Automatika

Journal for Control, Measurement, Electronics, Computing and

Communications

ISSN: 0005-1144 (Print) 1848-3380 (Online) Journal homepage: https://www.tandfonline.com/loi/taut20

Self-organizing inter-cell interference coordination

in 4G and beyond networks using genetic
algorithms

Aasia Kashaf, Moazzam Islam Tiwana, Imran Usman & Mohsin Islam Tiwana

To cite this article: Aasia Kashaf, Moazzam Islam Tiwana, Imran Usman & Mohsin Islam Tiwana
(2017) Self-organizing inter-cell interference coordination in 4G and beyond networks using
genetic algorithms, Automatika, 58:1, 48-54, DOI: 10.1080/00051144.2017.1337394

To link to this article: https://doi.org/10.1080/00051144.2017.1337394

© 2017 The Author(s). Published by Informa

UK Limited, trading as Taylor & Francis
Group

Published online: 16 Jun 2017.

Submit your article to this journal

Article views: 1131

View related articles

View Crossmark data

Citing articles: 1 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=taut20
AUTOMATIKA, 2017
VOL. 58, NO. 1, 48–54
https://doi.org/10.1080/00051144.2017.1337394

REGULAR PAPER

Self-organizing inter-cell interference coordination in 4G and beyond networks

using genetic algorithms
Aasia Kashafa, Moazzam Islam Tiwanaa, Imran Usmanb and Mohsin Islam Tiwanac
a
Department of Electrical Engineering, COMSATS Institute of Information Technology, Islamabad, Pakistan; bCollege of Computing and
Informatics, Saudi Electronic University, Riyadh, Kingdom of Saudi Arabia; cDepartment of Mechatronics Engineering, National University of
Sciences and Technology, H-12, Islamabad, Pakistan

ABSTRACT ARTICLE HISTORY

The design objective of the 4G and beyond networks is not only to provide high data rate Received 24 June 2016
services but also ensure a good subscriber experience in terms of quality of service. However, Accepted 10 May 2017
the main challenge to this objective is the growing size and heterogeneity of these networks. KEYWORDS
This paper proposes a genetic-algorithm-based approach for the self-optimization of Long-term evolution; radio
interference mitigation parameters for downlink inter-cell interference coordination parameter resource management; inter-
in Long Term Evolution (LTE) networks. The proposed algorithm is generic in nature and cell interference
operates in an environment with the variations in traffic, user positions and propagation coordination; genetic
conditions. A comprehensive analysis of the obtained simulation results is presented, which algorithm; self-organizing
shows that the proposed approach can significantly improve the network coverage in terms of networks; self-optimization
call accept rate as well as capacity in terms of throughput.

1. Introduction advent of Long Term Evolution (LTE), the research on

LTE self-optimization involves RRM parameters like
The 4G and beyond mobile access technology promises
interference mitigation using inter-cell interference coor-
to deliver high data rates from 100 Mbps to 1Gbps
dination (ICIC) [9,10], load balancing [11,12] and band-
along with a high quality-of-service (QoS) provision for
width allocation [13].
a variety of applications [1,2]. However, the increase in
Recently, reinforcement (machine) learning tech-
the size and the heterogeneity of these networks has
nique of fuzzy Q-learning (FQL) has been used for the
given a strong impetus to the research in self-organizing
optimization of RRM parameters in LTE. FQL has
networks [3]. Self-organization targets to reduce the
been used for the optimization of mobility parameters
capital expenditure (CAPEX) as well as operational
of both GSM Edge Radio Access Networks (GERAN)
expenditure (OPEX) of these networks. While at the
[14] and LTE network [15]. More recently, FQL has
same time, the QoS constraints on various telecom serv-
been used for coverage/capacity optimization of LTE
ices in terms of coverage and capacity must also be sat-
networks by adjusting the vertical tilt angle of the
isfied. The autonomic self-organization functions
antenna employed at eNodeB (eNB) [16–18]. Despite
comprise of self-configuration, self-optimization and
the widespread use of FQL, all these works lack any
self-healing. This work focuses on self-optimization
discussion on the criteria when to switch from explora-
where performance measurements from the network
tion to exploitation mode. This problem is known as
known as key performance indicators (KPIs) are used
exploration/exploitation dilemma. Furthermore, FQL
to optimize the radio resource management (RRM)
algorithm is quite susceptible to trapping in local
parameters. This leads to an optimal network perfor-
minima.
mance. Hence, autonomic self-optimization reduces
GA is a machine learning technique that has been
OPEX of the network by reducing the workload of the
used in many different fields to solve the complex opti-
site survey and network performance analysis teams.
mization problems [19]. It is an efficient stochastic
The industry and academia have done some signifi-
search technique that has the tendency to quickly
cant work on the self-optimization in radio access net-
attain optimal/near-optimal solution in large solution
works (RANs) during the last decade [4,5]. In spite of
spaces. With appropriate fitness function defined and
the early research work and industrial interest in self-
right parameters taken into consideration, this solution
optimization of GSM and UMTS [6–7], self-optimization
could be generalized. GA is an evolutionary algorithm
was not included as a part of these standards. Later, the
that models the biological processes of evolution,
focus of research shifted towards self-optimization in het-
through successive selective breeding over generations
erogeneous networks using load balancing [8]. With the

