D48 Vol. 12, No.

9 / September 2020 / Journal of Optical Communications and Networking Research Article

Future optical access network enabled by

modularization and softwarization of access and
transmission functions [Invited]
Jun-ichi Kani,* Jun Terada, Tomoya Hatano, Sang-Yuep Kim, Kota Asaka, AND
Takashi Yamada
NTT Access Network Service Systems Laboratories, NTT Corporation, 1-1 Hikarinooka, Yokosuka, Kanagawa 239-0847, Japan
*Corresponding author: [email protected]

Received 25 February 2020; revised 21 April 2020; accepted 18 May 2020; published 16 June 2020 (Doc. ID 391544)

This paper discusses the future direction of optical access networks and proposes a new optical access architec-
ture that will support the evolution of optical access systems from a fiber-to-the-home infrastructure to a common
optical access platform that connects various devices and systems to anywhere they want to reach. We describe two
technologies for its implementation. One is “modularization,” which is expected to be realized before 2030, and
the other is “full softwarization,” which is the longer-term goal of drastically increasing network flexibility. For
each technology, we summarize related technical trends and discuss a reasonable system configuration focusing
on the access nodes. © 2020 Optical Society of America
https://doi.org/10.1364/JOCN.391544

1. INTRODUCTION connected to the communication network. Furthermore, a

Optical broadband access services, with the so-called fiber-to- high degree of convergence between cyberspace (virtual space)
the-home (FTTH) infrastructure, have been widely adopted and physical space (real space) is expected to be achieved by
since around 2005, and the numbers of subscribers in Japan connecting end users to computers in the network and linking
and in China now exceed 30 million and 100 million, respec- the computers with super-high-speed and super-low-latency
tively [1,2]. FTTH deployment relieved the bottleneck that connections [9]. The role of optical access systems will change
existed between the subscribers and the central office, so that from the FTTH infrastructure to the common optical access
Internet services such as high-resolution video streaming ser- platform that connects various devices and systems to any
vices and various cloud computing services have been greatly location necessary. This change is expected to further accelerate
advanced. The optical access networks used in this deployment the advancement of information and communication services.
are currently shifting from 1 Gbit class to 10 Gbit class to sup- To realize this, a new access network architecture that breaks
port the increasing bandwidth demands created by imminent through several restrictions created by the current optical access
communication and video services. The FTTH trend is seen network architecture is needed.
not just in Asia but also across the world; it is expected to serve This paper first discusses requirements for the optical access
50% of fixed broadband subscribers worldwide by the early platform to further accelerate the advancement of information
2020s [3]. and communication services, and then it proposes a new opti-
Meanwhile, mobile Internet services have been progressing
cal access architecture that is fundamentally different from the
since around 2010, and the number of Long-Term Evolution
current architecture. To realize the new optical access architec-
(LTE) subscribers exceeded 100 million in 2017 in Japan.
ture, it is mandatory to improve the transmission performance
In the mobile communication network with LTE and later
standards, mobile antennas are more densely deployed than through the use of advanced functions such as adaptive forward
those for the conventional network, and these antennas are error correction (FEC) as well as digital-coherent transmission
connected to the central office by optical fibers so that higher and reception. The key point is to employ such functions
speeds can be supported [4,5]. Given this, optical access sys- flexibly, i.e., only when needed. This paper describes two tech-
tems that can effectively accommodate many mobile antennas nologies to achieve this. One is “modularization”; it allows the
are now an urgent requirement [6–8]. flexible addition and replacement of functions. The other is
In addition to this background, everything, including power “full softwarization”; it further enhances flexibility and realizes
networks, automobiles, factory machines, and robots, will be the convergence between the network and the computing

1943-0620/20/090D48-09 Journal © 2020 Optical Society of America

Authorized licensed use limited to: University of Melbourne. Downloaded on June 23,2020 at 10:31:57 UTC from IEEE Xplore. Restrictions apply.
 Research Article Vol. 12, No. 9 / September 2020 / Journal of Optical Communications and Networking D49

