symbols.Source: ANALYSIS OF COVERT OPTICAL COMMUNICATION · p. 3
• Q-ary pulse position modulation addresses the difficulty of building efficient error-correcting codes for OOK by sacrificing a constant fraction of throughput.Its low-photon-number performance and symbol symmetry enable use of many efficient codes.Source: ANALYSIS OF COVERT OPTICAL COMMUNICATION · p. 4

EXPERIMENTAL RESULTS

• The test bed sent 1 ns, 1550.2 nm laser pulses through a free-space channel split between Bob and Willie, whose detectors were gated to match pulse arrival.Alice and Bob used Q-ary PPM frames with Q = 32 and a first-order Reed-Solomon code; n ranged from 3.2×10^6 to 3.2×10^7.Source: Figure 2 · p. 4, EXPERIMENTAL RESULTS · p. 4
• The experiment emulated receiver noise by raising APD gate voltage to increase dark-click probability, rather than injecting optical pulses during detector gates.The study reports measured and targeted detector dark-click probabilities and Bob’s mean detected photon count in Table I.Source: EXPERIMENTAL RESULTS · pp. 4–5
• Bob’s (31, 15) Reed-Solomon code achieved 45%–60% of maximum throughput in the “careful Alice” regime and 55%–75% in other regimes.The authors report that this basic code demonstrates the theoretical scaling at reasonable error rates.Source: EXPERIMENTAL RESULTS · p. 5, Figure 3 · p. 5
• Willie’s error probability stayed constant as n increased under square-root scaling, but dropped when Alice used asymptotically larger ζ, illustrating the converse.Figure 4 compares experimental estimates, Monte Carlo simulations, and Gaussian approximations.Source: Figure 4 · p. 5, EXPERIMENTAL RESULTS · p. 6

DISCUSSION

• Practical excess noise, including blackbody radiation and detector dark counts, permits covert communication, as demonstrated in the proof-of-concept optical system.The authors contrast this with pure loss alone, which does not permit covert communication.Source: DISCUSSION · p. 6
• The authors identify relevance to RF communication in light of advances in quantum-noise-limited microwave amplifiers and detectors.Source: DISCUSSION · p. 6
• The work provides impetus for developing covert optical networks that could scale privacy to interconnected systems.Source: DISCUSSION · p. 6

Covert Optical Communication Theorems

• Achievability arguments bound detection effectiveness and show reliable decoding with a shared secret, while converses fix Willie's detector and bound reliable covert information.The theorems classify achievability as a lower limit and converses as an upper limit.Source: Covert Optical Communication Theorems · p. 6
• Over a pure-loss channel, Alice cannot communicate reliably and covertly to Bob, even with an unbounded shared secret and quantum-optimal receiver.Willie is limited only by the laws of physics in choosing his receiver measurement.Source: Covert Optical Communication Theorems · p. 6
• With any positive thermal noise, Alice can send O(√n) bits in n modes even when Willie captures all photons missed by Bob and uses arbitrary quantum measurements.Bob can use a suboptimal coherent receiver such as homodyne detection.Source: Covert Optical Communication Theorems · p. 6
• With a pure-loss channel but nonzero dark counts in Willie's photon-number-resolving detector, Alice can covertly send O(√n) bits in n modes.The model treats dark counts per mode as a Poisson process with average λw.Source: Covert Optical Communication Theorems · p. 7
• Pulse-position modulation constrains on-off signaling to support efficient error-correction codes at a constant throughput cost, and the experiment uses random selected frames with a Reed-Solomon code.The experimental construction adds a random modulo-Q vector to the encoded message before transmission.Source: Covert Optical Communication Theorems · p. 7, Alice’s encoder · p. 7
• Under the same dark-count setting, structured Q-ary pulse-position modulation achieves Q log Q bits using n optical modes.The theorem specifies a Q-ary PPM constellation.Source: Covert Optical Communication Theorems · p. 7, §I · p. 18
• If Alice's n-mode codewords have total photon-number variance σ², transmitting ω(√n) bits forces either Willie to detect her or Bob to fail decoding reliably.The theorem states this as n tends to infinity.Source: Covert Optical Communication Theorems · p. 7
• The converse remains unproved for input states with unbounded photon-number variance per mode, although the stated restriction covers all well-known bosonic-mode states.The unbounded-variance case is identified as an open problem.Source: Covert Optical Communication Theorems · p. 7

