future proof optical network infrastructure for 5g...

Future Proof Optical NetworkInfrastructure for 5G Transport

Paola Iovanna, Fabio Cavaliere, Francesco Testa, Stefano Stracca, Giulio Bottari, FilippoPonzini, Alberto Bianchi, and Roberto Sabella

Abstract—The telecommunication community hasreached a broad consensus that the current radioaccess network (RAN) and underlying transport will notbe able to scale up to the traffic volume and qualityexpected in 5G. Thus, it is necessary to remove all thetechnological bottlenecks and operational rigidities to en-sure a painless migration from the existing radio scenarioto the 5G one. This article presents a transport architectureable to serve as a backhaul and fronthaul to convey radiotraffic on the same optical infrastructure. The cornerstonesof the solution are a novel photonic technology used toprovide optical connectivity, complemented by a dedicatedagnostic framing; a deterministic switching module; and aflexible control paradigmbased on a layered scheme and onthe slicing concept to facilitate optimal interactions oftransport and radio resources while preserving a well-demarcatedmutual independence. Simulations and experi-ments are presented to demonstrate the aforementionedfeatures.

Index Terms—Fiber optics; Photonic integrated circuits;Radio access networks; Switches.

I. INTRODUCTION

T he rising penetration of smart connected devices,promising compelling services anywhere and anytime,

is having a huge impact on the mobile broadband infra-structure and on the ability to provide a quality experienceto the final users, whether humans or connected things.

The new 5G radio access ecosystem, expected by 2020[1], will have to sustain an average traffic increase ofone order of magnitude and peak rates up to three ordersof magnitude higher than the current ones [2]. Many time-sensitive services will also demand extremely low trans-mission latency.

Radio technologies will evolve by allocating new bands(beyond 10 GHz), by leveraging higher-order multiple-input multiple-output (MIMO), carrier aggregation, andbeamforming techniques. In parallel, the transport net-work for backhaul (BH) and fronthaul (FH) applicationswill demand support of higher capacities to increase thenumber of transport clients, to support a wider rangeof performance requirements, and to provide increased

flexibility. This must be achieved in a cost-effective and sus-tainable manner.

Based on the radio architectures, it is possible to definemany deployment scenarios, ranging from the fully central-ized one, i.e., the cloud radio access network (CRAN), to theconventional scenario, in which all functions are replicatedat each radio site with monolithic radio base stations(RBSs). Moreover, as envisioned in [3], novel radio-splittingmodels are under development to meet 5G high-bandwidthdemands, leveraging different distributions of radio func-tions between radio unit nodes and centralized processingnodes. Realistic scenarios will see a mix of all mentionedradio architectures, with a combination of traffic types tobe transported among the radio devices.

In this paper, we define an Xhaul concept which canunify and enhance the traditional BH and FH segmentsby enabling a flexible deployment and the reconfigurationof the network elements and networking functions. Xhaulsets connectivity services through the implementation of acontrol plane, which provides a unified network model,supporting different underlying data planes and protocolsplit schemes.

Several implementations are possible for an Xhaul net-work. Our Xhaul solution is based on DWDM fiber ringsconnecting a central hub to remote nodes where radioand wireline clients are connected. Xhaul is enabled by anew type of photonic device having a deep integration ofoptical functions on a single chip and presenting a cost,footprint, and power consumption adequate for this targetnetwork segment.

This paper is organized as follow: Section II defines theXhaul network concept and illustrates the different radiotransportneeds. Section IIIdescribes theXhaularchitectureand itsbuildingblocks, includingoptical transmission, fram-ing, agnostic switching, and control. Section IV provides in-sight into the enabling photonic technologies. Section Vreports analytic performance evaluations. Section VI illus-trates a first practical demonstration of Xhaul in an exper-imental test bed built at Ericsson Research Labs.Concluding remarks are made in Section VII.

II. XHAUL NETWORK CONCEPT

In traditional radio access networks (RANs), radio units(RUs), performing radio functionalities, and digital unitshttp://dx.doi.org/10.1364/JOCN.8.000B80

Manuscript received June 1, 2016; revised October 14, 2016; acceptedOctober 20, 2016; published November 17, 2016 (Doc. ID 267398).

The authors are with Ericsson, Pisa, Italy ([email protected]).

B80 J. OPT. COMMUN. NETW./VOL. 8, NO. 12/DECEMBER 2016 Iovanna et al.

1943-0620/16/120B80-13 Journal © 2016 Optical Society of America

(DUs), performing baseband processing, are integrated in aRBS, which is typically backhauled by an Ethernet signal.Ethernet clients are transported across the BH segment. Inthe CRAN architecture, remote RUs (RRUs) and DUs aredivided into two separate nodes and connected across aFH network. The RRUs are located at the antenna side.TheDUs are separated fromRRUs and possibly aggregatedin a DU pool. Centralizing DUs enables improvedcoordination of radio capabilities across a set of RRUs,faster RAN deployment, and cost savings. CommonPublic Radio Interface (CPRI) [4] is the radio interface pro-tocol widely used for communication between RRUs andDUs.Differentmobile radio technologies canbe transportedby CPRI with different constraints on transport. For exam-ple, the tolerated delay between the RRU and DU (one way)is ∼200 μs for GSM/WCDMA traffic and ∼100 μs for LTEtraffic. Other constraints are imposed on latency imbalance(asymmetry downlink/uplink) and jitter.

It has been extensively accepted that in 5G, a new func-tional split [3] between DU and RRU will be required tomitigate the 5G bandwidth explosion. Some physical-layerradio functions, e.g., resource mapping, will migrate to theRRUs.AnRRUwill be in charge of generating theproper I/Qsignals and radio carriers for the antennas. A new radiopacket interface between the DU and RRUwill enable a ra-dio bandwidth of one order of magnitude higher, without adramatic increase in the rate between DUs and RRUs com-paredwith the current CPRI.Moving toward a radio packetinterface does not mean that conventional Ethernetswitches can be used to connect DUs and RRUs becausethe latency requirements will still be there and they willbe carefullymanaged. This new radio packet low-latency in-terface (RPLI) is still under discussion and with differentnames from different vendors.

In Xhaul, both the BH, CPRI FH, and 5G FH based onRPLI can benefit from a transport layer, based on DWDMoptical technologies, which ensures low propagation de-lays, high throughputs, and low power consumption, whilebeing an economical choice for exploiting fiber resources.However, mixing all these interfaces cannot not be triviallyconsidered as a problem of “tunneling” client traffic overdummy optical pipes. This is even truer if we consider apractical scenario in which mobile traffic can change overtime and space in distribution and load.

