Enhancing Physical Layer Security for NOMA Transmission in mmWave Drone Networks
Nadisanka Rupasinghe, Yavuz Yapici, Ismail Guvenc, Huaiyu Dai, Arupjyoti Bhuyan
EEnhancing Physical Layer Security for NOMATransmission in mmWave Drone Networks
Nadisanka Rupasinghe ∗ , Yavuz Yapıcı ∗ , ˙Ismail G¨uvenc¸ ∗ , Huaiyu Dai ∗ , Arupjyoti Bhuyan †∗ Department of Electrical and Computer Engineering, North Carolina State University, Raleigh, NC † Idaho National Laboratory, Idaho Falls, ID { rprupasi, yyapici, iguvenc, hdai } @ncsu.edu, [email protected] Abstract —Physical layer security (PLS) is critically importantfor emerging wireless communication networks to maintain theconfidentiality of the information of legitimate users. In thispaper, we investigate enhancing PLS in an unmanned aerialvehicle (UAV) based communication network where a UAV actingas an aerial base station (BS) provides coverage in a denselypacked user area (such as a stadium or a concert area). Inparticular, non-orthogonal multiple access (NOMA) togetherwith highly-directional multi-antenna transmission techniques inmmWave frequency bands are utilized for improving spectralefficiency. In order to achieve PLS against potential eavesdropperattacks, we introduce a protected zone around the user region.However, limited resource availability refrain protected zonebeing extended to cover the entire eavesdropper region. Hence,we propose an approach to optimize the protected zone shape(for fixed area) at each UAV-BS hovering altitude. The associatedsecrecy performance is evaluated considering the secrecy outageand sum secrecy rates. Numerical results reveal the importanceof protected zone shape optimization at each altitude to maximizeNOMA secrecy rates.
Index Terms —5G, drone, HPPP, mmWave, non-orthogonalmultiple access (NOMA), physical layer security (PLS), UAV.
I. I
NTRODUCTION
The importance of deploying unmanned aerial vehicle(UAV) based communication networks during temporaryevents and after disasters to provide on-demand coverage andenhance capacity has recently been explored in some real worddeployments and field trials [1], [2]. In order to reap maximumbenefits from such networks, enhancing the spectral efficiency(SE) is essential. To that end, integrating non-orthogonalmultiple access (NOMA) transmission to UAVs acting asaerial base stations (BSs) can be an effective solution [3], [4].While enhancing the SE with NOMA, it is equally importantto guarantee the confidentiality of communication going onbetween UAV-BS and legitimate users. Hence, introducingappropriate physical layer security (PLS) techniques to suchnetworks become paramount importance.A UAV based mobile cloud computing system is proposedin [5] where UAVs offer computation offloading opportunitiesto mobile stations (MS) with limited local processing capa-bilities. In that, just for offloading purposes between a UAVand the MSs, NOMA is proposed as one viable solution. Inour earlier work [3], [4], NOMA transmission is introducedto UAVs acting as aerial BSs to provide coverage over astadium or a concert scenario. In particular, leveraging multi-antenna techniques a UAV-BS generates directional beams, andmultiple users are served within the same beam employing
This research was supported in part by NSF under the grant CNS-1618692.
