1

Integrated Access and Backhaul in 5G mmWaveNetworks: Potentials and Challenges

Michele Polese, Student Member, IEEE, Marco Giordani, Student Member, IEEE, Tommaso Zugno,Arnab Roy, Member, IEEE, Sanjay Goyal, Member, IEEE, Douglas Castor, Senior Member, IEEE,

Michele Zorzi, Fellow, IEEE

Abstract—Integrated Access and Backhaul (IAB) is beinginvestigated as a means to overcome deployment costs of ultra-dense 5G millimeter wave (mmWave) networks by realizingwireless backhaul links to relay the access traffic. For the devel-opment of these systems, however, it is fundamental to validatethe performance of IAB in realistic scenarios through end-to-end system level simulations. In this paper, we shed light onthe most recent standardization activities on IAB, and comparearchitectures with and without IAB in mmWave deployments.While it is well understood that IAB networks reduce deploymentcosts by obviating the need to provide wired backhaul to eachcellular base-station, in this paper we demonstrate the cell-edge throughput advantage offered by IAB using end-to-endsystem level simulations. We further highlight some researchchallenges associated with this architecture that will requirefurther investigations.

Index Terms—5G, Millimeter Wave, Integrated Access andBackhaul, 3GPP, NR.

This paper has been submitted to IEEE for publication. Copyright may be transferred without notice.

I. INTRODUCTION

Recently, the 3rd Generation Partnership Project (3GPP)has completed, as part of its Release 15, the standardizationof a new Radio Access Technology (RAT), i.e., 3GPP NR,that introduces novel designs and technologies to complywith the ever more stringent performance requirements [1]for 5th generation (5G) networks. In addition to a flexibleframe structure and a revised core network design, NR fea-tures carrier frequencies up to 52.6 GHz. The large availablespectrum at millimeter wave (mmWave) frequencies, withbandwidths which are often much larger than in previousnetwork generations, offers the potential of orders of magni-tude higher transmission speeds than when operating in thecongested bands below 6 GHz, and improves security andprivacy because of the directional transmissions which aretypically established [2].

However, operating in the mmWave spectrum comes withits own set of challenges, severe path and penetration lossesbeing one of them [2]. One promising approach to overcomesuch limitations is using high gain antennas to help closethe link, thus introducing directionality in the communication,

Michele Polese, Marco Giordani, Tommaso Zugno and Michele Zorziare with the Department of Information Engineering (DEI), Univer-sity of Padova, Italy, and Consorzio Futuro in Ricerca (CFR), Italy.Email:{polesemi,giordani,zugnotom,zorzi}@dei.unipd.it.

Arnab Roy, Sanjay Goyal and Douglas Castor are with InterDig-ital Communications, Inc., USA. Email:{arnab.roy, sanjay.goyal, dou-glas.castor}@interdigital.com.

with electronic beamforming to support mobile users. Networkdensification is also used to improve the performance by re-ducing inter-site distance to establish stronger access channels.An ultra-dense deployment, however, involves high capitaland operational expenditures (capex and opex) for networkoperators [3], because high capacity backhaul connectionshave to be provided to a larger number of cellular base stationsthan in networks operating at lower frequencies.

Network disaggregation (i.e., the separation of the layers ofthe protocol stack into different physical equipments) [4] andvirtualization (i.e., the usage of software- and not hardware-based protocol stack implementations) [5] can lower capexand opex by reducing the complexity of individual basestations. Some researchers have also started investigating thefeasibility of Integrated Access and Backhaul (IAB), in whichonly a fraction of Next Generation Node Bases (gNBs, theNR term for a base station) connect to traditional fiber-likeinfrastructures, while the others wirelessly relay the back-haul traffic, possibly through multiple hops and at mmWavefrequencies [6]. The importance of the IAB framework asa cost-effective alternative to the wired backhaul has beenrecognized by the 3GPP. Indeed, it has recently completeda Study Item for 3GPP NR Release 16 [7], which investigatesarchitectures, radio protocols, and physical layer aspects forsharing radio resources between access and backhaul links.Although the 3GPP Long Term Evolution (LTE) and LTE-Advanced standards already provide specifications for basestations with wireless backhauling capabilities, the Study Itemon IAB foresees a more advanced and flexible solution, whichincludes the support of multi-hop communications, dynamicmultiplexing of the resources, and a plug-and-play designto reduce the deployment complexity. However, despite theconsensus about IAB’s ability to reduce costs, designing anefficient and high-performance IAB network is still an openresearch challenge.

