Mobility for Cellular-Connected UAVs: challenges for the network provider
Erika Fonseca, Boris Galkin, Marvin Kelly, Luiz A. DaSilva, Ivana Dusparic
11 Mobility for Cellular-Connected UAVs: Challengesfor the network provider
Erika Fonseca*, Boris Galkin*, Marvin Kelly † , Luiz A. DaSilva ‡ , Ivana Dusparic* ∗ CONNECT - Trinity College Dublin, Ireland, † Dense Air Ltd., ‡ Commonwealth Cyber Initiative, Virginia Tech, USA { fonsecae, galkinb, duspari } @tcd.ie, [email protected] “This work has been submitted to the IEEE for possible publication. Copyright may be transferred withoutnotice, after which this version may no longer be accessible.” Abstract —Unmanned Aerial Vehicle (UAV) technology is be-coming more prevalent and more diverse in its application. 5Gand beyond networks must enable UAV connectivity. This willrequire the network operator to consider this new type of userin the planning and operation of the network. This work presentsthe challenges an operator will encounter and should consider inthe future as UAVs become users of the network. We analysethe 3GPP specifications, the existing research literature, anda publicly available UAV connectivity dataset, to describe thechallenges. We classify these challenges into network planningand network optimisation categories. We discuss the challengeof planning network coverage when considering coverage forflying users and the PCI collision and confusion issues that canbe aggravated by these users. In discussing network optimisa-tion challenges, we introduce Automatic Neighbouring Relation(ANR) and handover challenges, specifically the number ofneighbours in the Neighbour Relation Table (NRT), and theirpotential deletion and block-listing, the frequent number ofhandovers and the possibility that the UAV disconnects becauseof handover issues. We discuss possible approaches to addressthe presented challenges and use a real-world dataset to supportour findings about these challenges and their importance.
Index Terms —Drone, UAV as end-user, handover, ANR, Neigh-bouring list.
I. I
NTRODUCTION
Unmanned Aerial Vehicles (UAVs) are expected to make useof Fifth Generation Mobile Networks (5G) connectivity whenperforming building inspections (roofs, chimneys, siding),security surveillance, search and rescue operations, mapping,agricultural surveys, delivery of goods, live streaming of showsand events, etc [1]. Although the regulatory bodies have not yetdefined how this integration will happen, UAV connectivity isthe focus of a number of research efforts in the 3 rd GenerationPartnership Program (3GPP) [2]–[4]. 3GPP Release 14 [2],for example, states that a UAV needs to maintain continuousconnectivity with the cellular network while flying at speedsof up to 300 km/h.Qualcomm has carried out several experiments to analysethe viability of using the existing cellular network for pro-viding connectivity to UAVs [5]. They report that the UAVscan have a connection to the cellular network through theside lobes of the antennas that have their main lobe pointedto the ground, where the Ground User Equipment (GUE) typically operates. Preliminary results show that the coverageis adequate for UAVs flying up to 120 m above ground [5].Obstacles between User Equipment (UE) and the cells candeteriorate the signal. The UAV coverage is adequate because,with the greater height of the UAVs, there are no obstaclesbetween them and the antennas. However, at greater heights,the increased Line-of-Sight (LoS) to multiple cells results inhigh levels of interference at the UAVs, which poses handoverand mobility management challenges [6].Handover is the process by which a UE changes its servingcell. It is typically triggered when the UE moves out of thecoverage area of its current serving cell. Ideally, the handovershould be seamless to the UE, such that it would not sufferany data interruption during the process. If a UE experiencesmultiple handovers, a handover delay might occur, resulting insubstantial deterioration to the UE Quality of Service (QoS)[7]. To proceed with a handover, the UE needs to detect pilotsignals from neighboring cells. The