Table 1. Features of the Future Optical Access

Platform
Current Optical Future Optical
Aspect Access Network Access Platform
Services (example) • Voice • Mobile fronthaul/
backhaul
• Data • IoT, big data (via
wireless antennas)
• Video • Direct computer
connections
• Factory machines,
medical devices
Capacity per ONU Up to 10 Gbit/s Up to 100 Gbit/s
(50 Gbit/s in the
future)
Fig. 1. FSAN Standard roadmap 2.0 [10]. End-to-end latency >10 ms <1 ms
Connection ONU to/from OLT Any to any
Distance ∼20 km As you go
system. This paper reviews the technical trends and discusses Power consumption ∼3.3 W × Less than the current
future perspectives for each of these technologies. number of subscribers network
Figure 1 shows the roadmap of optical access technology
evolution developed and released under the Full Service Access
solution in terms of minimizing the end-to-end latency. 1 ms
Network (FSAN) initiative [10]. This roadmap indicates that
should be a good target of the latency in terms of supporting
the future optical access system (FOAS) realized by disruptive
various time-sensitive applications such as a haptic communi-
technologies will appear after the enhancement of passive
optical network (PON) technologies such as gigabit-capable cation, a remote control of factory machines, and a real-time
PON (G-PON), 10 Gbit-capable symmetric PON (XGS- virtual-reality experience [14]. The distance to support the
PON), and 40 Gbit-capable PON (NG-PON2) [11–13]. We 1 ms latency will be ultimately limited by the deterministic
expect that the new architecture and the enabling technolo- latency created by the propagation speed of light in silica fiber,
gies described in this paper will support the realization of the which is 5 µs/km.
FOAS. Existing and future PON technologies are expected to Regarding the power consumption, the current IP-based
not only continuously support the existing FTTH infrastruc- communication network consumes 82.5% of its power in the
ture but also support the new architecture on demand through IP routers and aggregation switches if we focus on the power
modularization and softwarization. consumed inside the operators’ buildings [15]. Considering
that direct optical connection will drastically reduce the burden
of electrical routers and switches, the new architecture would
2. ARCHITECTUAL EVOLUTION be able to consume much less power than the conventional
The future optical access platform will connect various devices architecture depending on the traffic that cuts through the
and systems to anywhere they want to reach with the expecta- access node. A fair target would be to realize power consump-
tion of securing further advances in information and commu- tion less than that in the current network while drastically
nication services. Table 1 summarizes its key features compared increasing the network capability and flexibility as mentioned;
to the current optical access network. a perspective for the power consumption in the future optical
As the future optical access platform will offer converged access platform is further described in Section 4. As a refer-
access, it will cover mobile fronthaul/backhaul (MFH/MBH) ence, current IP-based communication equipment inside the
and various devices related to the Internet of Things (IoT) and operator’s buildings can consume ∼3.3 W × the number of
big data, such as smart phones/glasses/watches, human sensors, subscribers [15].
connected cars, surveillance cameras, through MFH/MBH Figures 2(a) and 2(b) illustrate typical optical access network
links. It should also cover direct computer connections configurations for the conventional and new architecture,
between any two locations; computers can be located at respectively.
user or cloud sites. Highly reliable services such as those needed In the conventional access network, all traffic is terminated
to control factory machines as well as medical devices should at the optical line terminal (OLT) at the access node, re-sorted
also be covered. The maximum capacity per optical network in the service node, and sent to the core network as illustrated
unit (ONU) will be 100 Gbit/s or more given the capacity in Fig. 2(a). PON systems are typically used to connect ONUs
increase in MFH [8] as well as direct computer connection at to an OLT. Access networks are typically connected to a metro
native bus speeds. network across edge routers, and metro networks are connected
End-to-end latency will be a very important factor in sup- to the core network across core routers.
porting and accelerating the evolution of information and On the other hand, in the proposed optical access network
communication services. The new architecture aims to allow based on the new architecture, the access node comprises
any-to-any direct optical connection, which is the ultimate the electrical processing part and the optical switching and

Authorized licensed use limited to: University of Melbourne. Downloaded on June 23,2020 at 10:31:57 UTC from IEEE Xplore. Restrictions apply.
 D50 Vol. 12, No. 9 / September 2020 / Journal of Optical Communications and Networking Research Article

across the architecture is important. In this regard, the pro-

posed architecture has the potential to offer advanced security
through quantum key distribution (QKD) using the direct
optical connection between any two locations; long-distance
QKD over several hundred kilometers as well as a QKD over a
power-splitter-based PON have been achieved [16,17].
At the access side, we can assume that new fiber infrastruc-
tures like the loop and bus topologies will emerge in addition
to the current point-to-point and star topologies for support-
ing higher reliability and more efficient user accommodation
[18,19] as various services with different requirements will
emerge. WDM will be utilized as a user multiplexing method
and/or a capacity increase method; each wavelength can be
dedicated to one ONU or shared among multiple ONUs via a
protocol as is done in PON systems in the various topologies
mentioned. In the case of using a PON protocol, e.g., in the
star and bus topologies, the OLT function is implemented in
the electrical processing part through the modularization or
softwarization technologies described in Sections 3 and 4.
Works related to the proposed architecture include the
architecture-on-demand (AoD) concept for transparent optical
networks [20] as well as the DISCUS project for access and
metro convergence using long-reach PONs [21]. The AoD
concept is to dynamically reconfigure the node architecture
by cross-connecting various optical functions such as optical
combiners/splitters, amplifiers, and spectrum-selected switches
through a large port-count optical switch so that new services
Fig. 2. Evolution of the optical access network: (a) current archi- and applications can be introduced more flexibly. This concept
tecture and (b) proposed new architecture. can be applied to the optical switching and WDM part in the
proposed architecture in Fig. 2(b), so that appropriate optical
functions, e.g., loss compensation, optical combine/split, and
wavelength multiplex/demultiplex, can be provided to each
wavelength-division multiplexing (WDM) part, and these
optical signal on demand. The DISCUS architecture assumes
two parts are completely separated. The optical switching and
the consolidation of electrical processing functions into the
WDM part drops only the traffic that needs to be processed at
DISCUS node, which covers subscribers up to ∼100 km
the access node to the electrical processing part, and it passes all from the node, using long-reach PONs, and assumes that a
other traffic to the core network without electrical termination, number of DISCUS nodes are connected through full-mesh
which thus realizes ultralow latency and lower power con- wavelength channels to form a flat and transparent core net-
sumption. Note that the metro networks in the conventional work. The DISCUS node employs an optical switch that
architecture will be consolidated to the core network in the cross-connects the wavelength channels in the core network as
new architecture in terms that all the nodes will be connected well as offering protection switching for the long-reach PONs
through optical paths, so the consolidated metro and core that are dual-parented to two neighboring DISCUS nodes.
network is simply shown as the core network in Fig. 2(b). The proposed architecture has an affinity with the DISCUS
Signals from an ONU are terminated at the electrical architecture as it allows full-mesh networking for the core
processing part at the access node, an electrical processing network and protection switching for the access network in a
function at another node in the core network, or at another similar way when needed.
ONU. The optical switching and WDM part uses wavelength This paper focuses on how to implement the electrical
multiplexers/demultiplexers or wavelength-selective switches processing part while leaving implementation of the optical
(WSSs) that multiplex/demultiplex signals to/from the access switching and WDM parts for further study, as the many
side as well as those at the core-network side as well as an opti- alternatives available need careful consideration.
cal switch that connects a port from an ONU to a port to the
electrical processing part or a port to the core network.
3. MODULARIZATION
Namely, the proposed architecture offers high-capacity
and low-latency connections through the direct optical con- A. Technical Trends
nection of any two locations. In addition, since the proposed In 2016, we proposed the concept of the Flexible Access
architecture accommodates a large number of services on the System Architecture (FASA); it makes it easier to add and
identical infrastructure, and since some of them can require change various functions by modularizing the access network
very high security, how to establish a secure communication equipment for the purpose of rapidly accommodating new

