Interference Coordination for Aerial and Terrestrial Nodes in Three-Tier LTE-Advanced HetNet
Abhaykumar Kumbhar, Hamidullah Binol, Ismail Guvenc, Kemal Akkaya
IInterference Coordination for Aerial and Terrestrial Nodes inThree-Tier LTE-Advanced HetNet
Abhaykumar Kumbhar , Hamidullah Binol , ˙Ismail G¨uvenc¸ , and Kemal Akkaya Dept. Electrical and Computer Engineering, Florida International University, Miami, FL, 33174 Dept. Electrical and Computer Engineering, North Carolina State University, Raleigh, NC, 27606
Abstract — Integrating unmanned aerial vehicles (UAVs) asuser equipment (UE) and base-stations (BSs) into an existingLTE-Advanced heterogeneous network (HetNet) can furtherenhance wireless connectivity and support emerging services.However, this would require effective configuration of system-level design parameters for interference management. Thispaper provides system-level insights into a three-tier LTE-Advanced air/ground HetNet, wherein the UAVs are deployedboth as BSs and UEs, and co-exist with existing terrestrialnodes. Moreover, this HetNet leverages on cell range ex-pansion (CRE), intercell interference coordination (ICIC),3D beamforming, and enhanced support for UAVs. ThroughMonte-Carlo simulations, we compare system-wide fifth per-centile spectral efficiency (5pSE) and coverage probability fordifferent ICIC techniques, while jointly optimizing the ICICand CRE parameters. Our results show that reduced powersubframes defined in 3GPP Rel-11 can provide considerablybetter 5pSE and coverage probability than the 3GPP Rel-10with almost blank subframes.
Index Terms — Cell range expansion, ICIC, LTE, UAV.
I. I
NTRODUCTION
Several of the telecommunications service providers areconsidering the use of unmanned aerial vehicles (UAVs) tomeet the mobile data and coverage demands, restore dam-aged infrastructure, and enable emerging service [1], [2].However, integration of these UAVs as aerial user equip-ment (AUEs) and unmanned aerial base-stations (UABSs),would require a system-level understanding to both modifyand extend the existing terrestrial network infrastructure.A vital goal while planning any air/ground heterogeneousnetwork (AG-HetNet) is to ensure ubiquitous data coveragewith broadband rates. To this end, existing works haveexplored the co-existence of terrestrial and aerial nodes ina network and assessed the performance this AG-HetNet interms of coverage probability and fifth percentile spectralefficiency (5pSE) as the two key performance indicators(KPIs).Despite the earlier works given in [3], [4], to the bestof our knowledge, there are no prior works that considerboth AUEs and UABSs to simultaneously co-exist withterrestrial nodes such as the macro base-stations (MBSs),pico base-stations (PBSs), and ground user equipment(GUEs) in LTE-Advanced AG-HetNet. To this purpose, wesimulate an AG-HetNet in public safety band class 14 as
This research was supported in part by NSF under CNS-1453678.
Fig. 1. The terrestrial nodes (MBS, PBS, and GUE) andaerial nodes (UABS and AUE) constitute the AG-HetNet.shown in Fig. 1; which leverages on 3GPP Rel-8 cell rangeexpansion (CRE), 3GPP Rel-10/11 intercell interferencecoordination (ICIC), 3GPP Rel-12 three-dimensional (3D)beamforming (3DBF), and 3GPP Rel-15 enhanced supportfor UAVs. Subsequently, we maximize the two KPIs ofthe network while mitigating intercell interference andjointly optimizing ICIC and CRE network parameters.Our simulation results show that a three-tier hierarchicalstructuring of reduced power subframes can effectivelyhelp mitigate interference in AG-HetNets.The rest of this paper is organized as follows. InSection II, we provide the AG-HetNet system model, 3Dchannel model, 3DBF, and definition of KPIs as a functionof network parameters. In Section III, we configure UABSsdeployment on a hexagonal grid and present ICIC networkparameters. In Section IV, through extensive computer sim-ulations, we analyze and compare the two KPIs of the AG-HetNet for various ICIC techniques and configurations.Finally, the last section provides concluding remarks.II. S
YSTEM M ODEL
