900 Vol. 15, No.

11 / November 2023 / Journal of Optical Communications and Networking Research Article

Throughput optimization in hybrid C-C+L-band

elastic optical networks enabled by hybrid
modulation and wavelength/modulation
conversion
Yuxin Xu,1 Hang Xing,1 Bin Chen,2 AND Maite Brandt-Pearce3, *
1
College of Information Engineering, Zhejiang University of Technology, Hangzhou, Zhejiang, 310000, China
2
School of Computer Science and Information Engineering, Hefei University of Technology, Hefei, Anhui, 230601, China
3
Charles L. Brown Department of Electrical and Computer Engineering, University of Virginia, Charlottesville, Virginia 22904, USA
*mb-p@virginia.edu

Received 8 August 2023; revised 5 October 2023; accepted 8 October 2023; published 24 October 2023

Elastic optical networks (EONs) have been proposed to address the increasing needs of future communications,
yet the capacity of these networks is expected to run out. Evolving the current C-band system to use the C+L-
band is a costly and time-consuming solution. This paper presents a stopgap solution to meet the increasing
communication needs in the short term by exploiting hybrid fused C- and L-band transmission before the full
evolution is achieved. The proposed physical-layer-impairments-aware high spectral efficiency (PAHSE) algo-
rithm jointly applies time domain hybrid modulation (TDHM) and modulation and wavelength conversion
techniques and uses accurate models to estimate physical layer impairments for the C- and C+L-band links. The
hybrid network scheme converts C-band signals to L-band at the regenerator nodes, which increases the system
throughput. The proposed PAHSE algorithm assigns the highest spectral efficiency to each transparent segment
of each traffic request while satisfying the quality of transmission requirements. Simulations are conducted on
two network topologies, one of large diameter and the other highly connected, using several types of traffic that
mimic demands expected in future communication scenarios. Compared to the benchmark, the proposed net-
work system and algorithm result in much higher network throughput, up to a 120% increase in the sum bit rate
in the NSFNET-24 topology, thus obviating the urgent need for full C+L deployment. © 2023 Optica Publishing
Group

https://doi.org/10.1364/JOCN.502763

1. INTRODUCTION [7]. Expanding the transmission to the L-band (C+L-band

Traffic demands in future communication systems are expected transmission) requires new transponders and additional optical
to be more heterogeneous and larger in volume because they amplifiers at every span. Compared to SDM transmission, the
originate from various applications, e.g., human-to-human, upgrade cost of C+L-band transmission is considerably lower,
human-to-machine, and machine-to-machine transmission. and thus has already captured the interest of the industry [8].
In the near future, elastic optical networks (EONs) using the A complete upgrade to C+L-band transmission is still costly
C-band over single-mode fiber, which outperform state-of-the- to deploy based on current network infrastructures [9,10]. The
art dense wavelength division multiplexing (DWDM) optical upgrade also takes time to accomplish, while communication
networks [1], will not have the capacity to cover expected demands are constantly increasing. In this paper, we propose a
communications needs for long [2]. Therefore, more sophisti- cost-effective compromise based on a hybrid of C- and C+L-
cated techniques have been proposed to increase the capacity band EON. The hybrid system uses the L-band over only a
of the networks, such as spatial-division multiplexing (SDM) few bottleneck links and serves as a stepping-stone solution
or C+L-band transmission [3–6]. SDM techniques that to address increasing communication needs while limiting
require multicore fibers are considered an ultimate solution cost during the infrastructure evolution from the C- to the
for enhancing the system’s capacity, yet they would incur C+L-band. Reference [11] is one of the only published works
trenching costs to bury new cables underground. Capital considering a hybrid C-C+L-band system. However, the
expenditure in labor and equipment will be immense con- work did not consider the compatibility between the C-band
sidering the dimension of continental-scale optical networks and L-band subsystems. Therefore, the C-band and L-band

1943-0620/23/110900-13 Journal © 2023 Optica Publishing Group

 Research Article Vol. 15, No. 11 / November 2023 / Journal of Optical Communications and Networking 901

subsystems are independent, and the performance of the Table 1. Contributions of This Work and Other
hybrid system is highly constrained. Our approach fuses the Published Works
two subsystems into one system to best accommodate traffic OEO Hybrid Fused
demands. Paper C+L TDHM Process C-C+L C-C+L
Signal regenerators have been deployed at network nodes
[10], 2018 3 7 7 7 7
to facilitate long-haul transmission while maintaining a sat-
[20], 2019 7 3 7 7 7
isfactory signal quality of transmission (QoT) [12]. While [18], 2020 3 7 7 7 7
regenerators are primarily used to perform optical-electrical- [5], 2022 3 7 7 7 7
optical (OEO) conversion to refresh the signal and negate the [6], 2022 3 3 7 7 7
noise, they can also enable wavelength and modulation conver- [8], 2022 3 7 3 7 7
sion at the network nodes, which greatly enhances flexibility [11], 2022 3 7 7 3 7
in signal transmission [13]. Utilizing modulation and wave- This work 3 3 3 3 3
length conversion enabled by reconfigurable optical add/drop
multiplexers (ROADMs) equipped with OEO circuits, our the capacity of the hybrid system by increasing the spectral
proposed hybrid system can perform signal transmission using efficiency.
both the conventional C-band and the upgraded L-band. Table 1 lists other published research in this area and their
Signals can be converted from the C-band to the L-band at the attributes. To the best of our knowledge, no other published
regenerator nodes [14]. work has considered all qualities we consider, and none has
Successful provisioning of traffic demands in a continental- proposed an algorithm to fuse the C- and L-bands using wave-
scale system requires accounting for physical layer impairments length conversion. The hybrid approach proposed in [11] is
(PLIs). PLIs harm the QoT, especially in long lightpaths closest to our work, and, therefore, we use it as a benchmark.
[15,16], and limit the signal transmission reach, constraining Through realistic simulations on the NSFNET-24, we show
the throughput. The enhanced Gaussian noise (EGN) model is that our algorithm can increase the network throughput from
one of the most accurate PLI estimates and accounts for differ- 76 Tbps using the simulation settings as in [11] to as high as
ent modulation schemes in C-band networks [17]. When the 168 Tbps with only 14 C+L-band links out of the 86 in the
transmission includes using the L-band, the PLI model must topology. Applying the proposed method in a denser topol-
consider the inter-band stimulated Raman scattering (ISRS). ogy, the COST239 network, the throughput increases from
The ISRS effect causes a power transfer from high-frequency 165 Tbps up to 291 Tbps given the aforementioned settings.
components to low-frequency components [18]. Yet, the algorithm has a low enough computational cost to be
Optimizing the network capacity should also consider applied for resource provisioning when all nodes communicate
the spectral efficiency (SE) of the modulation schemes used. with all the others in this continental-scale network.
Using a continuous baud rate and modulation format adap- The novelties and contributions of this article are
tive scheme is a low-cost opportunity to increase the network summarized as follows:
throughput. When the physical distance of an optical path • a design that fuses the C- and C+L-band networks
is between the transparent reach of two modulation formats, within one hybrid system,
traditional modulation selection schemes use the lower-level • an algorithm to optimize the throughput by maximally
modulation format to satisfy the signal-to-noise ratio (SNR) increasing the spectral efficiency and spectral assignment on
threshold (determined by the modulation type used). This each segment of each lightpath,
leads to over-provisioning of optical resources because con- • accurate PLI estimation models for C- and C+L-band
ventional modulation formats have discrete SNR thresholds links in the fused C-C+L-band system.
[19]. So-called hybrid modulation has been proposed to enable This paper is organized as follows. Section 2 introduces
flexible modulation formats to better utilize spectral resources TDHM and its application to fiber-optic networks. The
[20,21]. The time domain hybrid modulation (TDHM) models to estimate the PLIs of the C- and C+L-band links
technique can continuously adapt the spectral efficiency while are described in Section 3. Section 4 introduces the hybrid
satisfying the minimum uncoded bit error rate (BER) [22–24]. C-C+L-band EON. Section 5 then describes the proposed
It can be used to assign the most spectrally efficient modula- PAHSE algorithm using modulation and wavelength conver-
tion to match an optical channel’s physical layer transmission sion. Section 6 presents our simulation settings and numerical
condition and, therefore, achieve the most efficient spectrum results. Finally, we conclude the paper in Section 7.
utilization.
As its primary contribution, this paper optimizes the capac-
ity of a hybrid C-C+L-band EON by jointly considering the 2. TIME DOMAIN HYBRID MODULATION
following three aspects: a hybrid modulation format with fine TDHM achieves fine granularity in SE by manipulating the
spectral efficiency granularity, a state-dependent GN model for number of bits per symbol by interleaving two modulation
C- and C+L-band PLI estimation, and modulation and wave- formats in the time domain using different ratios [19]. By
length conversion at regenerator nodes so that the C-band and properly designing a time-division multiplexing frame and
L-band subsystems can be fused into one system. This paper allocating the time-slot occupancy ratios of different modu-
proposes a heuristic algorithm, which we call the PLI-aware lation formats in the frame, the TDHM technique can result
high spectral efficiency (PAHSE) algorithm that maximizes in a continuous value for the SE. TDHM can be applied to
 902 Vol. 15, No. 11 / November 2023 / Journal of Optical Communications and Networking Research Article