Authorized licensed use limited to: University of Melbourne. Downloaded on June 23,2020 at 10:31:57 UTC from IEEE Xplore. Restrictions apply.
 Research Article Vol. 12, No. 9 / September 2020 / Journal of Optical Communications and Networking D51

service requirements [22–24]. Its milestones include modular-

ization of the dynamic bandwidth assignment (DBA) function,
which determines the PON performance, as a software mod-
ule [25] as well as application programming interface (API)
standardization for DBA software modules [26,27]. Use cases
of the softwarized DBA include easier implementation of the
cooperative DBA with mobile systems for MFH [28] as well
as implementation of several different DBA algorithms in an
OLT for multipurpose PON sharing [29].
Around the same time, the Open Networking Lab, currently
the Open Networking Foundation (ONF), proposed the
concept of the Central Office Re-architected as a Datacenter
(CORD) [30]. Among several CORD projects, residential
CORD (R-CORD) aimed to disaggregate PON systems for Fig. 4. System configuration with the hardware OLT module and
FTTH services for the purpose of allowing telecommunication software DBA module based on the SEBA architecture.
operators to benefit from both the economics of scale as well
as the agility, and promoted the development of higher-layer
functions that had, up to then, been implemented as propri- SEBA assumes the use of both box-type OLTs and module-
etary designs inside the access system equipment as a set of type OLTs and assumes the optional implementation of
open source software. time-critical functions like DBA as software. Figure 4 illus-
How to split the conventional PON OLT functions into trates a system configuration in which the module-type OLTs
those kept as hardware and those implemented as software has are used and their DBA functions are placed at Compute,
several options as shown in Fig. 3. One option is to implement which is a local server collocated at the access node. Each
multi-point media access control (MPMC) and DBA func- module-type OLT is connected to a port of the aggregation
tions as software; the successful implementation of MPMC and service gateway (ASG) and controlled by the access node
(AN) driver via the ASG. An ASG is typically one or several
as software in a server was reported for gigabit Ethernet
electrical switches/routers to support aggregation in layer 2 as
PON (GE-PON) OLT as well as 10 Gbit Ethernet PON
well as service edge/broadband network gateway (BNG) func-
(10G-EPON) OLT [31,32]. Another option is to implement
tions in layer 3. The ASG is controlled by the ASG driver. The
MPMC as hardware and DBA as software; this can be realized
network edge mediator (NEM) supports FCAPS functions
by using the PON abstraction interface for time-critical appli-
and provides interfaces for the operation support system (OSS)
cations as specified by the Broadband Forum [26,27]. In this
and/or the orchestrator. Functions related to flow control, such
case, the DBA function can be located either on a server or in as the multicast proxy, are put on the SDN controller. The AN
a central processing unit (CPU) inside the OLT hardware; the and ASG drivers are typically put on Compute as well as the
former supports the use of the module-type OLT, while the time-critical functions like DBA [33].
latter supports the use of the box-type OLT as illustrated in
Fig. 3. Yet another option is to implement both MPMC and
DBA as hardware while moving higher-layer functions such B. Modularized Access Node Configuration for the
as the multicast proxy function, which allows the IP multicast Future Optical Access Platform
protocol to take advantage of the broadcast characteristic of To realize the future optical access platform shown in Fig. 2(b),
PON, and fault, configuration, accounting, performance, one requirement is to make it possible to flexibly add, replace,
and security (FCAPS) functions; this is the basic concept of and delete various access node functions on demand. The
R-CORD. first approach for this is to modularize these functions, i.e., to
As an evolved form of these pioneering efforts of access- implement each function as a software or hardware module.
system modularization and disaggregation, a new project called Studies to disaggregate transmission systems are progressing
Software Defined Networking (SDN)-Enabled Broadband in the Open and Disaggregated Transport Network (ODTN)
Access (SEBA) was launched in the ONF in 2018; SEBA project in the ONF as well as the Open Optical and Packet
Reference Design version 1.0 was issued in March 2019 [33]. Transport (OOPT) project in the Telecom Infra Project (TIP)
[34,35]. Accordingly, high-speed white-box switches and mul-
tiplexers that support pluggable digital-coherent transceiver
modules as well as wavelength-tunable transceiver modules
have been developed [36].
One important issue for realizing the future optical access
platform is how to integrate these disaggregated transmission
functions with the access-network functions described in
Section 3.A in the access node. Another important issue is how
to meet the requirements, which are likely to vary flow by flow.
To address this, we must realize “function chaining” to flexibly
Fig. 3. Implementation options of a modularized PON OLT. select the functions needed flow by flow.

