Quantum Key Distribution

A QKD link begins with a source producing attenuated pulses or entangled pairs, passes through fibre with exponential loss, and ends in detectors with finite efficiency and dark counts. The arrival rate sets the raw key rate; sifting halves it; reconciliation and privacy amplification reduce it further; finite-key statistics reduce it again for short sessions.

Computing that chain is the core engineering skill. A 50 km link at 0.2 dB/km has 10 dB loss, so one in ten photons arrives; detector efficiency and dark counts then determine the signal-to-noise ratio, which bounds achievable QBER and therefore secure rate. If the computed rate is below operational need, the architecture — not the algorithm — must change.

BB84 and its variants dominate deployed systems because they work with attenuated lasers and standard detectors. Decoy-state techniques bound leakage from multi-photon pulses. B92 uses two non-orthogonal states and is simpler but yields lower rate. E91 and entanglement-based schemes enable device independence in principle at the cost of source complexity.

Continuous-variable QKD encodes in field quadratures and can use standard coherent telecom hardware, achieving high rates at short range but with tighter loss tolerance. Measurement-device-independent QKD removes detector side channels by having an untrusted relay perform the measurement, at the price of network topology constraints.

Reconciliation corrects discrepancies between sifted keys, leaking information that must be accounted for. Cascade and LDPC methods trade interactivity against efficiency. Privacy amplification then compresses the key with universal hashing so any remaining adversary knowledge becomes negligible. Both stages consume key material and both appear in the final rate.

Authentication is the most frequently ignored requirement. The classical channel must be authenticated end to end, typically with pre-shared keys or post-quantum signatures, or a man-in-the-middle simply replaces the quantum link with two independent sessions. QKD does not solve authentication and cannot bootstrap trust from nothing.

Practical attacks target hardware. Detector efficiency mismatch allows an attacker to control which detector fires and blind it with bright light. Trojan-horse probing reflects light back from Alice's modulator to learn her basis choice. Source imperfections leak information about prepared states. Each attack requires specific countermeasures and continuous monitoring.

Architecturally, trust boundaries must be drawn explicitly. A trusted node holds keys in the clear, so a chain of twelve nodes has twelve points where compromise breaks confidentiality. Satellite links add ground station trust and availability constraints. The roadmap and the risk register must state all of this in the same language used for the rest of the estate.

A deployed QKD link is a chain: source, quantum channel, detector, classical authenticated channel, post-processing and key store. Loss in the quantum channel dominates the achievable rate, because rate falls steeply with distance and detector efficiency. The classical channel must be authenticated, since an unauthenticated classical channel lets a man-in-the-middle defeat the protocol entirely, and that authentication is classical cryptography.

Post-processing is computationally substantial. Basis reconciliation, error estimation, privacy amplification and authentication all consume time and are often the throughput bottleneck at high rates rather than the optics. Profiling the whole chain rather than the quantum layer alone is what reveals whether the system will deliver the key rate the application needs.

The key store completes the chain and is where operational reality bites. Keys must be buffered for outages, consumed at the application's rate and never reused. A link that produces keys faster than the encryptors consume them wastes capacity; one that produces slower creates dependence on the fallback path, and the fallback policy becomes the real security posture.

Free-space QKD, including satellite links, extends reach where fibre is impractical but introduces atmospheric loss, background light and pointing constraints. Satellite passes are short, so key generation is bursty and the buffer design changes accordingly. Availability is bounded by orbital geometry rather than by hardware reliability, which makes capacity planning a scheduling problem.

Short-range deployments inside data centres and campuses are easier to justify technically: controlled environment, known loss, dedicated fibre and a physical security perimeter you already manage. Their value is in establishing operational competence — monitoring, recalibration, integration with key management — before extending the technology to links where the environment is less controllable.

Ground station and infrastructure cost is frequently omitted from business cases. A satellite deployment needs optical ground stations, weather-aware scheduling and skilled operators; a campus deployment needs fibre, racks and cooling. The marginal cost of the quantum hardware is rarely the dominant term, and a business case that models only the hardware overstates the value.

A QKD link is an optical instrument with a security claim attached, and it drifts. Detector efficiency changes with temperature; fibre loss changes with patching and ageing; alignment shifts after physical work in the rack. Each of these shows up as an error-rate change that is indistinguishable from attack unless baseline behaviour is characterised and tracked.

