Two-Level Allocation for H-CRAN Architecture
Based in Offloading
Mariane de Paula da Silva Gonçalves Matheus Barros Leto
Operational Research Laboratory Institute of Technology
Federal University of Para Federal University of Para
Belém, Brazil
mariane.goncalves@itec.ufpa.br matheusleto08@gmail.com
Fabrı́cio José Brito Barros
Telecommunications and Communications Laboratory
Federal University of Para
Belém, Brazil
fbarros@ufpa.br
Abstract—The accelerated growth of data and applications
are significant challenges for the next generation of mobile
networks. Amongst them, it is highlighted the necessity for a co-
existence of new and old patterns during the transition between
architectures. Thus, this paper has investigated solutions for
offloading into a hybrid architecture, also known as H-CRAN
(Heterogeneous Cloud Radio Access Network Architecture), that
centralizes processing and searches for a better use of network
resources. The optimization strategy was performed through
the evolutive algorithm PSO (Particle Swarm Optimization) and
another based on FIFO (First In, First Out), for benchmarking
purposes. Maximum Bits Rate average and number of users
with or without connections were used as a measurement of
performance. Through the results, it was noticed an improvement
of approximately 60% in the Maximum Bit Rate, when compared
to the traditional approach, enabling a better service to the users.
Index Terms—Mobile Networks, 5G, H-CRAN, QoS, Offload-
ing.
I. I NTRODUCTION
In recent years, heterogeneous networks (HetNets) have
been implanted aiming to increase the cell networks capacity
in dense and high traffic demanding areas, filling the blanks in
the area of coverage and discharging of the macro stations data
traffic (MBSs) [1] [2]. However, a dense network increases
the Capital expenditure (CAPEX) and Operational expenditure
(OPEX), due to the network implantation and additional
maintenances. Also, the increase in the inter-cell interference
and the coordination of the signal transmission [3].
Several authors have proposed the use of a C-RAN archi-
tecture (Cloud Radio Access Network), which proposes to be
centered, scalable and flexible [3] [4]. Thus, the mitigation of
the non-centered architectures problems, such as inefficiency
of the use of resources, mainly in high traffic loads scenarios,
is expected. In this new architecture, which moves the process-
ing of the primary signal to the Band Base Units (BBU) of
the eNodesBs, named BBU-Cloud. The RRHs (Remote Radio
Rafael Fogarolli Vieira
Operational Research Laboratory
Federal University of Para
Belém, Brazil Belém, Brazil
fogarollirafael@gmail.com
Diego Lisboa Cardoso
Institute of Technology
Federal University of Para
Belém, Brazil
diego@ufpa.br
Heads), with less functions, keep the radio access in the cell
sites, enabling a dense implantation at a minimum cost [5].
However, the C-RAN still offer many challenges, amongst
which it is highlighted the rigorous latency control, jitter and
the high infrastructure cost, mainly on what concerns the
implantation of the fronthaul, which needs to be implanted
with high capacity technologies due to its latency limitations.
This process of migration is not traditionally performed
in a subtle way. When a network needs to be updated,
almost all the network equipment needs to be substituted and,
therefore, there is the necessity of a transition phase in order
to satisfy the coexistence of new and old patterns. Then, the
Heterogeneous Cloud Radio Access Network (H-CRAN), in
which both HetNets (non-centralized) and C-RAN (central-
ized) are combined to allow the implantation of dense and
heterogeneous networks, aiming to centralize the processing
and performing functions of wide scale cooperative signal
processing, the network functions, and then, the development
of the SE (Spectral Efficiency) and EE (Energy Efficiency) is
substantially improved [6].
As a result, H-CRAN has been recognized as promising
network architecture to 5G networks and has gained much
attention by the industry and the academics. In spite of all
that, there is a series of challenges to be faced inside H-CRAN,
including optimized resource allocation, energy consumption,
development of prior versions compatible patterns, high call
blocking events and Quality of Service (QoS) [7].
The proposed offloading technique, in addition to the net-
work load balance and the dynamic mapping between BBU-
RRH may be considered one of the solutions to these prob-
lems. Hence, [8] [9] investigates the benefits and challenges
