Content uploaded by Raúl Parada
Author content
All content in this area was uploaded by Raúl Parada on May 17, 2017
Content may be subject to copyright.
Cell discovery based on historical
user’s location in mmWave 5G
Ra´
ul Parada⇤and Michele Zorzi†
Dipartimento di Ingegneria dell’Informazione
Universit`
a degli Studi di Padova
Via Gradenigo 6/b, 35131, Padova, Italy
Email: ⇤rparada@dei.unipd.it, †zorzi@dei.unipd.it
Abstract—By 2020, the global mobile data traffic will reach
30.6 exabytes per month. Hence, microwave bands will become
saturated and insufficient to deliver that increment of data.
Millimeter wave (a.k.a. mmWave) is a promising band, from
30 to 300 GHz, to allocate that data. A drawback is the high
attenuation in non-line-of-sight scenarios because the millimeter
wavelengths are blocked by common obstacles. Opposite to Long
Term Evolution (LTE) networks which transmit in an isotropic
manner, mmWave small base stations (SBS) have to transmit
through directional antennas to achieve a sufficient signal to
noise ratio within a radius of up to 200 meters. Moreover, current
SBSs do not adapt to the environment to increase their efficiency,
since they are configured manually. In this paper, we propose a
learning Weight-based Algorithm to decrease the delay of finding
UEs by autonomously prioritizing those sectors where more users
are expected to be found according to previous experience. Our
results show an increment of the number of UEs found in the
first scan by over 19% and a delay reduction by over 84%, on
average for all SBS-UE distances, with respect to comparable
state-of-the-art approaches.
I. INTRODUCTION
By 2020, the global mobile data traffic will reach 30.6 ex-
abytes per month by a compound annual growth rate (CAGR)
of 53 percent [1]. In this scenario, the traditional microwave
cellular bands will become saturated and insufficient to deliver
that increment of data annually. Therefore, in the last few
years researchers have been investigating frequency bands in
the millimeter wave (mmWave) spectrum, i.e., from 30 to 300
GHz [2]-[3]. Two advantages of millimeter wave are the high
bit rate and the small antenna dimension [4]. Unfortunately, a
drawback is the high attenuation in non-line-of-sight (NLOS)
scenarios because the millimeter wavelengths are blocked by
common obstacles such as trees, buildings and pedestrians [5].
In a heterogeneous deployment, common Macro base stations
from typical Long Term Evolution (LTE) networks will be
complemented by small cells to provide the desired Gbps
speed rate. Opposite to LTE networks, which typically transmit
in an isotropic manner, mmWave small base stations (SBS)
have to transmit through directional antennas to increase the
signal to noise ratio (SNR), in order to be able to cover a
range of up to 200 meters. Since each small cell will reach
a radius of 200 meters, and considering the blockage caused
by obstacles, hundreds of small cells will be needed to cover
large areas, so that the cell search procedure [6] will be an
important issue in mmWave.
In LTE networks, macro cells search for new users in an
isotropic manner where all user equipment (UE) surrounding
the macro base station will receive the information to establish
a connection. However, since small cells in millimeter wave
networks are equipped with directional antennas, multiple
beams are required to cover the 360-degree space. Further-
more, there is a trade-off between the beamwidth of the
antenna beam (expressed in degrees) and the coverage range
(expressed in meters), whereby the wider the beamwidth the
fewer the antennas needed but the shorter the distance covered.
In addition, current cellular networks are not flexible, and
manual configuration is required to optimize them. Neverthe-
less, manual configuration of hundreds of small cells means a
huge cost in terms of time and money, and is not flexible to
changes. Therefore, self configuration of the network based on
the environment will provide flexibility and, as a consequence,
saving in manual operations and better performance.
The main problem about the cell search procedure in
millimeter wave is the delay until a UE is found. The narrower
the beamwidth, the higher the number of sectors to search
and, as a consequence, the higher the delay to scan the
360-degree space. Static approaches such as exhaustive and
iterative initial access procedures have been previously studied
[6]. However, we believe that using dynamic schemes based
on historical learning the initial access delay will decrease.
Hence, our overall goal is to reduce the delay of the cell
search procedure by avoiding fixed obstacles like buildings,
and prioritizing sectors according to the expected number of
users based on past experience. Specifically, we provide the
following contributions:
•A method based on weights determined according to
previous connections per sector.