In this paper, we review 3GPP standardization activities onIAB and evaluate the performance of the IAB architecturein realistic deployments. In particular, we compare networkscenarios in which a percentage of gNBs (i.e., the IAB-nodes)use wireless backhaul connections to a few gNBs (i.e., theIAB-donors) with a wired connection to the core networkagainst two baseline solutions, i.e., a network with only theIAB-donors, and one in which all the gNBs have a wiredconnection. We also investigate how to efficiently forward thebackhaul traffic from the wireless IAB-nodes to the core net-work and demonstrate the impact of topology setup strategies

arX

iv:1

906.

0109

9v1

[cs

.NI]

3 J

un 2

019

2

on the overall throughput and latency performance. To doso, we conduct end-to-end simulations using ns-3, an open-source network simulator which has recently been extended tofeature a detailed 3GPP-like protocol stack implementation ofIAB at mmWaves [8]. Unlike traditional performance analyses,e.g., [9], [10], [11], which are focused on Physical (PHY)or Medium Access Control (MAC) layer layer protocols, wealso investigate the impact of upper layers, thereby provid-ing a more comprehensive network-level analysis. Moreover,we consider both traditional User Datagram Protocol (UDP)services and more realistic applications including web brows-ing and Dynamic Adaptive Streaming over HTTP (DASH)for high-quality video streaming. Our results demonstratethat, while wired backhaul implementations deliver improvedoverall throughput in conditions of highly saturated traffic,the IAB configuration promotes fairness for the worst usersby associating to relay nodes (IAB-nodes) the UEs whichotherwise would have a poor connection to the wired donor.

Despite these encouraging features, real benefits of theIAB architecture and questions on network behavior underdifferent traffic conditions have been largely unanswered sofar. Accordingly, in this work we study the performanceof a typical network under different traffic considerationsand provide insights on the observed performance gains andshortcomings under different scenarios.

The remainder of the paper is organized as follows. InSec. II we describe the 3GPP activities related to IAB. Then,in Sec. III, we discuss the model and the results of the end-to-end performance evaluation, and identify IAB potentialsand challenges in Sec. IV. Finally, we conclude the paper inSec. V.

II. INTEGRATED ACCESS AND BACKHAUL IN 3GPP NR

The 3GPP recently finalized the Study Item on IAB [7],whose main objective was to assess the feasibility of integratedaccess and wireless backhaul over NR (i.e., the 5G radiointerface), and to propose potential solutions to ensure efficientbackhauling operations. This Study Item led to a Work Item,and is expected to be integrated in future releases of the 3GPPspecifications.

The Study Item considered fixed wireless relays with bothin-band (i.e., the access and the backhaul traffic are multi-plexed over the same frequency band) and out-of-band back-hauling capabilities (i.e., the access and the backhaul traffic useseparate frequency bands), with a focus on the former, which ismore challenging in terms of network design and managementbut maximizes the spectrum utilization. According to [7],IAB operations are spectrum agnostic, thus the relays can bedeployed either in the above-6 GHz or sub-6 GHz spectrum,and can operate both in standalone (SA) (connected to the 5Gcore network) or Non Stand Alone (NSA) modes (connected tothe 4G Evolved Packet Core (EPC)). The possible topologiesfor an IAB network are (i) a Spanning Tree (ST), in whicheach IAB-node is connected to a single parent, or (ii) aDirected Acyclic Graph (DAG), in which each IAB-node maybe connected to multiple upstream nodes.