of the load balance in centralized networks (CRAN), where
a model of optimized mapping to RRH-BBU is proposed,
aiming to maximize the number of connected users and,
thus, improve the QoS. The former aims at balancing without
worrying about the balance amongst the sectors, and the latter,
besides guaranteeing the KPI, aims to search the load balance
amongst the sectors of the BBUs, then, optimizing the network
balance.
In [10], it is proposed a slicing of the network to multi-
tenant H-CRANs, which considers the baseband resources,
fronthaul and backhaul capacities, QoS and interference. How-
ever, the efficiency of a balanced allocation in the BBUs
sections is not explored using offloading techniques in mobile
networks, which would avoid high traffic congestion in the
network and high events of calls blocking.
In [11] the author addresses the problem of wireless back-
haul in an H-CRAN network and an energy allocation is made
to multiple cells of different levels, taking into consideration
the energy consumption of different types of cells and wireless
backhaul. An interation algorithm is proposed to determine the
maximum number of cells that can be supported in H-CRAN.
Though, in the survey performed by the authors, it was
not found any work that investigates offloading solutions
aiming at network load balancing and improvement of QoS Fig. 1. H-CRAN Architecture Proposed
for H-CRAN architectures. The main contribution of this
paper is the proposal of a new schemata for offloading in
H-CRAN architectures, focusing on the balance of BBU
loads, a better use of network resources, maximization of the
capacity, improvement of the coverage and maximization of
the QoS for the final users, thus, promoting heterogeneity and,
consequently, improving the spectrum efficiency.
Besides this introductory section, this paper is divided as
follows: Section 2, which presents a model of the system;
Section 3, the Network Parameters; Section 4, the results of
the paper and ending up at Section 5, with the conclusion and
future works.
II. SYSTEM MODEL
Considering that the users are uniformly distributed, offload-
ing solutions for load balancing are investigated for H-CRAN
architectures. The levels of interference and QoS requirements
are collected aiming to reach the best allocation of the network
resources. In the modeling, 6 RRHs and a macro EnodeB were
used. The User Equipments were uniformly distributed in the
scenario, according to Fig. 1. To support the small cells, 1
BBU Pool composed of 2 BBUs, with 2 sectors each, each
one with maximum capacity defined as (HCmax = 30) [4]. Fig. 2. Algoritmo TLA
As mentioned before, the proposal is to maximize the
coverage and the QoS of the UEs and, to do so, it was
developed an evolutive algorithm PSO, in order to find an Thus, the values of the particles are updated, searching
optimal solution to the two level allocation in an H-CRAN for the best Gbest, after that, it is in a loop until it reaches
architecture. the stop criteria, defined as 1000 iterations, where at each
The decision-making of offloading is explained in Fig. iteration it updates its particles, calculates its fitness. When
2, where the analysis of KPIs and QoS are predetermined. the current solution is better than the previous one, the particle
The PSO starts with the random generation of particles, values are updated (PBest), if not, the solution is discarded and
with a linear distribution, where each particle is a possible the previous solution is maintained. For every iteration, it is
solution to the problem and represents the set of RRHs and verified if the best solution found remains the same, otherwise,
their respective sectors in which they are connected. In the it is updated by the best found (GBest).
sequence, each particle is evaluated by the objective function, Then, if there are users blocked in the BBU, the decision is
in which it returns how good the solution is. made to offload, where the UE that has the best SNR with the
EnodeB macro is allocated to it. With this, there is a better C. Intra-BBU Handover:
use of the network resources and maintenance of the levels of Performed when there is a transition of users from one
QoS. sector to the other within a single BBU, under the condition
The objective function of the implemented PSO algorithm that the sectors involved in such transition are treated entirely
is composed by a combination of KPIs, proposed in [4], which within the eNodeB.
were adjusted to the proposed scenario and that is described
below. Weights were assigned to obtain the maximum QoS, "
X X Ui X X
Rt+1 t+1


