Professional Documents
Culture Documents
Research Article
Abstract: One of the promising features of 5G networks is device-to-device (D2D) communication that enables direct
transmission between D2D user equipments (UEs). Besides the traditional cellular transmission mode, UEs can select between
the reuse and dedicated modes. In this study the authors consider a scenario where a communicating D2D pair and a cellular
UE that communicates with an evolved Node-B can use the same spectrum. It is assumed that the cellular UE can move in the
network while the D2D UEs are static. The movement of the cellular UE can affect the quality of the communication between the
D2D pair. Therefore, the transmission mode between the D2D UEs might change to keep the best quality. In this study the
authors propose a new mobility management and vertical handover algorithm that handles the transmission mode transition
during the D2D connection to maximise the overall throughput. The algorithm uses distance from the border and critical
direction set as mobility variables that are analytically determined. These variables are calculated using a mode selection map
that is derived analytically when pathloss and fading models are used. Finally, in order to analyse the performance of the
proposed handover algorithm, the authors analytically calculate handover rate and sojourn time metrics.
hce 2Pc
RR2 = log 1 + (8)
hse 2Ps + N0
1 hsd 2Ps
RD1 = log 1 + (9)
2 N0
Fig. 1 Network model – the coexistence of the D2D pair and the cellular 1 hce 2Pc
RD2 = log 1 + (10)
UE-eNB communications 2 N0
Rayleigh random variable, [23], for which the channel gain is hse 2Ps hed 2Pe
1
given. RC1 = min log 1 + , log 1 + (11)
Let hi j denote the channel gain between transmitter i and 4 N0 N0
receiver j being in distance di j. Then for the pathloss LoS channel
we have 1 hce 2Pc
RC2 = log 1 + (12)
2 N0
K
hi j = (1) It should be noted that in the reuse mode both ce and sd links
diαj
reuse the same frequency resources. However, we assume that for
the dedicated and cellular modes, the ce link uses half of the
where K is a unit-less constant that depends on the antennas
resources while the remaining half is used by the sd and sed links,
characteristics and α is the pathloss exponent. For the fading model 1 1
we have respectively. The 2 in (9), (10), and (12), and 4 in (11), refer to this
fact. It should be noted that for the sed link, it is assumed that the
1
Ki j se and ed links use 4 of total frequency resources and since they
hi j = (2)
diαj are cascaded with encoding at eNB, the total capacity of the sed
link is calculated as the minimum of the capacities of the se and ed
with Ki j defined as links as shown in (11).
2
Ki j = cSi j zi j (3) 2.3 Mobility model
The mobility model used in this paper is based on the random
where c is the propagation constant, Si j = 10si j /10 is the log-normal waypoint (RWP) model, [13], in which the moving entity randomly
shadow fading variable, where si j is the zero mean Gaussian shifts between two points called waypoints. At each waypoint the
random variable with standard deviation σs and zi j is the fast fading mobile node chooses (i) a random direction uniformly distributed
characteristic of the channel that is modelled as a complex in the range 0, 2π , (ii) a transition length from some distribution,
Gaussian random variable with standard deviation σz. and (iii) velocity from some distribution. The node may have a
In both equations, (1) and (2), i is either s, c, or e, and j is either random pause time at each waypoint.
e or d where the subscripts s, c, d, and e stand for UEs, UEc, UEd,
and eNB, respectively. 3 Problem formulation
We assume that in the reuse mode the resources for UEs–UEd As discussed earlier, the handover process relies on the values of
transmission are shared with the UEc–eNB transmission. specific decision variables that are computed based on the
Therefore, each of the receivers in the network is affected by the measurements in the network. In this section, we formulate the
interference from the transmitters as well as the background noise. decision variables to be used by the handover mechanism. These
The interference channel is shown in Fig. 1 as red-dotted lines. variables are: (i) distance from a region border in the mode
However, when the D2D UE are in the dedicated or cellular modes, selection map, (ii) the critical direction set. The distance variable is
half of the resources are assigned to the D2D UE, and the rest is the distance of the moving UEc from a border in the mode selection
assigned to UEc–eNB transmission, and hence there is no map and it is computed as the minimum of the set of all
interference at the receivers. perpendicular distances from UEc's location to the region border.
