You are on page 1of 5

2422 IEEE COMMUNICATIONS LETTERS, VOL. 23, NO.

12, DECEMBER 2019

Greenfield Design in 5G FWA Networks


Omar Kaddoura , José Outes-Carnero, Juan A. García-Fernández , Rocio Acedo-Hernández ,
Maria Cerón-Larrubia, Lorena Ríos, Juan J. Sánchez-Sánchez , and Raquel Barco

Abstract— With the arrival of 5G, Fixed Wireless Access, Until the present, FWA has been discarded as an alternative
which aims at providing households with broadband internet to fixed technologies since wireless technologies were not able
access using mobile technology, has become a relevant use case. to reach data rates provided by optical fiber. Nevertheless, with
Current greenfield design techniques are not optimal since they
target at optimizing the radiofrequency configuration for the
the advent of 5G and the standardization of New Radio (NR),
whole household, sometimes at the expense of other households. data rates that wireless technologies may reach can compete
However, optimizing only the spot where to install the antenna with those provided by fixed technologies. This is mainly due
of the customer premises equipment is enough for Fixed Wire- to two factors: better spectral efficiency achieved thanks to the
less Access networks. This letter presents an enhancement of appearance of smart antennas and the increase of the available
automatic greenfield design methodology so that, apart from bandwidth thanks to the use of millimeter waves [3].
finding the location where to deploy network base stations and
the optimal configuration of network parameters, the optimal Particularly, the deployment of FWA networks is relevant
location where to deploy customer premises equipment is also in suburban and rural areas, where the distances between
determined. A greenfield design exercise has been carried out in a homes are larger than in urban areas, so that the amount of
real scenario proving that the proposed method both improves the cable required to provide homes with broadband services is
final network performance and reduces the required deployment higher. In addition, suburban and rural areas are favorable to
economic costs compared with state-of-the-art methodology.
millimeter waves, since these require the absence of obstacles
Index Terms— Greenfield design, FWA, ACP, radio network
planning, radio network optimization. between transmitter and receiver [4]. The utilization of mil-
limeter waves also requires that FWA CPEs have their antennas
I. I NTRODUCTION mounted on the rooftops of the households and connected to

O VER the years, internet service providers have made


efforts to bring broadband services to homes and
increase service data rates. Nowadays, with regards to the
indoor 5G Wi-Fi routers.
In order to minimize the costs required to deploy the
infrastructure of a network, a proper planning exercise is
last mile of connectivity between the network architecture needed. This planning consists in selecting the locations where
and homes, the technologies that enable broadband internet to install the network elements and designing their correspond-
access can be classified into two main groups: fixed and ing configuration to maximize the final network performance.
wireless. In this context, fixed technologies use cables and A planning exercise with no legacy networks is known as
wireless technologies use air interface. Among these, fixed greenfield design. Greenfield design activities aim at finding
technologies have predominated over wireless technologies to the best network configuration to satisfy the demand of
bring broadband internet access to homes since they achieve users while minimizing the cost of the network deployment.
higher service data rates. Greenfield design is composed of two main and independent
Especially, the growing demand for multimedia services has stages: radio network planning (i.e., selecting the proper
been satisfied with the arrival of optical fiber, which has been locations where to install the base stations) and radio network
introduced into homes thanks to technologies like Fiber To The optimization (i.e., selecting appropriate values for network
Home (FTTH) [1]. However, bringing optical fiber to homes parameters like power allocation, height of antennas, azimuth
is expensive due to the cost associated with infrastructure of antennas, tilt of antennas, mobility thresholds, …).
deployment. Because of this, wireless technologies, which This letter proposes the enhancement of existing greenfield
eliminate the need of cabling, are considered as the best design methods so that they can also be applied to FWA
alternative to fixed technologies. Fixed Wireless Access (FWA) networks. In addition, in this letter, the optimality of the results
is one of the architectures designed to bring broadband internet of state-of-the-art methods is improved by adding an itera-
access to homes [2]. FWA consists in replacing the last mile tive algorithm which combines both radio network planning
of cables by air interface so that a single network transceiver and radio network optimization stages. Moreover, the pro-
provides broadband service to multiple homes provided with posed method suits operators’ demands to efficiently per-
specific Customer Premises Equipment (CPE). form greenfield design activities thanks to employing greedy
Manuscript received August 13, 2019; accepted September 1, 2019. Date of
algorithm.
publication September 5, 2019; date of current version December 10, 2019. This letter is organized as follows. Section II reviews state-
The associate editor coordinating the review of this letter and approving it for of-the-art greenfield design techniques for mobile communi-
publication was H. Zhang. (Corresponding author: Omar Kaddoura.) cation networks and highlights the problem of using these
O. Kaddoura and R. Barco are with the Department of Com-
munication Engineering, University of Malaga, 29071 Málaga, Spain to design FWA networks. In Section III, a greenfield design
(omar.kaddoura@ericsson.com). method customized for FWA networks is described. Finally,
J. Outes-Carnero, J. A. García-Fernández, R. Acedo-Hernández, Section IV addresses the goodness of the proposed method for
M. Cerón-Larrubia, L. Ríos, and J. J. Sánchez-Sánchez are with Ericsson AB,
29590 Málaga, Spain. greenfield design of NR FWA networks when applied to a real
Digital Object Identifier 10.1109/LCOMM.2019.2939470 suburban environment.
1558-2558 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: Universidad Nacional de Colombia (UNAL). Downloaded on April 06,2023 at 02:38:28 UTC from IEEE Xplore. Restrictions apply.
KADDOURA et al.: GREENFIELD DESIGN IN 5G FWA NETWORKS 2423