Authorized licensed use limited to: University of Melbourne. Downloaded on June 23,2020 at 10:31:57 UTC from IEEE Xplore. Restrictions apply.
 D52 Vol. 12, No. 9 / September 2020 / Journal of Optical Communications and Networking Research Article

support this failure recovery by switching the corresponding

port(s) to the backup port(s).
The mentioned functions are integrated through a switch
fabric or a high-speed bus, and they thus realize integrated
control as well as a flexible function chaining. Note that,
given the recent trend of copackaging high-speed optics with
Ethernet switch integrated circuits (ICs), the connections in
the switch fabric and the high-speed bus may become optical.
Two examples of the function chaining are described below.
The green line with circles in Fig. 5 shows an example of
function chaining for the MFH access. In 5G MFH, the
remote unit (RU), DU, and centralized unit (CU) conduct a
wireless-optical conversion, a lower-layer part of the baseband
processing, and a higher-layer part of the baseband process-
ing, respectively [8,39]; one reasonable implementation is to
collocate the CU and the DU at a central office and connect
the RU at a remote antenna site to the CU/DU with an optical
fiber. After the baseband processing in the CU/DU, signals
Fig. 5. Modularized access node configuration for the future opti-
are sent to the 5G core network (5GC). For this purpose,
cal access platform.
we need to chain the following functions in the access node:
the DU function, the CU function, a switching function to
forward the signals to an appropriate node in the 5GC, and a
Figure 5 shows an example of the modularized access node transponder, i.e., an appropriate transmission function to reach
configuration. It assumes the accommodation of various access the node. In Fig. 5, it is assumed that the CU function and
infrastructure topologies as mentioned in the previous section. the switching function are realized as software, i.e., vCU and
As an example, this figure shows the combined bus and star vSW in the figure, while it is assumed that the DU function
topology used to provide PON-type wavelength-shared ser- remains in hardware, e.g., through a fully programmable gate
vices for consumer users and IoT as well as the loop topology array (FPGA), considering the difficulty of virtualizing it as
to provide wavelength-dedicated services for business users, mentioned before.
MFH, and direct computer connections. Functions in a con- The red line with triangles in Fig. 5 shows the example of
ventional OLT are achieved by combining a hardware OLT function chaining for residential broadband access. In a typical
with software on the access-system controller. The large-scale broadband access network, optical signals from residential cus-
layer-2 aggregation switch (SW) is assumed to remain a hard- tomers are terminated with an OLT, aggregated with a layer-2
ware box in this example. The dedicated server provides various switch, and sent to a BNG, which provides layer-3 connectivity
software functions for data signals, such as a virtual centralized to each customer. The BNG is connected to a core router in the
unit (vCU) for MFH [8], a virtual BNG (vBNG), and a virtual IP core network. For this purpose, the following functions are
switch (vSW) like Open vSwitch, while controllers are used chained on the red line in Fig. 5: OLT, aggregation SW, BNG
to process control signals. A specific network interface card (vBNG in the figure), and a transponder. Note that a BNG
(NIC) is used for the MFH connection; it includes distributed can be implemented as hardware and can be consolidated in
unit (DU) functions, which may be difficult to fully virtual- a higher-level node depending on the scale of the service and
ize with an affordable margin and fair scalability in terms of other requirements; this example, however, assumes that it is
virtualized as a vBNG and set in the access node. A service-
current technology [37]. Wavelength and other transponder
layer process like edge computing becomes implementable
parameters can be controlled by the transmission controller
in the access node if the BNG function is set there, and one
through the transponder abstraction interface (TAI) [36]. We
use case is to put a vBNG at the access node only for a limited
assume that adaptive modulation and adaptive forward error
number of customers, i.e., customers of the highest class of
correction (FEC) are implemented in addition to wavelength broadband service, while consolidating BNG functions for the
tunability in each transponder, so that the modulation for- other customers in a higher-level node.
mat and FEC strength are adjusted to match the distance the Each ONU or remote node that directly accesses a higher-
transponder needs to reach. level node by passing through the optical switching and WDM
For the loop access network, a simple optical add/drop part employs a wavelength-tunable transmitter and optionally
multiplexer (OADM) architecture such as the one that uses employs a digital-coherent transmission function depending
2-by-2 optical couplers can be applied, and WDM access can on the maximum distance the ONU must reach.
be implemented on that; i.e., each remote node uses a different Note that it may not be possible to migrate currently
wavelength pair to communicate with the access node [38]. deployed PON systems to this new architecture because an
Loop networks are robust in general because each remote node optical switching device with loss of a few decibels (dB) must
can keep communication active by using the other direction of be inserted between the OLT and the existing outside plant.
the loop fiber if a fiber cut occurs in one direction. The optical While this proposal basically assumes new systems, it may
switching and WDM part in the proposed access node can be possible to accommodate existing systems after a careful

