Quantum Passive Optical Network (QPON)
- QPON is a quantum network architecture that uses passive optical splitting and shared infrastructure to distribute quantum keys across multiple users.
- It integrates quantum key distribution with existing GPON, EPON, and FTTH systems, ensuring secure communications while maintaining cost efficiency.
- Demonstrations exhibit secure key rates from kilobits to megabits per second, addressing challenges such as splitter loss, Raman noise, and wavelength coexistence.
Quantum Passive Optical Network (QPON) denotes a family of quantum access-network architectures that place quantum key distribution (QKD) or related quantum services on the splitter-based infrastructure of passive optical networks. In the literature, the term is used in two closely related senses: a passive optical network that carries quantum traffic over shared access infrastructure, and a “Quantum Protected Optical Network” in which QKD-generated keys secure ordinary optical access services. Across both usages, the defining features are passive optical distribution, multi-user sharing, and the attempt to preserve deployment economics by reusing GPON-, EPON-, coherent-PON-, or FTTH-style plants rather than dedicated dark fiber (Hentschel et al., 2014, Fröhlich et al., 2015, Hajomer et al., 2024).
1. Terminology, scope, and historical development
Early multi-user quantum networking work already exhibited the core QPON design logic: passive routing, shared infrastructure, and wavelength economy. A field-tested wavelength-saving QKD network built from passive optical elements used a real-time full-connectivity topology with 5 nodes supported with 2 wavelengths, and generalized the router construction to $2N+1$ ports with wavelengths (Wang et al., 2012). In that system, routing was implemented with 3-port circulators and WDM-based multiplexer/demultiplexer units rather than active switches.
A more explicit access-network formulation appeared in a differential phase-shift (DPS) PON proposal in which a centralized station containing all expensive components served many users over a passive binary tree with 50:50 beam splitters. The architecture was intentionally asymmetric: the central station held the laser, interferometer, and single-photon detectors, while user-side nodes were reduced to clock recovery, phase modulation, delay, attenuation, and basic polarization control. The stated target regime was local environments with moderate distances on the order of a few kilometers (Hentschel et al., 2014).
In GPON-oriented work, QPON was also framed as a “Quantum Protected Optical Network”: a standard GPON-style passive optical network upgraded so that downstream broadcast data could be encrypted with QKD-generated keys shared between the OLT and multiple ONUs. That formulation emphasized concurrent operation of multi-user QKD and full-power GPON traffic on the same access network, and introduced the dual-feeder implementation as a GPON-compatible remedy to splitter-induced imbalance between weak quantum signals and Raman noise (Fröhlich et al., 2015).
Subsequent work broadened the concept in two directions. One direction focused on coexistence engineering in lit access networks, especially GPON, NG-PON2, 10G-EPON, coherent PON, and FTTH testbeds (Vokić et al., 2020, Wang et al., 2021, Wang et al., 2024). The other direction developed genuinely point-to-multipoint downstream quantum access protocols, especially continuous-variable (CV) and coherent-state broadcast schemes, in which one transmitter can generate simultaneous independent keys with many receivers through a passive splitter (Hajomer et al., 2024, Pan et al., 2024, Yin et al., 30 Apr 2026). The literature therefore treats QPON less as a single standardized protocol than as an architectural class.
2. Architectural patterns
A dominant QPON pattern is the asymmetric upstream architecture in which the ONU or end user hosts a technologically lean quantum transmitter and the central office hosts a shared receiver. In the 2014 DPS-PON proposal, bright pulse packets were generated centrally, sent through the passive network, attenuated and phase-encoded at Bob, and returned to Alice for demodulation (Hentschel et al., 2014). Later lit-PON implementations moved to ONU-side quantum-state preparation using a directly modulated laser at 1 GHz, while retaining the delay interferometer and single-photon detector at the central office, explicitly centralizing complexity and cost (Vokiic et al., 2022). A related GPON access-network design placed quantum transmitters at the ONUs, a shared single-photon receiver at the OLT, and time-division multiplexed users so that only one transmitter’s photons arrived at the receiver at a time (Fröhlich et al., 2015).
A second pattern is the downstream broadcast architecture, closer to classical PON operation. In a practical downstream quantum access network over 10G-EPON, the QKD transmitter was located at the central office with the OLT and QKD receivers were placed at user nodes alongside ONUs (Wang et al., 2021). CV-QPON generalized this downstream model by preparing Gaussian-modulated coherent states at Alice, broadcasting them through a passive $1:N$ splitter, and letting every Bob perform coherent detection in every round (Hajomer et al., 2024). A related 16-node access network used a downstream point-to-multipoint coherent-state protocol in which one central transmitter served multiple receivers concurrently (Pan et al., 2024). The fully passive thermal-source architecture extended passive-state preparation from point-to-point CVQKD to point-to-multi-point operation by distributing a passively prepared downstream signal to multiple Bobs (Yin et al., 30 Apr 2026).