II. R ELATED W ORK AND P ROBLEM F ORMULATION According to the state of the art, greenfield design methods
are based on discrete population models for the traffic descrip-
FWA greenfield design methodology aims at finding the best
tion [11]. In these models, the locations where users can be
combination of deployed base stations, CPEs antennas location
found are discretized into bins known as demand nodes. This
and network parameter configuration. Therefore, finding the
discretization allows evaluating the performance of a given
best network configuration is translated into a combinatorial
network configuration as the aggregation of the performance
optimization problem.
in all the demand nodes into which the network is divided.
Among the total number of candidate locations L where to
Nevertheless, as analyzed by authors in [12], this model is
deploy base stations, any subset of l base stations corresponds
not valid in FWA networks, since the location of the CPE
to a possible network configuration. Therefore, the size of
antenna is unknown in advance. This is, the traffic demanded
the search space of all possible base stations configurations
by a given household cannot be assigned to a specific demand
is represented as follows:
node. In fact, since automatic greenfield design techniques
L  
L for FWA networks must find not only the optimum location
(1)
l and configuration of network equipment but also the best
l=0
spot where to install the CPE antenna, a new methodology
Regarding the configuration of parameters, being P the total is required. Moreover, radio network planning and radio net-
number of parameters to be configured and Npl the number of work optimization are executed as two independent phases
possible values that parameter p in base station l may have, of current greenfield design techniques. However, they could
the size of the search space of parameter configuration is be combined so they cooperate with each other to find the
described as follows: optimum network configuration.
P Therefore, the next section presents a modification of ACP
Npl (2) methods so that the location of CPEs antennas is also opti-
p=1 mized. In addition, next section also presents an iterative
algorithm which combines both radio network planning and
In conventional networks, (1) and (2) can be combined
radio network optimization to enhance state-of-the-art green-
to calculate the size of the search space. However, in FWA
field design techniques for FWA networks.
networks, the combinatorial problem is increased since the
possible location of CPEs antennas must be taken into con- III. P ROPOSED M ETHOD
sideration. Thus, being Nk the number of demand nodes in The automatic method herein described allows performing
household k and being K the total number of households, greenfield design activities in FWA networks. The proposed
the size of the search space of possible locations where install method finds proper locations where to install network trans-
network CPEs antennas is represented as follows: ceivers (base stations and CPEs antennas) as well as it opti-
K
mizes the configuration of network parameters.