•A comparison of misdetection probability and delay met-
rics with state-of-the-art approaches against our proposed
method.
The remainder of this paper is organized as follows: Section
II introduces the problem motivation and surveys the related
state of the art. The principle of the Weight-based cell search
procedure is explained in Section III. Simulation experiments
and results are described in Section IV. Finally, we conclude
the paper in Section V.
155
European Wireless 2017
ISBN 978-3-8007-4426-8
© VDE VERLAG GMBH ∙ Berlin ∙ Offenbach
II. STATE O F T H E ART
The problem of cell discovery in mmWave networks has
been studied by Jeong et al. [7], where a comparison between
omnidirectional and directional search is discussed. Barati
et al. [8] introduced the concept of Generalized Likelihood
Ratio Test (GLRT) where base stations periodically transmit
synchronization signals to establish communication. Capone
et al. presented cell discovery schemes based on location
information [9].
Giordani et al. compared different cell discovery schemes
[6] to find the best trade-off between delay and coverage.
These cell discovery schemes are:
•Exhaustive scheme: where all sectors are scanned to
find UEs. This scheme provides better coverage by using
narrow beams. However, narrow beams result in high
delay in covering the whole space since more sectors have
to be scanned.
•Iterative scheme: where different beamforming schemes
are generated to decrease the delay. This scheme reduces
the delay during cell discovery by fast generating wider
sectors while reducing the distance of coverage, but may
not discover UEs located farther from the Base Station
(BS). Nevertheless, once a UE is discovered, narrower
sectors can be used to increase the precision about the
UE’s position and increasing the SNR for a better channel
estimation and transmission rate between BS and UE.
•Context information based scheme: where Global Posi-
tioning System (GPS) coordinates from the closest small
base station are provided by the macro cell to the UE.
The UE also gets its GPS information to steer the beam
to the closest small base station. This approach reduces
the delay but increases the energy consumption.
Even though these approaches provide mechanisms to de-
crease the delay during the cell search procedure, they do not
take into account the presence of obstacles commonly found
in real scenarios (i.e., buildings), nor dynamic UEs and ob-
stacles increasing the chances of signal blockage. Thus, since
mmWave transmissions are blocked by obstacles, Capone et
al. introduced a learning approach in [10] to mitigate the
discovery time. However, this approach does not take into
consideration the number of users in a given sector which will
highly influence the average discovery time. Abbas and Zorzi
[11] present an analog beamforming receiver architecture
based context information to achieve a similar performance as
hybrid beamforming. Nevertheless, this work focuses uniquely
on the physical beamforming aspect of the cell search. Instead,
we propose a light and autonomously calibrated technique,
called Weight-based Algorithm, in which the SBS learns from
the environment to predict the next action and improve the
efficiency in dynamic scenarios.
III. WEIGHT-BAS ED CELL SEARCH PRINCIPLE
MmWave signals are blocked by obstacles because of their
short wavelength. Therefore, transmissions towards obstacles
will not reach users located behind them. Figure 1 represents
a scenario where one SBS (SBS1) is scanning the 2D space
to establish connection with UEs. SBS1 scans the space using
beams with 45owidth, thus, eight sectors are required to cover
the 360-degree space. The example in Figure 1 introduces the
Fig. 1. Cell search procedure with obstacles
idea of an intelligent SBS, which only steers its beams through
a known populated area for UEs instead of wasting time and
energy emitting towards a building. Nevertheless, current SBSs
are not capable of deciding which areas to cover or not,
nor which directions are more likely to contain many UEs.
Thereby, we present the Weight-based Algorithm, in which the
order of emitted sectors from the SBSs depends on previous
experience (learning) according to the number of detected
UEs. By implementing intelligent algorithms in the SBSs, the
delay in detecting UEs during the cell search procedure will
decrease and the number of UEs found will increase in the
first scan. We define the first scan as the sector where the
cell search procedure is initiated. Our learning Weight-based
algorithm, by prioritizing sectors with high percentage of UEs,
will detect a higher number of users in the first scan, reducing
the latency when a UE receives a signal from the base station.