Another aspect a cost-effective network should consideris that not all requirements are expected to be critical atthe same time. For example, NGMN [1] has definedtwenty-five use cases grouped in eight families.

The Xhaul infrastructure shall be open to providing dif-ferent connectivity services in response to different radioneeds. This concept is captured in Fig. 1.

On the left side of the picture, some of the most promi-nent use cases expected in the 5G scenario are listed. Onthe right side of the picture, a reference sketch of the Xhaulnetwork is illustrated. It is based on “remote nodes,” whereantennas are connected via wired or microwave links. Inaddition, wireline connections are connected to remotenodes. Xhaul is also based on “hub nodes,” which collecttraffic from the several remote nodes connected to the sameXhaul areas. The network in between is an optical networkbased on WDM technology.

In 5G systems, radio and transport networks will beabstracted into infrastructure slices. The slice is a connec-tivity service defined by a number of customizable soft-ware-defined functions that govern duration, capacity,speed, latency, robustness, security, and availability ina geographical coverage area. Through network slices,

Fig. 1. XHaul reference scheme.

Iovanna et al. VOL. 8, NO. 12/DECEMBER 2016/J. OPT. COMMUN. NETW. B81

operators can provide, on demand, infrastructure as a ser-vice in a programmable way.

Slices are agnostic and independent of the underlyingphysical resources and the used technologies. Infra-structure resources and functions can be dedicated to aslice or shared among multiple slices.

In this paper, a layered model is proposed where theradio and transport have a client-server relationship. Itallows both a tight interworking with the radio and fixedaccess, while keeping the clear separation of responsibility,troubleshooting, and services between the radio andtransport.

According to the layered model, it is possible to defineclasses of service at radio and transport layers identifyinglayer-specific requirements. The transport has someautonomy from the radio to organize its internal resourcesin services and expose them in corresponding slices. Thetransport can dynamically rearrange traffic and providea sharing of physical resources for different slices. It canalso optimize the physical resources without complicatingthe radio’s tasks.

In Fig. 1, some relevant 5G end-user services are re-ported, and the most critical requirement to consider foreach of them is shown. Broadband access in a dense areaand in a crowd, for example, will have the traffic densityper area as a critical parameter, possibly time bounded.Ultra-low latency applications, like the ones in critical ma-chine-type communications (MTCs), will impose stringentlimits on the end-to-end transmission delay without chal-lenging bandwidth needs. Legacy mobile and wireline willalso constitute use cases for Xhaul.

III. ARCHITECTURE AND BUILDING BLOCKS

Xhaul has a logical hub and spoke architecture, whichenables point-to-point (P2P) logical connections betweenremote nodes and a hub node through fiber rings. This al-lows the reduction of the intermediate steps of processingthe signals, thus limiting the latency required for trans-port. Any different physical topology where DWDM canbe used to establish P2P connectivity through dedicatedwavelengths also applies to Xhaul, for example, linearchains of optical add/drop multiplexers (OADMs), point-to-multipoint distribution infrastructures based on wave-length-selected or wavelength-distributed passive opticalnetworks (PONs), and meshed networks realized throughreconfigurable OADMs. Of course, implementation details,such as protection methods, vary accordingly.

Figure 2 illustrates the topology. Many radio and wire-line clients are connected to remote nodes, which act as en-try points in the Xhaul network. This particular networkarrangement allows for a better exploitation of the concen-tration of the traffic and the statistical multiplexing, andallows a better traffic balance among DUs. Daily andweekly variations and “traffic tides” moving among ageographical area are open to significant optimizationopportunities at the hub covering said area. For example,a single hub can serve a residential district and a business

district, which typically have complementary traffictrends during the 24 hours of switching on/off radio equip-ment, with savings in power consumption and transportresources.

The optical transmission and switching technologieswith the agnostic framing, the deterministic switch, andthe overall Xhaul control are important elements ofXhaul. The following sections discuss these elements indetail.

A. Optical Transmission

The introduction of 5G mobile systems is pushing highercapacities (100+ Gbit/s) in fiber transport networks.

As for the distance requirement, coherent interfaces canachieve thousands of kilometers, but the Xhaul segmentspans a maximum of a few tens of kilometers, so direct de-tection interfaces would bemore appropriate than coherentinterfaces and would come with lower cost and power con-sumption.

However, current 100 Gbit∕s direct detection transceiv-ers, e.g., those based on PAM4, cannot achieve a sufficientlink budget in the presence of a realistic number of opticaladd/drop nodes (OADMs) in the network. For example, a20 km ring with one hub node and 8 OADM sites wouldrequire about 18 dB of link budget considering the follow-ing loss values: 0.6 dB for channels passing through eachOADM, 3 dB for an added or a dropped channel, 5.5 dB for awavelength multiplexer/de-multiplexer at the hub, and a0.25 dB∕km fiber attenuation coefficient.

Fiber chromatic dispersion is another issue with100 Gbit∕s DWDM direct detection: in an Xhaul network,it is desirable to avoid dispersion compensation in line tosave on costs and not to introduce additional loss. But

Fig. 2. Xhaul topology.


direct detection interfaces cannot exploit electrical equali-zation like coherent interfaces.

A practical solution to increase the link budget andchromatic dispersion tolerance is to transmit 50 Gbit∕swavelengths instead of 100 Gbit∕s ones. While this halvesthe aggregate capacity, the total capacity value (a few ter-abits/s) is still sufficient for 5G transport.

In summary, an ideal modulation format for Xhaulshould satisfy the following features:

• direct detection, to decrease cost and power consumption,avoiding local oscillator and digital signal processing atthe receiver;

• no dispersion compensation up to 20 km;• receiver sensitivity sufficient to achieve a link budget ofabout 18 dB, with or without optical amplification.

Figure 3 comparesdirect detectionmodulation formats at50 Gbit∕s as regards the chromatic dispersion tolerance. Itis obtained by a numerical simulation, considering 0 dBmofchannel power transmitted in a standard single-mode fiber(SMF) with 0.22 dB∕km and −20 ps2∕km−1 fiber attenua-tion and chromatic dispersion coefficients, respectively.The fiber is modeled as a linear system, since non-linear ef-fects are negligible, due to the low transmitted power andshort distances in the Xhaul network segment. The opticalpenalty on the y-axis is defined as the receiver sensitivityrefers to the on–off keying (OOK) sensitivity at 0 km and10−3 BER, a value that can be corrected by common hard-decision forward error correction (FEC) codes.