Build a baseline over weeks, not hours. Record error rate, detection statistics, key rate and temperature together, then alarm on deviation from the established envelope rather than on absolute thresholds. Alarm on anomalous improvement too, because detector-blinding style attacks can make statistics look better than they should.

Change management must include the optical path. A routine fibre patch, a replaced transceiver or a rack reorganisation can silently degrade a link, and without a recorded change event the resulting alarm is investigated as a possible attack. Linking the change calendar to the monitoring system saves days of confused incident response.

QKD is appropriate when three conditions hold together: the link is fixed and physically controllable, the confidentiality requirement is long-lived enough that computational assumptions are uncomfortable, and the organisation can operate specialised optical equipment. When any of these fails, post-quantum cryptography over existing infrastructure is usually the better investment.

The comparison should be explicit and quantitative. QKD delivers information-theoretic key agreement over a specific link with significant fixed and operational cost. PQC delivers computational security everywhere at near-zero marginal cost and integrates with existing protocols. Most architectures need the second broadly and the first narrowly, and stating that boundary clearly is more valuable than advocating either.

Evaluate with a pilot that produces operational evidence: key rate at real loss, availability over a month including a deliberate outage, integration with the encryptor and key-management stack, and the operator hours consumed per week. Those four numbers settle the decision faster than any theoretical comparison, and they survive scrutiny from finance and security alike.

Compute secure key rate against distance for realistic devices.

Model loss, efficiency and dark counts, then plot secure rate from 10 to 120 km with and without decoy states.

Demonstrate abort behaviour under attack.

Simulate BB84 with an intercept-resend eavesdropper at varying interception fractions and plot QBER against the abort threshold.

Quantify assurance loss across relays.

Model a multi-node chain and compute the number of compromise points and the impact on end-to-end assurance.

Context: Protection against future decryption of recorded communications.

Method: Point-to-point QKD for high-value links combined with PQC elsewhere.

Findings and status: Hybrid estate: QKD for fixed high-value spans, PQC for the rest.

Transferable lesson: Segmentation of assurance level is the practical pattern.

Context: Intercontinental key exchange beyond fibre reach.

Method: Free-space links to ground stations during passes.

Findings and status: Demonstrated key exchange; availability limited by weather and orbit.

Transferable lesson: Availability engineering is as important as key rate.

Context: Data-centre interconnect protection.

Method: CV-QKD over short spans using telecom components.

Findings and status: High rates achieved at short distance; cost per link remained significant.

Transferable lesson: Short-range CV-QKD suits dense interconnect scenarios.

Context: Offering quantum-safe transport as a service.

Method: QKD key delivery integrated with key management and network orchestration.

Findings and status: Service viable only where customers valued physics-based assurance.

Transferable lesson: Market positioning depends on differentiated assurance, not raw speed.

Context: Post-processing throughput, not the optical layer, limited key delivery, so encryptors fell back to classical keys during peak replication.

Method: Profiled the full chain, sized the key buffer against peak consumption, and instrumented fallback so every reversion was logged and alerted.

Findings and status: Fallback became a visible, measured event rather than an invisible default, and capacity planning moved to the real bottleneck.

Transferable lesson: The bottleneck is usually post-processing; instrument the fallback or you will not know it happened.

Context: Routine fibre maintenance caused error-rate spikes that were repeatedly investigated as possible attacks.

Method: Correlated the change calendar with monitoring, established per-link baselines over a month and alarmed on envelope deviation.

Findings and status: False investigations dropped sharply and genuine anomalies stood out against a stable baseline.

Transferable lesson: Change correlation is as important as physics in operating a QKD link.

Work through these in order — each builds evidence the capstone requires. Record assumptions, code versions, shots or sample counts, and validation results; the reasoning is assessed, not only the answer.

A defence organisation requires a quantum-safe link between two sites and must justify the choice.

Problem: Design the link, model key rate and availability, document trust boundaries, and compare against a PQC-only architecture.

Objectives: Produce an engineering design with quantified assurance and a defensible recommendation.

Deliverables: Link design, rate model, availability analysis, trust register, cost comparison and recommendation.

Assessment: Physics modelling 30%, architecture 25%, assurance analysis 25%, recommendation 20%.