We consider a three-tier AG-HetNet deployment, whereall the MBS, PBS and UABS locations (in 3D) are capturedin matrices X mbs ∈ R N mbs × , X pbs ∈ R N pbs × , and X uabs ∈ R N uabs × , respectively, with N mbs , N pbs and N uabs denoting the number of MBSs, PBSs, and UABSswithin the simulation area ( A sim ). Similarly, the 3D dis-tribution of GUEs and AUEs are respectively captured inmatrices X gue and X aue . Assuming a fixed antenna height,the location of wireless nodes MBS, PBS, GUE, and AUEare modeled using a 2D Poisson point process (PPP), withintensities λ mbs , λ pbs , λ gue and λ aue , respectively. On theother hand, UABSs are deployed on a fixed hexagonal grid a r X i v : . [ c s . N I] O c t nd at two different heights (see also Table II).For an arbitrary n th UE, let d on , d pn , and d un bethe nearest distance from macrocell of interest (MOI),picocell of interest (MOI), and UABS-cell of interest(UOI), respectively. Then assuming Nakagami-m fadingchannel, the reference symbol received power from MOI,POI, and UOI is given by R mbs ( d on ) = P mbs A E ( φ, θ ) H ϕ ( d on ) / , R pbs ( d pn ) = P pbs A E ( φ, θ ) H ϕ ( d pn ) / ,R uabs ( d un ) = P uabs A E ( φ, θ ) H ϕ ( d un ) / , (1)where random variable H accounts for Nakagami-m fadingand is defined in (2) of [3]. Through shaping parameter m , received signal power can be approximated to achievevariable fading conditions. The value m > approximatesto Rician fading along line-of-sight (LOS) and m = 1 approximates to Rayleigh fading along non-LOS (NLOS).The variable A E ( φ, θ ) is the transmitter antenna’s 3DBFelement defined in (19)–(21) of [5]. Using 3DBF, thepower transmission from MBS ( P mbs ), PBS ( P pbs ), andUABS ( P uabs ) can be controlled for UEs in cell-edge/CREregion. This limits the power transmission into adjacentcells that causes intercell interference and subsequently im-proves signal-to-interference ratio (SIR) of desired signal.The variables ϕ ( d on ) , ϕ ( d pn ) , and ϕ ( d un ) are path-lossrespectively observed from MBS, PBS, and UABS in dB. A. Path Loss Model
Based on the type of communication link, i.e., ground-to-ground (GTG), any-to-air (ATA), and air-to-ground(ATG) between a UE and base-station (BS) of interest, weconsider distinct path-loss models for an accurate analysisof signal reliability.We consider Okumura-Hata path loss (OHPL) to es-timate the GTG communication link between GUE andterrestrial MBS and PBS. OHPL in an urban terrestrialenvironment is defined in (1)–(2) of [6]. In an urban-macrowith aerial scenario, we consider ATA communication linkbetween an AUE and any nearest BS. The average pathloss for ATA link is calculated over the probabilities ofLOS/NLOS defined in Table B-1, and path loss in TableB-2 of [7]. The average path loss for ATG communicationlink between GUE and UABS is calculated over theprobabilities of LOS/NLOS defined in (4) of [3].Fig. 2 illustrates the empirical path loss cumulativedistribution functions (CDFs), calculated for all distancesbetween base stations ( X mbs , X pbs , and X uabs ) and UEs( X gue and X aue ), using conditions defined in previousparagraph. Inspection of Fig. 2 reveals that the maximumallowable path loss is diverse for GTG, ATG, and ATAcommunication links. This variation is primarily due tothe environmental factors and LOS/NLOS probability ofcommunication link. Nevertheless, maximum allowablepath-loss for the models used in GTG, ATA, and ATG linkis approximately 255 dB, 216 dB, and 154 dB, respectively. Fig. 2. The CDF of path loss observed for the communi-cation link between UEs and base-stations.Fig. 3. Cell selection and UE association in USF/CSFsubframes of MBS, PBS, and UABS. B. Spectral Efficiency with 3GPP Rel.10/11 ICIC
We consider CRE at small cells such as PBS and UABSto extend the network coverage and increase capacity, byoffloading traffic from congested cells; nevertheless, anadverse side effect of CRE includes increased interferenceat UEs in cell-edge/CRE region.To address this intercell interference, both MBS andPBS are capable of using 3GPP Rel-10/11 ICIC tech-niques, wherein MBS and PBS can transmit radio framesat reduced power levels. The radio subframes with reducedpower are termed as coordinated subframes (CSF) andfull power as uncoordinated subframes (USF). The powerreduction factor is given by α mbs and α pbs at MBS andPBS. In particular, α mbs = α pbs = 0 corresponds to Rel-10 almost blank subframes eICIC, α mbs = α pbs = 1 corresponds to no ICIC, and otherwise corresponds to Rel-11 reduced power FeICIC. We