using the PSO. Intra-BBU HO = is .Rjs
i
Dij n
j6=i s∈SOS
#
 !##−1 X
Rt+1 t+1
"

X X − is .Rjs (6)
t+1


w1  max Ui Ris ) − HCs , 0 + s
s i
(
1 if Intra-BU HO = 0
hP P
t+1 t+1



i−1 KPIintra = −1 (7)
Ui P P
[Intra-BBU HO ] otherwise.
w2 i j6=i Dij 1− n sSOSn Ris .Rjs
hP P
Ui P P t+1 t+1


+ w3 i i6=j Dij n sSOS Ris .Rjs −
P t t+1
−1
s Ris .Ris D. Forced Handovers:
P P t t+1

−1
+ w4 s i Ris .Rjs Ui (1) It is used when the HC limit is exceeded.
XX
Rtis + Rt+1


A. Key Performance Indicator of Blocked Users (KPIBU ) fHO = is Ui (8)
s i
It evaluates the capacity of the BBUs and its sectors (HCs). (
1 if fHO = 0
When the capacity is exceeded, the users are blocked. The KPIf = −1 (9)
number of blocked users BU in the network in time t + 1, is [fHO ] otherwise.
given by [4]: III. NETWORK PARAMETERS
X
"
X
! ! # The SUI propagation model, which is widely used in liter-
t+1 ature for trajectories losses in cell phone networks operating
BU = max Ui Ris − HCs , 0 (2)
s i over 2GHz, was used. In which P LSU I (d), measured in dB
is found by [12] [13] [14]:
Where i = 1,2, ..., M and s = 1,2, ..., S. The KPI for blocked
users (KP IBU ) is then presented as:  
( PLSU I (d) = P L( d0) + 10n log10 dd0 + X(f c) + X(RX) +
1 if BU = 0 Xσ (10a)  
KPIBU = 1
(3) 4πd0
BU otherwise P L(d0) = 10n log10 (10b)
λ
c
n = a − b.hT X + (10c)
hT X
According to [4], the performance index of the network can  
be evaluated through 3 metrics: Inter-BBU Handovers, Intra fM HZ
Xf c = 6 log10 , fc > 2GH Z (10d)
BBU Handovers and Forced Handovers. The formulations 2000
used by the author are shown below:
 
hRX
XRX = −10.8. log10 (10e)
2
B. Inter-BBU Handover: • λ denotes the wavelength of the carrier in meters;
Necessary transfers from a BBU to another.
• P L(d0) represents the pathloss in free space, in dB at an
X X Ui X X
!! initial distance of d0 = 100meters;
t+1 t+1