In the following sections, we will review the main innova-tions introduced in [7] for the network architecture, the proce-

dures for network management, and the resource multiplexingthrough scheduling.

A. Architecture

As shown in Fig. 1, the logical architecture of an IAB net-work is composed of multiple IAB-nodes, which have wirelessbackhauling capabilities and can serve User Equipments (UEs)as well as other IAB-nodes, and IAB-donors, which have fiberconnectivity towards the core network and can serve UEs andIAB-nodes.

The Study Item initially proposed five different configu-ration options for the architecture, with different levels ofdecentralization of the network functionalities and differentsolutions to enable backhauling. The final version, selected forfuture standardization, was preferred because it had limitedimpact on the core network specifications, had lower relaycomplexity and processing requirements, and had more limitedsignaling overhead.

According to the chosen architecture, each IAB-node hoststwo NR functions: (i) a Mobile Termination (MT), used tomaintain the wireless backhaul connection towards an up-stream IAB-node or IAB-donor, and (ii) a Distributed Unit(DU), to provide access connection to the UEs or the down-stream MTs of other IAB-nodes. The DU connects to a CUhosted by the IAB-donor by means of the NR F1∗ interfacerunning over the wireless backhaul link. Therefore, in theaccess of IAB-nodes and donors there is a coexistence of twointerfaces, i.e., the Uu interface (between the UEs and the DUof the gNBs) and the aforementioned F1∗ interface.

With this choice it is possible to exploit the functional splitof the radio protocol stack: the CU at the IAB-donor holdsall the control and upper layer functionalities, while the lowerlayer operations are delegated to the DUs located at the IAB-nodes. The split happens at the Radio Link Control (RLC)layer, therefore Radio Resource Control (RRC), Service DataAdaptation Protocol (SDAP) and Packet Data ConvergenceProtocol (PDCP) layers reside in the CU, while RLC, MACand PHY are hosted by the DUs. An additional adaptationlayer is added on top of RLC, which routes the data acrossthe IAB network topology, hence enabling the end-to-endconnection between DUs and the CU.

B. Network Procedures and Topology Management

An important element to be considered in an IAB deploy-ment is the establishment and management of the networktopology. This is because the end-to-end performance of theoverall network strongly depends on the number of hopsbetween the donor and the end relay, on how many relaysthe donor needs to support, and strategies adopted for proce-dures such as network formation, route selection and resourceallocation. To ensure efficient IAB operations, it is necessaryto optimize the performance of various network proceduresinvolving topology and resource management.

The topology establishment is performed during the IAB-node setup, and is a critical step. When an IAB-node becomesactive, it first selects the upstream node to attach to. Toaccomplish this, the MT performs the same initial access

3

Fig. 1: Protocol stack and basic architecture of an IAB network. The Uu interface represents the interface between the User Equipment (UE) and the DistributedUnit (DU) in the IAB-node, while the F1∗ interface is used between the IAB DU and the upstream Central Unit (CU).

procedure as a UE, i.e., it makes use of the synchronizationsignals transmitted by the available cells (formally calledsynchronization signal block (SSB) in NR) to estimate thechannel and select the parent. Moreover, although not currentlysupported by the specifications, we argue that it would bebeneficial if the MT could retrieve additional information (e.g.,the number of hops to reach the donor, the cell load, etc.), andthen select the cell to attach to, based on more advanced pathselection metrics [12] than just the Received Signal Strength(RSS), as will be discussed in Sec. III. Then, the IAB-nodeconfigures its DU, establishes the F1∗ connection towards theCU in the remote IAB-donor, and is ready to provide servicesto UEs and other IAB-nodes. During this initial phase, theIAB-node may transmit information to the IAB-donor aboutits topological location within the IAB network.

The topology management function then dynamically adaptsthe IAB topology in order to maintain service continuity(e.g., when a backhaul link is degraded or lost), or for loadbalancing purposes (e.g., to avoid congestion). In addition tothe information provided during the initial setup procedure,the IAB-nodes may also transmit periodic information abouttraffic load and backhaul link quality. This allows the CUto be aware of the overall IAB topology, find the optimalconfiguration, and adapt it by changing network connectivity(i.e., the associations between the IAB-nodes) accordingly.