The procedure is described in Algorithm 1. First, we initial-
ize a matrix with the number of sectors and the number of UEs
found for each sector (1). Line 2 defines the number of UEs
generated during the simulation, locating them according to a
random Angle of Departure (3) and Angle of Arrival (4) and
distance with respect to the SBS (5), in meters. The algorithm
is user-configurable, where the user can define the number of
antennas for SBS (6) and UE (7). Depending on the number of
antennas, the beamwidth will vary. Thus, a higher number of
antennas will emit a thinner beam than obtained using fewer
antennas. Then, the beamwidth for SBS atx and UE arx is
defined in lines 8 and 9, respectively. Hence, the number of
sectors required to cover the whole 360ospace will depend
on those variables, resulting in stx (10) and srx (11) sectors.
For preliminary results, we consider a 2D space. Initially, we
assume that the SBS transmits sequentially (12), initializing a
vector vwith this order. The UE also transmits sequentially,
however, it does not vary the sequence since the algorithm
is based uniquely in the SBS. A user-configurable number of
simulations sim are set up (13). for (14) runs all iterations
until reaching the value of sim. In line 15, the function for
scans all the UEs defined in line 2. The function for, in line
16, processes all the sectors from the SBS, and line 17 those
sectors from the UE. Line 18 performs the given calculations
156
European Wireless 2017
ISBN 978-3-8007-4426-8
© VDE VERLAG GMBH ∙ Berlin ∙ Offenbach
Algorithm 1 Weight-based Algorithm
1: procedure UESDETECTED(X).Matrix with number of
UEs found per sector
2: set nnumber of UEs
3: AoD = runif(n, 0.0, 2*⇡).Assign randomly Angle
of Departure
4: AoA = -AoD .Assign Angle of Arrival
5: distance = sqrt(runif(n, 0.0, 150.02)) .Assign
randomly distance SBS-UE
6: set ntx number of transmit antennas (SBS)
7: set nrx number of receive antennas (UE)
8: set atx angle sector SBS .Beamwidth - based on ntx
9: set arx angle sector UE .Beamwidth - based on nrx
10: set stx number sectors SBS .360÷atx
11: set srx number sectors UE .360÷arx
12: set vvector sector order .Initially 1,2,3...s
13: set sim number of iterations
14: for qin 1: sim do
15: for hin1:ndo
16: for iin1:stx do
17: for jin1:srx do
18: Perform channel model operations [12]
19: if SNRij >Threshold then
20: X[i,UE found+1]
21: order X.Descending with respect to number of
found UEs
22: return v=X[1:stx,] .set new order for next
round
from a channel model [12], to obtain the resulting SNR. If
the SNR value is higher than a given threshold (see Table
I) (19) we increase the number of UEs found for that given
sector iby 1 unit (20). After scanning all UEs and sectors, the
matrix Xwill be ordered depending on the number of UEs
found per sector (21). Therefore, the new vector vwill be
updated according to the number of UEs found in each sector
in descending order. Thus in the next round, the SBS will
start emitting following the new order established in vector
v. Vector vwill be updated until the simulation ends in the
Function for, in line 14. We define as an iteration the fact
that the SBS and the UEs have scanned the whole 360-degree
space. The Weight-based Algorithm is repeated in each Round.
We defined the concept of Rounds to let the Weight-based
Algorithm collect enough data for re-ordering the emitting
vector properly.
Referring to the algorithm complexity, from prior ap-
proaches such as exhaustive and iterative, the weight-based
algorithm does not introduce any heavy complexity. The only
functionality added is the insertion of values into a matrix
O(1) and sorting it using the radix sorting algorithm O(nk) in
descending order, in the worst case [13]. Hence, considering
the overall complexity of our algorithm the added complexity
is low. As to the possible memory constraints to store the
matrix inside the SBS, a 16x2 matrix type is about 2.2 KBytes
(using function object.size in R). Since the matrix is created
independently in each SBS, our Weight-based Algorithm is
highly scalable. This work differs from the one proposed
in [10] because of the memory learned. While they keep
track of user positions and beamforming parameters of all
successful discoveries, our approach only considers the sectors
identification, reducing the memory size.
The intuition behind the Weight-based Algorithm is based
on the number of users found in each sector along each round.