Differential binaryphase-shiftkeying (DBPSK)anddiffer-ential quadrature phase-shift keying (DQPSK) are directlydetected by an interferometric receiver with balancedphotodiodes.

The combined amplitude phase shift (CAPS) modulationformats family is described in [5]. CAPS-1 is indicated

simply as CAPS. At the transmitter, CAPS-N requires a2N states digital encoder followed by an I/Q modulator,with the exception of CAPS-1, which uses a simpleMach–Zehnder modulator (MZM). At the receiver, allCAPS formats are directly detected as OOK, with no needfor digital signal processing.

Figure 3 shows that CAPS-3 outperforms the othermodulation formats for link distances higher than 8 km.

Discrete multi-tone (DMT) is an additional modulationformat using direct detection. Its performance is very im-plementation specific, depending on the design variable(number of sub-carriers, cyclic prefix, pilot tones, and soon). This is why DMT has not been included in Fig. 3.

Figure 4 reports the experimental results (dots) versusthe simulations (curves), comparing OOK, PAM-4, andCAPS-3 at 25 Gbit∕s (it was not possible to perform50 Gbit∕s experiments with the available instrumenta-tion). The experiments confirm CAPS-3 robustness to chro-matic dispersion. To obtain the results shown in Fig. 4, aMZM was used to generate OOK and PAM-4 signals,and a nested MZ I/Q modulator was used to generatethe CAPS-3 signal. The electrical signals at modulator’s in-put are obtained by means of a DAC with 13 GHz 3 dBbandwidth, operating at 64 GSample/s. The digital signal,at the input of the DAC, is generated through off-linedigital signal processing by encoding a periodic pseudo-random binary sequence (BRBS) of length 211 − 1, accord-ing to the considered modulation format.

B. Framing

Ethernet, CPRI, and RPLI clients can be separatelymapped over dedicated optical channels in the DWDMcomb. A set of optical channels is used to transportEthernet traffic originated by RBSs. A second set is usedto transport CPRI traffic between RRUs and DUs.Finally, a third set is used to transport a mix of Ethernetand CPRI traffic, wrapped in a novel framing structure.

Fig. 3. Direct detection modulation formats at 50 Gbit∕s.Fig. 4. Modulation formats comparison at 25 Gbit∕s: experi-ments (dots) versus simulations.


For illustration purposes, the following discussion will referto CPRI, but these considerations can be extended to RPLIor any kind of time-sensitive client signal.

The use of dedicated optical channels ensures easiermanagement and sharp client-type segregation, facilitat-ing shared/leased network scenarios. The use of sharedλs enables better bandwidth utilization and fewer opticaltransceivers.

Sharing is enabled by a simple and novel framing,providing FEC and operation and maintenance (OAM)for both CPRI and Ethernet, without the complexity andthe synchronization performance degradation introducedby standard protocols like OTN.

OTN [6] is an optical transport standard developed by theITU-T. It is also known as ITU G.709 and a “digital wrap-per.” In OTN, the ITU defines the payload encapsulation,OAM overhead, FEC, and multiplexing hierarchy. The re-sult is a transport standard that includes the benefits ofSDH (such as resiliency and manageability) but with im-provements for transporting data payloads. OTN standardsinclude a standard multiplexing hierarchy, defining exactlyhow the lower-rate signals map into the higher-rate pay-loads. This allows any OTN switch and any WDM platformto electronically groom and switch lower-rate serviceswithin 10, 40, or 100 Gbit∕s wavelengths.

CPRI mapping over OTN was recently included in theITU-T supplement [7]. The main challenge of transporting

CPRI, and any time-sensitive FH interface, over an OTN isto limit the jitter and wander introduced while mappingand de-mapping CPRI to OTN. An analysis on root-mean-square of the frequency offset (“jitter”) andmean timeinterval error was performed by ITU-T. ITU-T simulationsshow that to meet 2 ppb, as specified by the CPRI standard,a stringent de-synchronizer bandwidth is required, muchlower than the 300 Hz normally used in OTN. This wouldlead either to designing RRUs capable of tolerating higherinput noise or to redesigning the OTN equipment, includingthe very stable oscillators and sharp filters. Another issue,still under discussion, is the compensation of a possible im-balance of latency times up- and downstream. Today, thepractical use of CPRI over OTN is limited to the case of syn-chronous mapping of CPRI signals belonging to a singlesynchronization domain.

To overcome the issues that arise when mappingtime-sensitive FH interfaces over OTN, an alternative mul-tiplexing methodology is presented. Furthermore, we de-scribe how the methodology can be extended to Ethernetclient signals. Though the discussion focus is on opticalchannels, the methodology is agnostic to the propagationmedium and can be applied to wireless signals as well.For a cost-effective implementation, the proposed methodmakes the realistic assumption that FH signals are trans-ported over short-reach links (a few tens of kilometers), sothere is no need for advanced features such as a complexmultiplexing hierarchy and protection mechanisms.

Fig. 5. (a) Transmitter, (b) receiver, (c) framer, and (d) de-framer.


The proposed framing procedure is synchronous to theCPRI client to avoid any degradation of the synchroniza-tion accuracy.

Optional FEC is provided, based on the RS(255,239), butthe number of interleaved codecs can optionally be reducedto limit the additional latency. The first experiments haveshown a FEC latency lower than 4 μs with 9 interleavedcodes, with no appreciable degradation of the FEC gainwith respect to the OTN case.

The bandwidth efficiency compared to the CPRI is im-proved by replacing spectrally inefficient line codes, writtenas8B/10B,with codeswithmore efficient scramblers, e.g., us-ing as the generating polynomial 1� x� x3 � x12 � x16,leaving space for the FEC overhead and in-band signaling.

The transmitter scheme is outlined in Fig. 5(a), takingCPRI Option-7 (9.8304 Gbit∕s) as an example.

The clock signal is extracted from the received CPRI sig-nal and distributed to all the transmitter blocks. After se-rial-to-parallel conversion, the 8B/10B redundant bits areremoved, with the exception of the control bits, which iden-tify the K-codes. Then, a framer block [Fig. 5(c)] applies theFEC to the data, control, and any other OAM bit. If sodesired, different FEC codes or interleaving could beused for the data and OAM bits. The framer also includesa block for the compensation of the difference between theupstream and downstream delays, as required by the CPRIclient. This can be done by means of a buffer. Finally, theassembled frame is scrambled.