NOMA transmission. In [3], assuming the availability of userdistance information, a beam scanning strategy is proposedto maximize NOMA sum rates whereas in [4] we study theperformance of different feedback schemes for NOMA.PLS in wireless communication networks has recently at-tracted significant attention [6]–[8]. One of the main objectivesof the PLS is to increase the performance gap of the linkquality between the legitimate user and that of the eavesdrop-per (Eve) by exploiting the physical properties of the wirelessmedium [9]. To enhance the PLS in wireless ad-hoc networks,artificial noise (AN) aided multi-antenna transmission strat-egy is proposed in [6]. In [7], a protected zone is definedsurrounding the transmitter along with beamforming and ANtransmission to enhance the PLS in a multi-input-single-output(MISO) communication system. Within the protected zone itis guaranteed that no Eve exists. Considering single antennaand multi-antenna scenarios, PLS with NOMA transmissionin large-scale networks is investigated in [8]. In particular, forsingle antenna scenario Eve exclusion area is proposed whilefor multi-antenna scenario AN generation towards undesireddirections is introduced to enhance PLS.In this paper, we consider a similar scenario as in [3], [4],where a UAV-BS is employed to provide broadband connectiv-ity over a densely packed user area in a stadium. NOMA alongwith multi-antenna transmission is then introduced to improvethe SE. In particular, we consider there are Eves outside of theuser area trying to breach communication going on betweenlegitimate users and UAV-BS. In order to enhance the PLSof the UAV based communication network, we introduce aprotected zone around the user area [7], [8]. However, dueto physical constraints, protected zone may not be able toeliminate all the Eves distributed within the area. Hence, wepropose an approach to optimize the protected zone shapebased on UAV-BS hovering altitude such that the achievableNOMA sum secrecy rates are maximized.II. S
YSTEM M ODEL
A. Overview
We consider a mmWave-NOMA transmission scenariowhere a single UAV-BS equipped with an M element uniformlinear array (ULA) is serving single-antenna users in the DL.We assume that all the users lie inside a specific user region as shown in Fig. 1. A 3-dimensional (3D) beam is generatedby the UAV-BS which entirely covers the user region. Weassume that there are K users in total, and the users can berepresented by the set N U = { , , . . . K } . The user region a r X i v : . [ c s . I T ] D ec eamformingDirectionUAV-BS i-thuser j-thuserUser Region Eve RegionProtectedZone IProtected Zone II Fig. 1: System scenario where NOMA transmission serves multipleusers simultaneously in a single DL beam. is identified by an inner-radius L , an outer-radius L , andthe angle ∆ , which is the fixed angle within the projectionof horizontal propagation pattern of UAV-BS on the xy -plane.Note that it is possible to reasonably model various differenthot spot scenarios such as a stadium, concert hall, traffic jam,and urban canyon by modifying these control parameters.We assume that although the user region is free fromeavesdroppers, the surrounding region includes Eves trying tointercept the transmission between UAV-BS and the legitimateusers. We designate the bounded region around the user region,which includes Eves as Eve region . Similar to the user region,we identify the Eve region by the same inner radius L , anouter radius L maxE (greater than L ), and ∆ maxE (greater than ∆) , as shown in Fig. 1. We assume K E Eves in total, whichare represented by the set N E = { , , . . . K E } . Note thathorizontal footprint of the UAV-BS beam pattern covers theEve region (so that any Eve has nonzero channel to UAV-BS),as well, but the coverage over Eve region might be providedby the side lobes depending on the specific radiation pattern. B. Location Distribution and mmWave Channel Model