Hence, the higher the number of UEs in a certain sector,
the higher the priority to transmit earlier in that sector in
the next round. Figure 2 introduces the concept of Weight-
based Algorithm with a graphical example. Three scenarios
are presented: exhaustive (A), iterative (B) and the proposed
Weight-based Algorithm (C). Figure 2 presents the config-
uration of the transmission order along three Rounds (I,II
and III). Each SBS scans the whole 360-degree space through
eight sectors (S1-S8), with the exception of scenario Bwhich
performs a hierarchical scan. In addition, each sector contains
a number (in red) indicating the order to transmit from 1 to 8
(4 in scenario B), being 1 the first and 8 (or 4) the last. The
goal of the SBS is to find UEs as quickly as possible, thereby
decreasing the discovery delay. Note in Figure 2, how the SBS
in both scenarios Aand Bdoes not detect the obstacle located
in sector S4 (or part of S2) and still transmits through this
sector. Furthermore, along the rounds the SBS behavior does
not change and the transmitting order still follows a sequential
order. Conversely, the SBS in scenario C, with the Weight-
based Algorithm, has detected the building in Round Iand
self-configures to avoid this blocked sector in the following
rounds (II &III). Also, the SBS transmits depending on the
UEs found in Round II and autonomously calibrates in the
following rounds (III). Thus, each round is considered as the
calibration time where the SBS learns from the environment.
Moreover, in scenario B, because of its wider beam, the range
is shorter, missing some UEs located farther.
From Figure 2 we can observe how the Weight-based
algorithm (scenario C), prioritizes those sectors with higher
probability. The main function which allows the intelligent
characteristic of our approach is written in lines 20 and 21 of
Algorithm 1. As previously explained, line 20 uniquely inserts
an incremented value inside a matrix and line 21 sorts the
matrix in descending order. Therefore, the first position of the
matrix will correspond to the beam with the highest number
of UEs found. Thus, in the next round, the order in steering
beams will be performed based on that historical matrix from
all previous rounds. Nevertheless, this matrix is updated in
following rounds to optimize the prioritization of the beams.
IV. SIMULATION EXPERIMENTS AND RESULTS
This section presents the simulation experiments performed
using the described Algorithms 1 and the obtained results.
A. Setting up the environment
Algorithm 1 has been implemented in the statistical environ-
ment R [14] because of its extensive collection of packages to
perform all kinds of simulations, including machine learning
157
European Wireless 2017
ISBN 978-3-8007-4426-8
© VDE VERLAG GMBH ∙ Berlin ∙ Offenbach
S1
S2
S3
S4S5
S6
S7
S8
1
2
3
4
5
6
7
8S1
S2
S3
S4S5
S6
S7
S8
1
2
3
4
5
6
7
8
S1
S2
S3
S4S5
S6
S7
S8
1
2
3
4
5
6
7
8S1
S2
S3
S4S5
S6
S7
S8
1
3
5
2
6
4
7
A - EX HAUSTI VEC - WEIGHT
S1
S2
S3
S4S5
S6
S7
S8
1
2
3
4
5
6
7
8
S1
S2
S3
S4S5
S6
S7
S8
2
3
5
1
7
6
4
III III
III III
S1
S2
S3
S4
1
2
3
4
S1
S2
S3
S4
1
2
3
4
B - ITERATI VE
S1
S2
S3
S4
1
2
3
4
III III
Fig. 2. SBS emission order with and without Weight-based Algorithm
techniques. Table I lists the parameters used in the performed
simulations. Those parameters in Table I are considered to
TABLE I
SUMMARY OF THE PARAMETERS USED IN THE EXPERIMENTAL SETUP.
Parameter Value Description
sim 100 Number of iterations
n 10 Number of UEs
BW 1 GHz System bandwidth
DL-ULPTX 30 dBm Transmission power
NF 5 dB Noise figure
fc 28 MHz Carrier frequency
distance 0-150 m Cell radius
⌧-5 dB SNR threshold
SBS antenna 2 x 2, 4 x 4 or 8 x 8
UE antenna 2 x 2
SBS position (0, 0) m
UEs position varied Uniform in annulus
UEs speed 0 m/s No mobility
BF analog Beamforming architecture
Scan delay per sector 100µs
Propagation model LOS, NLOS, outage According to [12]
simulate a realistic system. The channel model implemented
in our paper is based on real-world measurements performed
in New York City at 28 GHz providing realistic mmWave-
propagation condition and loss. This model provides statistical
parameters to obtain three possible situations: line-of-sight
(LOS), non-line-of-sight (NLOS), and outage. Further details
of the channel model and its parameters can be found in [12].