A third axis of variation concerns the optical distribution network. Single-feeder networks maximize infrastructure reuse but expose the quantum channel to feeder-generated Raman noise and reflections. Dual-feeder networks separate the strongest classical traffic from the quantum path and are repeatedly used to suppress downstream backscatter noise (Fröhlich et al., 2015, Vokić et al., 2020). A distinct variant routes the quantum channel around the splitter at the splitting point using WDM filters, so the classical GPON traffic still traverses the splitter while the quantum signal bypasses splitter loss (Sun et al., 2016). More recent coherent-PON work showed that O-band QKD can also be overlaid on a single-feeder-fiber, 20-km, coherent PON without modifying the existing access infrastructure, provided the classical data remain in the C-band and receiver filtering is sufficiently narrow (Wang et al., 2024).
3. Protocol families and signal models
Discrete-variable QPON implementations are dominated by BB84-derived and DPS-derived schemes. In the DPS line, information is encoded in the phase difference between adjacent pulses,
and recovered with a delay interferometer and single-photon detection. DPS is attractive in this context because it uses one measurement basis and permits lean ONU-side transmitters based on directly modulated lasers or other simplified optical encoders (Vokiic et al., 2022). The earlier centralized DPS-PON proposal used random phases $0$ or , an unbalanced Mach–Zehnder interferometer with 1 ns delay, and attenuation to photons per pulse at the user node (Hentschel et al., 2014).
BB84-derived QPON work spans phase encoding, polarization encoding, and decoy-state operation. The GPON quantum-secured access network employed efficient phase-encoded BB84 with decoy states and biased basis choice, a 1550 nm quantum channel, and a 250 MHz optical clock broadcast downstream (Fröhlich et al., 2015). Practical 10G-EPON coexistence used polarization-encoding decoy-state BB84 at 625 MHz with signal, decoy, and vacuum settings of $0.4$, $0.1$, and 0, respectively (Wang et al., 2021). A coherent-PON overlay used O-band polarization-encoding decoy-state BB84 at 1310 nm, 25 MHz repetition rate, 200 ps pulses, and three intensities 1, 2, and 3 photons per pulse (Wang et al., 2024).
Continuous-variable QPON replaces single-photon click statistics with coherent-state splitting and quadrature detection. In CV-QPON, Alice prepares Gaussian-modulated coherent states, a passive 4 splitter distributes the signal, and each Bob performs coherent detection. The per-user effective transmittance is written as
5
and the asymptotic key rate is
6
The same work distinguishes untrusted and trusted broadcast protocols, with the trusted model lowering Eve’s assigned information by treating selected user modes as trusted rather than adversarial (Hajomer et al., 2024). The 16-node coherent-state point-to-multipoint protocol introduced the more explicit multiuser constraint
7
thereby requiring the final key for Bob 8 to be private not only against Eve but also against other Bobs (Pan et al., 2024). The fully passive thermal-source network used the analogous form
9
and reported that under standard experimental conditions $1:N$0, so the rate reduces to a point-to-point-like expression (Yin et al., 30 Apr 2026).
4. Coexistence physics and impairment mitigation
The central physical challenge in QPON is that passive splitters attenuate the quantum signal while leaving the dominant noise processes insufficiently reduced. In GPON-style networks this produces an imbalance between splitter loss and Raman noise. A 1×128 splitter was reported to add at least about 21 dB of extra loss to the quantum channel, while Raman noise generated by full-power classical traffic remained essentially unattenuated in the most harmful paths (Fröhlich et al., 2015). Splitter penalties also motivated splitter-bypass proposals in which the quantum channel is extracted at the splitting point and avoids the splitter altogether, increasing the quantum counting rate $1:N$1-fold for split ratio $1:N$2 and improving the signal-to-noise ratio by a factor $1:N$3 that ranges from 1 to $1:N$4 (Sun et al., 2016).
Raman scattering is the dominant coexistence impairment in most lit-QPON studies. In upstream DPS-QKD over lit GPON and NG-PON2, the strongest design conclusion was that the quantum channel should be placed in the upstream direction, because downstream-generated Raman contributions must traverse the splitter and are partially suppressed, while upstream classical traffic co-propagates with the quantum channel over the fiber span (Vokiic et al., 2022). In coherent-PON coexistence, downstream backscatter from the feeder fiber again dominated, whereas upstream forward-scattering noise was reported as negligible up to 3.2 dBm upstream launch power (Wang et al., 2024). In a realistic FTTH GPON testbed with O-band QKD, back-reflections from ONT signals, especially from the second 1:4 splitting stage and the 1:2 coupler before Bob, degraded the secure key rate by roughly 3 dB when the first ONT was activated (Makris et al., 2023).