for the PLI estimate that accounts for various modulation

schemes [27]. Several common assumptions are made. First,
the fiber links are dispersion uncompensated, i.e., only dig-
ital signal processing techniques are used for compensation.
Second, each polarization has the same PSD. Third, the fiber
attenuation loss is totally compensated for. Fourth, the NLI
PSD accumulates along the lightpath independently from span
to span. Lastly, the channel spectra are nonoverlapping and
rectangular in shape [15].
A network link is broken into fiber spans of length L, sepa-
rated by erbium-doped fiber amplifiers (EDFAs); one span is
on the order of 100 km long. Given a fiber loss of 0.2 dB/km,
Fig. 1. SNR versus SE for different pre-FEC BER requirements. the transmitted power is attenuated by about 20 dB at the end
of each span [15]. The optical amplification process causes ASE
current DWDM networks simply by updating the transceivers, noise, which can be modeled as additive Gaussian noise with
as discussed in [25]. It is equally suitable for EONs since they PSD per polarization in span s of lightpath i given by
can handle any demand data rate and accommodate arbitrary ASE
G i,s = (e αL − 1)hvnsp , (1)
baud rates using variable-bandwidth transponders.
The experimental curve in Fig. 2 of [22] shows the relation- where h is Planck’s constant, α is the fiber power attenuation
ship between the required optical signal-to-noise ratio and factor, n sp is the spontaneous emission factor, and ν is the light
the SE, given a pre-FEC (forward error correction) BER for frequency. As is commonly done, we assume that EDFAs have
a 35 Gbaud/s signal. Based on the results given in [22,23], a flat frequency gain over the C-band [29].
the SNR (dB) versus SE curve between each even integer SE The NLI effects for each span include self-interference
value for hybrid modulation is almost a straight line. Using effects and interference due to other co-propagating channels,
a linear interpolation between two integer SEs, the required named SCI and XCI, respectively [27]:
SNR versus SE can be approximated as shown in Fig. 1. Our NLI SCI XCI
heuristic algorithm, PAHSE, uses this approximation to reduce G i,s = G i,s + G i,s . (2)
its computational burden. NLI
G i,s represents the NLI PSD on span s of demand i’s light-
Note that the proposed PAHSE algorithm can work with SCI XCI
any modulation format, such as more advanced probabilistic path, per polarization; G i,s represents the SCI PSD; and G i,s
constellation shaping (PCS) or frequency domain hybrid represents the XCI PSD. The SCI caused by the channel itself,
modulation (FDHM), as long as the curve of the required SNR denoted as channel i, has a PSD approximated by
versus SE is given. 16 1 π β2 2
 2 
SCI
G i,s = (γ )2 (G i )3 ρi · · asinh Bi ,
27 2π |β2 | · α 2 α
3. PHYSICAL LAYER IMPAIRMENTS (3)
where γ is the fiber nonlinear coefficient; β2 represents the
Long-haul networks consisting of conventional C-band links group velocity dispersion parameter; and G i and Bi are the
and C+L-band links require custom PLI models for each signal PSD per polarization and bandwidth of demand i,
band to guarantee QoT. For links that transmit only in the respectively. The XCI noise depends on the difference between
C-band, the EGN model can be used due to its high accuracy the center frequencies and bandwidths of neighboring chan-
in estimating the PLIs when the signals use various modula- nels. The EGN model approximates the PSD of the XCI
tion formats [26,27]. For the links that transmit in both the caused by channel k sharing a link with channel i in span s as
C-C+L-bands, a closed-form PLI estimate considering ISRS is
Nk
necessary, as described below [18]. 32 2 X
XCI
G i,s = γ Gi ρk G 2k
27 k=1,k6=i
A. Enhanced-Gaussian Noise Model
asinh(π 2 |β2 |[ f k − f i + Bk /2]Bi /α)

Nonlinear noise, chromatic dispersion, and amplified sponta- ×
4π|β2 | · α
neous emission (ASE) noise are the main PLIs in single-core
long-haul fiber optic communications over the C-band. We asinh(π 2 |β2 |[ f k − f i − Bk /2]Bi /α)

assume that modern digital signal processing techniques can − , (4)
4π|β2 | · α
compensate for the chromatic dispersion, and optical ampli-
fiers can address the attenuation loss. Therefore, we only where f i is the center frequency of channel i; f k is the center
consider nonlinear fiber interference (NLI) and ASE noise as frequency of the neighboring channel k; Nk is the number of
contributing to PLI [15,28]. neighboring channels in the link; and n k and n i are the kth
The GN model is often used to estimate the PLI power and ith channels that share the lightpath, respectively. G k is
spectral density (PSD) analytically [15,28]. Because our con- the PSD of the signal of demand k, per polarization, and Bk is
tribution optimizes the modulation format to improve the the bandwidth of demand k. ρi and ρk are used similarly as in
spectral efficiency, we adopt the more advanced EGN model Eq. (7) of [27].
 Research Article Vol. 15, No. 11 / November 2023 / Journal of Optical Communications and Networking 903

In order to satisfy the required QoT, the PLIs must result C. Filtering Effects
in an SNR higher than the SNR threshold. Therefore, we can Signals that traverse ROADMs transparently undergo filter-
write the QoT constraint for each transparent segment as ing that narrows the spectrum and can affect the quality of
Gi transmission, especially when several filters are cascaded. To
  ≥ SNRth (SE), (5) address this issue, we utilize an adaptive optical channel guard-
ASE SCI XCI
P

G i,s + G i,s + G i,s band to compensate for bandwidth narrowing, as in [31]. The
s
guard-band, BiG , surrounding channel i is computed as
where the sum is over all spans in the transparent segment, and
− 2n1
SNRth (SE) is the SNR threshold for a certain SE.