Authorized licensed use limited to: University of Melbourne. Downloaded on June 23,2020 at 10:31:57 UTC from IEEE Xplore. Restrictions apply.
 Research Article Vol. 12, No. 9 / September 2020 / Journal of Optical Communications and Networking D53

processing units (GPUs) as an accelerator [40,41]. Regarding

the encryption encoding, which is not a physical-layer function
but needs high processing speeds similar to the functions in the
physical layer, 5.37 Gbit/s software operation of CTR-AES128
has been achieved with the use of GPU acceleration [42]. Note
that high-speed encryption encoding/decoding can be used in
combination with QKD in the new optical access architecture
as described in Section 2.
Furthermore, a challenge to implement a digital signal
processing (DSP) function as software for realizing a flexi-
Fig. 6. Cascaded modularization of a GE-PON OLT. ble digital-coherent transmission system has been reported;
the real-time coherent detection of a 5 Gbit/s quadrature
phase-shift keying (QPSK) signal has been demonstrated
consideration confirms whether the loss budget margin covers [43]. The softwarization of DSP functions has the potential to
the additional loss. dramatically increase the flexibility of the network as various
At the access side, optical transceivers can be modularized in DSP-based and DSP-assisted modulation methods including
addition to the OLT hardware so that each OLT port adapts QPSK, quadrature amplitude modulation (QAM), orthogo-
to different requirements, e.g., different loss budgets. Figure 6 nal frequency-division multiplexing (OFDM), and pulse
shows an implementation example of this cascaded modular- amplitude modulation (PAM) can be properly used for accom-
ization. The transceivers for WDM access (for business and modating different requirements. It has been studied to apply
for MFH in the figure) can be tunable so that the operator can the use of accelerators for softwarizing physical-layer functions
avoid preparing and managing the transceivers wavelength by of mobile access systems as well [44–46]. The physical-layer
wavelength. softwarization is especially suitable to the centralized and cloud
radio access network (C-RAN) architecture, which centralizes
4. FULL SOFTWARIZATION the baseband processing resources into a pool and virtual-
izes soft baseband units on demand [44]. Various types and
A. Technical Trends
generations of mobile access systems can be flexibly deployed
As challenges beyond the modularization efforts described in with adaptation to the changing needs through physical-layer
Section 3, full softwarization down to electrical processing in softwarization in the C-RAN architecture.
the physical layer has been studied for increasing the flexibility
and decreasing the application-specific or protocol-specific
B. Power-Consumption Performance Trend of
hardware more drastically. Figure 7 shows the main functions
Microprocessors
of higher-speed and longer-reach optical access termination (in
the future) as well as software implementation of each function Accelerating computing technologies powered by commer-
reported so far. cially available multicore microprocessors and accelerators
Main physical-layer functions such as framing, line coding, such as GPUs and FPGAs are key enablers of greater flexibility
and FEC encoding/decoding have been successfully oper- and minimal specific-purpose hardware as mentioned. Even
ated as software at 10.3125 Gbit/s in real time using graphic though performance sufficient to support state-of-the-art
optical systems such as high-speed PONs and digital-coherent
transmission systems has been achieved and demonstrated as
summarized in Section 4.A, a key aspect that must be consid-
ered for actual system deployment is the power consumption of
the access segment given the extensive use of microprocessors
as well as accelerators to handle large numbers of customers.
Microprocessors undergo continuous evolution toward
better power-consumption performance due to advances in
power control methods and enormous progress in microelec-
tronics. Figure 8 shows the power-consumption performance
trend, which is defined by the normalized ratio of power con-
sumption in thermal design power (TDP) to computation
performance [units of million instructions per second (MIPS)].
In fact, the results indicate the continuous improvement in
power-consumption performance; i.e., the average annual
rates are approximately 14% and 13% for Core and Xeon
series CPUs, respectively. Given these annual rates, the power-
Fig. 7. Main functions for higher-speed and longer-reach optical
consumption performance is expected to be improved by 5
access termination and software implementation status reported to 50 times in 10 and 25 years, respectively; this will ensure
so far. the positive prospects for the commercial deployment of full

Authorized licensed use limited to: University of Melbourne. Downloaded on June 23,2020 at 10:31:57 UTC from IEEE Xplore. Restrictions apply.
 D54 Vol. 12, No. 9 / September 2020 / Journal of Optical Communications and Networking Research Article

Fig. 8. Power-consumption performance of Intel Core and Xeon

processors.
Fig. 9. Access node configuration for the new access-network
architecture, with the full-softwarization approach.
softwarized access networks. The power-consumption per-
formance ratios of GPUs are dropping even more drastically The access node controller, the optical switching and WDM
[47]. Moreover, given the emergence of 5G mobile in conjunc- parts, and the inter-server connection can be inherited from the
tion with IoT and their upgraded variants, which will have a modularized configuration discussed in Fig. 5. This will allow
multiplier effect on the microprocessor industry in the near a smooth migration from the modularized configuration to the
future, full softwarization will be further accelerated towards fully softwarized configuration discussed in this section.
commercial deployment.