In case the IAB-nodes support a DAG topology with multi-connectivity towards multiple upstream nodes, it is also possi-ble to provide greater redundancy and load balancing. In thiscase, the addition/removal of redundant routes is managed bythe CU based on the propagation conditions and traffic load

of each wireless backhaul link.

C. Scheduling and Resource Multiplexing

For in-band IAB operations, the need to multiplex boththe access and the backhaul traffic within the same frequencyband forces half-duplex operations. This constraint has beenacknowledged in the 3GPP Study Item report [7], althoughfull-duplex solutions are not excluded. Therefore, the radioresources must be orthogonally partitioned between the accessand the backhaul, either in time (Time Division Multiplexing(TDM), which is the preferred solution in [7]), frequency(Frequency Division Multiplexing (FDM)), or space (SpaceDivision Multiplexing (SDM)), using a centralized or decen-tralized scheduling coordination mechanism across the IAB-nodes and the IAB-donor.

Despite the limitations imposed by the half-duplex con-straint, the IAB network is required to address the access trafficrequirements of all the users. For this reason, the availableresources should be allocated fairly, taking into account chan-nel measurements and topology-related information possiblyexchanged between the IAB-nodes. Furthermore, both hop-by-hop and end-to-end flow control mechanisms should be pro-vided to mitigate the risk of congestion on intermediate hops,which might arise in case of poor propagation conditions.

III. END-TO-END EVALUATION OF IAB

We have evaluated the end-to-end performance of an IABmmWave network using the simulator described in [8], whichimplements a full-stack model of the cellular network, with the

4

Fig. 2: Throughput (left) and latency (right) comparison of IAB path selection policies varying the percentage of IAB-donors p for a density of 45 gNB/km2

and a constant bitrate traffic.

3GPP channel model for mmWave frequencies and beamform-ing. Moreover, thanks to the integration with ns-3, it is possibleto study end-to-end scenarios with the TCP/IP stack [13] andrealistic applications, such as the 3GPP Hypertext TransferProtocol (HTTP) model. In the scenario we investigated, thebase stations are deployed following a Poisson Point Process(PPP) with density λ BS/km2, and a fraction 0 ≤ p ≤ 1of the N base stations have wired backhaul connections(i.e., the IAB-donors), while the others (i.e., the IAB-nodes)are wirelessly connected to the IAB-donors, perhaps overmultiple hops. The network implements in-band backhaul, at28 GHz, with TDM of the radio resources among the accessand the backhaul links. We consider uniform rectangularantenna arrays in the base stations and UEs, with 64 and 16elements, respectively, and the beamforming model describedin [14]. The base stations use the backhaul-aware round robinscheduler presented in [8]. The UEs are also deployed witha PPP with density λu = 10λ UE/km2, although we onlyevaluate the performance of the subset of users connected to atarget base station, which is either the first gNB deployed in abaseline scenario in which all nodes have a wired connectionto the core network, or the first IAB-node that performs theinitial access in an IAB scenario.

Backhaul path selection policies. The first set of results,reported in Fig. 2, sheds light on the impact of different back-haul path selection policies in an IAB setup. As introduced inSec. II-B, path selection refers to the procedure by which IAB-nodes find the path towards an IAB-donor, possibly throughmultiple hops. In our previous work [10], we investigated twodifferent policies to forward the backhaul traffic: (i) a Highest-quality-first (HQF) approach which selects, as a parent, thegNB with the highest quality, i.e., the highest Signal-to-Noise-Ratio (SNR), and (ii) a Wired-first (WF) approach whichselects a direct link to the IAB-donor with the best signal,even if an IAB-node with better channel quality is available,provided that some minimum channel quality criterion issatisfied. The first approach facilitates a best-quality wirelessbackhaul connection in the first hop but, in turn, may increase