Lf
BiG = Bi · − Bi , (10)
10 · (− ln 2) · log10 (e ) · N f
B. Physical Layer Impairments for the C+L-Band
where we use the same parameters as in [31]: L f = −5 dB
Using C+L-band transmission expands the available spectrum is the filter insertion loss, N f is the number of filters the
from 4.4 to 10 THz, resulting in non-negligible inter-channel transmitted signal passes through, and n = 6 is the effective
stimulated Raman scattering (ISRS). ISRS causes a transfer of Gaussian order of the filter model. Note that filtering changes
power from higher-frequency components to lower-frequency the spectral shape of the signal, violating the assumptions of a
components within the optical signals propagating in the rectangular spectrum needed for the EGN model, and effec-
fiber. Although this phenomenon is always present, it becomes tively acting as an upper bound on the PLIs. A more precise
particularly noticeable in communication systems operating
estimate can be obtained using the approach in [32].
within the C+L-band. We adopt the PLI estimation model
from [18] to calculate the PLIs for links where signals are
transmitted on both the C- and L-bands. 4. EVOLUTION TO A HYBRID C-C+L-BAND
The ISRS gain over one span of length L at frequency f can OPTICAL NETWORK
be approximated as
Expanding C-band networks to use the C+L-band requires
Ptot e −αL−Ptot Cr L eff f equipping the network with transceivers and amplifiers that
ρ(L, f ) = R . (6) support the L-band. However, upgrading from C-band hard-
G Tx (v)e −Ptot Cr L eff v dv
ware to C+L-band is costly and may take time to implement,
Ptot is the total input power transmitted on the same fiber link, especially considering continental-size networks. Additionally,
C r is the Raman gain slope, L eff is the effective length, and these long-haul networks need signal regenerators for data
G Tx is the launch power distribution across the optical channel transmission, particularly when bandwidth resources are
bandwidth [18]. extensively used. In this paper, we develop a network resource
Affected by ISRS, the estimates of the ASE and NLI noise provisioning plan to upgrade the network to the C+L-band by
need to be updated by considering the power transfer due to using strategically placed regenerators.
ISRS. The ASE noise PSD becomes To determine the best locations for the regenerators, we
ASE
G i,s = e αL+ρ(L, f ) − 1 hvn sp .


(7) need to identify the bottleneck links. For this purpose, we
first assume that every node in the network has a lightpath
The SCI affected by ISRS can be approximated as established to every other node using the shortest path routing

2 γ 2 π algorithm. We also assume that each node demands the same
Ti − α 2 φi Bi2
  
4
SCI
G i,s ≈ G i3 R iS · asinh bit rate transmission to each other node, and that enough
9 Bi2 φi 3α 2 α πα
spectrum is assigned on each route using the first fit spectrum
4α 2 − Ti

φi Bi2
 assignment algorithm. Given a particular modulation format
+ asinh , (8) and SNR threshold, the data rate of the demands is increased
2α 2πα
until the most congested link’s spectrum reaches 10 THz.
where φi = 23 π 2 β2 , and Ti = (2α − Ptot C r f i )2 . The XCI can We obtain estimates of the resulting PLIs using the models
be approximated as described in Section 3. We then determine the minimum num-
Nk  S 2 ber of regenerators that satisfy the QoT of each connection.
32 S 2

X 2 Rk γ2 This minimal regenerator deployment procedure is shown in a
XCI
G i,s ≈ G i R i (G k )
27 k=1,k6 =i
S
Ri Bk φi,k 3α 2 flowchart given in Fig. 2 using the notation defined in Table 2.
We illustrate our approach on the NSFNET-24 network.
Tk − α 2 φi,k Bi 4α 2 − Tk φi,k Bi We assume dual-polarization QPSK modulation, widely
    
· atan + atan , applied in continental backbone networks such as this one.
α α 2α 2α
(9) Following the algorithm described in Fig. 2, the system needs
seven regenerators to satisfy the QoT; their placement is
where R kS is the symbol rate of channel k, and φi,k = optimized using the MILP described in [33] for C+L-band
2π 2 ( f k − f i )β2 . The variation of the attenuation coeffi- transmission available in all 86 links, where the objective is to
cient over the C+L-band can be neglected [18]. In addition, minimize the number of regenerators. The resulting optimal
we neglect the variation in the dispersion. More details on the placement of the regenerators is shown in Fig. 3. Interestingly,
impact of ISRS on α and β2 can be found in [30]. for the fully deployed C+L-band network shown in Fig. 3(c),
 904 Vol. 15, No. 11 / November 2023 / Journal of Optical Communications and Networking Research Article

that originate in a C-band wavelength to an L-band wavelength

at the regenerator nodes, if needed. Modulation/wavelength
conversion is done electrically within the ROADMs via the
OEO process, and thus contributes no additional insertion
loss.

5. PLI-AWARE HIGH SPECTRAL EFFICIENCY

ALGORITHM
This section describes our proposed PAHSE algorithm for
offline cross-layer assignment of network resources to traffic
demands in a hybrid C-C+L-band network. As a first step,
Fig. 2. Flowchart of the regenerator deployment algorithm. the placement of regenerator nodes is derived for a given net-
work topology, as in Section 4. As a second step, the resource
allocation algorithm determines a route, spectrum assignment,
Table 2. Symbol Definitions and modulation format for each demand to efficiently allocate
D List of all demands the spectrum. Although we employ the regenerator placement
(N, L) Network topology consisting of a set of notes N and scheme described in Section 4, the proposed PAHSE algorithm
links L is compatible with any regenerator deployment. The objective
R List of demand bit rates of the algorithm is to increase R i , the bit rate for demand i,
P List of routes, initial value P = ∅ between all source and destination nodes until the network
S List of transparent segments system can no longer accommodate them [24,34]. For a bit
B List of demand (or segment, once regenerators are rate R i , the bandwidth, denoted as Bi , and the resulting SE,
deployed) bandwidths, initial value B = ∅ denoted as ηi , are variables depending on the modulation
F List of demand (or segment, once regenerators are format chosen. The problem can be modeled as
deployed) center frequencies, initial value F = ∅
Set of regenerator notes
P
G max i R i
Hm List of SEs at iteration m
s.t. (a) Signal spectra do not overlap
(b) Routes found from source to destination nodes
we can provision 150 GHz dual-polarized QPSK over all
sources to all destinations using this optimal solution. (c) The SNR ≥ SNR threshold for chosen SE
As an EON undergoes an upgrade from the C- to the C+L- (d) Wavelength and modulation conversion occurs
band, a hybrid network that contains some C-band and some only at regenerator nodes.
C+L-band links can maintain reliable transmission during (11)
the infrastructure upgrade, as illustrated in Fig. 3(b) for the Constraint (a) ensures that the signals are spectrally separated
NSFNET-24 topology. This allows for an evolution to the by at least the required guard-band within each lightpath to
point when the entire network consists of C+L-band-enabled maintain appropriate wavelength division multiplexing with-
links. The hybrid network only needs to upgrade the trans- out distortion. Constraint (b) ensures that each demand begins
ceivers at the regenerator nodes to the C+L-band and add at its source and ends at its destination using a route without
C+L amplifiers on the links between them. With wavelength loops. Constraint (c) guarantees the QoT requirements for
conversion performed by the OEO circuits in the regenerators, communications for the modulation scheme chosen by the
the hybrid network can accommodate C-band and C+L-band algorithm. Constraint (d) ensures that the OEO conversion,
transmission in the same network system; it converts signals i.e., demodulation and signal regeneration, occurs only at