coordinate USF/CSF dutycycle using β mbs and (1 − β mbs ) at MBS and β pbs and (1 − β pbs ) at PBS.Let Γ mbs , Γ mbscsf , Γ pbs , Γ pbscsf , Γ uabs , and Γ uabscsf denoteSIR at USF and CSF subframes of MOI, POI, and UOI,respectively. Then, using positive biased CRE τ pbs at PBSsand τ uabs at UABSs, these small cells can expand theirSIR coverage. Subsequently, during the process of cellselection, a UE always camps on the nearest BS thatyields the best SIR. Then an individual MBS, or PBS,ABLE I. SIR and SE definitions. Signal-to-interference ratio (SIR) SE in USF/CSF radio frames Γ mbs = R mbs ( d on ) R pbs ( d pn )+ R uabs ( d un )+ I agg C mbsusf = β mbs log (1+Γ mbs ) N mbsusf Γ mbscsf = αR mbs ( d on ) α pbs R pbs ( d pn )+ R uabs ( d un )+ I agg C mbscsf = (1 − β mbs )log (1+Γ mbscsf ) N mbscsf Γ pbs = R pbs ( d pn ) R mbs ( d on )+ R uabs ( d un )+ I agg C pbsusf = β pbs log (1+Γ pbs ) N pbsusf Γ pbscsf = α pbs R pbs ( d pn ) αR mbs ( d on )+ R uabs ( d un )+ I agg C uabscsf = (1 − β pbs )log (1+Γ uabs ) N pueusf Γ uabs = R uabs ( d un ) R mbs ( d on )+ R pbs ( d pn )+ I agg C mbsusf = ( β mbs + β pbs )log (1+Γ uabs ) N uueusf Γ uabscsf = R uabs ( d un ) αR mbs ( d on )+ α pbs R pbs ( d pn )+ I agg C uabscsf = (2 − ( β mbs + β pbs ))log (1+Γ uabscsf ) N uuecsf or UABS can schedule their UE in either USF/CSF radiosubframes based on their respective scheduling threshold ρ mbs , ρ pbs , ρ uabs . This association of a UE with the nearestBS and scheduling in USF/CSF subframes for six differentscenarios is summarized in Fig. 3.By following an approach similar to that of [4], wedefine the SIR and 5pSE experienced by an n th arbitraryUE for six different scenarios and are given in Table I.Therein, I agg is the aggregate interference at a UE fromall the BSs, except from MOI, POI, and UOI, while N mueusf , N muecsf , N pueusf , N puecsf , N uueusf , and N uuecsf are the number ofMBS-UE, PBS-UE, and UABS-UE scheduled in USF/CSF.III. K EY P ERFORMANCE I NDICATORS
In this article, 5pSE corresponds to the worst fifthpercentile UE capacity amongst all of the scheduled UEs.On the other hand, we define the coverage probability ofthe network as the percentage of an area having broadbandrates and capacity larger than a threshold of T C SE .In this study, we maximize the two KPIs of the network,while obtaining the best ICIC network configuration usinga brute force algorithm . However, the brute force algorithmis computationally infeasible to search for all possibleoptimal values in a large search space. Therefore, to reducethe system complexity and simulation runtime, we considerUABSs deployment on a fixed hexagonal grid and applythe same ICIC parameters across all MBSs, PBSs, andUABSs. With a feasible set of vectors, we determine thebest state, S (cid:48) KPI , out of all possible states S such that: S (cid:48) KPI = arg max S C KPI ( S ) , (2)where KPI ∈ (cid:0) , COV (cid:1) . The objective function C ( . ) denotes 5pSE and C cov ( . ) denotes coverage prob-ability for a given state S = (cid:104) X uabs , S ICICmbs , S ICICpbs , S ICICuabs (cid:105) .As defined previously, X uabs is the matrix representingthe location of the N uabs UABSs in three dimensions, S ICICmbs = [ α mbs , β mbs , ρ mbs ] ∈ R N mbs × is a matrixthat captures individual ICIC parameters for each MBS, S ICICpbs = [ α pbs , β pbs , ρ pbs , τ pbs ] ∈ R N pbs × is a matrixthat captures individual ICIC parameters for each PBS,and S ICICuabs = [ τ uabs , ρ uabs ] ∈ R N uabs × is a matrix thatoccupies individual ICIC parameters for each UABS. TABLE II. System and simulation parameters. Parameter Value
Simulation area ( A sim )
100 km MBS, PBS, GUE, AUE intensities 4, 12, 100, and 1.8 perkm Number of UABS 60MBS, PBS, and UABS transmit powers 46, 30, and 26 dBmHeight of MBS, PBS, and UABS 36 and 15mHeight of UABS 36 and 50 mHeight of GUE and AUE 1.5 and 22.5 mPSC LTE Band 14 center frequency 763 MHz for downlinkPower reduction factor α mbs and α pbs to USF Duty cycle β mbs , β pbs to %Scheduling threshold for MUEs ( ρ mbs ) dB to dBScheduling threshold for PUEs ( ρ pbs ) − dB to dBScheduling threshold for UUEs ( ρ uabs ) − dB to dBRange expansion bias for τ uabs , τ uabs dB to dB IV. S
IMULATION R ESULTS