According to the channel characteristics, the SBS cell radius
can reach between 100 and 200 meters. We consider a maxi-
mum SBS-UE distance of 150 meters in our results. Since the
mmWave 5G base stations transmit through directional beams
by implementing beamforming, we use analog beamforming
through a Uniform Planar Array (UPA). The array can consist
of 8x8,4x4and 2x2elements. We established the parameter
”Scan delay per sector” to obtain the total delay in a realistic
situation. 10 UEs were generated to better compare the impact
of implementing our Weight-based Algorithm. The SBS is
placed at coordinates (0,0) m. The UEs are generated, at each
iteration, within an annulus inside the given cell radius and a
distance in 5 m increments (from 0 to 150 m) from the SBS.
At each iteration, the SBS will steer its beams to find a user.
We assume that the UE is detected when the SNR computed
between SBS and UE is above ⌧=5dB. Otherwise, the UE
will be considered out of range. Each distance is performed
100 times with different seeds for statistical confidence.
In our study, we evaluate the performance of our proposed
learning algorithm with respect to state-of-the-art approaches
in terms of number of UEs found in the first scan, average
delay among all UEs, and misdetection probability.
B. Number of UEs found in First Scan
SBSs serve the network by providing connectivity to the
UEs located around them. In 3GPP-LTE, the eNB scans
continuously the space in an isotropic manner covering the
whole 360-degree space in a single scan. However, with
mmWave, the higher the time the SBS requires to find a UE
the less efficient the SBS because of the delay in scanning
the whole 360-degree space through directional beams. For
instance, if most UEs are located in a particular area and
this area is scanned first, the SBS can establish earlier the
appropriate connections. Therefore, we define as ”First Scan”
the first steered beam performed by the SBS when it has
initiated the cell search. Within the First Scan are included
all the sectors scanned by the UE in a single round. Hence,
those found UEs which meet the threshold, will be considered.
For instance, if the UE has a system configuration of 2x2
elements, it will scan the whole 360-degree space with four
sectors, thus, the First Scan in this case will be four sectors
from the UE’s point of view.
Figure 3 shows the percentage of UEs found (y-axis) vs.
the distance BS-UE, in meters (x-axis). We evaluated the three
described algorithms: exhaustive, iterative and weight-based.
Thus, in Figure 3, we represented the exhaustive 8x8(red-
solid line), weight 8x8(black-dashed), iterative 8x8(blue-
dotted line), exhaustive 4x4(green-dotdashed line), weight
4x4(orange-longdashed line) and iterative 4x4(brown-
twodashed line).
We can observe how all six lines start at 50% UEs found
and decrease with the distance. They start with 50% of UEs
found because of the back lobe cancellation, meaning that we
only consider front and side lobes from the array of antennas
within a 180-degree range, hence, since the UEs are uniformly
158
European Wireless 2017
ISBN 978-3-8007-4426-8
© VDE VERLAG GMBH ∙ Berlin ∙ Offenbach
0 50 100 150
0.0 0.2 0.4 0.6 0.8 1.0
Distance SBS−UE [m]
Percentage UE Found
Exhaustive 8x8
Weight 8x8
Iterative 8x8
Exhaustive 4x4
Weight 4x4
Iterative 4x4
Fig. 3. The weight algorithm detects a higher number of UEs with respect
to the state-of-the-art approaches
distributed along the 360-degree space, 50% of these UEs
will be within the beam. The Weight-based algorithm with
both the 8x8and the 4x4elements configuration performs
better than the state-of-the-art schemes. The Weight-based
algorithm, considering a system of 8x8elements, outperforms
by 18% and 20% the state-of-the-art iterative and exhaustive
algorithms, respectively. With a system of 4x4elements,
the improvement is over 8% and 19% with respect to the
iterative and exhaustive scheme, respectively. These results
were calculated averaging the increment of the number of UEs
found for all distances. However, we obtained a maximum
improvement over 40% and 50% with the configuration 8x
8elements at a distance SBS-UE of 150 m for the iterative
and exhaustive schemes, respectively. In case of a system
configured with 4x4elements, the increment of the number
of UEs found is over 22% and 41%, with a distance SBS-UE
of 150 m, compared to the iterative and exhaustive algorithms.