The inverse operation is performed at the receiver[Fig. 5(b)] with a de-framer [Fig. 5(d)].

An example of a frame structure is illustrated in Fig. 6.

The frame is 2390 octets long, arranged in 239 rows by10 columns. Columns 1 to 9 are for the payload, while col-umn 0 is reserved for overhead: frame alignment word (dis-tributed in rows 0 to 5), FEC codes for the payload (rows 10to 153), bit-interleaved parity (rows 6 to 8), generic commu-nication channel (row 9), OAM channel (rows 154 to 222),and FEC code for protected overhead (rows 223 to 238).

When the numbers and bit rates of the FH client signalsare not sufficient to “fill” a wavelength up to the maximumsupported bit rate, the unutilized bits can be used to trans-port another type of client. In particular, Ethernet is of in-terest so the same DWDM channel can serve both FH andBH connections.

The proposed frame cannot provide the entire set of fea-tures ensured by OTN, but it is intended to target a simplerscenario, where P2P logical connections are the most fre-quent case.

The basic concept is very simple and consists of allocat-ing two separate portions of the same frame to Ethernetand CPRI (or CPRI-like) clients. The size and position ofthe portions within the frame are known, as well as theframe size, making it very easy to separate Ethernet pack-ets from CPRI frames.

Portion size and position can be programmable, e.g., viacontrol interfaces, depending on the network configuration

and planned traffic load. The frame is synchronous to theCPRI clock in order to minimize the impact on delay- andjitter-sensitive CPRI frames while ingress Ethernet pack-ets are buffered to absorb differences in the clock value andaccuracy. A possible way to map Ethernet frames in thededicated timeslots of the frame is the use of the genericframing procedure (GFP).

C. Switching

The reference scheme of the Xhaul switch is illustratedin Fig. 7. Ethernet and RPLI clients, demanding low andhigh latencies, respectively, are first sent to a packet sched-uler, complemented by a set of buffers for time alignment.This block can be based on conventional Ethernet switchtechnology, assigning different priorities to different la-tency classes. Several implementations are possible, butdescribing them would be outside of the scope of the paper.Another block is in charge of multiplexing CPRI clients,according to the multiplexing protocol described in theCPRI specification [4]. The time-slotted switch maps allEthernet, CPRI, and RPLI clients according to the agnosticframing described in Subsection III.B. Framed traffic is

Fig. 6. Example of frame structure (239 rows by 10 columns).


then wrapped in optical channels and sent toward the hubvia the optical transport network. This can be performed byROADMs, based on standard technologies or new technol-ogies (described in Section IV).

The proposed switch architecture allows us to solve theissues that would arise with conventional packet or circuitswitches. Although in principle, packet-based interfacesare appealing for Xhaul applications because they promiseto manage both Ethernet and CPRI in the same off-the-shelf packet switching engine, packet delay variation(PDV) and delay control are difficult to manage for a longchain of switches or as soon as the ingress traffic ap-proaches the switch overload condition. IEEE 802.1Time-Sensitive Networking (TSN) [8] is studying this prob-lem, but there are issues with latency (in the case of multi-ple switching hops), with deterministic delay to be ensureddownstream and upstream, with the PDV, and with syn-chronization distribution.

On the other hand, circuit switches can better deal withjitter and delay control but lose any statistical multiplexinggain, which is a fundamental feature for dealing with 5Gtraffic loads. Statistical multiplexing is indeed highly ben-eficial in 5G and is characterized by a bursty distribution oftraffic where peaks can be as high as 10 Gbit∕s per sectorand average as low as 200 kbit∕s. Without statistical multi-plexing, a network serving 100 sectors would require1 Tbit∕s.

D. Control

Xhaul is a new transport network segment with specificrequirements, compared with the traditional accessand metro, where the level of dynamicity is limited andover-provisioning is typically applied. As explained inSection II, the control of the Xhaul network benefits froma layered model (slicing), through which the Xhaul controlhandles the transport resources by dynamically mappingthe traffic, organizing the slicing view for the RAN. The

layeredmodel also guarantees clear demarcation points be-tween the Xhaul network and its client RAN.

Figure 8 illustrates a schematic view of the control func-tions, showing a radio controller managing radio equip-ment (i.e., RRUs, DUs, RBSs) and an Xhaul controllerfor the transport network. In this model, the radio layeris extended with a new radio logic (RL) block, which re-ceives requests to set 5G services and translates theminto requests for the radio and to Xhaul controllers, as de-scribed in the layered slicing model, and performs theradio-transport orchestrating function.

For example, the RL may receive a traffic matrix withend-user mobile connection requests and service parame-ters. On this basis, the RL determines which RRU orRBS shall be connected to the hub, defining the connectiv-ity and its parameters (e.g., bandwidth, service type, resil-iency level, policy constraints).

The RL determines if coordination requirements for dif-ferent connected RRUs to use the same path should be im-posed. For example, if some cells require very tight phasealignment, as in MIMO, then the relevant FH flows shallbe transported together, minimizing the phase misalign-ments.RLalsodetermineswhichhigh-bandwidth traffic de-mand needs to be distributed on multiple optical channelsand which are the aggregation policy constraints.

The Xhaul controller receives the transport demands,grouped in trafficmatrixes, and determines the best possiblepath to serve each demand at the given time or in a plannedtime interval. For example, the Xhaul controller defines themapping of data traffic, as described in Subsection III.B, theaggregation in optical wavelengths, and the configuration ofthe switch described in Subsection III.C.

The Xhaul controller can split and recombine correlatedradio requests over several optical channels (multi-wavelength transport). For example, even if it is preferredthat coordinated radio traffic is conveyed on the same

Fig. 7. Switch reference scheme.

Fig. 8. Control architecture.


wavelength, different wavelengths can be considered if theradio constraints, e.g., latency, are satisfied.

As another example, support of time-sensitive traffic canrequire the Xhaul controller to re-route non-time-sensitivetraffic on alternative paths.

Finally, when the transport layer cannot serve trafficdemands with the desired quality of service, the Xhaulcontroller asks the RL to negotiate with the radio layerfor a relaxation of the radio performances, e.g., the co-ordination level.

There are two main advantages with this approach: (i) itkeeps the radio and transport layers separated, while pro-viding tight cooperation, and (ii) operator-specific policiescan be easily managed by introducing “administrative”parameters, enabling a multi-operator scenario.