Fig. 3. Evolution from the C-band to the C+L-band in the NSFNET-24 topology; the number adjacent to each link is the link length in km.
(a) Current C-band network, (b) hybrid C-C+L-band network, and (c) final C+L-band network.
 Research Article Vol. 15, No. 11 / November 2023 / Journal of Optical Communications and Networking 905

regenerator nodes; the OEO process negates the noise and Algorithm 1. PLI-Aware High Spectral Efficiency
allows the signal to be reformed, possibly using a different
center frequency and modulation format. Input: List of all demands, D, with their requested bit rates, R;
Due to the non-linearity of the EGN and ISRS models, network topology, (N, L), where N is the set of nodes and L is the set
solving problem Eq. (11) to maximize the total throughput of links; G is the set of regenerator nodes; dx and dy are SE
increments, where dx > dy.
while adhering to the constraints using an optimal technique
Definitions:
would be computationally prohibitive. It cannot be solved
i, the index for transparent segments;
using standard techniques, such as integer linear programming,
m, the iteration number with initial value m = 1;
even after linearization. This paper uses an iterative algorithm Hm , the list of SEs at iteration m, i.e., ηim ∈ Hm , i = 1, ... , |S|, with
to determine the modulation format, i.e., the SE, for each initial values H1 = {ηi1 = 2, i = 1, ... , |S|} and H0 = ∅;
demand. Link-based heuristic resource provisioning algorithms ηi∗ , a placeholder value used in optimization, with initial value
are too complicated for this cross-layer scenario because they ηi ∗ = 0, i = 1, ... , |S|.
require complex link-level algorithms, and thus a path-based procedure PAHSE ALGORITHM (N, L, D, Hm , G, R, dx)
algorithm compatible with wavelength conversion and modu- D = sort(D), based on descending R i
lation conversion enabled by the pre-deployed regenerators is P = Path-Select(N, L, G, D)
used. S = OEO Process(N, L, D, P, Hm , R, G)
In order to efficiently maximize the sum rate, the PAHSE (F, B) = First-Fit(N, L, S, P, Hm , R)
algorithm successively and iteratively maximizes the SEs for while Hm 6 = Hm−1 do
each traffic demand. PAHSE assigns network resources, such for i = 1 to |S| do
as a route, sufficient spectrum on each link, and modulation while ηi ∗ 6 = ηim do
formats (SE), to each transparent segment. It uses iterations ηi ∗ = ηim
to select the highest level of hybrid modulation format while ηim = SE-Maximization(ηi ∗ , N, L, F, P, Hm , G, dx)
satisfying the PLI requirements estimated by the EGN model end while
for C links and the ISRS model for C+L links. In this paper, ηim+1 = ηim − dx ∈ Hm+1
end for
a C+L-link-preferred shortest path algorithm that prioritizes
m := m + 1
C+L links (described in Algorithm 4 below) is used as the
end while
routing algorithm because the extensive usage of the C+L- while QoT not satisfied do
band links makes better use of available resources to increase (Hm , F, B) = Reassignment(N, L, S, P, Hm , G, R, dy)
system throughput and performance. The spectrum assign- end while
ment uses the first-fit algorithm, which prioritizes using the Output: Hm , F, P, B
C-band spectral slots. The L-band spectrum is chosen when
the C-band resources run out. This spectrum assignment
scheme is energy efficient because unused L-band amplifiers are
only turned on when needed. While only the C-band spectrum
is used, the ISRS effects can be neglected. The ISRS effects
appear on the entire spectrum when the L-band spectrum
begins to be occupied on the same lightpath where the C-band
spectrum is used. Other more sophisticated routing algorithms
could also be applied, as could other spectrum assignment
methods, yielding further network resource savings at the cost
of increased complexity.
Algorithm 1 and the flowchart given in Fig. 4 describe
our proposed algorithm: the SE is optimized by first using
a greedy SE maximization and then joint SE and spectrum
adjustments. As shown in Fig. 4, PAHSE initializes the process
by performing the routing assignment, the OEO process,
and the initial spectrum assignment for all SEs equal to 2
(BPSK). The OEO process effectively breaks each demand into
transparent segments that span the fiber between the regen- Fig. 4. Flowchart of the PLI-aware high spectral efficiency
erators along the route. The transparent segments for each (PAHSE) algorithm (notation defined in Table 2 and Algorithm 1).
demand are independent and thus can be assigned different
center frequencies and modulation formats. Because we use
modulation conversion, each transparent segment’s SE can
frequencies of the demands and fine-tunes the SE using step
be independently maximized. After initialization, PAHSE
obtains the largest SEs possible recursively using dx as the step size dy while ensuring that the QoT requirements remain
size. Increasing the SEs while leaving the center frequencies satisfied. Because we use wavelength conversion, the spectrum
unchanged shrinks the demands bandwidths, creating gaps of each transparent segment can be compacted independently.
between the demands-assigned spectra. Once PAHSE has The PAHSE algorithm outputs the routed demands, their SE,
greedily assigned initial SEs, it reallocates (compacts) the center and assigned spectra.
 906 Vol. 15, No. 11 / November 2023 / Journal of Optical Communications and Networking Research Article