This increment of the number of UEs found in the first scan
by the weight-based algorithm is due to the prioritization of
the beams once the SBS is calibrated (i.e., has learned from
historical experience).
C. Average Delay
A key characteristic over all network communications is the
delay (or latency from the user’s point of view). In our study,
the delay is computed as the interval of time, in µs, between
when the SBS steers the first beam and when the UE receives
the signal. The signal received should have an SNR above ⌧.
Figure 4 shows the average delay (y-axis) in µs vs. the BS-
UE distance, in meters (x-axis). We evaluated the exhaustive
and weight-based algorithm. Note that the iterative scheme is
0 50 100 150
0 500 1000 2000 3000
Distance SBS−UE [m]
Average Delay [µs]
Exhaustive 8x8
Weight 8x8
Exhaustive 4x4
Weight 4x4
Fig. 4. Overall, the weight scheme reduces the delay
not considered in this evaluation because it performs worse
than the exhaustive scheme [6]. Thus, in Figure 4, we rep-
resented the exhaustive 8x8(red-solid line), weight 8x8
(black-dashed), exhaustive 4x4(green-dotdashed line) and
weight 4x4(orange-longdashed line).
We can observe how for all the distances (x-axis), the
weight-based Algorithm performs better than the exhaustive
scheme, by 85% and 21% with the 8x8and 4x4elements
configuration, respectively. Note that previous results are the
average reduction of delay from all distances. Nevertheless,
we achieved a maximum reduction of delay, by 169% and
33%, at a distance SBS-UE of 15 meters, with the 8x8and
4x4element configurations, respectively.
D. Misdetection probability
The efficiency of a SBS can be evaluated based on the
percentage of successful connections with users during cell
search. We define the misdetection probability as the proba-
bility that a UE, located within the cell of the SBS when it
scanned the whole 360o-space, is not detected, i.e., its SNR is
below threshold.
Figure 5 compares the misdetection probability (y-axis) for
the presented schemes: exhaustive 8x8(red-solid line), weight
8x8(black-slashed line), exhaustive 4x4(green-dotdashed
line) and weight 4x4(orange-longdashed line) vs. the distance
SBS-UE, in meters, along the x-axis.
We can appreciate how, as expected, the Weight-based
algorithm has a similar performance, in terms of misdetection
probability, to the exhaustive scheme with both the 8x8and 4
x4element configurations. Therefore, our approach detects the
same number of UEs as the state-of-the-art schemes, indicating
its good performance.
159
European Wireless 2017
ISBN 978-3-8007-4426-8
© VDE VERLAG GMBH ∙ Berlin ∙ Offenbach
0 50 100 150
0.0 0.2 0.4 0.6 0.8 1.0
Distance SBS−UE [m]
Misdetection Probability
Exhaustive 8x8
Weight 8x8
Exhaustive 4x4
Weight 4x4
Fig. 5. The Weight-based algorithm reduces the misdetection probability with
respect to the state-of-the-art exhaustive scheme.
As a conclusion, implementing intelligence in the SBS
results in a reduction of the delay (latency from the UE’s point
of view) and an increment of the number of UEs found in the
first scan during the cell search procedure, with no degradation
in the misdetection probability.
V. C ONCLUSION AND FUTURE WORK
In this paper, we presented the Weight-based Algorithm
which prioritizes the sectors to scan according to previous
experience. The SBS is calibrated autonomously based on past
rounds, thereby increasing the efficiency in finding UEs with
the first scan and reducing the discovery time.
More specifically, we simulated a medium populated sce-
nario where ten UEs are randomly located surrounding a
SBS, to observe the impact of using our algorithm compared
to the state-of-the-art exhaustive and iterative schemes. Our
simulation experiments returned promising results with a delay
reduction over 84% (in average for all distances). Moreover,
we increased the number of UEs found in the first scan by over
19% (averaged on all distances), speeding up the cell search
procedure and increasing the efficiency of the SBS. Further-
more, we observe how our intelligent cell search approach can
be configured in any kind of scenario. The novelty of this work
resides in the implementation of a learning algorithm based
on the historical location of users for future prediction. This
paper introduced a promising technique to reduce the delay
in cell search procedure, where our preliminary evaluation
returned interesting results. A simple scenario was configured
to perform initial exploratory evaluations and to verify the
reliability of our proposed method. According to the results
returned in our simulations, the Weight-based Algorithm will
gain in more complex scenarios with more SBSs and UEs.