Another assumption is that near capacity channel codes are used This variable can be calculated by determining the tangent lines to
at the transmitters, hence the throughput of each transmission is the borders of the map regions from the location of the UEc, which
considered to be equal to the Shannon capacity of the channel is discussed later in this work. Fig. 2 shows tangent and
allocated for that transmission. Moreover, the overall throughput of perpendicular points on the mode selection map when the
the sub-network presented in Fig. 1 in the reuse, dedicated, and communication channel is pathloss or fading.
cellular modes can be calculated as the sum of the channel Moreover, for the purpose of analysis, we introduce the
throughputs of the concurrent transmissions for each mode: handover rate and sojourn time as performance metrics of the
handover.
RR = RR1 + RR2 (4)
3.1 Mode selection map
RD = RD1 + RD2 (5)
For the pathloss LoS channel scenario, the mode selection map
RC = RC1 + RC2 (6) boundaries are found by equating the overall throughput
expressions in (4)–(6), one by one as follows:
where RR1, RR2, RD1, RD2, RC1, and RC2 are the UEs–UEd throughput
in reuse mode, UEc–eNB throughput in reuse mode, UEs–UEd RR = RD (13)
throughput in dedicated mode, UEc–eNB in dedicated mode, UEs–
RR = RC (14)
eNB–UEd throughput in cellular mode, and UEc-eNB in cellular
mode, respectively. These throughput functions are defined as
X 2 = Xc − Xd 2 + Y c − Y d 2
(19a)
Y tn − Y i
tan−1 + π, if Xtn < Xi, Y tn > Y i (24a)
Xtn − Xi
Y tn − Y i
θtn = tan−1 , if Xtn > Xi (24b)
Xtn − Xi
Y tn − Y i
tan−1 − π, if Xtn < Xi, Y tn < Y i (24c)
Xtn − Xi
IET Commun., 2019, Vol. 13 Iss. 14, pp. 2173-2185 2177
© The Institution of Engineering and Technology 2019
Fig. 4 Different channels' throughput for reuse, dedicated, and cellular modes when UEc moves
(a) UEs–UEd throughput in reuse mode, (b) UEc–eNB throughput in reuse mode, (c) UEc–eNB throughput in dedicated mode, (d) UEc–eNB throughput in cellular mode
P dA(r, θ) is the probability of the moving entity being in the area KPc
dA(r, θ) after one movement from the origin. a4 = (40e)
N0
The expected value of period time can be computed by first
determining E[T] and E[S]. KPs KPe
a5 = min 1 + α
,1 + α (40f)
dse N0 ded N0
3.3.2 Sojourn time.: Sojourn time is defined as the expected time
duration that the mobile UEc movement would not result in a TM and x and y are defined as follows
change for the D2D UEs. Assuming that the mobile node does not
have the pause time with a constant velocity, i.e. V ≡ v, the sojourn 1
time in any coverage area C0 can be determined as x= (41a)
Xα
ST = E[T] . ∫ C0
P dA(r, θ) (36) y=
1
(41b)
Yα
a1 f x, y = a x y2 + b x y + c x = 0 (42)
RR = log2 1 + 1 + a3y (37)
1 + a2 x
where
1
RD = log2 a1′ 1 + a4y (38) a(x) = a32 a1′ + a2 x 2
(43a)
2
distance from the border. We found the critical direction set in (26) For the fading channel and equation (17), dy/dx is calculated as
as well as the perpendicular distance using (23). We computed follows:
these variables with the assumption that the tangent and
perpendicular points are given. These can be found using (20) and dRR1
dy E dx
(21). = − (50)
Dividing the derivation of (19a) and (19b) we have dx E
dRR2
−E
dRC2
dy dy
dY dY
dXc X Xc + Y c dXc while for (16), it is calculated as follows:
c
= . (44)
dX Y X −X + Y −Y dY c
dXc c d c d dXc dRR1
dy E dx
= − (51)
The left-hand side of (44) can also be calculated using (41a) and dx E
dRR2
−E
dRD2
dy dy
(41b)
dY
To determine dy/dx in (50) and (51), we have calculated
dXc Y x dy E[RR1 /dx], E[RR2 /dy], E[RD2 /dy], and E[RC2 /dy] numerically as
= . . (45) shown in Fig. 5. As discussed before E[RD1] and E[RC1] are
dX X y dx
dXc constant, so we have
Equating (44) and (45) results in dRD1 dRD1
E =E =0 (52a)
dY
dx dy
X Xc + Y c dXc Y x dy
c
. = . . (46) dRC1 dRC1
Y X −X + Y −Y dY c X y dx E =E =0 (52b)
c d c d dXc dx dy
dY c /dXc can be determined using the above equation as follows: Because of the non-linearity and irreversibility of the equations
in (48a) and (48b), we use a numerical method to find the tangent
Y2 x dy and perpendicular points, denoted as {(Xtn, Y tn) n = 1, …, Nt} and
dY c Xc − . y . dx . (Xc − Xd)
X2 {(Xpn, Y pn) n = 1, …, Np}, respectively.