the number of hops required to forward the traffic to an IAB-donor. The second approach, while minimizing the numberof end-to-end hops, may choose backhaul links with poorerchannel quality. The HQF policy may also leverage a functionthat biases the link selection towards gNBs with wired back-haul to decrease the number of hops to the core network. Thebias computed by the function is not fixed, but depends on thenumber of hops from the IAB-node to the candidate parent itis trying to connect to [10]. Moreover, both conservative andaggressive bias functions can be designed (aggressive HQFpolicies will progressively operate like WF policies). Fig. 2demonstrates that the WF approach should be preferred sinceit offers lower end-to-end latency and higher total throughputcompared to the other investigated policies. The results showthat minimizing the number of hops required to connect toan IAB-donor improves throughput and reduces latency byreducing relaying overhead and congestion at intermediateIAB-nodes.

IAB deployment scenarios. We also tested three differentdeployment scenarios. The best case is when all the N basestations in the network are equipped with a wired connection tothe core network (i.e., the all wired scenario). This representsthe most expensive solution, in terms of density of fiber drops,but permits the whole bandwidth to be used for access traffic.With the IAB-nodes option, pN base stations are IAB-donors,i.e., have a wired connection and (1 − p)N have wirelessbackhaul. Finally, the baseline is the one that 3GPP considersfor comparisons with IAB solutions, described in [7], i.e., adeployment with only pN wired base stations and no IAB-nodes (the only donors configuration).

UDP user traffic. In Fig. 3, we consider an IAB networkwhere each user downloads content from a remote server witha constant bitrate of 220 Mbps, using UDP as the transportprotocol, thus introducing a full buffer source traffic model.The flow of each end-to-end connection does not self-regulateto the actual network conditions, thus congestion arises. Thisexperiment aims to test the performance of an IAB setupin a saturation regime, where the access and backhaul links

5

(a) Fifth percentile throughput. (b) Throughput Cumulative Distribution Function (CDF) for p = 0.3.

Fig. 3: Fifth percentile and CDF of the throughput for the users of a target IAB-node, varying the percentage of IAB-donors p and the deployment strategies,for a density of 45 gNB/km2.

(a) Average rebuffering for DASH clients. (b) Average delay to retrieve an HTTP web page.

Fig. 4: Performance for users in a target IAB-node, with different applications, for a density of 30 gNB/km2.

are constantly used. As expected, the best performance isprovided by the all wired configuration, given that it providesthe same access point density as the IAB setup, but avoidsthe multiplexing of resources between access and backhaul.On the other hand, it is possible to identify two advantagesand one drawback of the IAB configuration with respect tothe only donors one. A higher throughput for the worst usersis achieved when using IAB-nodes, as shown by the fifthpercentile throughput plot in Fig. 3a. In particular, for p = 0.5(i.e., when the number of relays is equal to the number of IAB-donors), IAB has only 13% less fifth percentile throughputthan the all wired configuration. Moreover, the usage of IAB-nodes likely offloads the worst users from the IAB-donors,and this frees up resources for users with the best IAB-donor channel quality, thereby enabling a higher throughput,as illustrated in Fig. 3b. The IAB solution, however, requiresmultiplexing of the wireless resources between access andbackhaul. In a scenario where the links are always saturated,this results in a worse performance for the average users

connected to the relays, which are throttled on the backhaullinks by the round robin scheduler at the donors and have asmaller throughput than with the only donors setup.

DASH, HTTP user traffic. The next set of results considers amore common use case, in which the users either stream videousing DASH [15] or access web pages using HTTP from aremote server. This kind of source traffic is asynchronous andbursty, and, in the DASH case, the flow adapts itself to thevarying capacity offered by the network, after some delays dueto the signaling and convergence of the algorithm. Therefore,the network is not as stressed as in the previous experiment,and in this case the advantage of IAB is more visible. Indeed,thanks to the better channel seen on average by the user dueto more numerous nodes compared to the only donor case,and thanks to the asynchronous and independent nature ofthe traffic at each user, which provides greater multiplexinggains, the performance of the IAB network is not far fromthat of the network with all wired access points. In particular,

