Coexistence of Discrete-Variable QKD with WDM Classical Signals in the C-Band for Fiber Access Environments

Full text

ICTON 2019 Th.A5.4
978-1-7281-2779-8/19/$31.00 ©2019 IEEE 1
Coexistence of Discrete-Variable QKD with WDM Classical
Signals in the C-Band for Fiber Access Environments
Dimitris Zavitsanos1*, Giannis Giannoulis1, Adam Raptakis1, Christos Papapanos1, Fotini Setaki2,
Eleni Theodoropoulou2, George Lyberopoulos2, Christos Kouloumentas1,3, and Hercules Avramopoulos1
1School of Electrical and Computer Engineering, National Technical University of Athens, Athens, Greece
2COSMOTE Mobile Telecommunications S.A, Athens, Greece
3Optagon Photonics, Ag. Paraskevi 15341, Athens, Greece
*e-mail: dimizavitsanos@mail.ntua.gr
ABSTRACT
In this paper, a coexistence scheme between a Discrete-Variable Quantum Key Distribution (DV-QKD) and four
bidirectional classical channels in a Passive Optical Network (PON) topology is theoretically investigated. The
study aims to explore the imposed limitations considering the coexistence of weak quantum channels with
realistic traffic flows of classical streams through shared fiber infrastructures. Based on a ‘plug and play’ phase
coding DV-QKD implementation, we conducted numerical simulations of the QBER and the secure key rate for
fiber distances up to 10 km. The reported results suggest that in a fixed C-band grid, the spectral isolation
between classical and quantum channels is essential at dense grids. By removing the leakage noise through
stronger spectral isolation, the photons linked with the Raman scattering become the dominant noise source,
since this mechanism covers an ultra-broadband window and gets stronger as the propagation distance increases.
Keywords: QKD, discrete-variable, QBER, WDM, coexistence scheme, Raman noise, FWM.
1. INTRODUCTION
The densification of 5G mobile deployments with fiber-connected femto-cells requires a converged transport
topology which can serve diverse traffic patterns in deep fiber segment of the network. In this context, the PON-
based topologies become an attractive platform to meet the rising traffic demands at the optical edge [1].
Entering the era of Quantum Computer, the definition of a quantum-safe strategy becomes a top priority for the
telecom service providers to fortify these fiber optical segments. Whilst the development of quantum-resistant
cryptographic systems whose security relies on different, hard mathematical problems that are resistant to being
solved by a large-scale quantum computer is gaining more and more ground [2], the quantum-safe grade security
has only been proven to be guaranteed with the exchange of quantum keys through QKD systems [3].
The multiplexing of weak quantum channels with classical data channels in the PON topologies has been
proposed to enable provable secure communication allowing for fast updates of Advanced Encryption Standard
(AES) encryption keys [4]. The implementation of QKD concepts with existing Fiber-To-The-Home (FTTH)
architectures has also been proposed emphasizing on the actual deployment constraints [5]. The feasibility of the
coexistence between quantum and classical channels through the same fiber employing C-band Dense
Wavelength Division Multiplexing (DWDM) technology has been conducted for both Continuous-Variable
QKD (CV-QKD) and DV-QKD protocols [6]. Recently, the integration of a CV-QKD scheme was reported over
a 25Km WDM-PON network with intense classical channels [7].
In this work, a closer look on the coexistence scheme of DV-QKD is provided focusing on passive DWDM
networks with short fiber segments. Detailed theoretical results on performance evaluation of the QKD scheme
are provided, focusing on the actual topology parameters and the physical layer parameters of the dataplane.
2. DWDM INTEGRATION OF DV-QKD AND NOISE SOURCES
The study on coexistence scheme is performed for a passive network structure realized through WDM allocation.
For this purpose, in Fig. 1 we consider a generic network topology in which the DV-QKD system could be
deployed: the quantum channel is multiplexed with forward (from Alice to Bob) and backward propagating
channels by using passive filtering elements. In more detail, we assume a C-band 100 GHz
Multiplexer/Demultiplexer (MUX/DEMUX) setup with a Standard Single-Mode Fiber (SSMF) with an
attenuation of Ƚ=0.21dB/km, where the quantum channel is multiplexed along with four classical channels.
The channel separation between adjacent channels is assumed to be 100-200-400 GHz. In addition, the loss of
receiver internal components is assumed to be 2.65 dB. For the quantum channel, we choose a wavelength of
1552.5 nm on the International Telecommunication Union (ITU) C-band grid and a ȟɉ =0.8 nm quantum
receiver bandwidth. The post-processing steps required for the key distillation are performed over a standard
Ethernet connection.