Nk (3) A. FWA Cost Function
k=1 Being S a network state (i.e., a list of active cells together
with the parameter values assigned to each cell), automatic
Therefore, the size of the search space of FWA greenfield
radio network planning and network optimization methods
design is as follows:
⎡ l ⎤ K aim at finding the optimal network state Sbest . This is,
 L    P 
⎣ L the minimum number of base stations and the optimal config-
Npl ⎦ Nk (4) uration of network parameters which maximize the network
l p=1
l=0 k=1 performance. For this, these methods define cost functions to
describe a figure of merit of the network performance. Thus,
Since the evaluation of the full search space in mobile
the classical optimization problem is reduced to finding Sbest
communication networks is not feasible, initially, greenfield
which minimizes the cost function Φ (S) [5], [13]:
design was carried out by skilled engineers in the field of
Radio-frequency (RF), who evaluated the impact of different Sbest = argmin (Φ (S)) (5)
network configurations over the final network performance. S
N
 M

However, the great success of mobile technologies boosted
the growth of these networks so that manual design became Φ (S) = αi ωj δij (S) (6)
i=1 j=1
unfeasible. Therefore, to overcome the previous, computer
assisted methods known as Automatic Cell Planning (ACP) [5] 0, if objective for performance indicator is met
arose and played a crucial role. δ (S) =
1, otherwise
Regarding the automation of radio network planning, in [6],
(7)
three primitive alternatives were presented to find optimum
base station locations. However, over time, new technologies where N is the number of demand nodes in the network and
were standardized and radio network planning was modified M is the number of performance indicators (i.e., coverage,
to consider interference [7], throughput [8] or cell loading [9]. capacity, quality, …). The traffic demanded in demand node i
With respect to radio network optimization, metaheuristic is represented as αi . The relevance of performance indicator
algorithms as well as artificial neural networks were proposed j is indicated by weight ωj . δij (S) is the value of function
to find the best configuration of network parameters [10]. δ (S) for performance indicator j at demand node i.

Authorized licensed use limited to: Universidad Nacional de Colombia (UNAL). Downloaded on April 06,2023 at 02:38:28 UTC from IEEE Xplore. Restrictions apply.
2424 IEEE COMMUNICATIONS LETTERS, VOL. 23, NO. 12, DECEMBER 2019

Unlike classical greenfield design, finding the optimum CPE Algorithm 1 Automatic FWA Greenfield Design
antenna location is part of greenfield design in FWA networks. Input : Households demand nodes location
Therefore, final locations of households’ demand nodes are Candidate CPE antenna locations
unknown in advance. To overcome this, a new automatic Weights and objectives
algorithm needs to described so that the cost function is Output: Best network configuration Sbest
1 C ← ΦFWA (Sinitial ) // Initial cost
influenced only by the best CPE antenna location found in 2 S ← Sinitial // Initial network state
every household. This letter proposes a modification of the 3 repeat
cost function used to evaluate the network performance. Thus, 4 done ← true
for FWA networks, the cost function in (6) is transformed into /* Radio network planning */
the next: 5 A←∞
L foreach candidate base station do
 6
ΦFWA (S) = αk · min Tk (S) (8) 7 Sbs
cand := S adding candidate base station
k=1 8 B ← ΦFWA Sbs cand
9 if B < A then
where 10 A←B
 
best ← Scand
Sbs bs
Tk (S) = R1k (S) , R2k (S) , . . . , RN
k
k
(S) (9) 11
12 end
and 13 end
M
14 S ← Sbs

best
/* Radio network optimization */
Rik (S) = ωj δij (S) (10) 15 A←∞
j=1 16 foreach candidate value of every parameter do
where L is the number of households in the network, Nk is 17 Spcand := S including candidate value
18 B ← ΦFWA (Spcand )
the number of candidate demand nodes in household k and M 19 if B < A then
is the number of performance indicators. The traffic demanded 20 A←B
by household k is represented as αk . Vector Tk (S) represents 21 Spbest ← Spcand
the collection of costs associated with each candidate demand 22 end
node in household k, and Rik (S) represents the cost of 23 end
S ← Spbest
candidate demand node i in household k. 24
25 D ← ΦFWA (S)
Contrary to other state-of-the-art cost functions Φ (S), FWA 26 if D < C then
cost function ΦFWA (S) relies on a vector Tk (S) which rep- 27 C←D
resents all the candidate demand nodes of a household (i.e., 28 done ← false
the candidate CPE antenna locations). Hence, from all these 29 end
locations, the one providing the lowest cost is considered as 30 until done;
31 return S
the demand node of the entire household and then used in the
evaluation of the system cost function.