Algorithm 2. SE-Maximization Algorithm 4. Path-Select

Input:ηi∗ ; m
N; L; F; B; P; H ; G; dx. Input: N; L; G; D.
Definitions: Definitions:
SNRi is the SNR estimated for segment i; Pi , the selected path of ith demand, Pi ∈ P;
SNRth (ηi∗ ) is SNR threshold value at SE ηi∗ ; Ws , the weight of link s . Ws = 1, if link s is a conventional C-band
PLIi , the PLI of segment i. link; Ws = 0, if link s is a C+L-band link;
procedure SE-MAXIMIZATION (ηi∗ , N, L, F, B, P, Hm , G, dx) Wi, j , the total weight of the j th path of demand i.
estimate PLIi using F, B, P, Hm , G procedure PATH-SELECT (N, L, G, D)
estimate SNRi using PLIi for i = 1 to |D| do
based on ηi∗ , Return SNRth (ηi∗ ) Obtain the list of k-shortest paths labeled j = 1, ... , k
if SNRi > SNRth (ηi∗ ) then for j = 1 to
Pk do
ηim = ηi∗ + dx Wi, j = links s along path j Ws ;
end if end for
Output: ηim Pi = argmin j =1,...,k Wi, j
end for
Output: P
Algorithm 3. Spectrum Reassignment
hybrid C-C+L-band elastic optical network compared to the
Input: N; L; S; P; Hm ; G; R; dy.
benchmarks.
Definitions:
R i , the bit rate of segment i;
Bi , the bandwidth of segment i. A. Benchmarks
procedure REASSIGNMENT (N, L, S, P, Hm , G, R, dy)
To show the benefits of fusing the C- and C+L-band systems
(F, B) = First-Fit(N, L, S, P, Hm , R)
and of using TDHM, the following scenarios are simulated and
for i = 1 to |S| do
estimate PLIi using F, B, P, Hm , G
compared.
estimate SNRi using PLIi • Unfused C+L. We use the scheme described in [11] as
end for the published benchmark in this paper. It does not employ
∀ηim ∈ Hm , get SNRth (ηim ) any modulation or wavelength conversion, nor does it employ
for i = 1 to |S| do TDHM. Demands allocated in the C-band cannot be con-
if SNRi < SNRth (ηim ) then verted to the L-band and vice versa. The modulation format
ηim := ηim − dy applied is conventional modulation (CM), i.e., using an even
end if
integer SE.
Bi = R i /ηim
• Fused C+L. This scheme utilizes modulation conversion
end for
Output: Hm , F, B
and fully flexible wavelength conversions (across the C- and
L-bands) at the regenerators but not TDHM. The modulation
format applied is CM.
The algorithm parameters dx and dy can be adjusted based • Fused C+L w/TDHM. This is our proposed approach
on the dimension of the optical networks and the computa- that uses the TDHM technique to improve the SE through the
tional resources available. The smaller the dx and dy, the better PAHSE algorithm and wavelength and modulation conversion
the algorithm’s performance, at the cost of more computation. at the regenerators.
The two main steps in the PAHSE algorithm are detailed
in Algorithms 2–4. Algorithm 2 is a greedy algorithm for the
B. Simulation Parameters
coarse adjustment of SE. If the predicted SNR is higher than
the SNR threshold for the selected SE, the SE increases. The We assume a hybrid C-C+L-band elastic optical network as
process is performed on each demand recursively while satis- described in Section 4 comprising single-mode fibers with
fying the QoT for all demands until all SEs have converged. full elastic bandwidth, with 4400 GHz of C-band spec-
Algorithm 3 describes how the center frequencies are reallo- trum and 5600 GHz of L-band spectrum. The simulation
cated, and the SEs are recursively reassigned jointly to satisfy parameters are listed in Table 3. The SE ranges from 2 to 12,
the QoT requirements. Algorithm 4 describes the C+L-link- i.e., dual-polarization BPSK (PM-BPSK) at the low end and
preferred routing algorithm, which is based on the k-shortest dual-polarization 64 QAM (PM-64QAM) at the high end (all
transmissions use independent dual polarizations). The SE
path algorithm and prioritizes routes using C+L links.
tuning uses dx = 0.5 and dy = 0.25.

6. SIMULATION DESCRIPTION AND RESULTS C. Network Traffic

In this section, we provide results from numerous network- We simulate future communication scenarios where data
level simulations first on the NSFNET-24 topology and then transmissions are expected to be quite heterogeneous, as well as
on the COST239 network topology to demonstrate the ben- high volume [35]. We consider that each node in the network
efits of using the proposed PAHSE algorithm and the fused communicates with all the others, i.e., the total number of
 Research Article Vol. 15, No. 11 / November 2023 / Journal of Optical Communications and Networking 907

Table 3. System Parameters

β2 −21.7 ps2 /km
γ 1.32 (W · km)−1
n sp 1.58
ν 193.55 THz
Cr 0.028 (W · km · THz)−1
G 0.015 W/THz
L 100 km
α 0.22 dB/km
Target BER 4 × 10−3
k 5

demands is M × |N| × (|N| − 1), where M is the number of

transmissions requested per source–destination pair. In our
simulations, we use M = 1 for NSFNET-24 and M = 8 for
COST239; COST239 is denser and has a smaller diameter,
and thus, a higher data rate is achievable than can be supported
by a single demand. The scale of our simulation is therefore
significantly larger than in [25,36].
We consider three scenarios that emulate (a) heterogeneous
traffic, (b) homogeneous traffic, and, for the NSFNET-24,
(c) realistic population-based demands. For traffic type (a),
the bit rate of each demand is random and uniformly distrib-
uted across various ranges representing different degrees of
heterogeneity in future optical networks; all plots for traffic
type (a) show the mean values of the quantity measured over
100 trials and 95% confidence intervals, marked as “error
bars.” For traffic type (b), the demands have identical bit rates,
which represents future scenarios with a well-distributed and
decentralized traffic pattern. For traffic type (c) used on the
NSFNET-24, the demand volume is estimated based on the
US population served by pairwise network nodes, provided
by [16], with a scale factor that results in a total sum of data
transmission T. The demand volume is deterministic in both
scenarios (b) and (c). In all simulations, the minimum band-
width of any demand is 30 GHz to ensure the accuracy of the
EGN model [26,28,37].
Fig. 5. Maximum spectrum used on the NSFNET-24 network:
(a) heterogeneous traffic, (b) homogeneous traffic, (c) realistic
D. Numerical Results for NSFNET-24 population-based demands. The plots on the left show the spectrum
In this section, the NSFNET-24 network with |N| = 24 nodes usage of the links using the C-band; the plots on the right show the
spectrum usage of the links using the C+L-band.
and |L| = 86 links (the average topology degree is 3.58), as
shown in Fig. 3, is used to test the proposed algorithm.
The highest spectrum needed across the network to sat- Given the 4400 GHz C-band spectrum and the 10 THz
isfy all demands is a measure of the efficiency of a resource C+L-band spectrum, the proposed algorithm can accommo-
provisioning algorithm [38]. Figures 5(a)–5(c), represent date a larger traffic volume and more heterogeneous demands,
traffic types (a), (b), and (c), respectively. Each figure has two compared to the benchmark. Compared to Unfused C+L,
subfigures: on the left is the spectrum usage of the C-band Fused C+L w/TDHM saves up to 56.9% of spectral resources
links of the hybrid C-C+L network, and on the right is the for the traffic model (a) cases tested. For scenario (b), the
spectrum usage of the evolved links that use the full C+L-band proposed algorithm can save up to 56.7% more in spectral
spectrum. The plots show that the fused C+L-band saves a resources than the benchmark Unfused C+L for the cases
significant spectrum compared to the unfused C+L-band tested. For scenario (c), the proposed algorithm can accommo-
system. Also, applying an advanced modulation format, such date up to 127% more data than the benchmark Unfused C+L
as TDHM, results in a lower spectrum usage than the CM for the cases tested. When demand bit rates exceed a uniform
format. The Fused C+L w/TDHM scheme, which utilizes distribution over 200–990 Gbps in traffic type (a), or the total
the proposed PAHSE algorithm, has the best performance in data transmission T exceeds 218 Tbps for traffic type (c),
spectrum usage compared to the other benchmarks, allowing none of the techniques tested works, and the network needs
the highest network throughput. more capacity, such as using SDM transmission. Hence, we
 908 Vol. 15, No. 11 / November 2023 / Journal of Optical Communications and Networking Research Article