= − (47)
dXc Y2 x dy Having the tangent and perpendicular points, the distance from
Yc − 2 . y . dx . (Y c − Y d)
X the map and critical direction variables can easily be determined
using the formulas presented in Section 3.2.
Substituting dY c /dXc from (47) in (20) and (21), results in the
tangent and perpendicular points relationships respectively, as 4.3 Handover rate
follows:
The general procedure to find the handover rate was presented in
Y2 x dy 3.3.1. In this part, we present an analytical approach for finding the
Yc − Yi Xc − 2 . y . dx . (Xc − Xd)
= −
X
(48a) handover rate for D2D enabled network. As shown in (34), to
Xc − Xi Y2 x dy calculate RH, E[N] and E[T p] should be determined first.
Yc − . y . dx . (Y c − Y d)
X2 Based on the mode selection map derived for RR = RD and
RR = RC in pathloss channel case and E[RR] = E[RD] and
Y2 x dy
Yc − Yi Yc − . y . dx . (Y c − Y d) E[RR] = E[RC] in fading channel case, Fig. 6 shows the Cn zones
X2
= (48b) with different colors. Cn defines the region where if UEc moves
Xc − Xi Y2 x dy
Xc − 2 . y . dx . (Xc − Xd) from its current location to Cn, then TM for communication
X
between the D2D UEs changes n times. We note that depending on
dy/dx in above equations is calculated for the pathloss channel the UEc's location, the direction of movement, the transition length,
using (42) as follows: and the mode selection map's shape, the number of handovers for
D2D caused by UEc's movement, is different.
dy a′(x)y2 + b′(x)y + c′(x) As shown in Figs. 6a–d, as UEc moves in the direction shown
= − (49)
dx 2a(x)y + b(x) by the arrow, the maximum number of handovers are 4, 2, 3, and 4,
respectively. This depends on the transition length between two
where a′(x), b′(x), and c′(x) denote as derivation of (43a)–(43c), waypoints. In order to determine Cn, we need to compute the
respectively. intersection points of the moving UEc, as it moves from waypoint
(Xi, Y i) in the direction of θ, with the border shown for the two Using (53a)–(53c), the expected number of handovers for the
cases (two equations per case). For example for Fig. 6a, the part of movement from (Xi, Y i) in the direction of θ, as shown in (35) is
the line between the third and fourth intersection points, belongs to reformulated as
the C3.