list of neighbour cells isdefined on the Neighbour Relation Table (NRT) that is storedin the connected cell. In 5G this list is generated locallyby Automatic Neighbouring Relation (ANR), based on UEmeasurements of Reference Signal Received Power (RSRP)from nearby cells. ANR was introduced in Third GenerationMobile Networks (3G) and was shown to reduce planning andoperational costs for operators [8]. With the more stringentlatency and data rate requirements in 5G, seamless handoversare even more critical, which places a greater importance onthe efficient use of ANR.Unlike GUEs, UAVs will sense a large number of cells [5],which leads to a considerable increase in the size of the ANRat the serving cell and increase the complexity of handoverdecisions [9].This paper comprises the following sections: In Section II,we review mobility management in 5G networks. In SectionIII, we identify challenges that UAV mobility can bring tofuture networks and propose approaches to mitigate these chal-lenges. We illustrate the challenges using a publicly availableavailable dataset of measurements taken by an UAV flying inthe city of Dublin, and connecting to a two-tier network. InSection IV, we conclude the paper by discussing directionsfor future work. a r X i v : . [ c s . N I] F e b II. M
OBILITY MANAGEMENT IN ,and the UE. AMF is responsible for handling connection andmobility management for UEs. The gNB provides connectionto the UE; it has a connection to AMF via the NG interfaceand to other gNBs via the Xn interface. The last entity is theUE itself.In 2G and 3G networks, the NRT is deployed as part ofthe operations and maintenance system, which is equivalentto OAM (Operations Administration and Maintenance) in 5G.In 5G the gNB has the permission to create new entries in theNRT. The ANR determines which cell should be added basedon UE measurements and OAM updates. The UE can performmeasurements to check for new cells, measure signal quality,determine if it needs to make a handover, or add a new cellto the ANR [10].The purpose of this procedure is to transfer measurementresults from the UE to the network in order to allow thenetwork to decide how to improve performance for the UEsand the network itself. The UE can initiate the measurementsonly after successful security activation in the network.These measurements occur as often as determined by thegNB and vary based on the implementation of each operator.If the measurement is made in the same frequency band (intra-frequency) it can be done without any specific preparations tomake the measurements. If the measurements are in anotherfrequency (inter-frequency) the network needs to schedule ameasurement gap where the UE stops receiving and transmit-ting data, changes to the frequency where it has to make themeasurements, and senses it in order to find more suitableBSs. These gaps can affect the performance observed by theUE if the UE is in dedicated mode (transmitting and receivingdata). In idle mode, the UE can perform the inter-frequencymeasurements without impacting its QoS. The measurementsare sent to the serving cell, which uses them to check forevents to trigger a handover, or to add a new cell to the ANR,for example.The information regularly decoded from a measurement bythe UE includes the local identifier of the cell, named physicalcell identifier (PCI) in LTE and 5G. If the PCI is not in theNRT, then the serving cell can send a message instructingthe UE to sense the evolved cell global identifier (ECGI) ofthat cell, that is its global ID, in order to introduce this newcell into the ANR. To determine the ECGI, the UE needs todecode more data from the sensed BS, and to decode the ECGIit needs more than a single measurement gap. If the UE isin connected mode, actively receiving and transmitting data,the UE might not have time to perform the inter-frequencymeasurement and to decode the ECGI, as a result of whichthe UE might be disconnected. In this paper, we refer to the serving BS as gNB as we consider the UAVconnected to the 5G network. When we use the generic term BS we refer toany technology BS, not necessarily 5G.