Automatic FWA greenfield design is summarized in


B. FWA Greenfield Design Algorithm
Algorithm 1.
As previously explained, greenfield design is composed of
two main stages: radio network planning and radio network IV. R ESULTS
optimization. Commonly, these two phases are carried out The method proposed in this letter was evaluated in a subur-
separately. Firstly, all the locations where to deploy network ban 3x3 km2 area at King County, Washington (United States
base stations are identified. Secondly, optimization algorithms of America). This activity consisted in providing broadband
are used to configure network parameters to achieve superior access to 4520 households using NR as enabling air interface
network performance. technology. To achieve this goal, a total of 1373 lampposts
Unlike other state-of-the-art methods, this letter proposes were identified as candidate locations where to install base
an automatic iterative greedy algorithm which combines stations. Trisector antennas with fixed azimuth separation
both radio network planning and radio network optimization. of 120 degrees were chosen to be deployed at candidate base
In every iteration, firstly the algorithm searches the best station locations. RF spectrum was configured to be deployed
location where to add one new base station and secondly, all in frequency band n257 [14], using 28 GHz carrier frequency
network parameters are optimized taking into consideration the for both DL and UL and 200 MHz bandwidth. Drive tests were
new addition. The algorithm finishes once the cost function carried out to perform RF propagation measurements used to
cannot be improved, returning the best locations where to calibrate Planet® 3D Model [15] so that losses from candidate
deploy CPEs antennas and network base stations as well as base station locations to candidate CPE antenna locations were
the optimum configuration of network parameters. estimated. Finally, drive test data were rasterized in 2D grids
One might think that this algorithm does not guarantee per candidate base station. The demand nodes of the grids
global optimality. However, as stated in [10], since networks represent RF measurements in the rooftops of the households.
are designed to present low cell coupling so that their per- 2x2 m2 was considered as high enough resolution to assume
formance is maximized, greedy algorithm yields very good that propagation conditions will not change inside every
results faster than other metaheuristic algorithms. demand node.

Authorized licensed use limited to: Universidad Nacional de Colombia (UNAL). Downloaded on April 06,2023 at 02:38:28 UTC from IEEE Xplore. Restrictions apply.
KADDOURA et al.: GREENFIELD DESIGN IN 5G FWA NETWORKS 2425

TABLE I
O PTIMIZATION PARAMETERS

TABLE II
Fig. 1. Evolution of cost function and performance indicators during the
O PTIMIZATION O BJECTIVES optimization.