5. SUMMARY
C. Fully Softwarized Access Node Configuration for
the Future Optical Access Platform This paper discussed the future direction of the optical access
Given the trends described in the above subsections, we can network and proposed a new optical access architecture with
expect the full softwarization of electrical functions in the focus on its access node configuration. The new architecture
optical access network by the 2030s so that the future access will support the evolution of the optical access system from
nodes can connect various devices and systems to anywhere FTTH infrastructure to the common optical access platform
they want to reach more flexibly, i.e., with the minimum usage that allows unlimited connection of various devices and sys-
and replacement of purpose-specific hardware. Figure 9 shows tems. We described two technologies for its implementation;
an example of the fully softwarized access node configuration. one is “modularization” and the other is “full softwarization.”
The only application-specific hardware needed is the optical Modularization is expected to be realized before 2030, while
transceiver; PON and point-to-point optics will be typically full softwarization can be considered a long-term challenge
used at the access side while intensity-modulation direct- for more drastic evolution after 2030. For each technology, we
detection (IM-DD) optics and analog coherent optics (ACO) reviewed related technical trends and elucidated a reasonable
will be typically used at the core-network side with wavelength access node configuration. Further studies are encouraged
tunability for WDM transmission. The access-side optics can towards the realization of the new optical access architecture
be modularized, or can be tunable for WDM access as men- that will serve as the base permitting further acceleration in the
tioned in the previous section. These optical transceivers will evolution of information and communication services.
be directly connected to servers. As mentioned in Section 3.B,
the interfaces to/from the switch fabric or the high-speed bus REFERENCES
can be optical as well. The difference between the internode 1. https://www.statista.com/statistics/821488/japan-broadband-
and intranode communication interfaces may become just subscription-number-by-type/.
optics through further innovation of optical interconnection 2. https://www.statista.com/statistics/1021346/ftth-customers-of-
technologies for computers. china-telecom/.
3. https://broadbandtrends.wordpress.com/2017/01/18/ftth-to-
Employing accelerators such as GPUs and FPGAs will be
represent-nearly-50-of-fixed-broadband-subscribers-by-2022/.
mandatory for the server at the core-network side to support 4. X. Ge, S. Tu, G. Mao, C.-X. Wang, and T. Han, “5G ultra-dense cellu-
FEC and DSP processing. Such accelerators will also be used lar networks,” IEEE Wireless Commun. 23, 72–79 (2016).
for conducting various processing functions in the other serv- 5. A. Benjebbour, K. Saito, Y. Saito, and Y. Kishiyama, “5G radio
ers, such as high-speed wireless signal processing [44–46] and access technology,” NTT DOCOMO Tech. J. 1, 16–28 (2016).
6. T. Pfeiffer, “Next generation mobile fronthaul and midhaul architec-
lower-layer PON functions [40–42]. The DU functions will tures,” J. Opt. Commun. Netw. 7, B38–B45 (2015).
be softwarized in addition to the CU functions for the MFH 7. J.-I. Kani, “Solutions for future mobile fronthaul and access-network
application. convergence,” J. Lightwave Technol. 35, 527–534 (2017).

Authorized licensed use limited to: University of Melbourne. Downloaded on June 23,2020 at 10:31:57 UTC from IEEE Xplore. Restrictions apply.
 Research Article Vol. 12, No. 9 / September 2020 / Journal of Optical Communications and Networking D55