For the quantum transmission, we consider a BB84-QKD system, based on a ‘plug and play’ phase coding
implementation, where the need for compensation between the two arms is circumvented and the birefringence
effects and polarization dependent losses are compensated, so that the visibility can be set at 99% [8,9].

ICTON 2019 Th.A5.4
2
The optical pulses are generated at Bob’s site, with a repetition rate set to f୰ୣ୮ =5MHz, propagate through an
unbalanced Mach-Zehnder interferometer and a delay line of 24 km. Then, they are reflected at Alice’s site,
where they are phase modulated and propagate back to Bob. The photons are assumed to be detected by a pair of
Infrared Single-Photon Detectors (APD1 and APD2) operated in gated mode with a time-window of 1ns for the
detection. The Single Photon Avalanche Diode (SPAD) unit is assumed to exhibit a dead time of ɒୢୣୟୢ =0.ͳɊs,
an afterpulse probability of 0.8%, a dark count rate of 4×10
ି଺nsିଵ and quantum efficiency of Ʉୗ୔୅ୈ =0.1.
Figure 1. DV-QKD integration in the passive network structure of deep fiber access operated by C-band DWDM.
The performance of the quantum channel suffers from the noise photons generated from the nonlinear response
of the fiber medium across the classical layer. Besides, the leakage photons linked with the crosstalk of the
neighboring channels (LCXT) at the DEMUX unit introduce severe distortion in the quantum link.
The nonlinear effects in the transmission fiber are the Spontaneous Raman Scattering (SpRS) and the Four-
Wave Mixing (FWM). FWM arises from the third order susceptibility ɖ(ଷ) of the fiber where source photons
interact in such a way that they create new ones (f୧୨୩) at different frequencies from those the initial ones (f୧,f
୨,f
୩)
[10]:
f୧୨୩ =f
୧+f
୨െf୩, k ്i, j (1)
The power level of the FWM product is given by:
P
୧୨୩ =஗ୢమஓమ୔౟୔ౠ୔ౡୣషಉై
ଽ஑మ(1 െeି஑୐)ଶ, (2)
where P
୧,P
୨,P
୩ are the power of the input classical channels, L is the propagation distance, ɀ is the fiber
nonlinearity and d is the FWM degeneracy factor; The FWM efficiency Ʉ is given by:
Ʉ=஑మ
஑మା୼ஒమ[1 + ସୣషಉై ୱ୧୬మ(౴ಊై
మ)
(ଵିୣషಉై)మ], (3)
where the phase matching factor ȟȾ depends on the fiber dispersion D=17 ps/(nmήkm) of the SMF at
1550 nm and the channel spacing between classical channels. To evaluate the effect of this nonlinear
mechanism, we count the FWM products which are generated into the quantum channel’s spectral passband [11].
The Raman interaction is associated with inelastic scattering of the strong classical signals with fiber material,
where the scattered photons may arise at longer wavelengths (Stokes) or shorter (anti-Stokes) compared to the
pump wavelength. The scattered Raman power is proportional to the initial classical power P
୧ and gets stronger
as the propagation distance increases [12]:
P
୰ୟ୫,୤=P
୧Lɏ(ɉ)ȟɉeି஑୐ (4)
P
୰ୟ୫,ୠ=P
୧
sinh(ȽL)
Lɏ(ɉ)ȟɉeି஑୐, (5)
where P
୰ୟ୫,୤ and P
୰ୟ୫,ୠ is the forward and the backward SpRS respectively, ɏ(ɉ) is the effective Raman cross-
section and ȟɉ is the quantum receiver spectral bandwidth. The forward SpRS noise is maximized at a specific
propagation distance (~20.7 km) [10], after which decreases depending on the fiber attenuation value. On the
contrary, backward SpRS is a monotonically increasing function of fiber length in its entire domain. Since we
consider ultra-dense spectral allocation in our scheme, we assume a constant value of effective Raman cross-
section of ɏ(ɉ)=2.3×10
ିଽ(km ήnm)ିଵ to calculate the power levels of Raman noise from both the co-
propagated channels (forward SpRS) and the counter-propagated channels (backward SpRS) [8].