The slicing scheme, implemented in the control architec-ture in Fig. 8, is based on software-defined networking(SDN)andnetwork function virtualization (NFV)principles[9,10], but with specific peculiarities. The south-bound in-terface is based on signaling protocols that are transmittedin-band, exploiting available bits in the agnostic framingde-scribed in Subsection III.B and implemented in the demon-strator illustrated in Section V. This approach avoids aparallel and dedicated network for control.

IV. ENABLING PHOTONIC TECHNOLOGIES

Xhaul calls for a new type of photonic devices with amuch lower cost, a higher level of device miniaturization,and a lower power consumption compared to the currentoptical modules designed for metro or long haul.

Two types of devices are crucial for 5G optical transport:optical switches and multi-wavelength transceivers, and

photonic integration is definitely the enabling technologyfor both types of devices. Silicon photonics is most suitablefor large-scale integrated switching devices due to its charac-teristics of easy integration with control electronics, highminiaturization, mass producibility, potential high yield,and low cost due to the use of a well-established CMOSproduction infrastructure. As far as integrated multi-wavelength transceivers are concerned, relevant technologi-caladvanceshavebeenmadewith twodifferent technologicalapproaches: InPmonolithic-integratedDWDM transceivers,and silicon photonics-based transceivers.

For optical switching, a new type of system-on-chip de-vice, referred to in the following as Mini-ROADM, is underdevelopment in our labs, and its architecture is depicted inFig. 9(a). A dual-polarization gain control (DPGC) is fol-lowed by a semiconductor optical amplifier (SOA) for theamplification of the DWDM comb. The DPGC adapts therandom polarization of the input signal to the device’s mainpropagation mode. The polarization controller was experi-mentally characterized for 50 random polarization states.All the arbitrary polarization states were compensatedwithin a 1 dB accuracy. Optical channels are added/dropped by silicon micro-ring resonators (MRR). The localchannel ports are coupled to the ring by DPGCs and 1 × 2switches, which allows us to revert the propagation direc-tion and implement ring protection functionalities.

In the prototype, the maximum drop loss difference fromthe first to the last channels is about 1.5 dB. The measuredchannel isolation from adjacent channels is larger than30 dB.

This device is especially designed for use in a double-ringnetwork, with one ring transmitting downlink signals fromthe hub to the remote nodes and the other ring transmit-ting upstream signals from the remote nodes to the hub.This device is simpler than the conventional ROADMs

Fig. 9. (a) Architecture of the Mini-ROADM and (b) details of the switch elements.


used in metro networks. It has two line ports and 12 localports to add and drop an equal number of 100 GHz spacedoptical channels. With the current silicon photonics tech-nology, it is possible to expand the number of ports up to24, but supporting a denser spacing (e.g., 50 GHz) wouldrequire technological advances to improve the wavelengthstability of the micro-rings.

The fundamental functions performed by the Mini-ROADM are DWDM channels multiplexing/de-multiplex-ing at the DU site and reconfigurable add/drop of selectedlocal channels at the RRU site. Additional auxiliary net-working functions have been added to the Mini-ROADM:channel direction switching and add-channel power regu-lation. The direction switching, which can be activatedautomatically, is implemented on each single wavelengthfor both add and drop functions, and it is used to re-establish the connection in the case of a fiber break. Sowhen some of the nodes distributed along the ring becomeisolated, the wavelength signals used to communicate withthose nodes change direction to overcome the fiber break.The fiber paths are then protected without the need to du-plicate the optical interfaces at each add/drop port. Opticalattenuators are implemented at the input of the add chan-nel ports to regulate the optical power depending on thefiber span length. This can be beneficial in keeping thereceiver sensitivity degradation due to crosstalk low,because it avoids a situation in which the drop switchelements’ high-power bypass signals are transmitted to-gether with the weak signal to be dropped.

For utilization of the Mini-ROADM in a real system, itmust be polarization insensitive and low loss, preferablylossless. Its design has been based onmicro-ring resonatorsas wavelength-selective switching elements [11–13] be-cause of its small size (few μm), low loss, and low powerconsumption (few mW).

The Mini-ROADM architecture presents two indepen-dent circuit structures, one used for adding local channelsto the network, and one used for dropping channels fromthe network. Each circuit structure has a polarization di-versity architecture in which the two TE and TM polariza-tion components of the signals are separated, at the deviceinput, by a polarization splitter and rotator block (PSR)(implemented here by a double-polarization grating cou-pler) and are guided inside two separate optical buses,to which a number of micro-ring resonators are coupled.When the micro-ring resonator switches are activated,the signals are dropped or added. At the output ports,the two polarization components are recombined beforecoupling to the optical fibers.

The arrays of 1 × 2 optical switches at the input/outputports are used to provide the desired bi-directionality toadd and drop functions for each wavelength. The primaryand backup paths are illustrated in Fig. 9(b).

For supporting a data rate transmission up to 28 Gbaud,an optical bandwidth of the micro-ring resonator switchelement>40 GHz is specified and a two-coupled ring archi-tecture has been implemented in order to provide the speci-fied bandwidth together with the desired adjacent channelisolation (>20 dB).

An array of optical attenuators is used to regulate thepower of the added channels injected into the network,and optical detectors are placed in a strategic place insidethe chip for monitoring.

Finally, III-V SOA dies are hybrid-integrated into thesilicon-on-insulator (SOI) substrate in order to compensatefor the device’s internal loss and fulfill the lossless charac-teristics as much as possible.

The Xhaul network also envisages compact, low-power,and low-cost multi-wavelength transceivers.

InP monolithic integration technology [14] has made sig-nificant progress, and it is a possible candidate for the im-plementation of integrated multichannel DWDM TX andRX. Being based on the same material used today for pro-ducing discrete components, it is useful for the realizationof optical devices in the near term, while the monolithicintegration of many channels in the same chip allows acost reduction compared with the traditional discrete-component approach.

An example of the implementation of a complete trans-ceiver with 10 DWDM channels, 100 GHz spaced and work-ing at 10 Gbit∕s in a CFP2 package, has been developedwith the architecture shown in Fig. 10. The evolution to-ward a 28 Gbit∕s data rate is under development.