8. J. S. Wey and J. Zhang, “Passive optical networks for 5G transport: 30. L. Peterson, A. Al-Shabibi, T. Anshutz, S. Baker, A. Bavier, S. Das,
technology and standards,” J. Lightwave Technol. 37, 2830–2837 J. Hart, G. Palukar, and W. Snow, “Central office re-architected as a
(2019). data center,” IEEE Commun. Mag. 54(10), 96–101 (2016).
9. Y. Shiroishi, K. Uchiyama, and N. Suzuki, “Society 5.0: for human 31. M. Tadokoro, K. Nishimoto, T. Mochida, T. Tanaka, T. Yamada, A.
security and well-being,” Computer 51, 91–95 (2018). Takeda, and T. Inoue, “Design of softwarized EPON OLT and its
10. https://www.fsan.org/roadmap/. transmission jitter suppression techniques over MPCP,” in Optical
11. “Gigabit-Capable Passive Optical Networks (GPON),” ITU-T Fiber Communication Conference (OFC), San Diego, California,
Recommendation G.984 series. USA, 2017, paper W2A.28.
12. “10-Gigabit-Capable Symmetric Passive Optical Networks 32. K. Nishimoto, Y. Koyasako, T. Yamada, J.-I. Kani, and A. Otaka,
(XGS-PON),” ITU-T Recommendation G.9807 series. “Software module-based 10G optical line terminal for enhancing
13. “40-Gigabit-Capable Passive Optical Networks (NG-PON2),” ITU-T flexibility of access networks,” in 4th IEEE Conference on Network
Recommendation G.989 series. Softwarization and Workshops (NetSoft), June 2018, pp. 500–505.
14. M. Bennis, M. Debbah, and H. V. Poor, “Ultrareliable and low- 33. “Reference Design—SDN Enabled Broadband Access (SEBA)
latency wireless communication: tail, risk, and scale,” Proc. IEEE Version 1.0,” ONF TS-100 (2019).
106, 1834–1853 (2018). 34. Open Networking Foundation (ONF), “Open and disaggre-
15. A. Otaka, “Power saving ad-hoc report,” presented at the IEEE gated transport network (ODTN),” https://www.opennetworking.
802.3 Interim Meeting, Seoul, South Korea, 15–18 September org/odtn/.
2008. 35. Telecom Infra Project (TIP), “Open optical and packet transport
16. A. G. Boso, D. Rusca, C. Vulliez, C. Autebert, M. Caloz, M. (OOPT),” https://telecominfraproject.com/oopt/.
Perrenoud, G. Gras, F. Bussières, M.-J. Li, D. Nolan, A. Martin, 36. V. Lopez, W. Ishida, A. Mayoral, T. Tanaka, O. G. de Dios, and J. P.
and H. Zbinden, “Secure quantum key distribution over 421 km of Fernandez-Palacios, “Enabling fully programmable transponder
optical fiber,” Phys. Rev. Lett. 121, 190502 (2018). white boxes,” J. Opt. Commun. Netw. 12, A214–A223 (2020).
17. B. Fröhlich, J. Dynes, M. Lucamarini, A. W. Sharpe, S. W.-B. Tam, 37. F. Giannone, K. Kondepu, H. Gupta, F. Civerchia, P. Castoldi, A.
Z. Yuan, and A. J. Shields, “Quantum secured gigabit optical access Franklin, and L. Valcarenghi, “Impact of virtualisation technolo-
networks,” Sci. Rep. 5, 18121 (2016). gies on virtualised RAN midhaul latency budget: a quantitative
18. M. Gunkel, M. Schneiders, S. Vorbeck, W. Weiershausen, R. Leppla, experimental evaluation,” IEEE Commun. Lett. 23, 604–607 (2019).
F. Rumpf, R. Herber, V. Furst, and M. Rodenfels, “Aggregation net- 38. K. Iwatsuki, J.-I. Kani, H. Suzuki, and M. Fujiwara, “Access and
works: cost comparison of WDM ring vs. double star topology,” in metro networks based on WDM technologies,” J. Lightwave
Optical Network Design and Modeling, July 2008. Technol. 22, 2623–2630 (2004).
19. A. T. Kawakita, K. Hara, Y. Kimura, K. Horikawa, H. Furukawa, 39. A. Umesh, T. Yajima, T. Uchino, and S. Okuyama, “Overview of
Y. Suzuki, and S. Ikeda, “Design for long-reach coexisting O-RAN fronthaul specifications,” NTT DOCOMO Tech. J. 21, 46–59
PON in consideration of area characteristics with wavelength (2019).
selective asymmetrical splitters,” in 24th OptoElectronics and 40. T. Suzuki, S.-Y. Kim, J.-I. Kani, and J. Terada, “Demonstration of
Communications Conference (OECC) and 2019 International fully softwarized 10G-EPON PHY processing on a general-purpose
Conference on Photonics in Switching and Computing (PSC) server for flexible access systems,” J. Lightwave Technol. 38,
(2019). 777–783 (2020).
20. N. Amaya, G. S. Zervas, and D. Simeonidou, “Architecture on 41. T. Suzuki, S.-Y. Kim, J.-I. Kani, T. Hanawa, K.-I. Suzuki, and A.
demand for transparent optical networks,” in 13th International Otaka, “Demonstration of 10-Gbps real-time Reed–Solomon
Conference on Transparent Optical Networks (ICTON), Stockholm, decoding using GPU direct transfer and kernel scheduling for
Sweden, 2011, pp. 1–4. flexible access systems,” J. Lightwave Technol. 36, 1875–1881
21. M. Ruffini, M. Achouche, A. Arbelaez, R. Bonk, A. Di Giglio, N. J. (2018).
Doran, M. Furdek, R. Jensen, J. Montalvo, N. Parsons, T. Pfeiffer, 42. T. Suzuki, S. Kim, J. Kani, K. I. Suzuki, A. Otaka, and T. Hanawa,
L. Quesada, C. Raack, H. Rohde, M. Schiano, G. Talli, P. Townsend, “Parallelization of cipher algorithm on CPU/GPU for real-time
R. Wessaly, L. Wosinska, X. Yin, and D. B. Payne, “Access and software-defined access network,” in Asia-Pacific Signal and
metro network convergence for flexible end-to-end network design Information Processing Association Annual Summit and Conference
[Invited],” J. Opt. Commun. Netw. 9, 524–535 (2017). (APSIPA) (2015), pp. 484–487.
22. NTT Press Release, “Introducing the new FASA concept for 43. S. Kim, T. Suzuki, J. Kani, A. Otaka, and T. Hanawa, “Coherent
future access systems,” 2016, https://www.ntt.co.jp/news2016/ receiver DSP implemented on a general-purpose server for full
1602e/160208a.html. software-defined optical access,” in Optical Fiber Communication
23. M. Yoshino, H. Ujikawa, T. Harada, R. Yasunaga, T. Mochida, K. Conference, OSA Technical Digest (online) (Optical Society of
Asaka, J. Kani, and K. Suzuki, “New access system architecture America, 2018), paper Tu3L.2.
(FASA)—enabling provision of services flexibly and promptly,” NTT 44. C.-L. I, J. Huang, R. Duan, C. Cui, J. Jiang, and L. Li, “Recent
Tech. Rev. 14, 1–6 (2016). progress on C-RAN centralization and cloudification,” IEEE Access
24. J. Kani, M. Yoshino, K. Asaka, H. Ujikawa, T. Yamada, K. Nishimoto, 2, 1030–1039 (2014).
K. Suzuki, and A. Otaka, “Flexible access system architecture 45. R. Doost-Mohammady, O. Bejarano, L. Zhong, J. R. Cavallaro, E.
(FASA) to support diverse requirements and agile service creation,” Knightly, Z. M. Mao, W. W. Li, X. Chen, and A. Sabharwal, “RENEW:
J. Lightwave Technol. 36, 1510–1515 (2018). programmable and observable massive MIMO networks,” in 52nd
25. K. Asaka, H. Ujikawa, H. Uzawa, H. Nakamura, J. Kani, A. Otaka, Asilomar Conference on Signals, Systems, and Computers, Pacific
and J. Terada, “Disaggregation of time-critical applications in Grove, California, USA, 2018, pp. 1654–1658.
flexible access system architecture,” J. Opt. Commun. Netw. 11, 46. K. Li, M. Wu, G. Wang, and J. R. Cavallaro, “A high performance
A33–A39 (2019). GPU-based software-defined basestation,” in 48th Asilomar
26. “Functional model for PON abstraction interface,” Broadband Conference on Signals, Systems and Computers, Pacific Grove,
Forum Technical Report 402 (2018). California, USA, 2014, pp. 2060–2064.
27. “PON abstraction interface for time-critical applications,” 47. S.-Y. Kim, T. Suzuki, J.-I. Kani, and J. Terada, “Exploiting general
Broadband Forum Technical Report 403 (2018). purpose hardware in optical access systems,” J. Opt. Commun.
28. H. Uzawa, H. Nomura, T. Shimada, D. Hisano, K. Miyamoto, Y. Netw. 12, A182–A189 (2020).
Nakayama, K. Takahashi, J. Terada, and A. Otaka, “Practical
mobile-DBA scheme considering data arrival period for 5G mobile Jun-ichi Kani (M’98-SM’18) received M.E. and Ph.D. degrees in applied
fronthaul with TDM-PON,” in IEEE/OSA European Conference on physics from Waseda University, Tokyo, Japan, in 1996 and 2005, respec-
Optical Communication (ECOC), September 2017, paper M1B. 2. tively. In 1996, he joined NTT Optical Network Systems Laboratories, where
29. N. Afraz and M. Ruffini, “A sharing platform for multi-tenant PONs,” he was engaged in research on optical multiplexing and transmission tech-
J. Lightwave Technol. 36, 5413–5423 (2018). nologies. Since 2003, he has been with NTT Access Network Service Systems