As future work, we plan to enhance our proposed Weight-
based Algorithm by extending the simulation scenario with
multiple SBSs, adding transmission error ratios, considering
static/dynamic obstacles and mobile UEs, simulating a large
population of UEs that go in and out of the SBS’s range, and
considering more sophisticated learning techniques.
REFERENCES
[1] Cisco Systems, “Cisco Visual Networking Index: Global
Mobile Data Traffic Forecast, 2015–2020,” 2016, (Accessed
Apr. 2017) Available at: http://www.cisco.com/c/
en/us/solutions/collateral/service-provider/
visual-networking- index-vni/vni- forecast-qa.pdf.
[2] Z. Pi and F. Khan, “An introduction to millimeter-wave mobile broad-
band systems,” IEEE Communications Magazine, vol. 49, no. 6, pp.
101–107, June 2011.
[3] F. Al-Ogaili and R. M. Shubair, “Millimeter-wave mobile communica-
tions for 5G: Challenges and opportunities,” in 2016 IEEE International
Symposium on Antennas and Propagation (APSURSI), June 2016, pp.
1003–1004.
[4] S. Rangan, T. S. Rappaport, and E. Erkip, “Millimeter-wave cellular
wireless networks: Potentials and challenges,” Proceedings of the IEEE,
vol. 102, no. 3, pp. 366–385, March 2014.
[5] T. Rappaport, R. Heath, R. Daniels, and J. Murdock, Millimeter
Wave Wireless Communications, ser. Prentice Hall Communications
Engineering and Emerging Technologies Series from Ted Rappaport.
Pearson Education, 2014. [Online]. Available: https://books.google.it/
books?id=0Jh6BAAAQBAJ
[6] M. Giordani, M. Mezzavilla, and M. Zorzi, “Initial Access in 5G
mmWave Cellular Networks,” IEEE Communications Magazine, vol. 54,
no. 11, pp. 40–47, November 2016.
[7] C. Jeong, J. Park, and H. Yu, “Random access in millimeter-wave
beamforming cellular networks: issues and approaches,” IEEE Commu-
nications Magazine, vol. 53, no. 1, pp. 180–185, January 2015.
[8] C. N. Barati, S. A. Hosseini, S. Rangan, P. Liu, T. Korakis, S. S. Panwar,
and T. S. Rappaport, “Directional Cell Discovery in Millimeter Wave
Cellular Networks,” IEEE Transactions on Wireless Communications,
vol. 14, no. 12, pp. 6664–6678, Dec 2015.
[9] A. Capone, I. Filippini, and V. Sciancalepore, “Context Information for
Fast Cell Discovery in mm-wave 5G Networks,” in Proceedings of the
21th European Wireless Conference, May 2015, pp. 1–6.
[10] A. Capone, I. Filippini, V. Sciancalepore, and D. Tremolada, “Obstacle
avoidance cell discovery using mm-waves directive antennas in 5G
networks,” in IEEE 26th Annual International Symposium on Personal,
Indoor, and Mobile Radio Communications (PIMRC), Aug 2015, pp.
2349–2353.
[11] W. B. Abbas and M. Zorzi, “Context information based initial cell
search for millimeter wave 5G cellular networks,” in 2016 European
Conference on Networks and Communications (EuCNC), June 2016,
pp. 111–116.
[12] M. R. Akdeniz, Y. Liu, M. K. Samimi, S. Sun, S. Rangan, T. S.
Rappaport, and E. Erkip, “Millimeter Wave Channel Modeling and
Cellular Capacity Evaluation,” IEEE Journal on Selected Areas in
Communications, vol. 32, no. 6, pp. 1164–1179, June 2014.
[13] E. Rowell, “Big-O Cheat Sheet,” 2013, (Accessed Apr. 2017) Available
at: http://bigocheatsheet.com/.
[14] R. Ihaka and R. Gentleman, “The R Project for Statistical Computing,”
1993, (Accessed Apr. 2017) Available at: https://www.r-project.org/.
160
European Wireless 2017
ISBN 978-3-8007-4426-8
© VDE VERLAG GMBH ∙ Berlin ∙ Offenbach