We assume that users and Eves are uniformly, ran-domly distributed within their specified regions follow-ing homogeneous Poisson point process (HPPP) with thedensities λ and λ E , respectively. The number of users(Eves) in the user (Eve) region is therefore Poissondistributed, i.e., P ( k users in the user region ) = µ k e − µ k ! with µ = ( L − L ) ∆2 λ .We assume that all the users have line-of-sight (LoS) pathssince i) UAV-BS is hovering at relatively high altitudes, andii) LoS path is much stronger than non-LoS (NLoS) paths inmmWave frequency band [3], [10]. The channel h k betweenthe k -th user and the UAV-BS is therefore given as h k = √ M α k a ( θ k ) (cid:104) PL (cid:16)(cid:112) d k + h (cid:17)(cid:105) / , (1) UserRegion Eve Region
ProtectedZone IProtectedZone II
Fig. 2: Footprint of protected zone represented by angle-distance pair (∆ E , L E ) , which is free from any eavesdroppers. where h , d k , α k and θ k represent UAV-BS hovering altitude,horizontal distance between k -th user and UAV-BS, small scalefading gain (i.e., complex Gaussian with CN (0 , ), and angle-of-departure (AoD), respectively. In addition, a ( θ k ) is thesteering vector associated with AoD θ k , and PL ( x ) representsthe path loss (PL) over the distance x . Note that the channelbetween (cid:96) -th Eve in the Eve region (i.e., (cid:96) ∈ N E ) and UAV-BScan also be given using (1). C. Protected Zone Approach for Physical Layer Security
The overall transmission scheme between the UAV-BS andlegitimate users presented in Fig. 1 is highly prone to the Eveattacks, and the PLS is accordingly impaired. In this study,we consider protected zone approach to enhance the secrecyrates of the network [7], [8]. In the proposed approach, anadditional area (i.e., protected zone) around the user region(and inside the Eve region) has been cleared from Eves bymeans of some measures, as shown Fig. 2. This protectedarea is actually a fraction of the complete Eve region, and wedenote this fraction by q with q ≤ . Note that since clearingEves in the protected zone requires certain resources beingspent on the ground, our goal is to keep this area as small aspossible. In addition, we consider to optimize the shape of theprotected zone to enhance secrecy rates while keeping its areathe same, which is the main problem we tackle in this study.The protected zone can be represented by an angle-distance (radius) pair (∆ E , L E ) with ∆ minE ≤ ∆ E ≤ ∆ maxE and L ≤ L E ≤ L maxE . Note that ∆ minE is the minimum angle valuewhich occurs when L E = L maxE . We can therefore represent ∆ minE as follows ∆ minE = q (cid:104)(cid:0) ( L maxE ) − L (cid:1) ∆ maxE − ( L − L )∆ (cid:105) ( L maxE ) − L . (2)As sketched in Fig. 2, it is possible to have different shapesfor protected zone for a fixed q value. Note that whenever wehave ∆ E ≤ ∆ , L E should be sufficiently greater than L (e.g.,“Protected Zone I” in Fig. 2) to have a nonzero protected zone.When ∆ ≤ ∆ E ≤ ∆ maxE , L E might however be smaller (e.g.,“Protected Zone II” in Fig. 2) or greater than L depending onthe area of the user region and particular q choice. Specifically, L E can be parametrically expressed as follows L = L + q ∆ E (cid:104)(cid:0) ( L maxE ) − L (cid:1) ∆ maxE − ( L − L )∆ (cid:105) , (3)for ∆ minE ≤ ∆ E ≤ ∆ . Whenever we have ∆ < ∆ E ≤ ∆ maxE , L = L + q ∆ E (cid:104)(cid:0) ( L maxE ) − L (cid:1) ∆ maxE + 1 − qq ( L − L )∆ (cid:105) , (4)rovided L ≥ L , and L E is otherwise expressed as L = L + q ∆ E − ∆ (cid:104)(cid:0) ( L maxE ) − L (cid:1) ∆ maxE − ( L − L )∆ (cid:105) . (5)III. S ECURE
NOMA
FOR
UAV-BS D
OWNLINK
In this section, we consider NOMA transmission in UAV-BSdownlink (DL) to enhance the SE, and evaluate the associatedsecrecy rates in the presence of protected zone.
A. Secrecy Outage and Sum Secrecy Rates
We assume that UAV-BS generates a beam b where therespective projection in the azimuth domain is in the directionof θ with θ ∈ [0 , π ] [3]. Assuming critically spaced array,the effective channel gain of user k ∈ N U for beamformingdirection θ can be given using (1) as follows [3], [4] | h H k b | ≈ | α k | M × PL (cid:16)(cid:112) d k + h (cid:17) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) sin (cid:16) πM ( θ − θ k )2 (cid:17) sin (cid:16) π ( θ − θ k )2 (cid:17) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) , = | α k | PL (cid:16)(cid:112) d k + h (cid:17) F M ( π [ θ − θ k ]) , (6)where F M ( · ) is the Fej´er kernel. Similarly, the effectivechannel gain of the most detrimental Eve, g E is given as, g E = max k E ∈N E | h H k E b | (7)where h k E is the channel gain of the k E -th Eve.When deriving secrecy rates in NOMA