The LCXT comes from the leakage of a small fraction of the classical signal into the quantum channel’s
passband due to imperfect isolation of the DEMUX. We consider state-of-the-art WDM 4th order Butterworth

ICTON 2019 Th.A5.4
3
filtering elements so that the isolation of spectrally adjacent channels is 60 dB (for 100 GHz spacing), 80 dB (for
200 GHz spacing) and 90 dB (for 400 GHz spacing) [8]. The insertion loss of the DEMUX module is set to 1.5 dB.
In order to evaluate the Quantum Bit Error Rate (QBER) and the secure key rate, Rୱୣୡ, we expressed the power
levels of the above noise sources in terms of noise photons per ns (gate duration time). The QBER is defined as
the ratio between the number of false detections and total detections and for the BB84 protocol becomes:
QBER = 0.5 ୮౩౟ౝ౤౗ౢ(ଵି୚)ାଶ୮ీి ା୮౗౦ା୮౓ీ౉
୮౩౟ౝ౤౗ౢାଶ୮ీిା୮౗౦ ା୮౓ీ౉ , (6)
where V is the visibility and pୱ୧୥୬ୟ୪, pୈେ, pୟ୮ and p୛ୈ୑ =p
୊୛୑ +p
ୗ୮ୖୗ +p
୐େଡ଼୘ are the signal detection
probability, the dark count probability, the afterpulse probability and the WDM probability due to the
coexistence scheme respectively. The secure key rate is given by [12]:
Rୱୣୡ =R
ୱ୧୤୲ (1 െrୣୡ)(1 െr୮ୟ), (7)
where Rୱ୧୤୲ is the sifted key rate and rୣୡ and r୮ୟ are the fractions of bits used for error correction and privacy
amplification respectively. In our study, we considered a CASCADE algorithm to implement the error correction
after the key sifting. The calculation of Rୱୣୡ was carried out via the theoretical treatment in [12] while
performing a numerical optimization of the average number of photons per pulse ȝ to achieve the maximum Rୱୣୡ .
3. RESULTS AND DISCUSSION
We calculated the QBER and the Rୱୣୡ for the above setup by varying the channel spacing while the fiber length
is limited to 10 km, suited for deep fiber use cases. Through our studies, we adopted the 9% QBER threshold,
below which the CASCADE error correction algorithm cannot distil secret keys. It should be also mentioned that
our calculations overestimate Rୱୣୡ, especially for short fiber lengths and small classical power levels where the
key rates become higher, since they ignore interruptions of the key exchange during key distillation and various
experimental manipulations, such as fiber length measurements [8,12].
Figure 2. Contour plots of Rsec and QBER as functions of power level per classical channel and fiber length, for 100 GHz
spacing (60 dB isolation).
Figure 2 illustrates the Rୱୣୡ and QBER for the case of 100 GHz channel separation between the classical and
quantum channel. It is evident that the poor spectral isolation between the adjacent channels (60 dB) is the main
noise source and dominates over SpRS and especially FWM, which at these low power levels becomes
negligible. The acceptable QBER values can be obtained only for classical power levels below the -25 dBm.
Figure 3. Contour plots of Rsec and QBER as functions of power level per classical channel and fiber length, for 200 GHz
spacing (80 dB isolation).

ICTON 2019 Th.A5.4
4
As the channel spacing increases to 200 GHz (Fig. 3) and the spectral isolation becomes stronger (80 dB),
secure key exchange can be established at even higher power levels of classical flows. For example, at 2 km
fiber length, slightly lower values than -10 dBm are permitted. Since the leakage from the adjacent channels
significantly decreases, the SpRS impact is revealed, especially at higher propagation lengths where it becomes
dominant.
Figure 4 illustrates the results for the scenario where the channel spacing is set to 400 GHz. There is no
significant change in the kind of contour plot diagrams between the 200 GHz and the 400 GHz case, as the
isolation is decreased by only 10dB. In this case, the phase matching condition is not fulfilled, leading to even
lower FWM efficiency values. The LCXT value decreases even more and the dominant noise source is linked
with the Raman scattering. Quantum link transmission with acceptable QBER values is possible for classical
power levels of -5 dBm at short fiber lengths (~0.5 km).
Figure 4. Contour plots of Rsec and QBER as functions of power level per classical channel and fiber length, for 400 GHz
spacing (90 B isolation).
We also considered a scenario using ideal filtering operation, where the channel spacing is 100GHz but LCXT
is zero (infinite isolation). Figure 5b illustrates the QBER for this use case (right figure), which is compared to
a setup where the quantum channel is multiplexed along with only two classical channels (Fig. 5a). By using
two classical streams, the FWM generates two side band products spectrally located out of the quantum
passband.
The presence of FWM effect is evident at short distances and high classical channels power levels in Fig. 5b.