together, require more resources than available and cause


congestion issues.
Thus, these objectives were used to evaluate the network
performance in all the tests carried out. For this, the calibrated
Planet® 3D Model along with Ericsson Air 5121 radiation
Table I summarizes the configuration and constraints of the patterns were used to calculate RF changes produced by
main optimization parameters. By applying these to (4), one modifications in antenna azimuths. The values and weights
can observe that the search space consists of approximately for all these objectives are summarized in Table II.
1.89 · 108475 possible network configurations among which
to find the best one. Therefore, the need for optimization A. Test Case A. FWA Cost Function
algorithms is highlighted as extremely significant. This test case consisted in performing the classical green-
An Ericsson internal optimization tool was used to carry out field design waterfall algorithm where radio network planning
the exercise. Two different test cases were defined to validate and radio network optimization are executed separately using
the improvement achieved by the proposed method. The first two different cost functions: the classical cost function in (6)
test case allowed evaluating the network performance in FWA and the proposed FWA cost function in (8). Unlike FWA cost
networks when using the proposed cost function. The second function where traffic is evaluated per household, in classical
test case allowed assessing the proposed iterative algorithm cost function, household traffic was uniformly distributed
that combines both radio network planning and radio network among all its demand nodes (i.e., candidate CPE antenna
optimization. In order to perform the evaluation of the results, locations) and, finally, the one in which the cost function is
the demand node presenting the best cost per household was minimized was chosen as the final CPE antenna location.
chosen as the selected CPE antenna location. Fig. 1 depicts the evolution of the cost and the performance
Three different boolean performance indicators were used indicators during the optimization exercise. According to its
to evaluate the cost function in this activity: weight, the capacity indicator drove the optimization and,
• Coverage. Good coverage is determined by a threshold therefore, it correlates with the evolution of the cost. The
which specifies the minimum required received signal traffic demanded per household in this exercise was forecast
strength. Coverage is used to identify areas where the to 100 Mbps and every base station was composed of 3 cells
received signal strength is enough so connection estab- of which capacity objective was set to 2.5 Gbps. Therefore,
lishment can be guaranteed. a maximum of 75 households (i.e., 1.66% of the total)
• Dominance. Good dominance is determined by two con- could be successfully served per base station. This effect was
ditions. The first condition is the minimum received observed during the first iterations of radio network plan-
signal strength of a cell to be considered as interfering ning in which the capacity indicator increased 1.66% every
the best server. The second condition is the maximum iteration. However, mainly due to the constraints imposed to
number of cells interfering the best server. Dominance is base stations locations (i.e., the number of predefined locations
used to identify how cells in the network interfere to each and the minimum distance among the selected ones), the cost
other. Specially, how the best server (i.e., the cell received function converges logarithmically. Finally, a step can be
with highest signal strength) is interfered by others. Since observed at the end of the curves of this test case due to the
network performance is usually limited by interference, azimuth optimization after the radio network planning phase.
dominance is useful to estimate the quality of service. Once the radio network planning and radio network
• Capacity. Good capacity is determined by a peak data optimization finished, resulting network configurations were
rate value supported by a cell. Capacity is used to evaluated in terms of target performance indicators as sum-
identify cells providing coverage to demand nodes which, marized in Table III. All coverage, dominance and capacity

Authorized licensed use limited to: Universidad Nacional de Colombia (UNAL). Downloaded on April 06,2023 at 02:38:28 UTC from IEEE Xplore. Restrictions apply.
2426 IEEE COMMUNICATIONS LETTERS, VOL. 23, NO. 12, DECEMBER 2019