transmission, weassume that UAV-BS knows the effective channel gains ofdesired users while those of Eves are unknown. Without anyloss of generality, we also assume that the users in set N U arealready indexed from the best to the worst with respect to theireffective channel gains as represented by (6). Defining β k tobe the power allocation coefficient of k -th user, we thereforehave β ≤ . . . ≤ β K such that (cid:80) Kk =1 β k = 1 . The transmittedsignal is generated by superposition coding as x = (cid:112) P Tx b K (cid:88) k =1 β k s k , (8)where P Tx and s k are the total DL transmit power and k -thuser’s message, respectively. The received signal at the k -thuser is then given as y k = h H k x + v k = (cid:112) P Tx h H k b K (cid:88) k =1 β k s k + v k , (9)where v k is zero-mean complex Gaussian additive white noisewith variance N .With the received signal as in (9) in hand, each user firstdecodes messages of all weaker users (allocated with largerpower) sequentially in the presence of stronger users’ mes-sages (allocated with smaller power). Those decoded messagesare then subtracted from the received signal in (9), and eachuser decodes its own message treating the stronger users’messages as noise. This overall decoding process is knownas successive interference cancellation (SIC), and k -th userdecodes its own message after SIC with the following SINR: SINR k = P Tx | h H k b | β k (1 − δ k ) P Tx k − (cid:80) l =1 | h H k b | β l + N , (10) where δ k is the Kronecker delta function taking if k = 1 ,and otherwise. Assuming that Eves have powerful detectioncapability [6], [8], the most detrimental Eve decodes k -th usermessage with the SINR given as SINR E k = P Tx β k g E (1 − δ k ) P Tx k − (cid:80) l =1 β l g E + N E0 , (11)where N E0 is the associated noise variance.Considering SINR in (10), the instantaneous rate at k -th user is given by R NOMA k = log (1 + SINR k ) . Similarly,considering (11), the instantaneous rate at the most detri-mental Eve for decoding the k -th user message is givenas R NOMA k, E = log (1 + SINR E k ) . The secrecy rate for k -thlegitimate user can therefore be given as [8], [11] C NOMA k = (cid:2) R NOMA k − R NOMA k, E (cid:3) + , (12)where [ x ] + = max { x, } . As (12) implies, the secrecy ratesare always strictly positive [12].Assuming that R k denotes desired secrecy rate for theuser k ∈ N U , we define the secrecy outage event occurringwhenever C NOMA k < R k with the respective secrecy outageprobability P o k = P { C NOMA k < R k } . As a result, outage sumsecrecy rate with NOMA transmission can be given as R NOMA = K (cid:88) k =1 (1 − P ok ) R k . (13)For performance comparison, we also consider outage sumsecrecy rate with OMA transmission. B. Shape Optimization for Protected Zone
In this section, we discuss optimization of the protectedzone shape to enhance the secrecy rates while keeping itsarea (i.e., q ) the same. We note that any particular subregionwithin the Eve region does not equally impair the achievablesecrecy rates even if the subregion areas are the same and theEves are equally capable. This is basically due to the varyingeffective channel gain between UAV-BS and Eve with differentsubregions, which is a function of not only the distance butalso the relative angle (i.e., angle offset from the beamformingdirection) associated with each Eve.Considering (12), the subregion involving the most detri-mental Eve has the largest impact on the secrecy rates. Hence,instead of choosing the subregions arbitrarily to form theprotected zone, it is more meaningful to include (i.e., protect )subregions which result in better effective channel gain forpotential eavesdroppers, and hence is likely to involve the mostdetrimental Eve.As we will show in Section IV-A, the location distribution ofthe most detrimental Eve depends also on the hovering altitudeof UAV-BS. In particular, the most detrimental Eve is likelyto be present in a subregion where ∆ E ≥ ∆ and L E ≤ L ,which is represented by “Protected Zone II” in Fig. 2, whenthe altitude is low. In contrast, the region including the mostdetrimental Eve becomes closer to “Protected Zone I” ofFig. 2 with ∆ E ≤ ∆ and L E ≥ L when the altitude is high.We therefore conclude that the shape of the protected zoneshould be optimized taking into account the UAV-BS hoveringltitude. Hence, at a particular altitude and for a given q , theoptimal shape of the protected zone can be identified as ∆ ∗ E , L ∗ E = argmax ∆ E , L E R NOMA (14)s.t. ∆ minE ≤ ∆ E ≤ ∆ maxE ,L E is computed by (3) − (5) , where R NOMA is given in (13).IV. N