Authorized licensed use limited to: University of Melbourne. Downloaded on June 23,2020 at 10:31:57 UTC from IEEE Xplore. Restrictions apply.
 D56 Vol. 12, No. 9 / September 2020 / Journal of Optical Communications and Networking Research Article

Laboratories, where he has been engaged in R&D and standardization of Sang-Yuep Kim received a Ph.D. degree in electronics engineering from
optical communication systems for access and metro applications, and he Kwangwoon University, Seoul, South Korea, in 2004. From 2004 to 2007,
currently heads the Access Systems Technology Group. Dr. Kani has been he was with the University of Tokyo, Japan, under a Postdoctoral Foreign
serving as Associate Rapporteur of Question 2 (optical systems for fiber access Researcher Fellowship. In 2008, he joined NTT Access Network Service
networks) in ITU-T Study Group 15 since 2009 and as Chair of the Full Systems Laboratories, NTT Corporation, Chiba, Japan. He is currently
Service Access Network initiative (FSAN) since 2015. He served as a Program researching DSP technologies for future optical access systems.
Co-Chair of the Optical Fiber Communications Conference (OFC) 2019.
Kota Asaka received B.S. and M.S. degrees in electrical engineering from
Jun Terada received a B.E. degree in science and engineering and a M.E. Waseda University, in 1996 and 1999, respectively, and a Ph.D. in physics
degree in computer science from Keio University, Yokohama, Japan, in 1993 from Kitasato University in 2008. He joined NTT Photonics Laboratories
and 1995, respectively. In 1995, he joined NTT LSI Laboratories, where he in 1999 and engaged in research on several photonics integrated circuits.
was involved in the research and development of low-voltage analog circuits, From 2009 to 2012, he worked on developing low-cost and small optical
especially A/D and D/A converters. In 1999, he was involved in developing subassemblies for access networks and served as the WG secretary of IEC
small and low-power wireless systems for sensor networks. In 2006, he was SC86C/WG4 for standardization of fiber optic active components. Since
involved in researching high-speed front-end circuits for optical transceivers. 2012, he has been with NTT Access Network Service System Laboratories,
He is currently a Senior Research Engineer and a Supervisor with NTT where he is engaged in R&D of next-generation optical access networks
Access Network Service Systems Laboratories, Yokosuka, Japan, where he is such as NG-PON2, OFDM-PON, and SDN/NFV in access. He has been
responsible for research and development management for optical and wire- participating in ITU-T/FSAN since 2012 as well as BBF since 2016.
less converged access networks. He is a member of the Institute of Electronics,
Information and Communication Engineers (IEICE) of Japan. He has served Takashi Yamada (M’17) received B.E., M.E., and Ph.D. degrees in electrical
as a member of Technical Program Committee of the Symposium on VLSI engineering from Hokkaido University, Sapporo, Japan, in 1997, 1999,
Circuits. He has been serving the Asian Solid-State Circuits Conference and 2003, respectively. In 2003, he joined NTT Access Network Systems
(A-SSCC) since 2012. Laboratories, Chiba, Japan, where he was involved with the research of
Tomoya Hatano received B.E., M.E., and Ph.D. degrees from Keio energy-efficient optical access network systems. From 2013 to 2016, he
University, Tokyo, Japan, in 2002, 2004, and 2007, respectively. In 2004, was with the Research and Development Planning Department, NTT
he joined NTT Access Service Systems Laboratories. He is engaged in the Information Network Laboratory Group, where he was involved in pub-
research and development of optical communication systems and network lic relations and information strategy. His current research interests are
virtualization. virtualization of optical access networks.

Authorized licensed use limited to: University of Melbourne. Downloaded on June 23,2020 at 10:31:57 UTC from IEEE Xplore. Restrictions apply.