Professional Documents
Culture Documents
farther from the UE and the availability of another MRN problem of finding the destination vehicle to deliver a packet
whose coverage area includes the UE in question. On the to in Vehicular Ad hoc Networks (VANETS). In this respect,
other hand, the advantages of our approach are numerous, at Road Side Units (RSUs) in VANETs are proposed in as
the top of which is the ability to serve users at the edge of the message routers, and with the information they hear from
cell without deploying expensive equipment (i.e., fixed and vehicles in their ranges, the RSUs can estimate the location of
dedicated relay nodes), or complex techniques that reduce the a packet destination. When an RSU needs to send a packet P
capacity of the network, like CoMP. to a vehicle D, it is required to specify Ds current location
since D is moving. For this, each vehicle sends periodically
II. RELATED WORK Hello packets to neighbors (including any RSUs in its
The use of MRNs was motivated by considering an coverage area) containing information about its position,
interference-limited scenario [5], in which MRNs operate at a speed, direction and timestamp. This enables the RSU to
much lower transmission power as compared to macro eNBs. estimate the location of D, and hence, chooses the best carrier
Both half-duplex and full duplex MRNs were considered. It of P. This way, the packet is forwarded from a vehicle to
was reported that the cell-edge performance while using another until it reaches the destination. We will borrow ideas
MRNs is better when compared to the performance where UE from this work, mostly to determine the connection times
devices are served directly by macro eNBs. The performance associated with the MRNs whose current coverage areas
improvement is due to the better propagation conditions include the concerned UE.
provided by the backhaul link (link between the eNBs and the An analysis is performed in [7] to study the data gain provided
relays) as contrasted to the direct links between the macro by a mobile relay node deployed on a public transportation
eNBs and the UEs at the extremity of the cell. vehicle and serving UEs inside the vehicle. The penetration
Since the relays are characterized by low transmission power, loss between the outdoor and the inside of the vehicle has
their coverage areas are relatively small. A related issue been used as the decision factor to determine the data gain.
concerns UEs that are near the edge of the coverage area of The results have shown that the gain in the data rate provided
the relay, which will wrongly connect to the macro cell base by an MRN is not always guaranteed since a UE attached to a
station (due to its higher received power) instead of relay and close to donor eNB could achieve lower data rate
connecting to the relay. To solve this issue and improve the gains than a UE directly attaches to an eNB.
relay performance, a solution was proposed in [6] by The notion of Coordinated and Cooperative Relay Systems
introducing a bias to cell selection and handover thresholds. (CCRS) is proposed in [8]. The CCRS system is introduced to
When the bias is applied, the relay cells can be extended and provide enhanced cellular coverage in highly populated public
an appropriate load balance between the macro cell and the transportation. The paper in [8] describes two architecture
relay nodes is achieved. Consequently, more users will alternatives for effective realization of CCRS. One of the
connect to the Relay Nodes. alternatives is based on the idea of interconnecting individual
The relay acts like an eNB from the UEs point of view, but it RNs deployed together to cooperate and share the capacity of
is still controlled by the eNB. Without relaying, the only individual mobile wireless backhaul links. The second
handover that is required is the handover from one eNB to alternative is based on adopting a scalable RN equipped with a
another. However, when relays are introduced [7], there is distributed antenna system. According to the results, CCRS
now handover between the eNB and the RNs and between the allows a large number of cellular users on board of the
RNs themselves. In [4], the handover between the relays is a transportation vehicle to be served, while reducing the
point of interest since an MRN serving static or a slow complexity from both the UE and the network elements.
moving UE (outside the MRNs vehicle) requires a dynamic Finally, the authors of [9] have proposed a concept
UE-MRN association. The handover scenario treated in [7] is similar to mobile femtocells stations. They have considered
based on the measurements sent from the UE to the mother two scenarios: fixed femtos with mobile UEs, and Mobile
eNB, which controls the RN serving the UE and decides when femtos with mobile UEs. It was obvious in the results that
to trigger a handover. The same eNB or the target eNB adding fixed or mobile femtocells has improved the
commands the target RN to allocate the necessary resources performance of the UEs. However, it was shown that mobile
for the new connection. The eNB then sends a handover femtos provide better performance than the fixed ones since
command to the UE via the RN causing the UE to detach from they can reach areas that are unreachable by fixed femtos.
the source RN and start synchronizing with the target RN. As can be seen from the above coverage, an area that
Meanwhile, packets that are in flight destined to this UE are is left untreated in the literature is the selection of MRNs by
buffered at the eNB until the handover is complete. cell edge UEs. More specifically, the issue concerns the
MRNs move on roads, and their coverage obviously moves selection of the most suitable MRN to associate with.
with them, and hence, the problem of finding an MRN to Accordingly, our work that we describe in this paper proposes
connect to is similar from the perspective of the UE to the a framework for selecting the most appropriate MRN, given
60
2015 Eight International Workshop on Selected Topics in Mobile and Wireless Computing
61
2015 Eight International Workshop on Selected Topics in Mobile and Wireless Computing
In our model, we assume that the users are in the most part 1
SNRi
2
pedestrians and hence their relative speed is very small when ROCcurrent (2)
n
compared to those of the MRNs. We therefore consider the
cell edge users to be relatively static to the MRNs, and in Where SNRi is the SNR value recorded between the UE and
accordance with [5], we assume that the vehicle velocity and each MRN in its (MRNi). The decision for associating the UE
direction are relatively constant over a 5 ms period [11]. From to MRNi is done periodically, each period T defined as follow:
previous systems studied in LTE, 5 ms is the time required for
accurate channel information and feedback delays at the eNB. speedavg ROCn
T Tn (3)
Moreover, even though the SNR is instantaneous, we consider speedcurrent ROCcurrent
it to remains constant during the 5 ms time period.