Note that from Eq. (3), FWM efficiency has a sinusoidal dependence on phase mismatch ǻȕand fiber length
which is reflected in the Rୱୣୡ and QBER plots (see Fig. 5b). Moving towards longer fiber links, the FWM effect
is eliminated and the quantum link is mainly contaminated by Raman scattering photons linked with the
bidirectional propagation of four classical channels.
Figure 5. Contour plot of QBER as a function of power level per classical channel and fiber length, for two hypothetical
scenarios where the channel spacing is 100GHz but LCXT is zero (infinite isolation), for: a) a setup where the quantum
channel is multiplexed along with only two classical channels (left figure), b) a setup where the quantum channel is
multiplexed along with four classical channels (right figure).

ICTON 2019 Th.A5.4
5
Based on the findings of our study, it is evident that the channel separation is an essential parameter since
a small amount of leakage power can lead to practically unfeasible key establishment, as in the first case we
considered (Fig. 2). On the contrary, an ideal scenario where the spectral isolation eliminates the leakage photons
(Fig. 5a), allows for the key establishment with increased power budgets for the classical streams. Therefore, the
use of appropriate filtering elements with enhanced spectral isolation is a key parameter to make the classical
and quantum coexistence in C-band feasible. This improved spectral isolation removes the contamination of
QKD link at shorter fiber lengths where the number of Raman scattering photons is still low. In the same way,
there isn’t any substantial difference at longer fiber lengths, because the impact of leakage becomes weaker due
to fiber losses (Figs. 3, 4). For this kind of longer fiber lengths, the key rate establishment becomes challenging
due to the significant increase of Raman noise photons. Regarding the photons associated with the FWM effect,
this noise source can affect the performance of the QKD link only at short distances and high classical channels
power levels.
4. CONCLUSIONS
The physical layer limitations of the coexistence between weak quantum channels and classical traffic flows in
the C-band are discussed through this research. The reported results imply that accumulated Raman scattering
photons become the dominant noise source for distance extension scenarios even with low incoming classical
power levels. The generated photons due to FWM effect become an essential noise source for intense classical
streams (> -10 dBm per channel) at short fiber lengths (< 3 km). The QBER calculations denote also that the
strong spectral isolation (> 80 dB) between classical and quantum signals is a key parameter to employ classical
channels with adequate power levels (> -10 dBm per channel) even for short fiber paths.
ACKNOWLEDGEMENTS
This work has been supported by the European Commission through H2020 FET-Flagship project UNIQORN,
Contract Number 820474.
REFERENCES
[1] S. Bidkar et al., "First demonstration of an ultra-low-latency fronthaul transport over a commercial TDM-
PON Platform,” in Proc. OFC 2018, San Diego, CA, USA, paper Tu2K.3.
[2] https://www.advaoptical.com/en/newsroom/press-releases/20180613-adva-achieves-world-first-100g-
quantum-safe-transport-over-2800km
[3] “Quantum Technologies for 5G”, Application Note on Quantum-Safe Security from ID Quantique,
Feb. 2019.
[4] B. Frohlich et al., “Quantum secured gigabit passive optical networks,” in Proc. OFC 2015, Los Angeles,
CA, USA, paper W4F.1.
[5] I. Choi et al., “Quantum information to the home,” New Journal of Physics 13 (2011): 063039.
[6] B. Qi et al., “Feasibility of quantum key distribution through a dense wavelength division multiplexing
network,” New Journal of physics 12 (2010): 103042.
[7] R. Kumar et al., “Coexistence of continuous variable QKD with intense DWDM classical channels,”New
Journal of Physics 17 (2015): 043027.
[8] M. Mlejnek, N. A. Kaliteevskiy, and D. A. Nolan, “Reducing spontaneous Raman scattering noise in high
quantum bit rate QKD systems over optical fiber,” arXiv preprint arXiv:1712.05891 (2017).
[9] M. Mlejnek et al., “Modelling high quantum bit rate QKD systems over optical fiber,” Quantum
Technologies 2018. vol. 10674, International Society for Optics and Photonics, 2018.
[10] N. A. Peters et al., “Dense wavelength multiplexing of 1550 nm QKD with strong classical channels in
reconfigurable networking environments,” New Journal of physics 11.4 (2009): 045012.
[11] B. Goebel and No. Hanik, “Analytical calculation of the number of four-wave-mixing products in optical
multichannel communication systems,” Technische Universität München, Technical, German (2008).
[12] P. Eraerds et al., “Quantum key distribution and 1 Gbps data encryption over a single fibre,” New Journal
of Physics 12.6 (2010): 063027.