Table 4. Maximum Achievable Network Throughput for NSFNET-24

Total Maximum Throughput (Tbps)
Algorithms (a) Heterogeneous Traffic (Average) (b) Homogeneous Data-Rate Demands (c) Population-Based Demands
Fused C+L w/TDHM 164 168 102
Fused C+L 135 141 87
Unfused C+L 76 76 45

conclude that the proposed algorithm will support an increase

in communication throughput, up to a certain point.
Table 4 shows the maximum achievable throughput of the
hybrid C-C+L-band network. The Fused C+L algorithm
enables higher throughput than the Unfused C+L algorithm.
Furthermore, the throughput of the Fused C+L w/TDHM
approach is higher than Fused C+L because of the TDHM
technique and the PAHSE algorithm. For traffic model (a),
the network throughput is increased by the proposed Fused
C+L w/TDHM algorithm from 76 Tbps for Unfused C+L
to 164 Tbps for Fused C+L w/TDHM , a 116% perform-
ance increase. For traffic model (b), the proposed Fused C+L
w/TDHM has a maximum total throughput of 168 Tbps,
which is 92 Tbps higher compared to the benchmark Unfused
C+L (121% higher). Scenario (c), which has uneven traffic
distribution over the network, has the highest increase in total
throughput compared to the Unfused C+L (127% higher).
Figure 6 shows the distributions of the SE used for traffic
demands in the network. For the benchmark Unfused C+L,
only even positive integer SEs are observed since conventional
modulation is used. On the contrary, our heuristic PAHSE
optimization algorithm obtains a finer SE granularity for
the proposed Fused C+L w/TDHM . As a result, the overall
Fused C+L w/TDHM SEs are higher than Unfused C+L.
Note that an SE of 12, the PM-64QAM modulation format,
is never used in the NSFNET-24 network for the simulation
parameters tested. Figures 6(a) and 6(b) show many demands
modulation format SEs in the range of 6–6.25 because these
have a transmission reach of slightly more than 1000 km.
Figure 6(b) shows an SE peak in the range of 8–8.25 due to
the minimum limit of 30 GHz [37]. SE cannot be so high
that the demand bandwidths are narrower than 30 GHz.
Figure 6(c) shows more low-level modulation formats used
than in Figs. 6(a) and 6(b) because traffic model (c) has many
small volume demands due to the geographically uneven
distribution of the US population.
Figure 7 shows the average SE per demand. The pro-
posed Fused C+L w/TDHM has an average SE of up to
6.53 bits/symbol for traffic model (a), which is between dual- Fig. 6. Histogram of spectral efficiency across the NSFNET-24
polarized 8-QAM and 16-QAM. For model (b), the overall network: (a) heterogeneous case for demand bit rates uniformly
trend in average SE is very similar to model (a); the average SE distributed from 200 to 300 Gbps, (b) homogeneous data-rate
demands of 250 Gbps, (c) realistic population-based demands with
of Fused C+L w/TDHM is 6.60 bits/symbol, a 17% improve-
T = 90 Tbps.
ment compared to Unfused C+L. For model (c), the average
SE increases when the total data transmission increases. The
shapes of the curves in cases (a), (b), and (c) are different due
to the significant unevenness in the US population and com- requirements compared to the benchmark Unfused C+L for
munication needs across the NSFNET-24 network for (c), the cases tested, and thus leads to significant savings in trans-
where there are many smaller data flows. Figures 7(a)–7(c) mission resources. Moreover, both the TDHM technique and
show that the proposed algorithm has a maximum of 15.7%, modulation/wavelength conversion contribute greatly to the
16.5%, and 16.0% higher SE without violating the QoT increase in the average SE.
 Research Article Vol. 15, No. 11 / November 2023 / Journal of Optical Communications and Networking 909

The hybrid C-C+L-band COST239 network is simulated

using the same parameters as in Section 6.D. Figure 9 shows
the spectrum usage of two different traffic scenarios described
in Section 6.C: (a) heterogeneous traffic and (b) homogeneous
traffic. The results show that for scenario (a), our proposed
algorithm Fused C+L w/TDHM can save up to 47% of spec-
trum resources compared to the Unfused C+L benchmark.
For scenario (b), Fused C+L w/TDHM can save up to 45%
spectrum resources compared with Unfused C+L. This shows
that our algorithm is robust to the choice of topology and is
suitable for dense networks such as COST239.
Table 5 shows the maximum achievable throughput of
the hybrid C-C+L-band COST239 network. Fused C+L
Fig. 7. Average spectral efficiency over all demands for the w/TDHM outperforms Unfused C+L in network throughput.
NSFNET-24 network: (a) heterogeneous traffic, (b) homogeneous
traffic, (c) realistic population-based demands. The legend in (b)
For traffic model (a), compared with Unfused C+L, using
applies to all figures. the proposed Fused C+L w/TDHM can increase the net-
work throughput by 108 Tbps, which is a 72% performance
enhancement. In the scenario of traffic model (b), the proposed
E. Numerical Results for COST239 algorithm has even more significant benefits, reaching a 77%
enhancement in network throughput.
In this section, the COST239 network with |N| = 11 nodes
Figure 10 shows the average SE per demand within the
and |L| = 52 links (the average topology degree is 4.73), as
COST239 network. The results’ pattern of both traffic mod-
shown in Fig. 8, is considered. The COST239 network is els (a) and (b) are similar. The proposed Fused C+L w/TDHM
denser compared to the NSFNET-24 topology. Overall, the approach again has the highest SE, with an average SE around
results simulated based on the COST239 network and the 9, which is between PM-16-QAM and PM-32-QAM. The
NSFNET-24 have similar patterns and trends, further vali- average SE of Fused C+L is slightly higher than PM-16-QAM,
dating the applicability and usability of our proposed stopgap and Unfused C+L is the lowest, ranging from PM-8-QAM to
solution. PM-16-QAM. Compared with the benchmark Unfused C+L,
To deploy regeneration nodes, we use the approach in traffic model (a), the average SE of Fused C+L w/TDHM is
described in Section 4. Because the COST239 network is increased by 18%. In traffic model (b), the SE is increased by
smaller in diameter than the NSFNET-24, it can support 19%.
a modulation of higher order than QPSK; we use an SNR Figure 11 shows the distributions of the SE used for traf-
threshold higher than for QPSK so that the SE can be increased fic demands in the COST239 network. For Fused C+L
by using a few regenerators to overcome the PLI and fuse w/TDHM , the results of the COST239 network are more
the hybrid network. By adjusting the SNR threshold, we can uniform compared to the NSFNET-24 results given in Fig. 6
identify which links in the topology become congested. This because the network topology of COST239 contains transpar-
algorithm results in a hybrid C-C+L network with six links ent segments of widely different lengths. Compared with the
upgraded to the C+L-band and with three regeneration nodes, benchmark Unfused C+L, the SE distribution of the proposed
as shown in Fig. 8(b). Fused C+L w/TDHM has a higher proportion of high-order

Fig. 8. Evolution from the C-band to the C+L-band in the COST239 topology; the number adjacent to each link is the link length in km.
(a) Current C-band network, (b) hybrid C-C+L-band network, and (c) final C+L-band network.
 910 Vol. 15, No. 11 / November 2023 / Journal of Optical Communications and Networking Research Article