UMERICAL R ESULTS
In this section, we present numerical results to showthe importance of shape optimization of the protected zoneand its impact on the achievable sum secrecy rates withvarying UAV-BS hovering altitudes. Considering Fig. 1,we assume that L = 100 m, L = 25 m, L max E = 1 . L m, ∆ = 0 . rad (1 . ◦ ) , ∆ max E = 2∆ , ¯ θ = 0 ◦ , and M = 100 . Userdistribution is based on HPPP with λ = 1 , and user targetsecrecy rates are R j = 4 bits per channel use (BPCU) and R i = 1 BPCU, respectively. The power allocation ratios are β j = 0 . and β i = 0 . while P Tx = 10 dBm and N = − dBm. We assume two user NOMA transmission with j = 1 and i = 20 after ordering users with respect to theireffective channel gains. The path-loss model is assumed to bePL ( (cid:112) d k + h ) = 1+ (cid:16)(cid:112) d k + h (cid:17) γ with γ = 2 [4], [10], andthe UAV-BS altitude is h ∈ [10 , m. A. Location of the Most Detrimental Eavesdropper
We present the angle and distance distributions of thelocation of the most detrimental Eve in Fig. 3 and Fig. 4,respectively, for two different altitudes of h = { , } m, andHPPP densities of λ E = { . , } . In Fig. 3, we observe thatthe most detrimental Eve is very likely to have a relative anglewhich is greater than ∆2 at a lower altitude of h = 10 m. Inparticular, relative angle of the most detrimental Eve exceeds ∆2 all the time for λ E = 1 while it drops to approximately of the time for λ E = 0 . . When the altitude becomes higher(i.e., h = 100 m), the relative angle of most detrimental Evebecomes smaller than ∆2 . In Fig. 4, we observe that the PLdistance of the most detrimental Eve is smaller (greater) than m at lower (higher) altitudes, i.e., h = 10 m ( h = 100 m).We therefore conclude that the most detrimental Eve tends tohave larger relative angles and smaller PL distances at loweraltitudes in comparison to those at higher altitudes . B. Impact of the Protected Zone Shape on Secrecy Rates
In Fig. 5, we depict the sum secrecy rates along with theprotected zone angle (i.e., ∆ E ) at altitudes of h = { , } massuming q = 0 . . We observe that while the secrecy rates getmaximized at ∆ E ≈ . ◦ ( > ∆) for h = 10 m, the optimalangle turns out to be ∆ E ≈ . ◦ ( < ∆) at h = 100 m. Thisobservation is consistent with the discussion in Section IV-Ain the sense that the most detrimental Eve has a relative anglegreater (smaller) than ∆2 at low (high) altitudes.Similarly, Fig. 6 presents the secrecy rates along with theprotected zone distance (i.e., L E ) for the same settings as ofFig. 5. We observe that while the optimal distance maximizingthe secrecy rates is L E ≈ m at h = 10 m, it turns out tobe L E ≈ m at h = 100 m. As before, this observation alsonicely agrees with our discussions in Section IV-A regarding Angle [deg] CD F E = 0.1, h=10 m E = 1, h=10 m E = 0.1, h=100 m E = 1, h=100 m h = 100 m h = 10 m/2 = 0.573 ° Fig. 3: Angle distribution of the most detrimental eavesdropper.
20 40 60 80 100 120 140
Distance [m] CD F E = 0.1, h=10 m E =1, h=10 m E = 0.1, h=100 m E =1, h=100 m
100 m10 m
Fig. 4: Distance distribution of the most detrimental eavesdropper. e [deg] S u m s e c r e cy r a t e s ( h = m ) [ BP CU ] S u m s e c r e cy r a t e s ( h = m ) [ BP CU ] h = 100 mh = 10 m q=0.2 Fig. 5: Sum secrecy rates of NOMA along with the protected zoneangle (i.e., ∆ E ) for h = { , } m, q = 0 . , and λ E = 0 . . the distance distribution of the most detrimental Eve. Thisshows the importance of optimizing the protected zone shapeat different hovering altitudes to maximize sum secrecy rates. C. Secrecy Rates Variation with Altitude
In Fig. 7, we present sum secrecy rates of NOMA and OMAtransmission along with varying altitude of h ∈ [10 , mand for different protected zone sizes (i.e., q ∈ { , . , . } ). L e [m] S u m s e c r e cy r a t e s ( h = m ) [ BP CU ] S u m s e c r e cy r a t e s ( h = m ) [ BP CU ] h = 100 mh = 10 m q=0.2 Fig. 6: Sum secrecy rates of NOMA along with the protected zonedistance (i.e., L E ) for h = { , } m, q = 0 . , and λ E = 0 . .