B. SNR Variation Where is the nominal LTE scheduling period where the
smallest scheduling unit in LTE refers to 1 ms [[10]
Any wireless channel is subject to fading and pathloss. For
speedcurrent is the current speed of MRNi, ROCn is the SNR
this we use the channel coefficients Hi,j to capture the effects
nominal rate of change computed at average speed, and
of path-loss and shadowing. We assume that those coefficients
ROCcurrent is calculated in Equation 2.
are varying slowly relative to the signaling interval. We define
the Signal to Noise Ratio (SNRi,j) between transmitter MRNi
PFG
and receiver UEj at the UE as follow:
H i, j Pi
SNRi, j (1)
N i W E G GG
Where Pi is the transmitted power from the MRNi, W is the
bandwidth and Ni represents the spectral densities of the PBG
additive noise. EE
PGB
In order to model the SNR variation, we propose to use a
Markov Chain with three states that describe the channel B
quality at the UE through the SNR value: E (Excellent), G
(Good), and B (Bad). Each state corresponds to a range of
values, where the SNR can be classified. BB
After collecting SNR values over many wireless channels, the
Fig. 2. Markov Chain Model adopted in this work
mean and the variance can be computed to provide the
operator the behavior of the SNR variation. From these C. Time of Connection
statistics, the operator can subsequently infer a threshold value After deriving the SNR, we now turn our attention to compute
for the SNR. In the Markov Chain we assume that the the UEs time of connection with the MRN, which is the
transitions from and to non-adjacent states do not occur second metric employed by the UE to derive a weighted score
because the channel transitions are normally continuous. for each MRN it is in range with, and subsequently decide on
We define Im as the measurement interval time and assume which MRN to attach to. To help make the derivations clearer,
that each state has the same average duration. To derive the we use the diagram of Figure 3, which is a zoom-in version of
steady-state probability pss, we divide the time for which the Figure 1. MRNk is currently covering the UE and is
SNR is between [li, hi] by I where li and hi are the limits of the approaching Intersection I1, after which it will take one of
allowed SNR values of State i. On the other hand, the three possible paths, as shown in the figure. In our current
transition probabilities are computed by dividing the number work, we assigned equal probabilities to the possible paths
of state changes by the number of SNR measurements that a vehicle will take upon reaching an intersection, but in
collected during Im. The state diagram in Figure 2 shows the future work, we can assign probabilities that are functions of
states, and allows us to predict the next SNR value. some measure, like the average volume of traffic on each of
In the Markov model, we define the period of computing the the outgoing paths of an intersection. Such data may be
SNR as ISNR such that ISNR < Im and for every ISNR a new SNR obtained from a traffic authority database, or learned by Road
value is available. That is, during Im several SNR values are Side Units, which normally receive beacons from passing by
collected, whereas every ISNR a new SNR value is computed, vehicles using the 802.11p protocol a designated protocol
thus allowing for deriving the rate of change of the SNR from for vehicle-to-vehicle and vehicle-to-infrastructure
one state to another as follow: communication. In the latter case, the RSUs can send this
information periodically via 802.11p to the MRNs, which in
62
2015 Eight International Workshop on Selected Topics in Mobile and Wireless Computing
turn ccan relay themm to the UEs. OnO the other han nd, if an MRNN show
w the intersectiions outside thhe area but leaad to the roads
know ws with some certainty whicch path it willl take (learned d coverred by the areaa.
from previous trips)), it can include this informattion along with
h
the oother data discussed above to o the UE, thuus allowing thee
latterr to associate more
m definite probabilities
p to
o the outgoing
g
pathss of an intersection for this paarticular MRN.
63
2015 Eight International Workshop on Selected Topics in Mobile and Wireless Computing
threshold usually defined by the network operators. Some network without using Mobile Relay Nodes, or were suffering
channels, for example, start to operate when the SNR is as low from a very low SNR (<5 dB). Hence with Mobile Relay
as 5 dB [11], whereas others consider 10 dB to be a minimal Nodes the cell-edge users are connected to the network with a
acceptable SNR. better SNR. Moreover, Figure 6 shows that increasing the
number of the MRNs can result in a significant improvement
Algorithm MRNselect of the SNR and this is an expected result of increasing the
Define: MRNs density since the cell-edge users have more chance to
(xu, yu): UEs position on the map
R: MRNs transmission range find the suitable mobile relay to associate itself to it.