Generation of transmitted symbols

• Alice encodes random Q-ary data with a Reed-Solomon code, then adds the secret vector k modulo Q to produce coded symbols.The coded sequence has length n after mapping to transmitted and empty modes.Source: Generation of transmitted symbols · p. 8
• Each coded Q-ary symbol selects the pulse-bearing mode in its PPM frame; modes in frames outside S remain empty.Occupied and unoccupied modes become binary 1s and 0s, respectively.Source: Generation of transmitted symbols · p. 8
• For the experiment, Alice alternates the length-n signal with n zeros, giving Willie silent-channel observations while Bob discards those zeros.The interleaving supports estimating false-alarm and missed-detection probabilities under the same conditions.Source: Generation of transmitted symbols · p. 8

Bob’s decoder

• Bob randomly selects one detected pulse per selected PPM frame, marks frames with no pulse as erasures, subtracts k modulo Q, and applies the Reed-Solomon decoder.The shared offset is not subtracted from erasures.Source: Bob’s decoder · p. 8
• The experiment reports mean successfully decoded bits over 100 runs per (ζ, n) pair and symbol error rate as lost data symbols divided by transmitted data symbols.Undecoded codewords are discarded from the decoded-bit total.Source: Bob’s decoder · p. 8
• Willie's log-likelihood ratio compares click records under silence and transmission, with a threshold chosen to minimize detection error.The study also estimates detection error by Monte Carlo simulation and a Gaussian approximation using measured channel characteristics.Source: Willie’s detector · p. 8
• Confidence intervals for the estimated error probability use the Dvoretzky-Kiefer-Wolfowitz inequality.The inequality relates a distribution function to its empirical distribution function.Source: Willie’s detector · p. 9

Theorem 1 (Insufficiency of pure-loss for covert communication)

• Reliable covert communication is impossible over a pure-loss channel, even with an unbounded shared secret, unattenuated observation for Alice and Bob, and a quantum-optimal receiver.The impossibility holds when Willie’s measurement choice is limited only by physical laws.Source: §I · pp. 11, 15

Remark

• An SPD is asymptotically optimal for detection over a pure-loss channel, and a photon-number-resolving receiver can mimic it with a one-photon threshold.The SPD’s discrimination error is at most twice that of an optimal discriminator.Source: §I · p. 15

Theorem 3 (Square root law for the thermal noise channel)

• O(sqrt(n)) bits can be transmitted reliably and covertly in n modes with positive thermal noise, even when Willie may use an arbitrary receiver.Alice and Bob share a sufficiently long secret, while Bob can use a sub-optimal coherent receiver such as homodyne detection.Source: §I · pp. 15, 17

Theorem 7 (Dark counts yield square root law)

• O(sqrt(n)) bits can be sent in n modes when Willie has non-zero dark current, captures photons not reaching Bob, and Alice and Bob share a sufficient-length secret.Willie’s channel from Alice is pure-loss, but his receiver is limited by the non-zero dark current.Source: §I · p. 17

Dark counts yield square root law under structured modulation

• With a shared secret, Alice can lower-bound Willie’s detection error while reliably sending O(Q log Q) bits in n modes via Q-ary PPM, when Willie’s pure-loss interception is limited to a nonzero-dark-current PNR receiver.The scheme uses a secret-selected subset of PPM frames and a secret random shift modulo Q.Source: §I · p. 18

Converse of the square root law

• ω(√n) mean photons let Willie detect Alice’s codewords with error probability at most ε for any ε > 0, given sufficiently large n.The converse’s detection argument uses Willie’s noisy PNR receiver and a threshold on the total photon count.Source: §I · p. 20
• ω(√n) bits cannot be transmitted in n optical modes both covertly and reliably.The converse applies to Alice’s codewords under the theorem’s stated photon-number-variance condition.Source: Covert Optical Communication Theorems · p. 7, §I · p. 21

• The section provides the mathematical details of the experimental methods.Source: §II · p. 22

Calculation of Bob’s maximum throughput

• Bob’s maximum throughput is computed from the symmetric channel’s Shannon capacity Cs = I(X;Y) under uniform inputs, scaled by ζn.The channel model includes correct detections, erasures, and other output symbols; Q is the expected number of PPM frames selected for transmission.Source: Calculation of Bob’s maximum throughput · p. 22

Derivation of the log-likelihood ratio test statistic

• Willie’s likelihood-ratio test compares click-record likelihoods for silence and transmission, accounting for randomized frame selection and possible pulse detections or dark clicks.Source: Derivation of the log-likelihood ratio… · pp. 22–23
• For large n, a Gaussian approximation of the click count yields an analytical approximation of Willie’s error probability; the total click count explains the test’s limited sensitivity to channel variations.Source: Derivation of the log-likelihood ratio… · pp. 23–24