Determining the Cn zones analytically is complex, therefore we Nc(θ)
use the fact that for a specific direction of movement, i.e. θ, for E[Nθ] = ∑ n . PC (θ)
n
(54)
n=1
UEc located in waypoint (Xi, Y i), depending on the transition
length, l, in one movement period, the next waypoint could be To compute the expected number of handovers for UEc located
located in any of Cns. Knowing (Xi, Y i) and θ, we can find the
in Xi, Y i we determine E[Nθ] over θ which is a random variable
intersection points with the border using (28). The set of
intersection points is {(Xcn(θ), Y cn(θ))}n = 1, …, Nc(θ), where Nc(θ) is the with uniform distribution over [0, 2π]. Presenting a close form for
PC0(θ), PCn(θ) for n = 1, …, Nc(θ) − 1, and PCNc(θ)(θ) as a function
maximum number of intersections when UEc moves in the
direction of θ. Then using (29), we find {rcn(θ)}n = 1, …, Nc(θ), i.e. the of θ is very complex, since it first requires calculation of
intersection points {(Xtn(θ), Y tn(θ)) n = 1, …, Nt(θ)}, which can not
distance of intersection points from (Xi, Y i). We assume that rcn(θ)s
be determined analytically. Therefore, to find the expected number
are ordered, that is for n, m ∈ {1, …, Nc(θ)} if n < m then
of handovers, we use discrete values of θ. We only apply the values
rcn(θ) < rcm(θ). The probability of UEc, being in Cn, in one of θ for which, if UEc takes those values as the direction of
movement period in direction of θ is denoted as PCn(θ) and is movement, it will intersect with the border. In other words for
calculated using (30) as follows: θ ∈ ∪n = 1, …, Nd ψ n, which ψ n is defined in (27). The discrete value
of θ is denoted as θkl and it is defined as
PC0(θ) = P(l < rc1(θ))
(53a)
= 1 − exp −λπrc21(θ) βl − αl
θkl = αl + k , k = 1, …, K and l = 1, …, Nd (55)
K+1
PCn(θ) = P(rcn(θ) < l < rc(n + 1)(θ))
n = 1, …, Nc(θ) − 1 where αl and βl are lower and upper bounds of ψ n in (27), and K is
2
= exp −λπrcn (θ) (53b) denoted as the total number of discrete directions in each
−exp −λπrc2(n + 1)(θ) ψ l = [αl, βl] interval.
It should be noted that for the values of
PCNc(θ)(θ) = P(rcNc(θ)(θ) < l) θ ∈ {[0, 2π] ∖ ∪n = 1, …, Nd ψ n}, E[Nθ] = 0, since UEc will not
2
(53c) intersect with the border and PCn(θ)n = 1, …, Nc(θ) = 0.
= exp −λπrcN c(θ)(θ)
The following probabilities are used later for developing
equations to find the expected number of handovers and sojourn
time:
Fig. 8 Handover for the moving UEc when the TM to be selected is either the reuse or cellular
(a) Mobility of UEc and handover, (b) Zoom on handover area
The algorithm is executed periodically every T d units of time. modes, and (ii) reuse or dedicated modes. For both scenarios we
Provided the mode selection map and (Xi, Y i), the algorithm starts apply the parameters presented in Table 1.
by first finding the perpendicular points, i.e. {(Xpn, Y pn)}n = 1, …, Np by We assume that UEc starts moving from (Xc, Y c) = (1, 1) and
solving (48b). Then the resulting coordinates are substituted in (22) takes 10 steps using the RWP model as described in Section 2.3
to determine the distance set, H. Then the distance of UEc from the and 3.3 with E[L] = 50m. We also assume that UEc velocity is
map, i.e. dp, is computed using (23). The algorithm then checks if constant, V = 5m/s. The simulation results for scenarios 1 and 2,
dp < dpT. This condition checks the proximity of the moving UEc are shown in Figs. 8 and 9, respectively. Based on the parameter
values in Table 1, the distance threshold is calculated to be
with respect to the borders of the mode selection map. If the dpT = μ ⋅ V = (1 s) . (5 m/s) = 5m.
condition is satisfied then, the mobility management entity checks
if the moving UEc intersects with the border when it moves in the Fig. 8a, shows simulation result of the handover algorithm
when the movement of the UEc causes TM changes between the
direction of θ by solving (28) and finding the solutions as
reuse and cellular for the D2D UEs. As observed in this scenario
{(Xcn(θ), Y cn(θ))}n = 1, …, Nc(θ). The handover starts only when this
UEc intersects with the borders of mode selection map in several
equation has at least one solution, i.e. Nc(θ) > 0, meaning UEc
points. A larger view of the intersection part is shown in Fig. 8b.