{MRN1, MRN2, , MRNM}: set of MRNs that UE is in range with
(xm, ym), sm, dm, SNRm: position of MRNm, speed, direction, SNR
G: graph representing the roads within a distance R of UE, and all
intersections connecting to these roads (illustrated in Figure 3).
V: vertices of G (intersections)
E: edges of G (roads with weights = associated distances)
|V| = number of vertices in V={vj, j=1, 2, , |V|}
|E| = number of edges (roads) in E={ek, k=1, 2, , |E|}
Visited = {visitedk, k=1, 2, , |E|} // flags for traversed edges
P={p(ej), j=1, 2, , |V|}: set of exit points
p(ej)=1 if the circle centered at (xu, yu) with radius R intersects ej
p(ej)=0 if the circle does not intersect ej
//in Figure 4, exit points are on roads R5, R6, R7, R8, R9, R10, R11
for each interval time
for each MRNm in the set of in-range MRNs
Initialize elements of Visited to 0s
identify ep on which MRNm is driving, given (xm, ym) and dm
determine vq that MRNm is heading into
Visitedp = 1
S(empty) //define a stack and empty it
dist=0 //average distance travelled by MRN
prob=1
z = vq //z will be used as a variable intersection
dist = distance ( (xm,ym), z) //distance to first intersection
//push into the stack (LIFO) a set of equal probabilities Fig. 6. CDF of average SNR for different number of MRNs
for count(downstream ens of z)
push (S, 1/count(downstream ens of z))
end for Regarding the transmission range of the MRNs, we observe
call Boundaries() from Fig. 7 that the MRN having the highest transmission
Tm = dist sm // connection time of UE with MRNm
end for each range can provide better SNR to the UEs. Decreasing the
Select MRNh with highest (Th/ T) + (1- ) (SNRh/SNRthreshold) range results in reducing the MRNs coverage area which
function Boundaries()
means that the number of UEs that might be connected to this
for each en downstream of z MRN will get smaller and tthat he SNR values will decline.
prob = pop(S) //remove most recent probability in S
if Visitedn is equal to 1 The expected Time of connection is the second criterion of
continue //go back to the top of loop (road already visited)
end if
UE-MRN association, where our goal is to have adequate time
Visitedn = 1 of connection. For this, the MRN is supposed to give time
if p(en) == 0 //there is no exit point on this road slots to each UE based on its weighted SNR and weighted
//distance(en) is the length of road en
dist = dist + prob distance(en) expected Time of connection. Fig. 8 shows that the average
z = downstream vertex of en time of connection varies between 20 and 170 seconds, and
for count(downstream ens of z)
push (S, prob 1/count(downstream ens of z)) that the probability for the time of connection to be greater
end for than 90 seconds is equal to 0.55. This time is significant in
call Boundaries()
else
LTE since it is needed for the System Information Blocks to
dist = dist + prob distance(z, ep(en)) carry relevant information for the UE to access a cell, or
end if perform cell re-selection [13]. We note that the time of
end for each
end function connection directly affects the handover rate, because the
larger the time of connection is, the less likely it is for the UE
Fig. 5. Algorithm for selecting most suitable MRN by the UE
to lose connection with the MRN while a session is going on;
and hence, the smaller the handover rate becomes.
According to Figure 6, and for the scenario where 10 MRNs
are emloyed, the probability of having an SNR>5 dB is 0.95,
while the probability of having SNR>10 dB is 0.9. These
values are significant since the UEs were not attached to the
64
2015 Eight International Workshop on Selected Topics in Mobile and Wireless Computing
1
Estimated SNR for Different Transmission Ranges
0.9 400m
550m
0.8 700m b
800m
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-80 -60 -40 -20 0 20 40
Average SNR
0.9
We now propose a single metric L as a weighted average of
0.8
SNR and expected time of connection. We normalize the
0.7
expected SNR and the expected time. The metric L is defined
Estimated CDF
0.6
next in (4), where is a weight factor. Fig. 11 plots L in 3D
0.5 Pr(T<90s=0.45 with =7/8. We note that the parameter can be set by the
0.4 network operator experimentally in such a way that grants
0.3 UEs sufficient time of connection with the MRN that provides
0.2 high SNR levels. Moreover, the value of this parameter can be
0.1
learned by the UE subject to some criteria through a learning
0
algorithm that accounts for historical values and current
0 20 40 60 80 100 120
Expected Time of Connection (s)
140 160 180 situational conditions.
Fig. 8. CDF of expected time of connection (4)
L 1
time I
Figures 9 and 10 show that the Decision Interval Time (Eq.
(3)) decreases when the current speed (Figure 9) and SNR rate
of change increase. This is logical since as the current speed
increases the UE should update its association more often. The
same obviously applies to the ROC (Figure 10).
1.0
a
0.8
0.92
0.6
0.85
0.4
0.76
0.2
0.0 0.67
0.2
0.4 0.48
0.35
0.15
0
0.2
0.4
0.5
0.6
0.7
0.8
0.9
1
0.4
0.2
Normalized SNR
65
2015 Eight International Workshop on Selected Topics in Mobile and Wireless Computing
66