6

Fig. 4a reports the average duration of a rebuffering eventfor a DASH stream, for all the users in a target base station.The rebuffering happens when the DASH framework does notadapt fast enough to the network conditions, or if the networkcapacity is not sufficient to sustain even the minimum videoquality available in the DASH remote server. As can be seen,the only donors setup has the worst performance, with a 5 and2 times higher rebuffering than the all wired configuration, forp = 0.3 and 0.5, respectively. The IAB deployment, instead,degrades the performance of the all wired only by 1.4 and1.3 times, for p = 0.3 and 0.5, respectively. Likewise, Fig. 4bshows the average time it takes to completely download a webpage, from the first client HTTP request to the reception ofthe last object, and, as can be seen, the trend is similar to thatof the DASH rebuffering. Finally, for this kind of traffic, theimprovement introduced by the densification of IAB-donors(i.e., by increasing p from 0.3 to 0.5) is less marked than withthe constant bitrate traffic shown in Fig. 3.

IV. POTENTIALS AND CHALLENGES OF IAB

As highlighted by the results presented in Sec. III, IABnetworks present both benefits and limitations with respect todeployments where the radio resources are not multiplexedbetween the access and the backhaul. First, the IAB solutionmay present lower deployment costs and complexity with re-spect to the all wired setup, but, at the same time, splitting theavailable resources between access and backhaul traffic makesthe overall network performance worse than in the all wiredcase under heavily loaded network scenarios. However, forbursty traffic the performance of the IAB solution approachesthat of the all wired case. This shows that when evaluatingthe performance of IAB networks it is important to considerthe specific use case and end-to-end applications that run ontop of the network. Moreover, the results suggest that the mainadvantages of an IAB deployment, when compared to the onlydonors setup, come from an improvement in channel qualityfor cell edge users, on average, which consequently improvesthe area spectral efficiency.

On the other hand, the deployment of an IAB networkpresents challenges related to the design and interactions atdifferent layers of the protocol stack. An important issueis related to the enforcement of Quality of Service (QoS)guarantees in single and multi hop scenarios, so that mixedIAB traffic flows for end-to-end applications can safely co-exist. Additionally, the resources in the IAB network arelimited and shared between the access and the backhaul.Therefore, the admission and bearer configuration should takethis into consideration, in order to avoid overbooking theavailable resources and introducing congestion in the network.As shown in Fig. 3b, this may indeed worsen the experienceof the average users. Similarly, during the setup phase, inwhich the IAB-nodes join the network by performing initialaccess to their IAB parents, it is important to consider theattachment strategies to avoid overloading some IAB-donors,or excessively increasing the number of hops. Even thoughwe demonstrated in Fig. 2 that reducing the number ofrelay operations is always beneficial in terms of both end-to-end latency and throughput, how to design path selection

strategies which are robust to network topology changes andend terminals’ mobility is still an open research challengewhich deserves further investigation.

Most of these system-level challenges are strictly related tothe design of ad hoc scheduling procedures at the MAC layer,able to efficiently split the resources between the access andthe backhaul and provide interference management. Anotherimportant challenge is related to cross-layer effects emergingfrom retransmissions at multiple layers, and the configurationof RLC and transport layer timers may need to account for theadditional delays related to the retransmissions over multiplehops and the reordering of packets at the receiver. At thephysical layer, it will be interesting to evaluate the gain ofthe spatial multiplexing of the access and the backhaul, byusing digital or hybrid beamforming, which could avoid thetime or frequency multiplexing that are needed when usingsingle-beam analog beamforming.

Overall, these challenges represent promising research di-rections to enable self-configuring, easy-to-deploy and high-performing IAB networks, which could represent a cost-effective solution for an initial ultra-dense NR deployment atmmWave frequencies.