The mobility events defined by the 3GPP [10] that canhappen after the measurements are made and passed to thegNB are described below. They are divided into intra-RadioAccess Technology (RAT), denoted as events A, and inter-RAT, denoted as events B. • Event A1: The serving cell signal becomes better thanan operator-defined signal quality threshold, i.e. the cellis providing good signal quality. This event is commonlyused to cancel an ongoing handover procedure, to avoida ping-pong effect from the handover. • Event A2: The serving cell signal becomes worse thanan operator-defined signal quality threshold, i.e. the cellis not providing a good signal quality. This event cantrigger Inter-RAT measurements, for example, as newconnectivity options must be considered for the UE. • Event A3: The neighbour cell signal becomes better thanthe serving cell signal by a certain offset. This event cantrigger the handover process to the neighbour cell. • Event A4: The neighbour cell signal becomes better thanan operator-defined signal quality threshold. This event iscommonly used to trigger a handover. In this event, thehandover is not triggered by the radio-signal conditions,but due to a network strategy specified by the operator,such as load balancing across cells, for example. • Event A5: The operator defines 2 thresholds, refereed toas threshold1 (with lower value) and threshold2 (withhigher value) in 3GPP. This event occurs when theserving cell signal becomes lower than threshold1 and theneighbour cell signal higher than threshold2. This eventcan trigger a handover based on the absolute measuredsignal strength values. This time-critical handover can beuseful if the UE is leaving the serving cell coverage areaand needs to handover, even if the target cell is not betterby an offset than the serving cell to trigger an event A3. • Event A6: The neighbour cell signal becomes higher byan offset than the serving secondary cell signal. In thecase the UE has a multi-connection to more than one BS,and it can trigger a handover from its current secondarycell to a new one. • Event B1: An inter-RAT neighbour provides a strongersignal than an operator-defined signal quality threshold.This event may trigger a inter-RAT handover. • Event B2: The operator defines 2 thresholds, referred to asthreshold1 (with lower value) and threshold2 (with highervalue). The signal from the serving cell becomes lowerthan threshold1, and an inter-RAT neighbour provides asignal higher than threshold2. This event can trigger aninter-RAT handover.Although the UE measurements are already described inthe 5G standard [10], in LTE there are two other ways inwhich a cell may be added to an ANR [11], which mightbe adopted in 5G depending on the individual operator. Thefirst alternative to the UE taking measurements is the UEtransmitting an Uplink ID, which should be unique locally.The cells that detect the signal above a certain threshold willadd the serving cell of the source UE into their NRT. Anotherpossible solution is to add a cell to the table once a UE loses
AMF
Source gNB Target gNB
Xn Interface N G i n t e r f a c e N G i n t e r f ace CoreRAN
Fig. 1: Entities involved in a UE handover.connection and re-connects in a new cell. The new cell wouldadd the last cell to which the UE was connected into its NRT.As this method makes use of a UE disconnect, it cannot beapplied if the operator wants to provide seamless handover atall times. III. UAV M
OBILITY C HALLENGES
UAVs were introduced as a new type of user of the cellularnetwork in Long Term Evolution (LTE) and they are expectedto increase in numbers and applications in 5G networksand beyond. The requirements in 3GPP release 15 for UAVconnectivity to the network are summarized in Table I. Tomeet these requirements and the even stricter requirements infuture releases of 5G, the network will need to adapt to able toserve the connected UAVs. One of the biggest challenges forconnected UAVs is the presence of simultaneous LoS channelswith several cells which may be far away. In [5], authorsdemonstrated via simulations and experiments that a UAVcan sense significantly more cells than a GUE. The fact thatUAVs can detect a larger number of cells across a greater areameans that the network should treat the UAV UE differentlyfrom a GUE, in terms of mobility management. The principalmobility-related challenges that a UAV can introduce to 5Gnetworks operators are discussed in this section.
Parameters Value
Latency for traffic 50msUL/DL data rate 200kbpsApplication data rate (UL) up to 50 MbpsUAV UE height up to 300 mUAV UE velocity up to 160 km/h
TABLE I: UAV requirements in 3GPP Release 15.
A. Network Planning challenges
Before the cellular network starts its operation, the operatorsneed to plan the geographic locations of the gNBs, along withconfiguration parameters such as their antenna azimuth andmechanical tilt. If UAVs become a significant user of thenetwork, they need to be taken into consideration from theplanning stage of network deployment. This section discusseschallenges encountered at the planning stage.
Network coverage planning
Network coverage planning is essential to avoid interferenceand unnecessary handovers. For the previous generations ofcellular networks, the coverage was planned only for GUEs,and the main lobe of the BS antennas was often the onlyone taken into account. For the next generations of cellularnetwork, the coverage needs to be planned to also include UAVUEs, and needs to consider what kind of network coveragewill be provided in the air. A common way to plan a cellularnetwork is by using software tools that consider 3D maps ofa given area and antenna radiation patterns. To integrate UAVusers, the tools used to plan the network coverage need to beadapted to consider antenna side lobes and should also projectthe signal propagation into the sky.Another critical part of network planning that becomesharder with the introduction of a flying UE is the PCIdistribution. The flying UEs can exacerbate PCI confusionand collision, which have been reported in LTE networksand persist for 5G networks. Usually, the PCI planning ismade to allocate concurrent PCIs to BSs that are distant fromeach other, to ensure that a UE will be unlikely to detect thesame PCI being transmitted by more than one BS at a time.However, considering connected UAVs, it will be necessaryto understand the air coverage in advance to plan the PCIdistribution. Next, we discuss PCI confusion and collisionchallenge and why UAV users aggravate it.