intersects with the border. The last condition ensures that even if The dashed lines show the distance between the UEc locations and
the moving UEc is close to the border of the map, the handover
the border when it is less than the threshold (dp < dpT). For each of
would not be applied until an intersection with a border is seen. If
these UEc locations, the algorithm checks if during the next T d
any of the two conditions are not satisfied, the algorithm is
repeated in T d units of time. units of time, UEc intersects with the border. If there is at least one
intersection the handover is applied. In the figure, these handovers
are indicated by the cyan color of the UEc trajectory. If the
6 Numerical results handover is not performed, the trajectory color remains black.
In this section we present first our simulation results for using the The same explanation stands for the simulation results
handover algorithm presented in Section 5. Then we evaluate the presented in Figs. 9a and b. In this scenario, the mode selection is
performance of the algorithm using the performance metrics between the reuse and dedicated modes. Considering the same
defined in Sections 3.3.1 and 3.3.2. We assume that UEs an UEd are movement pattern as applied in the previous case, the moving UEc
static. Moreover the transmission power of UEs and eNB is causes only one handover, i.e. it intersects with the border in one
considered to be constant. point.
First, we apply the handover algorithm in two different To evaluate the performance of our handover algorithm we use
scenarios where TM is selected between the (i) reuse or cellular the metrics presented in Sections 4.3 and 4.4. The handover rate
and sojourn time are location-based functions with values
dependent on the first waypoint. For simplicity, we consider a distance from the border, dp is minimum. For Y c0 > 80, the
simulation scenario where UEc's first waypoint could be any point handover rate decreases, and limYc0 → ∞ RH = 0. For the curve
on y-axis between 0 and 250 m. This is shown in Figs. 10a and c related to RR = RD, the handover rate increases slightly until
for pathloss and fading channels, respectively, where the blue line Y c0 ≃ 25. This is close to the first intersection point of the blue line
represents the set of locations for first waypoint. Figs. 10b and d in Fig. 10a with the border which corresponds to dp ≃ 0. After this
shows the distance of each waypoint from the border for the two
cases where channel is pathloss or fading. point, the handover rate decreases sharply until it reaches a local
Having mode selection map defined, for each waypoint on y- minimum at Y c0 ≃ 100. Then again the handover rate increases
axis, denoted as Y c0, with its value changing between 0 and 250, until Y c0 ≃ 210 where dp ≃ 0 and the blue line intersects with the
the handover rate is calculated using (34), (57), and (58). For border. Between these two intersections dp starts from zero and
sojourn time (61b) is used. We take K = 30 discrete values of after reaching a maximum arrives at zero again at the other
θ ∈ ψ n. All other initialisation parameters remain the same as in intersection, as seen in Fig. 10a. Finally when Y c0 exceeds 210, the
Table 1. The simulation results corresponding to the performance handover rate declines sharply. One difference between the two
metrics of the algorithm are shown in Fig. 11. handover curves for RR = RC and RR = RD, is that the latter comes
As can be seen from Fig. 11a, for the curve related to RR = RC, with sharp changes and higher values for handover rate, however
the handover rate increases with a low slope until Y c0 ≃ 80. the former changes smoothly. This is due to the fact that the
Referring to Fig. 10b, we realise that this is the point where the waypoint line in the latter intersects with the border, which causes
the handover rate to abruptly inrease as one get closer to border evaluate the performance of the handover, we analytically derive
points. The same explanation is valid for handover rate derived for handover rate and sojourn time for a D2D enabled network. We
the fading case shown in Fig. 11c. consider a scenario where the first waypoint of the moving cellular
Fig. 11b shows the sojourn time as the other performance UE is located on the y-axis and present the performance metrics
metric of the handover algorithm. By comparing curves in Fig. 11b when the first waypoint changes. We noticed that the handover rate
with the results in Fig. 11a, it is seen that the changes in sojourn increases as the first waypoint gets closer to the border and
time follow the opposite changes in handover rate. In other words, decreases as the first waypoint gets farther from the border. The
when the handover rate increases the sojourn time decreases and behaviour of the sojourn time is opposite of the handover rate and
vice versa. This behaviour can be explained also from Fig. 10b. has a direct relation to the distance of the waypoint from the map.