Wavelength planning is therefore fundamental. In GPON-like coexistence, classical downstream traffic around 1489 nm and classical upstream traffic around 1310 nm motivated a C-band quantum placement around 1550.12 nm; in NG-PON2-like coexistence, classical downstream and upstream channels in the C- and L-bands motivated an O-band quantum channel at 1310.55 nm (Vokiic et al., 2022). O-band quantum placement carries higher fiber loss than C-band, but repeatedly appears as the preferred compromise in heavily loaded access networks because it is less exposed to Raman noise from dense C/L-band loading (Schrenk et al., 2019, Wang et al., 2024).
Filtering is the main receiver-side countermeasure. In lit DPS-QKD coexistence studies, the receiver chain used R/B waveband filters, DWDM add/drop filters, FBG filters, and LAN-WDM filters; the FBG filter improved rejection of Raman noise by about 23 dB compared with a broader CWDM-type approach (Vokiic et al., 2022). In a related NG-PON2 loading experiment, Raman contribution after dark-count subtraction was 1730 counts/s with LAN-WDM filtering and $1:N$5 counts/s with an FBG filter (Vokić et al., 2020). In O-band DPS-QKD over a 52-channel loaded PON, a custom O-band DWDM add/drop filter with 1.22 nm FWHM improved noise rejection by 11.9 dB compared with CWDM filtering (Schrenk et al., 2019).
5. Demonstrated performance and scaling
The reported performance of QPON systems spans kilobit-per-second discrete-variable overlays to multi-megabit-per-second continuous-variable access networks. Representative results are summarized below.
| System | Configuration | Reported result |
|---|---|---|
| DPS-QKD in lit PON | 13.5 km, 2:16 split, 19 classical channels | $1:N$6 secure bits/pulse, QBER $1:N$7, penalty $1:N$8 (Vokiic et al., 2022) |
| O-band DPS-QKD in loaded PON | 16 km, 2:16 split | $1:N$9 secure bits/pulse; QBER increase 0 for 20 C-band downstream channels and 1 for 1 upstream C-band channel (Schrenk et al., 2019) |
| Quantum-secured GPON | dual feeder, up to 128 users | positive secure key rate up to 2 splitting; average secure key rate per user around 0.5 kbps in the 128-user case (Fröhlich et al., 2015) |
| Practical downstream QAN over 10G-EPON | single feeder, 21 km, 16 users | 1.5 kbps secure key rate for each of 16 users with 9 dB attenuation (Wang et al., 2021) |
| O-band QKD over loaded FTTH GPON | single feeder, 1:16 split, up to 9 ONTs | 10.1 kbps SKR and 5.11% QBER with 9 ONTs; operation for ~60 hours uninterrupted (Makris et al., 2023) |
| CV-QPON | 8 users, 11 km access link | 1.5 Mbits/s total network key rate in the untrusted protocol and 2.1 Mbits/s in the trusted protocol (Hajomer et al., 2024) |
| 16-node quantum access network | 3 splitter, 6 km fiber | average secret key rate around 2.086 Mbps between the transmitter and each user (Pan et al., 2024) |
| Fully passive thermal-source QPON | 4-to-5 passive splitter after 5 km fiber | record final key generation rate of 19.48 Mbps per quantum network unit (Yin et al., 30 Apr 2026) |
Within the discrete-variable coexistence literature, several results are particularly illustrative. In the 13.5 km, 2:16 split lit-PON DPS experiment, the quantum channel coexisted with up to 19 classical channels under a 93.8 dB power difference between launched classical and quantum signals, yet the reported error-ratio penalty was only 0.52% (Vokiic et al., 2022). In the O-band DPS coexistence experiment over a 16 km, 2:16 split PON, baseline operation without classical channels yielded 2.7 kb/s raw key rate and 3.77% QBER, while secure-key generation remained possible under substantial downstream and upstream loading because the overall QBER stayed below the stated 5% threshold (Schrenk et al., 2019).