PCI challenges
In Section II, we introduced the events triggered after themeasurement reports. The first piece of information a UEsenses about a neighbouring gNB is the PCI, that is thelocal cell identifier. Each cell in 5G or LTE has its ownPCI. If the PCI assignment is poorly planned, it can affectthe handover process and delay the downlink synchronisation.Another possible consequence is increased Block Error Rate(BLER) and decoding failures of physical channels. In LTE,there are 504 unique PCIs, compared to 1008 in 5G. If thereare different tiers of the network, the network needs to dividethe PCIs for each tier.Consider a two tier network with macro-cells and small-cells, for example. The PCI values contained in set A will bereserved to the macro-cells and those in set B for small-cells.A and B have no intersection. This rule cannot be violatedinside the same network. This division decreases the numberof possible PCIs for each tier, which can aggravate the issue ofPCI availability. Due to the fact that the GUEs usually connectto cells that are close to them, with good network planning itis possible to avoid most cases of PCI collision and confusionfor GUEs.Figure 2 illustrates a well-planned network, where concur-rent PCIs have a significant distance between them, whichmeans that PCI confusion is not likely to happen for GUEs.The main issue occurs when a UAV flies overhead, as it sensesmore distant cells that can have the same PCI as the servingcell, which results in PCI collision, or be already on the NRTof the serving cell, which results in the PCI confusion. Bothissues are detailed below. gNB1PCI 8 gNB2PCI 25 gNB23PCI 8
Fig. 2: PCI confusion/collision challenge.
PCI Confusion:
PCI confusion happens when the detectedPCI is in the NRT of the serving cell. The serving cell assumesthat the sensed cell is already in the NRT and does not requesta check of the ECGI. The situation is made worse in thescenario where the UAV tries to handover to this concurrentcell because all of the handover configuration will be carriedout with the wrong cell and the UE could have its connectionbroken. The opposite can also happen: if a UAV adds a distantcell to the list and a GUE senses a closer cell with the samePCI the closer one would not be added to the NRT, whichwould result in the handover configuration being sent to thefar away cell. It may even result in the concurrent cell beingadded to a block-list, as many attempts to handover to thiscell would fail. A neighbour should be block-listed if thereare repeated attempts of unnecessary connections, and onceblock-listed, the cell is not an option for handover anymore.As an example, assume that in Figure 2 the UAV isconnected to the gNB2 . In the gNB2
NRT, the gNB1 is aneighbour, and its PCI is saved in the table correspondingto gNB1 . Once the UAV flies and senses a strong signal from gNB23 , it detects its PCI. As the PCI of gNB23 is the same asthat of gNB1 , the serving gNB, gNB2 , decides that the signalsensed by the UAV is from gNB1 and does not ask the UAVto verify the ECGI. If the UAV tries to handover to gNB23 ,all of the configuration for handover will be sent to gNB1 , andthe network might not be able to detect that there is a problembefore the UAV disconnects.