Another technology that still needs some technologicaladvances but holds promise, in the long run, of an evenlower cost and higher level of miniaturization, is the hybridlaser technology based on the integration of III-V light gen-erator dies with silicon photonic chips, as shown in Fig. 11.With this approach, all the optical circuits, including laserexternal cavities, modulators, and photodetectors, aremonolithically integrated in an SOI substrate, while theoptical gain block is implemented in the III-V material.The laser wavelength is set by controlling the wave-length-selective elements inside the external cavity.Significant progress toward the realization of low-cost hy-brid lasers in silicon photonics has been shown in [15,16],where the III-V integration is realized at the wafer levelwithout the need for an active alignment of each single-light generator chip to the silicon substrate. These lastdevelopments indicate that a practical realization of low-cost, mass producible, high-speed multi-channel DWDM

Fig. 10. Realization of InP-based monolithically integratedDWDM transceivers: (a) multi-channel transmitter and (b) multi-channel receiver, courtesy of Effect Photonics.


transceivers for transmission up to 20 km in silicon pho-tonic platforms is realistic.

V. PERFORMANCE EVALUATIONS

Calculations to evaluate the performances have ad-dressed the cases of mixed BH and RPLI-based FH. Wehave considered 5G radio access technologies (RATs) withthe following assumptions: 200 MHz bandwidth in air,125 μs time transmission interval (TTI), 256 QAM, andbeamforming (8 × 8 antennas). As a new functional split,it is assumed that the 5G radio unit is in charge of function-alities from the physical layer to resource element map-ping. With these hypotheses, the peak throughput forthe RPLI interface is estimated to be 34.1 Gbit∕s.

It is also assumed that the maximum tolerated end-to-end latency between radio units and baseband processingunits is of the order of one TTI, here assumed 125 μs. IntheXhaul network, considering the uplink direction, this la-tency budget is spent for buffering/scheduling/framing inthe remote node, for transmission in the fiber, and for de-framing in the hub node. Similar considerations hold forthe downlink direction.

For illustrative purposes, the case of a single 100 Gbit∕swavelength from the remote node to the hub node has beenconsidered. The switch at the remote node, illustrated inFig. 7, connects two 5G radio units, via RPLI, andEthernet clients for RBS backhauling.

The buffer of the switch enables handling the unfortu-nate case of simultaneous arrivals of two radio traffic peaksfrom the two RPLI clients, delaying one of them by one TTI.Thanks to the buffer, the transport bandwidth for thetransmission of the two RPLI clients is lower than thesum of the two traffic peaks, i.e., included between 34.1and 68.2 Gbit∕s.

However, the use of a buffer is only possible if the fiberpropagation delay does not absorb a significant portion ofthe end-to-end latency budget. If it is not possible to use thebuffer, the transport bandwidth shall be dimensioned tothe sum of the two peaks, i.e., 68.2 Gbit∕s.

In summary, the switch architecture in Fig. 7 allows atradeoff between the required transport bandwidth andthe latency.

In the 100 Gbit∕s wavelength, the bandwidth that is notutilized for RPLI can be allocated for Ethernet BH traffic,which has no stringent latency constraints. Figure 12 re-ports the bandwidth allocated for the RPLI (blue line)and the Ethernet (orange line) versus the fiber link length.

In the baseline case of 0 km, the RPLI can fully benefitfrom buffering, so a transport bandwidth of 34.1 Gbit∕s issufficient for the two clients. In the extreme case, in whichall the 125 μs end-to-end latency budget is absorbed by fi-ber propagation, 68.2 Gbit∕s needs to be allocated. Thishappens at 25 km. In all the intermediate cases, the trans-port bandwidth required for the two RPLI clients linearlyincreases with the distance.

As a final remark, these results are an upper bound ofthe transport bandwidth for RPLI because it has been as-sumed that the radio traffic sources can peak simultane-ously. A higher statistical multiplexing gain is expectedin real scenarios.

VI. DEMONSTRATION

The Xhaul architecture, described in Section III, hasfound a first practical demonstration in an experimentaltest bed built at Ericsson Research Labs. The purposeof the demonstration is to test concurrent BH and FH trans-port and switching over dedicated or shared optical channels(Subsections III.B and III.C, respectively) and the controlarchitecture illustrated in Subsection III.D. Currently, thedemonstration uses state-of-the-art optical technologies.The technologies, described in Subsections III.A and IV,will be included in the next demonstration.

Fig. 11. DWDM transceiver based on an array of hybrid lasersintegrated with a silicon photonics chip.

Fig. 12. RPLI-based FH and BH transport bandwidth versus fi-ber link length in a 100 Gbit∕s wavelength.


The general topology illustrated in Fig. 2 is implementedin the test bed, schematically illustrated in Fig. 13. It con-sists of two remote nodes, located at the antenna side, and ahub node. Figure 14 shows a picture of the demonstratorsetup in the lab.

Two RRUs and an Ethernet switch are connected to nodeRemote 1. One other RRU is connected to node Remote 2.

Each RRU generates a CPRI Option 3 (2.4576 Gbit∕s)signal. The Ethernet switch generates two GbE signalsto emulate the BH signals from a RBS.

The signals connected at each remote node are aggre-gated into 10 Gbit∕s wavelengths, according to the framingprotocol described in Subsection III.B, and sent to the hub.The protocol includes a Reed Solomon RS(255,239) FEC forbit error monitoring, as well as a dedicated in-band OAMchannel (Fig. 6). Limiting the number of interleaved codecsto 9, the latency contribution, introduced by the FEC, is4 μs for each direction, with no appreciable performancedegradation compared to standard OTN channels.

At each remote node, fixed OADMs are used for wave-length add/drop. The wavelengths are duplicated for

protection. At the hub, arrayed waveguide gratings(AWGs) are used for wavelength multiplexing and de-mul-tiplexing.

The ring is realized with a double fiber, one for each di-rection. The fiber span from the Hub to Remote 1 is 14 km;fromRemote 1 to Remote 2, it is is 4 km; and fromRemote 2to the Hub, it is 6 km.

The switch described in Subsection III.C is implementedbymeans of an optical-electrical-optical structure. Referringto the upstream direction, for the sake of simplicity, accord-ing to this structure, the 10 Gbit∕s DWDM channels areconverted from optical to electrical, cross-connected by anagnostic crossbar matrix, and de-framed, as described inSubsection III.B. The CPRI and Ethernet clients are sentto a CPRI mux/de-mux and to an Ethernet switch.

The DUs have their S1 interfaces connected to the corenetwork functions.

The Xhaul controller is realized with a rack-mounted PCrunning Linux OS, connected to the hub via USB.

The demonstrator is able to showcase the followinguse cases:

(1) Dynamic DU-RRU association

Following time-variant traffic loads, it is useful to con-solidate baseband processing in a reduced number ofDUs in low-load hours. Some DUs can be switched off, re-sulting in power savings. To prove this use case, theEthernet switch has been switched off and the CPRI traffichandled by Remote 1 has been moved to Remote 2 with noservice interruptions.