Inter-BBU HO = 1− Ris .Rjs
i
Dij n s∈SOSn • Xσ , is the fading caused by the shadowing
j6=i
(4) areas, assuming values in the following range
8.2 < σ < 10.6dB;
(
1 if Inter-BU HO = 0 • The parameters a, b and c are constants, used to model
KPIinter = −1 (5)
[Inter-BBU HO ] otherwise. the terrain types. For mountainous and dense vegetation
 terrains the parameter A is used, assuming the following
values (a = 4.6 / b = 0.0075 / c = 12.6);
• fM HZ is the carrier frequency (fc) in MHz;
• hT X is the height of the base station, in a range from
10 to 80 meters;
• n is the exponent of losses;
The size of the cells that contain the equipment of H-CRAN
architecture were determined considering the maximum prop-
agation loss obtained by the Downlink (DL) and Uplink(UL).
For each link, the propagation loss was represented by the
diminishing of the EIRP (Effective Isotropic Radiated Power)
and by the RSL (Received Signal Level), considering the
maximum power of the transmitter element and the minimal
power of the receptor element. The limiar power of the
equipments was considered according to [15].
The maximum propagation loss, considering the links, is
represented by the following equation:
P L(max) = min[P Lmax(DL) ; P Lmax(U L) ] (11)
The cell radius corresponds to distance of the SUI model,
when the model is equalized to the maximum propagation
loss given by (11). The equation that represents the radius is
given as follows:
P Lmax(DL) P L(d0 )
R = (d0 .10) (12)
10n
The UEs are only connected if they are contained within the
coverage of the RRH or macro EnodeB. The power received
inside each cell is given by formula (12), where it is performed
the dBm conversion to Watts:
P oti −LSU I
10
P otw = (13)
1000
Where P otw is given in Watts, Poti is the irradiated power,
in dBm, and LSU I is the signal fading. After calculating
the power, it is possible to calculate the value of the SNR
and, then, the maximum throughput of the connected UE, as
described below:
P otw
SN R = (14)
I a linear behavior due to the addition of users in the scenarios
Vmax = B ∗ log2 (1 + SN R) (15)
So that P otw is the potency, while is the value of the
white noise at formula (14).As for formula (15), it represents
the maximum rate of transmission of bit without errors by
noisy channel, where B is the band length available to the
users [16]. The parameters used in the model are shown in
the table below: [17] [18]:
TABLE I
S IMULATION PARAMETERS
Macro EnodeB RRH
Frequency 3.5 GHZ 2.5 GHZ
Power Transmission 43 dbm 33 dbm
Radius Coverage 1000 m 150 m
Bandwidth 20 MHZ
IV. RESULTS
MATLAB® was used to evaluate the performance of the
proposal. An approach based on FIFO was used as benchmark.
In this method, the resources are allocated without any type of
priority. The first user to require connection will be serviced
by the RRH that offers the best signal-to-noise ratio. As for
the TLA proposal, the connection in the RRHs is prioritized,
until the exhaustion of network resources, aiming at servicing
the users with maximum network capacity while maintaining
the QoS levels.
7 scenarios were generated containing, respectively, 200,
225, 250, 275, 300, 325, 350 users. All scenarios were
simulated 50 times and the results presented as their average.
The confidence interval was 95%.
Fig. 3. Average Connected Users
The average of disconnected users is around 30% higher in
FIFO, and the average of connected users is 25% higher in
TLA, according to Fig. 3 and Fig. 4, respectively. It happens
due to the offloading scheme that opens up the network traffic
and discharge the users that had been blocked in the BBUs to
the macro EnodeB.
The number of disconnected and connected users presents
(which are linear), the allocation of the network resources is
re-distributed in an efficient way, ensuring a better connection
of users in the network in TLA.
According to Fig. 5, which represents the maximum bit
rate,the minimum QoS levels were maintained for each sce-
nario, where both TLA and FIFO satisfy the minimal QoS
requirements. The former has guaranteed a better connection
when compared to the latter, both in constant growth. In the
beginning, the minimal QoS requirements of the UEs are