TABLE III
T EST C ASES R ESULTS

performance indicators presented better figures when the The proposed method has been evaluated using data from
proposed cost function was used. Moreover, the proposed a real suburban environment. The results obtained have
cost function resulted in the selection of fewer base stations demonstrated that the final network performance is improved
than when using classical cost function, which decreases with respect to state-of-the-art methods while the deployment
the deployment costs. In addition, despite the reduction in costs (i.e., the number of required network base stations)
the number of base stations, the number of congested cells are reduced. Therefore, the proposed method is proven to
also decreased when the proposed cost function was used. overcome cutting-edge methods to perform greenfield design
Therefore, the proposed cost function resulted in the best activities.
network performance. Moreover, this method may be subsequently used to adapt
The classical cost function aims at improving network FWA networks to the eventual changes of the RF environment.
performance assuming users are located in all demand nodes
of the household. However, since the proposed cost function is R EFERENCES
focused on improving the network performance in final CPE
[1] FTTH Council. (Feb. 2018). FTTH Handbook. Accessed:
antenna locations, the results meet the expectations. Oct. 30, 2018. [Online]. Available: http://www.ftthcouncil.eu/documents/
Publications/FTTH%20Handbook_2017_V8_FINAL.pdf
[2] W. Webb, “Broadband fixed wireless access as a key component of the
B. Test Case B. FWA Greenfield Design Algorithm future integrated communications environment,” IEEE Commun. Mag.,
This test case consisted in evaluating the combined algo- vol. 39, no. 9, pp. 115–121, Sep. 2001.
[3] J. Du, D. Chizhik, R. Feick, G. Castro, M. Rodríguez, and
rithm proposed in Algorithm 1. For this, the classical R. A. Valenzuela, “Suburban residential building penetration loss at
greenfield design waterfall algorithm was compared with the 28 GHz for fixed wireless access,” IEEE Wireless Commun. Lett., vol. 7,
proposed one where radio network planning and radio network no. 6, pp. 890–893, Dec. 2018.
[4] D. Soldani, P. Airas, T. Hoglund, H. Rasanen, and D. Debrecht, “5G to
optimization are executed iteratively. the home,” in Proc. IEEE 85th Veh. Technol. Conf. (VTC Spring),
Since, test case A proved the proposed cost function to Jun. 2017, pp. 1–5.
provide better results than the classical approach, then the [5] X. Huang, U. Behr, and W. Wiesbeck, “Automatic cell planning for
proposed cost function for FWA networks was used to evaluate a low-cost and spectrum efficient wireless network,” in Proc. IEEE
Global Telecommun. Conf. Conf. Rec. (GLOBECOM), vol. 1, Nov. 2000,
the network performance in this test case. pp. 276–282.
As for test case A, Fig. 1 depicts the evolution of the [6] L. J. Ibbetson and L. B. Lopes, “An automatic base site placement
cost and the performance indicators during the optimization algorithm,” in Proc. IEEE 47th Veh. Technol. Conf., vol. 2, May 1997,
pp. 760–764.
exercise. In overall, all the performance indicators as well as [7] W. Mai, H.-L. Liu, L. Chen, J. Li, and H. Xiao, “Multi-objective
the cost presented better figures when using the proposed cost evolutionary algorithm for 4G base station planning,” in Proc. 9th Int.
function and the combined algorithm than when not using Conf. Comput. Intell. Secur., Dec. 2013, pp. 85–89.
[8] F. Gordejuela-Sanchez and J. Zhang, “LTE access network planning and
them. Moreover, the step observed at the end of the curves optimization: A service-oriented and technology-specific perspective,”
when radio network planning and radio network optimization in Proc. IEEE Global Telecommun. Conf. (GLOBECOM), Nov. 2009,
were executed separately disappeared when both phases were pp. 1–5.
combined into a single iterative algorithm. [9] J. Zhang, J. Yang, M. E. Aydin, and J. Y. Wu, “Mathematical modelling
and comparisons of four heuristic optimization algorithms for WCDMA
Regarding the final performance indicators, again, Table III radio network planning,” in Proc. Int. Conf. Transparent Opt. Netw.,
shows that all coverage, dominance and capacity performance vol. 3, Jun. 2006, pp. 253–257.
indicators decreased significantly. Moreover, the number of [10] J. Ramiro and K. Hamied, Multi-Technology Self-Optimization.
Hoboken, NJ, USA: Wiley, 2011. [Online]. Available: https://ieeexplore.
required base stations to achieve these figures is lower than ieee.org/xpl/articleDetails.jsp?arnumber=8042858
the best obtained in test case A. These results have proven [11] K. Tutschku, “Demand-based radio network planning of cellular mobile
that the combined algorithm outperforms the results obtained communication systems,” in Proc. IEEE INFOCOM Conf. Comput.
Commun., vol. 3, Mar. 1998, pp. 1054–1061.
by the classical waterfall algorithm. [12] S. M. Allen, S. Hurley, R. K. Taplin, and R. M. Whitaker, “Automatic
cell planning of broadband fixed wireless networks,” in Proc. IEEE VTS
V. C ONCLUSION 53rd Veh. Technol. Conf., vol. 4, May 2001, pp. 2808–2812.
[13] S. Hurley, “Planning effective cellular mobile radio networks,” IEEE
This letter has presented an automatic method to perform Trans. Veh. Technol., vol. 51, no. 2, pp. 243–253, Mar. 2002.
greenfield design activities for FWA networks using NR [14] 3rd Generation Partnership Project; Technical Specification Group
technology which, unlike other state-of-the-art methods, also Radio Access Network; NR; Base Station (BS) Radio Transmission and
Reception, document 3GPP TS 38.104 v15.2.0, Jun. 2018.
finds the optimal CPE antenna locations that improve the final [15] Planet. Accessed: Sep. 13, 2019. [Online]. Available: https://www.
network performance. infovista.com/planet/rf-planning-optimization

Authorized licensed use limited to: Universidad Nacional de Colombia (UNAL). Downloaded on April 06,2023 at 02:38:28 UTC from IEEE Xplore. Restrictions apply.

You might also like