V. CONCLUSIONS AND FUTURE WORK

High-density deployments of 5G cells operating atmmWaves call for innovative solutions to reduce capital andoperating costs without degrading the end-to-end networkperformance. In this context, IAB has been investigated as anapproach to relay access traffic to the core network wirelessly,thereby removing the need for all base stations to be equippedwith fiber backhaul. In this work we have reviewed thecharacteristics of IAB capabilities that are currently beingstandardized in 3GPP NR Release 16 and evaluated theperformance of IAB networks for different applications andtraffic types such as Internet browsing (i.e., HTTP) and videostreaming (i.e., DASH). We showed that IAB represents aviable solution to efficiently relay cell-edge traffic, althoughthe benefits decrease for more congested networks. We havealso highlighted the limitations of the IAB paradigm andprovided guidelines on how to overcome them.

IAB standardization is, however, still an on-going process.As part of our future work, we will validate wireless backhaulsolutions considering recently proposed 3GPP scenarios andinvestigate the impact of mobility and network reconfigurationon the network performance.

REFERENCES

[1] 3GPP, “NR and NG-RAN Overall Description - Rel. 15,” TS 38.300,2018.

[2] S. Rangan, T. S. Rappaport, and E. Erkip, “Millimeter-Wave CellularWireless Networks: Potentials and Challenges,” Proceedings of theIEEE, vol. 102, no. 3, pp. 366–385, March 2014.

[3] D. Lopez-Perez, M. Ding, H. Claussen, and A. H. Jafari, “Towards 1Gbps/UE in Cellular Systems: Understanding Ultra-Dense Small CellDeployments,” IEEE Communications Surveys and Tutorials, vol. 17,no. 4, pp. 2078–2101, Fourth quarter 2015.

[4] N. Makris, C. Zarafetas, P. Basaras, T. Korakis, N. Nikaein, andL. Tassiulas, “Cloud-based Convergence of Heterogeneous RANs in5G Disaggregated Architectures,” in IEEE International Conference onCommunications (ICC). IEEE, 2018.

7

[5] R. Mijumbi, J. Serrat, J.-L. Gorricho, N. Bouten, F. De Turck,and R. Boutaba, “Network function virtualization: State-of-the-art andresearch challenges,” IEEE Communications Surveys and Tutorials,vol. 18, no. 1, pp. 236–262, Sept 2016.

[6] H. S. Dhillon and G. Caire, “Wireless Backhaul Networks: CapacityBound, Scalability Analysis and Design Guidelines,” IEEE Transactionson Wireless Communications, vol. 14, no. 11, pp. 6043–6056, Nov 2015.

[7] 3GPP, “NR; Study on integrated access and backhaul; Release 15,” TR38.874, 2018.

[8] M. Polese, M. Giordani, A. Roy, S. Goyal, D. Castor, and M. Zorzi,“End-to-End Simulation of Integrated Access and Backhaul atmmWaves,” in IEEE 23rd International Workshop on Computer AidedModeling and Design of Communication Links and Networks (CAMAD),Sep. 2018.

[9] C. Saha, M. Afshang, and H. S. Dhillon, “Bandwidth Partitioning andDownlink Analysis in Millimeter Wave Integrated Access and Backhaulfor 5G,” IEEE Transactions on Wireless Communications, vol. 17,no. 12, pp. 8195–8210, Dec 2018.

[10] M. Polese, M. Giordani, A. Roy, D. Castor, and M. Zorzi, “Dis-tributed Path Selection Strategies for Integrated Access and Backhaulat mmWaves,” in IEEE Global Communications Conference (GLOBE-COM), Dec 2018.

[11] A. Ometov, D. Moltchanov, M. Komarov, S. V. Volvenko, and Y. Kouch-eryavy, “Packet Level Performance Assessment of mmWave Backhaul-ing Technology for 3GPP NR Systems,” IEEE Access, vol. 7, pp. 9860–9871, 2019.

[12] M. Giordani, M. Polese, A. Roy, D. Castor, and M. Zorzi, “A Tutorialon Beam Management for 3GPP NR at mmWave Frequencies,” IEEECommunications Surveys and Tutorials, vol. 21, no. 1, pp. 173–196,First quarter 2019.