20 40 60 80 100 120 140h [m]00.511.522.533.54 S u m s e c r e cy r a t e s [ BP CU ] NOMA, q=0.2OMA, q=0.2NOMA, q=0.5OMA, q=0.5NOMA, q=0OMA, q=0NOMA, q=0.2 (fixed)NOMAOMA q=0.2
Fig. 7: Sum secrecy rates for NOMA and OMA along with UAV-BShovering altitude, where q ∈ { , . , . } , and λ E = 0 . . For a nonzero protected zone (i.e., q (cid:54) = 0 ), considering shapeoptimization as discussed in Section III-B, sum secrecy ratesare identified. In addition, Fig. 7 also captures sum secrecyrate variation with q = 0 . for a fixed shape (optimal shapeat h = 10 m). As can be observed, the fixed protected zoneshape yields sum secrecy rates comparable to that of optimizedprotected zone shape only around h = 10 m and performsworse at all the other altitudes. Further, we observe that thesecrecy rates improve if large portion of the Eve region canbe covered by the protected zone (i.e., q increases). Based onthe target sum secrecy rate and the operational altitude, thesmallest q can also be determined. By this way, the desiredsecrecy rates can be achieved optimally by designating lessarea as the protected zone which would relieve the burden ofclearing any unnecessary region free from Eves. Note also thatthe secrecy rates associated with NOMA is much larger thanthose of OMA especially at lower altitudes.In Fig. 8, variation of the optimal shape of the protectedzone is captured for q = 0 . , . . In particular, Fig. 8a showsthe optimal angle, ∆ ∗ E variation whereas Fig. 8b depicts opti-mal distance L ∗ E variation with UAV-BS hovering altitude. Ascan be observed from Fig. 8, ∆ ∗ E decreases with altitude (seeFig. 8a) while L ∗ E increases with altitude (see Fig. 8b). Thisobservation aligns nicely with the discussion in Section IV-A
20 40 60 80 100 120 140 h [m] E * [ deg ] q = 0.2q = 0.5 (a) Optimal angle, ∆ ∗ E .
20 40 60 80 100 120 140 h [m] L E * [ m ] q=0.2q=0.5 (b) Optimal distance, L ∗ E . Fig. 8: ∆ ∗ E and L ∗ E variation with varying UAV-BS hovering altitudes,Here q = 0 . , . . which tells us that at lower altitudes the most detrimentalEve tends to have a larger relative angle and smaller distancewhereas at higher altitudes this is vice versa.V. C ONCLUDING R EMARKS
In this paper, we investigate the secrecy rates of UAVbased mmWave communication network considering NOMAtransmission. In particular, we consider protected zone ap-proach to enhance the secrecy rates. Towards this end, wefirst investigate the distribution of the location of the mostdetrimental Eve which impairs secrecy rates the most. Wethen consider the protected zone which is free from any Eve,and the associated optimal shape of it to enhance the secrecyperformance. We show that the optimal shape of the protectedzone should cover the most detrimental Eve. In addition, wealso show that the optimal shape highly relies on the UAV-BShovering altitude such that the protected zone should be wider(narrower) in angle and shorter (longer) in distance at lower(higher) altitudes. R
IEEE Trans. Commun. , pp. 1–1, 2018.[4] ——, “Comparison of limited feedback schemes for noma transmissionin mmwave drone networks,” in
IEEE Int. Workshop on Sig. Process.Adv. in Wireless Commun. (SPAWC) , June 2018, pp. 1–5.[5] S. Jeong, O. Simeone, and J. Kang, “Mobile edge computing via a UAV-Mounted Cloudlet: Optimization of bit allocation and path planning,”
IEEE Trans. Vehic. Technol. , vol. PP, no. 99, pp. 1–1, 2017.[6] X. Zhang, X. Zhou, and M. R. McKay, “Enhancing secrecy with multi-antenna transmission in wireless ad hoc networks,”
IEEE Trans. Inf.Forensics Security , vol. 8, no. 11, pp. 1802–1814, Nov. 2013.[7] N. Romero-Zurita, D. McLernon, M. Ghogho, and A. Swami, “Phy layersecurity based on protected zone and artificial noise,”
IEEE Sig. Process.Lett. , vol. 20, no. 5, pp. 487–490, May 2013.[8] Y. Liu, Z. Qin, M. Elkashlan, Y. Gao, and L. Hanzo, “Enhancing thephysical layer security of non-orthogonal multiple access in large-scalenetworks,”
IEEE Trans. Wireless Commun. , vol. 16, no. 3, pp. 1656–1672, Mar. 2017.[9] J. M. Hamamreh, H. M. Furqan, and H. Arslan, “Classifications andapplications of physical layer security techniques for confidentiality: Acomprehensive survey,”
IEEE Commun. Surveys Tuts , pp. 1–1, 2018.[10] Z. Ding, P. Fan, and H. V. Poor, “Random beamforming in millimeter-wave NOMA networks,”
IEEE Access , no. 99, pp. 1–1, 2017.[11] M. Bloch, J. Barros, M. R. D. Rodrigues, and S. W. McLaughlin, “Wire-less information-theoretic security,”
IEEE Trans. Inf. Theory , vol. 54,no. 6, pp. 2515–2534, June 2008.[12] N. Yang, H. A. Suraweera, I. B. Collings, and C. Yuen, “Physical layersecurity of tas/mrc with antenna correlation,”