Fig. 9. Maximum spectrum used on COST239 for M = 8

demands per node pair: (a) heterogeneous traffic, (b) homogeneous
traffic. The plots on the left show the spectrum usage of the links
using the C-band; the plots on the right show the spectrum usage of Fig. 11. Histogram of spectral efficiency across the COST239
the links using the C+L-band. network for M = 8: (a) heterogeneous case for demand bit rates uni-
formly distributed from 250 to 350 Gbps, (b) homogeneous data-rate
demands of 350 Gbps.
Table 5. Maximum Achievable Network Throughput
for COST239
modulation formats, i.e., the proportion of SE between 10
Total Maximum Throughput (Tbps) and 12 of the proposed Fused C+L w/TDHM is significantly
(a) Heterogeneous (b) Homogeneous higher than the benchmark Unfused C+L.
Algorithms Traffic (Average) Data-Rate Demands
Fused C+L 257 291 7. CONCLUSIONS
w/TDHM
Fused C+L 229 258 In this paper, we proposed a technique to fuse the C- and
Unfused C+L 149 165 L-bands using the PAHSE algorithm as a resource pro-
visioning algorithm and jointly applied the TDHM and
modulation/wavelength conversion techniques. The PLIs are
estimated by the EGN model for the C-band and the GN
model with Raman scattering for the evolved C+L-band for
the hybrid C-C+L-band system. The goal is to design an EON
resource allocation algorithm and consider the throughput
improvement obtained by TDHM with fused C+L over only
bottleneck links, which meets our society’s increasing short-
term communication needs and prepares for the C+L-band
upgrade of the entire network.
A large-scale simulation shows that the proposed algo-
rithm saves significant network resources compared to the
benchmark algorithm. For the NSFNET-24 network, the
proposed algorithm saves up to 56.9% of spectral resources
while increasing the network throughput by 116% for random
Fig. 10. Average spectral efficiency over all demands for
traffic, compared to the benchmark for the cases tested. The
the COST239 network for M = 8: (a) heterogeneous traffic,
(b) homogeneous traffic.
algorithm is also effective at increasing the throughput of the
COST239 network, evidencing its wide applicability. We
 Research Article Vol. 15, No. 11 / November 2023 / Journal of Optical Communications and Networking 911

conclude that the proposed algorithm can be used as a stopgap 15. P. Poggiolini, G. Bosco, A. Carena, V. Curri, Y. Jiang, and F.
measure before high-cost capacity expansion techniques are Forghieri, “The GN-model of fiber non-linear propagation and its
applications,” J. Lightwave Technol. 32, 694–721 (2014).
needed.
16. M. Batayneh, D. A. Schupke, M. Hoffmann, A. Kirstaedter, and B.
Using more advanced modulation techniques within the Mukherjee, “On routing and transmission-range determination of
PAHSE algorithm, such as PCS, as well as more sophisticated multi-bit-rate signals over mixed-line-rate WDM optical networks
routing and spectrum assignment methods, is the subject of for carrier Ethernet,” IEEE/ACM Trans. Netw. 19, 1304–1316 (2011).
future studies. 17. A. Carena, G. Bosco, V. Curri, Y. Jiang, P. Poggiolini, and F.
Forghieri, “EGN model of non-linear fiber propagation,” Opt.
Funding. National Natural Science Foundation of China (62171175). Express 22, 16335–16362 (2014).
18. A. Mitra, D. Semrau, N. Gahlawat, A. Srivastava, P. Bayvel, and A.
Lord, “Effect of channel launch power on fill margin in C+L band
REFERENCES elastic optical networks,” J. Lightwave Technol. 38, 1032–1040
(2020).
1. S. Talebi, F. Alam, I. Katib, M. Khamis, R. Salama, and G. N. 19. Q. Zhuge, M. Morsy-Osman, X. Xu, M. Chagnon, M. Qiu, and D.
Rouskas, “Spectrum management techniques for elastic optical V. Plant, “Spectral efficiency-adaptive optical transmission using
networks: a survey,” Opt. Switching Netw. 13, 34–48 (2014). time domain hybrid QAM for agile optical networks,” J. Lightwave
2. E. Agrell, M. Karlsson, A. Chraplyvy, D. J. Richardson, P. M. Technol. 31, 2621–2628 (2013).
Krummrich, P. Winzer, K. Roberts, J. K. Fischer, S. J. Savory, B. 20. G. Bosco, “Advanced modulation techniques for flexible optical
J. Eggleton, M. Secondini, F. R. Kschischang, A. Lord, J. Prat, I. transceivers: the rate/reach tradeoff,” J. Lightwave Technol. 37,
Tomkos, J. E. Bowers, S. Srinivasan, M. Brandt-Pearce, and N. 36–49 (2019).
Gisin, “Roadmap of optical communications,” J. Opt. 18, 063002 21. J. Cho and P. J. Winzer, “Probabilistic constellation shaping for opti-
(2016). cal fiber communications,” J. Lightwave Technol. 37, 1590–1607
3. R. G. Van Uden, R. A. Correa, E. A. Lopez, F. Huijskens, C. Xia, G. (2019).
Li, A. Schülzgen, H. De Waardt, A. Koonen, and C. M. Okonkwo, 22. Q. Zhuge, M. Reimer, A. D. Shiner, A. Borowiec, D. Charlton, F.
“Ultra-high-density spatial division multiplexing with a few-mode Zhang, Q. Meng, W. Wei, D. V. Plant, and M. O’Sullivan, “Flexible
multicore fibre,” Nat. Photonics 8, 865–870 (2014). modulation and nonlinear tolerance for coherent transceivers,”
4. R.-J. Essiambre, G. Kramer, P. J. Winzer, G. J. Foschini, and B. in Opto-Electronics and Communications Conference (OECC)
Goebel, “Capacity limits of optical fiber networks,” J. Lightwave (Institute of Electrical and Electronics Engineers, 2015).
Technol. 28, 662–701 (2010). 23. M. Reimer, S. O. Gharan, A. D. Shiner, and M. O’Sullivan,
5. N. Deng, L. Zong, H. Jiang, Y. Duan, and K. Zhang, “Challenges “Optimized 4 and 8 dimensional modulation formats for variable
and enabling technologies for multi-band WDM optical networks,” capacity in optical networks,” in Optical Fiber Communication
J. Lightwave Technol. 40, 3385–3394 (2022). Conference (Optica Publishing Group, 2016), paper M3A.4.
6. N. Guo, Y. Yu, K. Zhang, N. Deng, T. Chang, and G. Shen, 24. L. Yan, E. Agrell, and H. Wymeersch, “Sensitivity comparison
“Power optimization for C+L-band signal transmission,” in of time domain hybrid modulation and rate adaptive coding,” in
Asia Communications and Photonics Conference (ACP) (2022), Optical Fiber Communication Conference (Optica Publishing Group,
pp. 1051–1055. 2016), paper W1I.3.
7. A. Ferrari, E. Virgillito, and V. Curri, “Band-division vs. space-division 25. H. Dai, Y. Li, and G. Shen, “Explore maximal potential capacity
multiplexing: a network performance statistical assessment,” of WDM optical networks using time domain hybrid modulation
J. Lightwave Technol. 38, 1041–1049 (2020). technique,” J. Lightwave Technol. 33, 3815–3826 (2015).
8. R. Sadeghi, B. Correia, A. Souza, N. Costa, J. Pedro, A. Napoli, 26. P. Poggiolini and Y. Jiang, “Recent advances in the modeling of the
and V. Curri, “Transparent vs translucent multi-band optical net- impact of nonlinear fiber propagation effects on uncompensated
working: capacity and energy analyses,” J. Lightwave Technol. 40, coherent transmission systems,” J. Lightwave Technol. 35, 458–480
3486–3498 (2022). (2017).
9. G. M. Saridis, D. Alexandropoulos, G. Zervas, and D. Simeonidou, 27. M. R. Zefreh, A. Carena, F. Forghieri, S. Piciaccia, and P. Poggiolini,
“Survey and evaluation of space division multiplexing: from tech- “Accurate closed-form GN/EGN-model formula leveraging a large
nologies to optical networks,” IEEE Commun. Surv. Tutorials 17, QAM-system test-set,” IEEE Photon. Technol. Lett. 31, 1381–1384
2136–2156 (2015). (2019).
10. D. Soma, Y. Wakayama, S. Beppu, S. Sumita, T. Tsuritani, T. 28. P. Johannisson and E. Agrell, “Modeling of nonlinear signal distor-
Hayashi, T. Nagashima, M. Suzuki, M. Yoshida, K. Kasai, M. tion in fiber-optic networks,” J. Lightwave Technol. 32, 3950–4544
Nakazawa, H. Takahashi, K. Igarashi, I. Morita, and M. Suzuki, (2014).
“10.16-peta-B/s dense SDM/WDM transmission over 6-mode 29. L. Yan, E. Agrell, H. Wymeersch, and M. Brandt-Pearce, “Resource
19-core fiber across the C+L band,” J. Lightwave Technol. 36, allocation for flexible-grid optical networks with nonlinear channel
1368–1375 (2018). model,” J. Opt. Commun. Netw 7, B101–B108 (2015).
11. Q. Yao, H. Yang, B. Bao, J. Zhang, H. Wang, D. Ge, S. Liu, D. Wang, 30. D. Semrau, R. I. Killey, and P. Bayvel, “A closed-form approxima-
Y. Li, D. Zhang, and H. Li, “SNR re-verification-based routing, band, tion of the Gaussian noise model in the presence of inter-channel
modulation, and spectrum assignment in hybrid C-C+L optical net- stimulated Raman scattering,” J. Lightwave Technol. 37, 1924–1936
works,” J. Lightwave Technol. 40, 3456–3469 (2022). (2019).
12. B. G. Bathula, R. K. Sinha, A. L. Chiu, M. D. Feuer, G. Li, S. L. 31. N. Guo, S. K. Bose, B. Mukherjee, and G. Shen, “Impact of fiber
Woodward, W. Zhang, R. Doverspike, P. Magill, and K. Bergman, attenuation and effective area on spectrum efficiency of elastic
“Constraint routing and regenerator site concentration in ROADM optical networks,” J. Lightwave Technol. 40, 2200–2213 (2022).
networks,” J. Opt. Commun. Netw. 5, 1202–1214 (2013). 32. Y. Xu and M. Brandt-Pearce, “Physical-layer impairment estima-
13. M. Jinno, T. Takagi, and K. Kiyokawa, “Minimal virtualized- tion for arbitrary spectral-shaped signals in optical networks,” Opt.
elastic-regenerator placement and least congestion resources Express 30, 17351–17360 (2022).
assignment for translucent elastic optical networks,” in Optical 33. Y. Xu, E. Agrell, and M. Brandt-Pearce, “Cross-layer static resource
Fiber Communication Conference (Optica Publishing Group, 2015), provisioning for dynamic traffic in flexible grid optical networks,”
paper Th3J.2. J. Opt. Commun. Netw. 13, 1–13 (2021).
14. A. Napoli, N. Calabretta, J. K. Fischer, N. Costa, S. Abrate, J. Pedro, 34. J. Zhao, L. Yan, H. Wymeersch, and E. Agrell, “Code rate optimiza-
V. Lopez, V. Curri, D. Zibar, E. Pincemin, S. Grot, G. Roelkens, tion in elastic optical networks,” in European Conference on Optical
C. Matrakidis, and W. Forysiak, “Perspectives of multi-band Communication (ECOC) (2015).
optical communication systems,” in 23rd Opto-Electronics and 35. M. Jinno, “Elastic optical networking: roles and benefits in beyond
Communications Conference (OECC) (2018). 100-Gb/s era,” J. Lightwave Technol. 35, 1116–1124 (2016).
 912 Vol. 15, No. 11 / November 2023 / Journal of Optical Communications and Networking Research Article