(2) Transport links resiliency

Using FEC and pre-configured BER thresholds, acontinuous monitoring of the transport link quality is per-formed, and degradations are detected in a timely manner.Before the number of errors becomes too high to be cor-rected by the FEC, the protection path is activated. To em-ulate link degradation, a variable optical attenuator (VOA)is placed in line.

(3) BH and FH sharing the same optical channel

At Remote 1, the Ethernet and CPRI are time-domainmultiplexed in the same wavelength using the framing pro-tocol described in Subsection III.B. No appreciable CPRIdegradation, i.e., jitter, was detected due to multiplexingwith the Ethernet.

(4) Service slicing

The use case shows the capability to expose a sliced viewof 5G services, hiding from users and from the service pro-vider how said services are actually mapped onto transportresources. This use case can be applied to share an Xhaulinfrastructure among different radio operators. As a practi-cal proof, two video sources, generating GbE traffic, areconnected to the Ethernet switch at Remote 1. As far asthe slicing of resources is concerned, these two clients areseen as two independent P2P Ethernet services by theservice provider connected to the hub. The physical resourcemapping of said slices is hidden from the upper layers. Inthe demonstration, one Ethernet service is mapped on a

Fig. 13. Sketch of the demonstration setup.

Fig. 14. Picture of the demonstration.


wavelength shared with the CPRI, and the other one istransported by a dedicated wavelength on a different path.

(5) Service-on-demand

The request for service can be done on demand and ne-gotiated. In the demonstration, the radio control asks toactivate two CPRI Options 3 at Remote 1. The Xhaul con-troller (Fig. 8) checks the availability of bandwidth resour-ces with the required latency. If the check is passed, thevideo is activated. If not, e.g., the controller could not finda path with acceptable latency, less stringent requirementsare proposed by the Xhaul controller.

VII. CONCLUSIONS

An Xhaul network architecture, based on optical tech-nologies, agnostic time-deterministic switching, and co-operative radio-transport control, has been discussed.The switch performance with a new 5G functional splithas been numerically assessed.

The Xhaul architecture has a logical hub and spoke top-ology, which provides P2P logical connectivity, regardless ofthe physical network layout. It allows support of significant5G use cases, such as dynamic association of DUs andRRUs, sharing of transport resources between FH andBH, service slicing, and service provisioning on demand.A first practical experimental setup has been preparedin Ericsson Research Labs to demonstrate the use cases.

Xhaul, in the near future, can exploit novel photonictechnologies that contribute to keeping costs, power con-sumption, and footprint low. A simulation and an experi-ment have been presented.

ACKNOWLEDGMENT

The authors would like to thank Prof. Enrico Forestieri ofScuola Superiore Sant’Anna (Pisa) for the results inFig. 3 and Dr. Gianluca Meloni of Consorzio NazionaleInteruniversitario per le Telecomunicazioni (CNIT) andDr. Francesco Fresi of Scuola Superiore Sant’Anna (Pisa)for the results in Fig. 4.

REFERENCES

[1] NGMN, “5G White Paper,” Feb. 2015 [Online]. Available:https://www.ngmn.org/uploads/media/NGMN_5G_White_Paper_V1_0.pdf.

[2] Ericsson, “Ericsson Mobility Report,” Nov. 2015 [Online].Available: http://www.ericsson.com/res/docs/2015/mobility‑report/ericsson‑mobility‑report‑nov‑2015.pdf.

[3] Ericsson, “Cloud RAN,” White Paper, Sept. 2015 [Online].Available: https://www.ericsson.com/res/docs/whitepapers/wp‑cloud‑ran.pdf.

[4] Common Public Radio Interface, CPRI Specification version7.0, Oct. 2015.

[5] E. Forestieri and G. Prati, “Novel optical line codes tolerant tofiber chromatic dispersion,” J. Lightwave Technol., vol. 19,pp. 1675–1684, Nov. 2001.

[6] ITU-T, “Interfaces for the optical transport network (OTN),”ITU-T Recommendation G.709, Feb. 2012.

[7] ITU-T, “OTN transport of CPRI signals,” ITU-TRecommendation G.Sup56, July 2015.

[8] IEEE 802.1 Time Sensitive Networking (TSN) Task Group,http://www.ieee802.org/1/pages/tsn.html.

[9] D. McDysan, “Software defined networking opportunities fortransport,” IEEE Commun. Mag., vol. 51, no. 3, pp. 28–31,Mar. 2013.

[10] ETSI, “Network functions virtualisation (NFV); architecturalframework,” ETSI GS NFV 002, 2015.

[11] P. Contu, P. Pintus, F. Testa, A. D’Errico, and F. Di Pasquale,“Analysis and design of micro-ring based switch elements insilicon photonics for optical interconnection,” in IEEE OpticalInterconnect Conf., 2013, paper TuB3.

[12] P. Dong, W. Qian, H. Liang, R. Shafiiha, N.-N. Feng, D. Feng,X. Zheng, A. V. Krishnamoorthy, and M. Asghari, “Low powerand compact reconfigurable multiplexing devices based onsilicon microring resonators,” Opt. Express, vol. 18, no. 10,pp. 9852–9858, 2010.

[13] V. Sorianello, F. Testa, P. Velha, S. Doneda, andM. Romagnoli,“Experimental evaluation of Residual Added Signal Crosstalkin a silicon photonics integrated ROADM,” in OpticalFiber Communication Conf. and Exhibition (OFC), 2014,paper Th2A.30.

[14] M. Smit, X. Leijtens, H. Ambrosius, E. Bente, J. van der Tol,B. Smalbrugge, T. de Vries, E.-J. Geluk, J. Bolk, R. vanVeldhoven, L. Augustin, P. Thijs, D. D’Agostino, H.Rabbani, K. Lawniczuk, S. Stopinski, S. Tahvili, A.Corradi, E. Kleijn, D. Dzibrou, M. Felicetti, E. Bitincka, V.Moskalenko, J. Zhao, R. Santos, G. Gilardi, W. Yao, K.Williams, P. Stabile, P. Kuindersma, J. Pello, S. Bhat, Y.Jiao, D. Heiss, G. Roelkens, M. Wale, P. Firth, F. Soares, N.Grote, M. Schell, H. Debregeas, M. Achouche, J.-L.Gentner, A. Bakker, T. Korthorst, D. Gallagher, A. Dabbs,A. Melloni, F. Morichetti, D. Melati, A. Wonfor, R. Penty,R. Broeke, B. Musk, and D. Robbins, “An introduction toInP-based generic integration technology,” Semicond. Sci.Technol. vol. 29, 083001, 2014.