Fig. 4. Average Disconnected Users
Fig. 5. Average Maximum Bit Rate (kbps)
satisfied and, later, the resources are re-distributed, obtaining
a higher number of users in the process.
Thus, it is a fact that the TLA proposal has obtained a
satisfactory performance, 60% higher when compared to the
FIFO approach, satisfying all the established requirements,
maximizing its capacity and maintaining the minimal levels
of QoS and network coverage.
It is possible to observe, through the merit figures used for
the performance prediction, that the load balance allows sig-
nificant network performance improvements when compared
FIFO algorithm-based techniques available in the literature.
V. CONCLUSION
The increasing data and applications volume are signifi-
cant challenges of the 5G mobile networks. This way, the
mobile network operators are investing in new technologies
to improve the network capabilities, mitigate interference and
improve the QoS of the final users.
In this paper, it was proposed schemata of offloading for H-
CRAN architectures, aiming at improving the use of network
resources while maintaining its minimal levels of QoS.
Through the results, it was noticed that the proposal of TLA
has obtained a satisfactory performance when compared to
FIFO, improving the Maximum Bit Rate by about 60%. This,
in turn, provides a better connection to all users of the network,
efficiently distributing all the network resources.
As future work, new scenarios will be modeled, with
frequency changes, BBU capacity variations, varying the num-
ber of antennas and their arrangements and analyzing other
performance measurements.
R EFERENCES
[1] Cisco. 2017. “Cisco Visual Networking Index: Global Mobile Data
Traffic Forecast Update, 2016 – 2021.”
[2] DUAN, Xiaoyu; WANG, Xianbin. Authentication handover and privacy
protection in 5G hetnets using software-defined networking. IEEE
Communications Magazine, v. 53, n. 4, p. 28-35, 2015.
[3] RAZA, Muhammad Rehan et al. Demonstration of dynamic resource
sharing benefits in an optical C-RAN. Journal of Optical Communica-
tions and Networking, v. 8, n. 8, p. 621-632, 2016.
[4] KHAN, Muhammad; ALHUMAIMA, Raad S.; AL-RAWESHIDY,
Hamed S. QoS-Aware Dynamic RRH Allocation in a Self-Optimized
Cloud Radio Access Network With RRH Proximity Constraint. IEEE
Transactions on Network and Service Management, v. 14, n. 3, p. 730-
744, 2017.
[5] WU, Jun et al. Cloud radio access network (C-RAN): a primer. IEEE
Network, v. 29, n. 1, p. 35-41, 2015.
[6] ZHANG, Biling et al. Resource Allocation for 5G Heterogeneous Cloud
Radio Access Networks with D2D Communication: A Matching and
Coalition Approach. IEEE Transactions on Vehicular Technology, 2018.
[7] MAROTTA, Marcelo A. et al. Design considerations for software-
defined wireless networking in heterogeneous cloud radio access net-
works. Journal of Internet Services and Applications, v. 8, n. 1, p. 18,
2017.
[8] KHAN, M.; ALHUMAIMA, R. S.; AL-RAWESHIDY, H. S. Quality of
service aware dynamic BBU-RRH mapping in cloud radio access net-
work. In: Emerging Technologies (ICET), 2015 International Conference
on. IEEE, 2015. p. 1-5.
[9] DA PAIXÃO, Ermı́nio Augusto Ramos et al. Optimized load balancing
by dynamic BBU-RRH mapping in C-RAN architecture. In: Fog and
Mobile Edge Computing (FMEC), 2018 Third International Conference
on. IEEE, 2018. p. 100-104.
[10] LEE, Ying Loong et al. Dynamic Network Slicing for Multitenant
Heterogeneous Cloud Radio Access Networks. IEEE Transactions on
Wireless Communications, v. 17, n. 4, p. 2146-2161, 2018.
[11] TRAN, Huu Q. et al. On the energy efficiency of NOMA for wireless
backhaul in multi-tier heterogeneous CRAN. 2017.
[12] PENG, Mugen et al. System architecture and key technologies for 5G
heterogeneous cloud radio access networks. IEEE network, v. 29, n. 2,
p. 6-14, 2015.
[13] CASTRO, Bruno Souza Lyra et al. Modelo de propagação para redes
sem fio fixas na banda de 5, 8 GHZ em cidades tı́picas da região
amazônica. 2010.
[14] SULYMAN, Ahmed Iyanda et al. Radio propagation path loss models
for 5G cellular networks in the 28 GHz and 38 GHz millimeter-wave
bands. IEEE Communications Magazine, v. 52, n. 9, p. 78-86, 2014.
[15] SULYMAN, Ahmed Iyanda et al. Directional radio propagation path
loss models for millimeter-wave wireless networks in the 28-, 60-, and
73-GHz bands. IEEE Transactions on Wireless Communications, v. 15,
n. 10, p. 6939-6947, 2016.
[16] DAHLMAN, Erik; PARKVALL, Stefan; SKOLD, Johan. 4G: LTE/LTE-
advanced for mobile broadband. Academic press, 2013.
[17] Shannon C E, “Communication in the Presence of Noise”. Proceedings
of the IRE, vol. 37 pp. 10-21, 1949.
[18] PHAIWITTHAYAPHORN, Prapassorn et al. Cell throughput based
sleep control scheme for heterogeneous cellular networks. In: Electrical
Engineering/Electronics, Computer, Telecommunications and Informa-
tion Technology (ECTI-CON), 2017 14th International Conference on.
IEEE, 2017. p. 584-587.
[19] ZHANG, Biling et al. Resource Allocation for 5G Heterogeneous Cloud
Radio Access Networks with D2D Communication: A Matching and
Coalition Approach. IEEE Transactions on Vehicular Technology, 2018.