PCI collision:
PCI collision happens when two cells thatcover the same area are allocated with the same PCI. In thissituation, the UE connected to one of them will not sensefor another cell with the same PCI, which can result in theUE not being connected to the best serving cell. For example,consider that the UAV is going in the direction of the hill andis connected to gNB1 . Even if gNB23 has a strong signal andis the only gNB available in that direction, the UAV will notconsider it as as option and will disconnect before trying toconnect to gNB23 .A possible consequence of PCI confusion and collision isthat the network has to be updated with more appropriate PCIsonce these issues happen. To update the PCI of a cell, the gNBneeds to be restarted, which can take more than one hour.To solve the PCI distribution issue, one possible solutionwould be for UAVs to have two radios for communication and measurements. Radio one (R1), would be used for com-munication, but its priority would be sensing. Radio two(R2), would be used for communication only. When the UAVneeds to sense and make measurements, we propose thatthe UAV would always sense the ECGI directly to avoidPCI confusion/collision. During the measurements, R1 shouldstop any communication that could be using the radio. R2would not stop its transmission and data reception at any timeduring the measurement reports. This method would ensurethat UAV does not lose connection during the measurements.The drawback of this approach is that having two radios ismore expensive and takes up additional space on the device.Nevertheless, the use of two radios should be considered byvendors and regulators.To support our earlier claim that the UAV UE should betaken into account by network operators during the varioussteps of network planning, we made use of the dataset avail-able in [12] with signal to noise power measurements madeby a UAV-mounted handset. The network is a two-tier cellularnetwork in Dublin city centre that operates in the 3.6GHzband. The discussion in this paper focuses on the small-cellmeasurements.Typically, for GUEs it is a fair assumption that the UEwill be connected to the closest cell, a common assumptionmade by the research community [13], [14]. This analysisinvestigates how often the UAV sensed the strongest signal ascoming from the geographically closest cell during its flight.Figure 3 illustrates the most potent sensed cell relative to itsdistance to the UAV, for four different altitudes, 30m, 60m,90m, and 120m. At 30m and 60m, the UAV senses morethan 50% of the time the strongest signal as coming fromclosest cell. The same does not happen at higher altitudes:when the UAV is at 90m and 120m, it senses the closest cellas the strongest for around 40% of the time; for almost 30%of the time, it senses the signal from the fourth closest cell asbeing the strongest one. The behaviour presented in the resultsclearly differs from the expected behaviour from a GUE.Figure 3 reinforces the idea that the coverage in the air needsto be considered before deploying new gNBs, as the UAVs canconnect to much more distant gNBs. It also highlights that theresearch community’s assumptions that the UE will connect toits closest gNB is no longer holds in the case of UAVs.
B. Network optimisation challenges
Once the cellular network is deployed, there is still theneed to further optimise the network due to changes in theenvironment or traffic load, or to increase its performance.Self-Organising Network (SON) is an automation approachintroduced in LTE that is also available in 5G, designed tomake the planning, configuration, management, optimisation,and error-correction of the cellular radio access network morestraightforward and rapid. Once a gNB turns on, SON willautomatically configure its PCI, transmission frequency andpower [15]. Our analysis into the UAV behaviour and the3GPP specifications show that SON should be adapted tooptimise the network to accommodate UAV users. In thissection we introduce the optimisation challenges that the UAVas a user can bring to the network operators.
Fig. 3: Percentage of time the UAV was sensing the n-thclosest cell as the strongest signal cell.
ANR challenges
ANR is a critical feature to deal with UE mobility man-agement and therefore it also needs to be adapted to the newreality of the UAV network users. In this section, we presentthe challenges that should be addressed for the ANR to allowthe network to meet the service requirements of UAVs.
Number of neighbours in NRT:
Authors in [16] report thata UAV senses more cells compared to GUEs, which can resultin an increase in the number of neighbours in the NRT. Thiscan be detrimental to all of the UEs connected to that cell, asa UE usually needs to sense all the cells in the neighbour tablebefore performing a handover [9]. If the list is too long andthe UE moves fast, the UE might not have time to sense all thecells in the NRT and may lose connection before performingthe handover. The need to sense an excessive number of cellsalso goes against the ultra-lean principle, whereby the networkis designed to significantly improve energy efficiency andavoid unnecessary measurements [11].
Block-listing neighbours:
Once a far away neighbour isadded by the UAV into the NRT, there is a small chancethat this cell will be sensed by a GUE. If it frequentlyhappens that a GUE cannot sense the far cell, this cell will bedeleted from the NRT frequently. Depending on the NeighbourRemoval Function’s implementation, that can also result in thisneighbour cell being added to the block-list of the serving cell.A cell can be block-listed if it is being removed frequentlyfrom the NRT. A block-listed cell is not an option for handoverfor any UE in the serving cell. If the cell is block-listed, UAVsthat could benefit from a handover to that cell will no longerhave this possibility.A possible solution to the ANR problems presented abovemight be having a separate NRT for flying users. This wouldensure they do not interfere with the GUE connectivity andvice-versa, and it would not deteriorate their service. It wouldallow network operators to design a more fine-tuned solutionto the NRT for flying users, which is a subject that has notbeen explored by the research community.