36. X. Wang, M. Brandt-Pearce, and S. Subramaniam, “Impact of wave- Engineering, Eindhoven University of Technology (TU/e), Eindhoven, The
length and modulation conversion on translucent elastic optical net- Netherlands. He is currently an associate professor at Hefei University of
works using MILP,” J. Opt. Commun. Netw. 7, 644–655 (2015). Technology. He is an author or co-author of more than 70 publications
37. R. Dutta and G. N. Rouskas, “Traffic grooming in WDM networks: and conference presentations and holds more than 10 patents. His research
past and future,” IEEE Netw. 16, 46–56 (2002). interests include developing practical and novel schemes for digital commu-
38. M. Yaghubi-Namaad, A. G. Rahbar, and B. Alizadeh, “Adaptive nication systems related to coded modulation, information theory, channel
modulation and flexible resource allocation in space-division- coding, and signal processing. Dr. Chen’s research has received multiple
multiplexed elastic optical networks,” J. Opt. Commun. Netw. 10, awards, including the CSC Scholarship, the 2018 Asia Communications
240–251 (2018).
and Photonics Conference (ACP2018) Best Paper Award, and the 2019
Optoelectronics and Communications Conference (OECC2019) Best Paper
Award. He also served on technical program committees (TPCs) and was
Yuxin Xu is an assistant professor at Zhejiang
workshop/session co-chair of several international conferences.
University of Technology. He received his Ph.D.
degree at the University of Virginia in 2019. He
was a postdoctoral researcher at the University
of Virginia in 2020. He was a visiting scholar at Maite Brandt-Pearce is a professor of Electrical
Nanyang Technological University in 2022. His Engineering and vice provost for faculty affairs at
research interests are fiber optic communications the University of Virginia (UVa). She joined UVa
and machine learning. after receiving her Ph.D. in electrical engineering
from Rice University in 1993. Her research inter-
ests include free-space optical communications,
visible-light communications, non-linear effects
Hang Xing received a B.E. degree in commu-
in fiber optics, and cross-layer design of optical
nications engineering from Jiliang University
networks subject to physical layer degradations.
of China, Hangzhou, in 2020. He is currently
Dr. Brandt-Pearce is the recipient of an NSF
pursuing his M.S. at the Zhejiang University of
CAREER Award and an NSF RIA. She is a co-recipient of Best Paper Awards
Technology. His research interests are resource
at ICC 2006 and GLOBECOM 2012. She has served on the editorial
allocation for optical networks and modeling of
boards of IEEE Transaction of Communications, IEEE Communications
physical layer impairments.
Letters, the IEEE/OSA Journal of Optical Communications and Networks, and
Springer’s Photonic Network Communications. She was a Jubilee Professor
at Chalmers University, Sweden, in 2014. After serving as General Chair of
the Asilomar Conference on Signals, Systems & Computers in 2009, she
served as Technical Vice-Chair of GLOBECOM 2016. She is a member
Bin Chen (S’14–M’17–SM’23) received the of Tau Beta Pi, Eta Kappa Nu, and a Fellow of the IEEE. In addition to
B.Sc. degree in electronic information science and co-editing a book entitled Cross-Layer Design in Optical Networks in the
technology from Hefei University of Technology, Springer Optical Networks Series in 2013, Prof. Brandt-Pearce has over 200
Hefei, China, in 2010 and the Ph.D. degree in technical publications.
electronics and communications engineering from
University College Dublin, Dublin, Ireland, in
2015. From 2016 to 2019, he had a postdoctoral
position with the Signal Processing Systems (SPS)
Group and the Electro-Optical Communication
Systems (ECO) Group, Department of Electrical