Single-feeder coexistence has also been demonstrated in increasingly realistic access environments. A carrier-grade FTTH GPON testbed with total attenuation 21 dB and QKD budget 24 dB reported 20 kbps SKR and 3.13% QBER in the no-ONT baseline, then average values of 9 kbps and 5.8% with 1 ONT, 6 kbps and 6.15% with 5 ONTs, and 10.1 kbps and 5.11% with 9 ONTs. The improvement from 1 to 9 ONTs was attributed to GPON Power Levelling Sequence upstream and DBA, which reduced per-ONT launch powers and improved back-reflection levels from 6 dBm to 7 dBm (Makris et al., 2023). In a 20-km, 32-user coherent PON, O-band decoy-state BB84 QKD achieved at least 8 bits/pulse, about 175 b/s, at 0 dBm downstream launch power, and QKD remained possible at 9.2 dBm for downstream wavelengths at and longer than Ch13 (9 THz, 0 nm) (Wang et al., 2024).
The continuous-variable literature changes the scaling picture. Instead of time-sharing a scarce detector, CV-QPON exploits the fact that coherent states can be split deterministically and measured in parallel. The 8-user, 11 km demonstration reported 1.5 Mbits/s total network key generation in the untrusted model and 2.1 Mbits/s in the trusted model, with Bob1 achieving 375.4 kbit/s and a 46% increase under the trusted protocol (Hajomer et al., 2024). The 16-node access network pushed the average secret key rate to around 2.086 Mbps per user, reported as two orders of magnitude higher than previous demonstrations (Pan et al., 2024). The thermal-source passive-state-preparation architecture then reported a record final key generation rate of 19.48 Mbps per quantum network unit and an average asymptotic SKR of 19.29 Mbps over a 20 km equivalent fiber channel (Yin et al., 30 Apr 2026).
6. Operational considerations, applications, and unresolved issues
QPON research consistently treats deployment economics as a first-order design constraint. Centralized receivers, shared detectors, directly modulated lasers, passive splitter trees, standard telecom filters, and compatibility with existing GPON or FTTH hardware all serve the same objective: adding quantum security without abandoning the basic cost structure of access networks (Vokiic et al., 2022, Fröhlich et al., 2015). This also explains why many demonstrations emphasize lit networks, carrier-grade launch powers, full-power GPON traffic, or “without modifying existing PON infrastructures” rather than only raw quantum performance (Wang et al., 2024).
Security models and performance claims are not uniform across the literature. The early DPS-PON proposal explicitly limited its security discussion to individual attacks for weak coherent pulses (Hentschel et al., 2014). The field-tested wavelength-saving passive network computed secure key rate with the GLLP formula while ignoring statistical fluctuations (Wang et al., 2012). The coherent-PON overlay adopted asymptotic infinite-key analysis, justified by continuous operation (Wang et al., 2024). CV point-to-multipoint systems make inter-user trust assumptions explicit: untrusted and trusted broadcast protocols in CV-QPON (Hajomer et al., 2024), correlation-limited key-rate expressions in coherent-state PTMP CV-QKD (Pan et al., 2024), and the observation in fully passive thermal-source QPON that inter-Bob mutual information was far below Eve’s Holevo bound under the reported conditions (Yin et al., 30 Apr 2026). A common misconception is therefore that “QPON” implies a single security model; the literature instead spans substantially different protocol assumptions.
Applications extend beyond secure access transport. A proposal for online voting used point-to-multipoint QKD via TDM-PON and WDM-PON to distribute secret keys from a voting authority to many participants, with Raman-scattering analysis, O-band quantum placement around 1310 nm, and dual-feeder or mode-switching variants proposed to reduce coexistence noise (Huberman et al., 27 May 2025). More broadly, the access-network setting implies that QPON must coexist with classical control, timing, and QoS mechanisms. Related PON scheduling work has formulated bursty delay-constrained traffic management as model predictive control using virtual queues and polynomial-time optimization; that work is not a quantum-network paper, but it explicitly describes the framework as transferable to a hypothetical QPON carrying latency-sensitive classical control traffic alongside quantum-related signaling (Roy et al., 2021).
The main unresolved issues remain architectural rather than conceptual. Single-feeder systems preserve infrastructure reuse but are highly sensitive to feeder-generated Raman backscatter and reflections; dual-feeder systems suppress the dominant noise source but add fiber cost; splitter-bypass architectures remove splitter loss for the quantum channel but alter the splitting-point optics; and CV broadcast protocols improve aggregate throughput but rely on coherent receivers, calibration strategies, and user-trust models that differ sharply from discrete-variable access systems (Sun et al., 2016, Hajomer et al., 2024). This suggests that QPON is best understood not as one network design, but as a design space defined by passive distribution, multi-user access, and the technical problem of making fragile quantum channels coexist with the power levels, splitter losses, and service expectations of modern optical access networks.