Handover challenges
Once the PCI confusion and collision issues are resolved,additional challenges related to handover need to be addressedto ensure that UAV UEs do not overload the network and donot unnecessarily disconnect. We discuss those below.
Frequency of handovers for UAVs:
Authors in [17] reportedthat UAVs perform, on average, five times more handoverswhen compared to a GUE. These values show that the mobilityof a UAV tends to generate more signalling overhead in thenetwork and that the parameters used to trigger Event A3 needto be adjusted for UAVs.
Connection interruption time:
Authors in [16] show thatsometimes the handover does not start for UAV users becausethe RSRP measured by the UAV from neighbouring cells doesnot have a minimum difference of 3dB between the servingcell and the possible handover target cell. As a result, the UAVUE does not send event A3, which is required to trigger thehandover. A consequence of this is that UAVs will experiencemore frequent disconnection from the network than GUEs[16]. Once the UAV moves, it moves between side lobes andantennas nulls quickly, and there is no time to make a seamlesshandover, resulting in disconnection when the UAV enters thenulls of the antenna [18].This indicates that the network parameters to start eventA3 in GUE are not suitable for UAVs. The event can triggerhandovers when they are not needed, resulting in a ping-pongeffect, or not trigger handovers at the right time, resultingin disconnection. It is, therefore, necessary to introduce anadaptive threshold to start event A3 for UAVs. The thresholdneeds to be designed for this type of user and needs to takeinto account the changes in the environment to adapt quicklyto the new situation.Using the data provided in [12], we carry out an additionalanalysis of the small-cell deployment, by looking at how oftenthere is a change of the strongest cell when the UAV isflying through the network. Figure 4 illustrates how often theUAV experiences a change of strongest cell during its path.The collected values show that the strongest cell fluctuatesdramatically across different heights. This is due mainly to theplanned service area for the network being primarily at groundlevel. At other heights there are no dominant cells and henceseveral cells are received with similar signal levels. Furtherinvestigation is needed on how the handover performance canbe optimised at these heights.IV. C
ONCLUSION
This paper presented and examined the main challengesthat network operators may encounter when UAVs becomecommon users of the network, and proposed directions tosolve some of these challenges. We divided the challengesinto network planning and network optimisation challenges.To support our claims regarding the challenges we analyseddata from a publicly available dataset which contained mea-surements from a UAV user connecting to a small cell networkin an urban environment.We presented the new coverage planning challenges whenconsidering UAV UEs. Existing network tools used for cov-erage planning are focused on GUEs and do not project how
Fig. 4: Number of strongest cells sensed per minute duringthe path per altitude.the coverage from the antenna main and side-lobes projectsinto the air. These tools need to be adapted to consider aircoverage; to achieve this, it will be necessary to run air drive-tests to access the coverage for UAVs. During the networkplanning, it is also vital to consider UAVs when designing thePCI assignment, to avoid PCI collision and confusion. Thetypical strategy to avoid collision and confusion is to allocateconcurrent PCIs to cells as distant to each other as possible.However, as UAVs can sense far away cells, this might not besufficient to avoid the PCI collision/confusion problem. Wepropose the implementation of two radios on the UAV, whereone would prioritise sensing ECGIs, which would avoid thementioned issues.We also presented challenges that can occur during theoptimisation of the network. The ones related to the ANRconcern the large number of neighbours in the list, and theblock-listing of cells. We suggested a possible direction tosolve this issue by implementing a separate NRT for