In this section, with the help of computer simulation andsystem parameters set to the values given in Table II, wecompare the two KPIs of the network with and withoutICIC techniques. The 3D surface plot in Fig. 4 and Fig. 6illustrates the combined effect of CRE at PBSs and UABSs(along x- and y-axes) on the coverage probability and 5pSE(along the z-axis) of the wireless network. In an initialinspection of Fig. 4 and Fig. 6, we can intuitively concludethat FeICIC performs better when compared to eICIC andwithout any ICIC techniques.When UABS are deployed at the same height as MBSor height higher than MBS; in the absence of any ICICmechanism, the optimal value of the CRE for coverageprobability and 5pSE is observed at around 0 dB as seenin Fig. 4 and Fig. 6. However, as the CRE increases, theinterference also increases at scheduled UEs. As a result,the performance of the two KPIs starts to decline.When the UABS are deployed at the same height asMBS, with eICIC, the two KPIs are seen to perform betterwhen CRE at PBSs is between − dB and at dB forUABSs. With FeICIC, the two KPIs are seen to performbetter when CRE at PBSs is at dB and between − dBfor UABSs. Whereas, when we deploy UABS at a heighthigher than MBS, with eICIC, the two KPIs are seen toperform better when CRE is between − dB for bothPBSs and UABSs. With FeICIC, the two KPIs are seen toperform better when CRE at PBSs is between − dB a) Coverage prob. vs. CRE. (b) Peak 5pSE vs. CRE. Fig. 4. The effects of combined CRE at PBS and UABSon the two KPIs of the network, with and without ICIC;when UABS are deployed at height of 36 m.Fig. 5. Performance comparison of the two KPIs; whenUABS are deployed at height of 36 m.and between − dB for UABSs.The comparative analysis of Fig. 4 and Fig. 6 revealsthat the improvement in coverage probability is less signifi-cant but the 5pSE improvement is significant. When UABSare deployed at the same height as MBS, for coverageprobability eICIC sees an improvement of . in theabsence of any ICIC, and FeICIC sees an improvement of . over eICIC. For 5pSE, eICIC sees an improvementof . in the absence of any ICIC, and FeICIC seesan improvement of . over eICIC. Whereas, whenthe UABS are deployed at a height higher than MBS,for coverage probability eICIC sees an improvement of . in the absence of any ICIC, and FeICIC sees animprovement of . over eICIC. For 5pSE, eICIC seesan improvement of . in the absence of any ICIC,and FeICIC sees an improvement of . over eICIC.Finally, we also observe, as the deployment height ofUABS increases, the LOS of UABS also increases. As aresult, interference at scheduled UEs increases, and there isa sparse decrease in the KPI values of the wireless network.V. C ONCLUSION
This paper gives system-level insights into the LTE-Advanced AG-HetNet. Through simulations, we maxi-mized the coverage probability and 5pSE of the network, (a) Coverage prob. vs. CRE. (b) Peak 5pSE vs. CRE.
Fig. 6. The effects of combined CRE at PBS and UABSon the two KPIs of the network, with and without ICIC;when the UABS are deployed at height of 50 m.Fig. 7. Performance comparison of the two KPIs; whenthe UABS are deployed at height of 50 m.while addressing the intercell interference and optimizingthe ICIC network parameters using a brute force technique.Our analysis shows that the HetNet with reduced powersubframes (FeICIC) yields better coverage probability and5pSE that than with almost blank subframes (eICIC).R
EFERENCES[1] A. Kumbhar, F. Koohifar, I. Guvenc, and B. Mueller, “A survey onlegacy and emerging technologies for public safety communications,”
IEEE Commun. Survery Tuts. , vol. 18, pp. 97–124, Sep. 2016.[2] S. Chandrasekharan, K. Gomez, A. Al-Hourani, S. Kandeepan,T. Rasheed, L. Goratti, L. Reynaud, D. Grace, I. Bucaille, T. Wirth, et al. , “Designing and implementing future aerial communicationnetworks,”
IEEE Commun. Mag. , vol. 54, no. 5, pp. 26–34, 2016.[3] M. M. Azari, F. Rosas, A. Chiumento, and S. Pollin, “Coexistenceof terrestrial and aerial users in cellular networks,” in
Proc. IEEEGlobal Commun. Conf. , Singapore, Singapore, 2017, pp. 1–6.[4] A. Kumbhar, I. G¨uvenc¸, S. Singh, and A. Tuncer, “ExploitingLTE-Advanced HetNets and FeICIC for UAV-assisted public safetycommunications,”
IEEE Access , vol. 6, pp. 783–796, 2018.[5] A. Kammoun, H. Khanfir, Z. Altman, M. Debbah, and M. Kamoun,“Preliminary results on 3D channel modeling: From theory to stan-dard,”