[15] G. Li, T. Creazzo, E. Marchena, P. K. L. Yu, and S. Krasulick,“A CMOS wafer-scale, monolithically integrated WDM plat-form for TB/s optical interconnects,” in Optical Fiber Commu-nication Conf. and Exhibition (OFC) 2014, paper Th1C.2.

[16] G.-H. Duan, C. Jany, A. Le Liepvre, A. Accard, M. Lamponi, D.Make, P. Kaspar, G. Levaufre, N. Girard, F. Lelarge, J.-M.Fedeli, A. Descos, B. B. Bakir, S. Messaoudene, D. Bordel,S. Menezo, G. de Valicourt, S. Keyvaninia, G. Roelkens, D.Van Thourhout, D. J. Thomson, F. Y. Gardes, and G. T.Reed, “Hybrid III–V on silicon lasers for photonic integratedcircuits on silicon,” IEEE J. Sel. Top. Quantum Electron.,vol. 20, no. 4, pp. 158–170, July/Aug. 2014.

Paola Iovanna received her degree in elec-tronic engineering from the University ofRoma “Tor Vergata” in 1996. From 1995to 1997, she collaborated with the FUBresearch center of Rome, working on fiberoptic communications and optical network-ing. From 1997 to 2000, she worked atTelecom Italia, where she was involved inexperimentation with new services basedon different access technologies (e.g., XDSL,frame relay, optical). In 2000, she joined

Ericsson in the Research Department, where she dealt withnetworking and design solutions for packet and optical techno-logy (GMPLS, MPLS, and Ethernet). From 2009 to 2012, she was


responsible for carrying out research projects on packet and opticalrouting, control planes, and path computation solutions. In theframework of such activities, she realized demonstrators andprototypes as well. In 2012, she was responsible for definingand prototyping SDN solutions for multi-domain transport incollaboration with customers. Since 2014, she has led a researchteam to define transport networking and control solutions for5G. She is actively involved in European projects and is on theTechnical Program Committees of international conferences likeECOC. She holds more than 50 patents in routing, traffic engi-neering systems, and PCE solutions for packet-opto networksbased on GMPLS, and multi-domain SDN transport, fronthaul,and backhaul solutions for 5G, and is the author of several tensof publications in either international scientific journals orconferences.

Fabio Cavaliere is with Ericsson, Italy,where he has been since 2006. As an expertin photonic systems and technologies, hecontributes to developing strategies for opti-cal transport and access networks. Beforejoining Ericsson, he was with Marconi,United Kingdom, from 1998 to 2006, inthe Photonic Network and System DesignAuthority. He is co-editor of ITU-TRecommendation G.metro and a memberof the Board of Stakeholders of Photonics

21. He has authored several publications and holds patents inoptical communications.

Francesco Testa received his degree inelectronic engineering, summa cum laude,from the University of Rome. In 1982, hewas granted a scholarship fromFondazione Bordoni to work on integratedoptics. In 1985, he worked at Alcatel-Faceon coherent optical systems. He joinedEricsson in 1991 to work on the first demon-strations of WDM and later in research onarchitectures and technologies for opticalnetworks. He is currently working as a

Principal Researcher at Ericsson Research, focusing on photon-ics-integrated technologies and applications.

Stefano Stracca received anM.Sc. degree inelectronic engineering from the Universityof Rome in 1988. He joined Ericsson in 1990,assuming the role of HWand system designerfor telecommunication products. Thereafter,he held the roles of technical coordinator, sys-temmanager, project manager, and line man-ager in different product areas. Currently, heis a Senior Researcher and Deputy Managerat the Ericsson Research Optical Systemsbranch in Pisa, dealing mainly with optical

networking solutions for 5G. He is also the Project Manager ofEuropean Commission FP7 STREP IRIS (Integrated Reconfigu-rable Silicon Photonic-Based Optical Switch). He is an inventor andhas filed several international patents.

Giulio Bottari received a telecommunica-tion engineering degree from the Universityof Pisa in 1998. He then joined Marconi,Genoa,wherehewasasystemdesignerofpho-tonic communication equipment. Since 2006,he has been with Ericsson in Pisa. Currently,he is a Senior Researcher and TechnologyIntelligence Driver in Ericsson Research, fo-cusing on transport architectures for 5G radionetworks, IoT scenarios, and synchronization.Hehas filed 70patents and is a co-author of 40

works in international refereed journals and conferences.

Filippo Ponzini was born in Piacenza,Italy, in 1973. He received a master’sdegree in telecommunications engineeringfrom the University of Parma, Italy. Hewas a researcher in optical technologies withScuola Superiore Sant’Anna, Pisa, Italy. In2007, he joined Ericsson Research. He isnow mainly involved in optical networksand systems for radio access networks, inparticular, for radio heterogeneous networksand their evolution toward 5G. He is the

author of more than 30 publications and holds more than 20international patents.

Alberto Bianchi received a Laurea degreein electronic engineering from theUniversityof Pisa, Pisa, Italy. In 1998, he was in theElectronic System Department of theUniversity of Pisa. Since 1999, he has beenwith Ericsson Telecommunications, wherehe is now a Senior Engineer. From 1999 to2006, he was in the R&D Laboratory,Rome, working on HW design and the devel-opment of Ericsson Switch Systems. From2006 to 2007, he was in the R&D

Laboratory in Milan, working on radio MINI-LINK products. In2007, he joined Ericsson Research in Pisa. His interests are inthe design of optical systems and networks, photonics switchingbased on integrated photonics, mobile backhaul/fronthaul net-works, and silicon photonics applications. He has authored severalpublications and holds several patents.

Roberto Sabella is the manager of theItalian branch of Ericsson Research. Hegot his D.Eng. degree in electronic engineer-ing in 1987 and then joined Ericsson, wherehe gained experience in packet-optical net-works and technologies, traffic engineeringand routing, and telecommunications net-works. He has authored more than 100 pa-pers for international journals, magazines,and conferences, as well as two books on op-tical communications. He holds more than

20 patents and was an adjunct professor at the SapienzaUniversity of Rome. He has guest edited many special issues inseveral journals and magazines.


future proof optical network infrastructure for 5g...

Documents