UAVs.The presented challenges with the handover process includedthe greater number of handovers for UAVs, compared toGUEs, and the connection interruption time which UAVsmight experience due to flying into the nulls of the gNBantennas. As a possible solution to both handover challenges,we proposed an adaptive threshold to trigger the handover forUAVs.The inclusion of a new type of user in the network requiresproper implementation from the initial planning stages ofnetwork deployment, up to its operation and optimisation. Thechallenges presented in this paper highlight the need for opera-tors to take steps to prepare the network for the introduction ofUAV users, otherwise the network may experience QoS issuesfor both air as well as ground users.In our future work, we intend to investigate the solutionsto the issues presented in this paper through simulations,mathematical tools and real-world measurements. A
CKNOWLEDGEMENTS
This work was supported by a research grant from ScienceFoundation Ireland (SFI) and the National Natural ScienceFoundation Of China (NSFC) under the SFI-NSFC PartnershipProgramme Grant Number 17/NSFC/5224, as well as SFIGrants No. 16/SP/3804. It was also supported by the Com-monwealth Cyber Initiative (CCI). The authors would like tothank Conor Duff and Gavin Lee from DenseAir for providingthe Dublin UAV measurement dataset.R
EFERENCES[1] J. Stanczak, I. Z. Kovacs, D. Koziol, J. Wigard, R. Amorim et al. ,“Mobility challenges for unmanned aerial vehicles connected to cellularlte networks,” in . IEEE, 2018.[2] “3rd Generation Partnership Project Technical Specification Group Ra-dio Access Network,” 3GPP, Tech. Rep., March 2017.[3] “TR 36.777: Technical Specification Group Radio Access Network;Study on Enhanced LTE Support for Aerial Vehicles (Release 15),”3GPP, Tech. Rep., 2018.[4] “Ts 36.331 : Technical specification group radio access network; evolveduniversal terrestrial radio access (e-utra); radio resource control (rrc);protocol specification (release 16),” 3GPP, Tech. Rep., 2020.[5] “LTE Unmanned Aircraft Systems,” Qualcomm, Tech. Rep., May 2017.[6] A. Mohamed, M. A. Imran, P. Xiao, and R. Tafazolli, “Memory-fullcontext-aware predictive mobility management in dual connectivity 5Gnetworks,”
IEEE Access , vol. 6, 2018.[7] T. Bilen, B. Canberk, and K. R. Chowdhury, “ Handover managementin software-defined ultra-dense 5G networks,”
IEEE Network , vol. 31,no. 4, 2017.[8] A. Dahlen, A. Johansson, F. Gunnarsson, J. Moe, T. Rimhagen et al. ,“Evaluations of LTE Automatic Neighbor Relations,” in , 2011.[9] H. Claussen, “Efficient self-optimization of neighbour cell lists inmacrocellular networks,” in . IEEE, 2010.[10] “TS 138.331: Technical Specification. 5G; NR; Radio Resource Control(RRC); Protocol specification; (3GPP TS 38.331 version 15.3.0 Release15) ,” 3GPP, Tech. Rep., 2020.[11] P. Ramachandra, F. Gunnarsson, K. Zetterberg, R. Moosavi, M. Amirijoo et al. , “ On automatic establishment of relations in 5G radio networks,”in . IEEE, 2016.[12] SATORI Project. (2021) SATORI – SmArt neTwOrking in the eRa ofAI . 2021-02-17. [Online]. Available: http://satori-project.net/[13] E. Fonseca, B. Galkin, L. A. DaSilva, and I. Dusparic, “Adaptive heightoptimisation for cellular-connected uavs using reinforcement learning,” arXiv preprint arXiv:2007.13695 , 2020.[14] R. Amer, W. Saad, and N. Marchetti, “ Mobility in the sky: Performanceand mobility analysis for cellular-connected UAVs,”
IEEE Transactionson Communications , vol. 68, no. 5, 2020.[15] “Telecommunication management; Self-Organizing Networks (SON);Concepts and requirements (Release 8) ,” 3GPP, Tech. Rep., 2015.[16] S. Euler, H.-L. Maattanen, X. Lin, Z. Zou, M. Bergstr¨om et al. , “Mobilitysupport for cellular connected unmanned aerial vehicles: Performanceand analysis,” in . IEEE, 2019.[17] A. Fakhreddine, C. Bettstetter, S. Hayat, R. Muzaffar, and D. Emini,“Handover challenges for cellular-connected drones,” in
Proceedingsof the 5th Workshop on Micro Aerial Vehicle Networks, Systems, andApplications .[18] M. Mozaffari, W. Saad, M. Bennis, and M. Debbah, “Unmanned aerialvehicle with underlaid device-to-device communications: Performanceand tradeoffs,”