When dp decreases it means that the waypoint gets closer to the The proposed analytical approach provides precise borders of
border, therefore the handover rate increases but the sojourn time different regions with significantly smaller execution time when
decreases since the average time that UEc moves without TM compared with relevant numerical approaches proposed in the
change to the D2D UEs is reduced. However, when dp increases, literature.
the handover rate decreases and the sojourn time increases. This is
because with higher distance from the border, when the velocity is 8 References
constant, UEc moves further without intersecting with the border. [1] Asadi, A., Wang, Q., Mancuso, V.: ‘A survey on device-to-device
The same explanation is valid for sojourn time derived for the communication in cellular networks’, IEEE Commun. Surv. Tutor., 2014, 16,
fading case shown in Fig. 11d. (4), pp. 1801–1819
[2] Mumtaz, S., Rodriguez, J.: ‘Smart device to Smart device communication’
To validate our analytical calculation of the handover rate and (Springer, United States, 2014)
sojourn time, for each specific point on the blue line in Fig. 10a, [3] Song, L., Niyato, D., Han, Z., et al.: ‘Wireless device-to-device
we place UEc there and we generate a random direction, θ, and communications and networks’ (Cambridge University Press, United States,
transition distance, L, based on their distribution and the 2015), Cambridge Books Online.
[4] Panagopoulos, A.D.: ‘Handbook of research on next generation mobile
probabilistic characteristics as summarised in Table 1. Then using communication systems’ (IGI Global, United States, 2015)
(28) we calculate the intersection points and their total number. [5] Doppler, K., Yu, C., Ribeiro, C.B., et al.: ‘Mode selection for device-to-device
Having these values, the handover rate and sojourn time are communication underlaying an LTE-Advanced network’. 2010 IEEE Wireless
calculated using (34) and (36), respectively. We iterate this Communication and Networking Conference, Sydney, NSW, 2010, pp. 1–6
[6] Liu, Z., Peng, T., Xiang, S., et al.: ‘Mode selection for device-to-device (d2d)
procedure for the current location 106 times and then we average communication under LTE-advanced networks’. 2012 IEEE Int. Conf. on in
out over the resulted values for handover rate and sojourn time. By Communications (ICC), Ottawa, ON, June 2012, pp. 5563–5567
applying these steps for all points on the blue line, we obtained the [7] ElSawy, H., Hossain, E., Alouini, M.S.: ‘Analytical modeling of mode
selection and power control for underlay d2d communication in cellular
simulation results shown in Figs. 11a–d. These results validate the networks’, IEEE Trans. Commun., 2014, 62, (11), pp. 4147–4161
analytical results. [8] Feng, D., Yu, G., Xiong, C., et al.: ‘Mode switching for energy-efficient
device-to-device communications in cellular networks’, IEEE Trans. Wirel.
Commun., 2015, 14, (12), pp. 6993–7003
7 Conclusion [9] Wang, L., Tang, H.: ‘Device-to-device communications in cellular networks’
(Springer, United States, 2016)
In this paper, we address the problem of vertical handover in D2D [10] Huang, Y., Nasir, A.A., Durrani, S., et al.: ‘Mode selection, resource
enabled 5G networks. Due to cellular UE's movement, the optimal allocation, and power control for D2D-enabled two-tier cellular network’,
TM between the D2D UEs may be changed in order to maximise IEEE Trans. Commun., 2016, 64, (8), pp. 3534–3547
the overall network throughput. We generate the mode selection [11] Tang, H., Ding, Z.: ‘Mixed mode transmission and resource allocation for
D2D communication’, IEEE Trans. Wirel. Commun., 2016, 15, (1), pp. 162–
map of a D2D enabled network and use the distance between the 175
cellular UE and its trajectory intersection with the region border as [12] Akyildiz, I.F., McNair, J., Ho, J.S.M., et al.: ‘Mobility management in next-
a decision parameter in the handover algorithm. Using this generation wireless systems’, Proc. IEEE, 1999, 87, (8), pp. 1347–1384
parameter, we develop a vertical handover algorithm. Then to