[13] T. R. Henderson, M. Lacage, G. F. Riley, C. Dowell, and J. Kopena,“Network simulations with the ns-3 simulator,” SIGCOMM demonstra-tion, vol. 14, no. 14, p. 527, 2008.

[14] M. Mezzavilla, M. Zhang, M. Polese, R. Ford, S. Dutta, S. Rangan, andM. Zorzi, “End-to-End Simulation of 5G mmWave Networks,” IEEECommunications Surveys and Tutorials, vol. 20, no. 3, pp. 2237–2263,Third quarter 2018.

[15] T. Stockhammer, “Dynamic Adaptive Streaming over HTTP: Standardsand Design Principles,” in Proceedings of the Second Annual ACMConference on Multimedia Systems (MMSys), 2011.

Michele Polese [S’17] received his B.Sc. (2014) and M.Sc. (2016) inTelecommunication Engineering from the University of Padova, Italy. SinceOctober 2016 he has been a Ph.D. student at the University of Padova. Hevisited New York University (NYU), AT&T Labs, and Northeastern Univer-sity. His research focuses on protocols and architectures for 5G mmWavenetworks.

Marco Giordani [S’17] received his B.Sc. (2013) and M.Sc. (2015) inTelecommunication Engineering from the University of Padova, Italy. SinceOctober 2016, he has been a Ph.D. student at the University of Padova. Hevisited New York University (NYU), USA, and TOYOTA InfotechnologyCenter, Inc., USA. In 2018 he received the “Daniel E. Noble FellowshipAward” from the IEEE Vehicular Technology Society. His research focuseson protocol design for 5G cellular and vehicular networks at mmWaves.

Tommaso Zugno received his B.Sc. (2015) and M.Sc. (2018) in Telecom-munication Engineering from the University of Padova, Italy. From Mayto October 2018 he was a Postgraduate researcher with the Departmentof Information Engineering, University of Padova, under the supervision ofProf. Michele Zorzi. Since October 2018 he has been a Ph.D. student at thesame university. His research focuses on protocols and architectures for 5GmmWave networks.

Arnab Roy received his B.E. in Electronics Engineering from Universityof Mumbai, India, and his M.S. and Ph.D. in Electrical Engineering fromPenn State University. He is currently with InterDigital, where he is workingon technology development for next generation cellular and WLAN systems.Before joining InterDigital he held research and product development rolesin the wireless communications industry. His current interests include com-munication technologies related to millimeterwave frequencies, unlicensedspectrum and automobiles.

Sanjay Goyal received his B.Tech. (2009) in communication and computerengineering from the LNM Institute of Information Technology, India, and thePh.D. (2016) and M.S. (2012) degrees in electrical engineering from NYUTandon School of Engineering, New York. Currently, he is working withInterDigital Communications, involved in technology development for nextgeneration communication systems. He is a co-winner of a Best Paper Awardfrom IEEE ICC 2016. His current research interests include NR in unlicensedspectrum, millimeter wave communications, full duplex cellular systems.

Douglas Castor [SM’19] received his B.S. in Electrical Engineering fromPennsylvania State University in 1992 and his M.S. in Electrical Engineeringfrom University of Pennsylvania in 1995. He is currently a Senior Directorof R&D at InterDigital, where he leads the incubation and development ofemerging wireless and sensor technologies. Since joining InterDigital in 2000,he has supported roles in both product development and research innovationsfor 3G, 4G, 5G cellular and other wireless technologies.

Michele Zorzi [F’07] is a Full Professor in the Information EngineeringDepartment of the University of Padova. His present research interests focuson various aspects of wireless communications. He was Editor-in-Chief ofIEEE Wireless Communications from 2003 to 2005, IEEE Transactions onCommunications from 2008 to 2011, and IEEE Transactions on CognitiveCommunications and Networking from 2014 to 2018. He served as a Member-at-Large of the ComSoc Board of Governors from 2009 to 2011, and asDirector of Education and Training from 2014 to 2015.