CONTACT Moazzam Islam Tiwana [email protected]

© 2017 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted
use, distribution, and reproduction in any medium, provided the original work is properly cited.
 AUTOMATIKA 49

[20]. It is a gradient-free technique that is very robust

against getting stuck in the local optima. Owing to
these reasons, we choose GA to solve the LTE network
optimization problem by optimizing the inter-cell
interference mitigation parameters.
The proposed scheme is valid both for LTE as well
as LTE-Advanced (LTE-A) as it is implemented in the
management plane of the network. The basic network
architecture of LTE and LTE-A is the same apart from
some new features in LTE-A-like carrier aggregation,
enhanced MIMO and Co-ordinated Multipoint
(CoMP) Transmission. The self-optimization algo-
rithms can be implemented in the operation and main-
tenance centre (OMC) of a network. In the paper, an
eNB and a base station represent one and the same
thing, and will be used interchangeably.
The paper is organized as follows: Section 2 presents
LTE ICIC model used in our case study. Section 3
presents the proposed GA-based self-optimization
framework. Section 4 deals with the simulation envi-
ronment and provides the numerical results of the pro-
posed scheme. Section 5 concludes this paper by
presenting conclusion and future directions.

Figure 1. System model.

2. System model
Consider a downlink OFDMA-based cellular network where Prbu and Prb 0 u represent the mean pilot signal
with ICIC. The system uses a frequency reuse factor of power received by u from the base stations b and b ,
0

1. Hence, all the available bandwidths are reused in respectively. The users with the worst quality metric hu
each cell. Physical resource block (PRB) is the smallest are allocated resources from the edge band, and hence
time-frequency resource unit that can be allocated to a they get benefit of maximal transmission power P of
user. base stations. When the edge sub-band is full, the users
Figure 1 presents a seven-adjacent-cell layout. The are allocated resources from the centre band. The
frequency in each cell is subdivided into three sub- handover for the users from one base station to the
bands. One is the edge band and the other two are the other is dependent on the received power-level differ-
centre bands. The users with worst channel quality are ence between the two base stations. For a user u,
allocated at the edge band/protected band. These users hard handover will be performed to move from the
are mostly located at cell edges but could also be closer 0
serving base station b to b , if the following condition
to the base station and experiencing deep fading condi- is satisfied:
tions. If the edge band of a cell is completely occupied,
the remaining users are allocated PRBs in the centre Prb 0 u  Prbu > Thyst (2)
bands. As evident from Figure 1, the main interference
is between the edge band users of a cell with centre where Thyst is the fixed hysteresis margin for all the
band users of the neighbouring cells. P is the maxi- base stations and is set to 6 dB in this study.
mum transmit power for each sub-carrier. However,
the transmit power in the centre band is reduced to
aP, in order to reduce interference with the neighbour- 3. Proposed architecture
ing edge bands. A priority scheme is used to allocate In the proposed scheme, we exploit the hidden depen-
the PRBs to the users on the edge band. This priority dencies present in the RRM parameters, and optimize
scheme is based upon the calculation of a quality met- them in order to improve the KPIs of the network. The
ric as explained here. For a user u with the serving base RRM parameters optimized in this case are the
station b, the quality metric hu is calculated using pilot a-parameters of individual eNBs. The KPIs optimized
channel signal strengths: are called accept rate, file transfer time (FTT) and the
average bit rate (ABR) of all the mobiles in the net-
Prbu
hu ¼ P (1) work. Owing to the mobility of mobiles, their changing
b
0 Pr 0 þ s 2n
6¼b b u distribution, fading phenomenon and interference etc.,
50 A. KASHAF ET AL.

Table 1. RRM parameter optimization using genetic algorithm.

Step 1: Create z randomly generated solutions to form first generation P1.
FOR each Ps: s = 1, …, v and
Step 1.1: Compute fitness value of all individuals
in Ps using LTE simulator.
Step 1.2: Evaluate fitness using the predefined
fitness function.
Step 1.3: Check the following conditions:
IF Termination Criteria is reached, go to step 2.
ELSE Create new generations using genetic
operations
Step 1.3.1 Crossover
Step 1.3.2 Mutation
Step 1.3.3 Replication
Step 1.3.4 Select solutions based on their
fitness using genetic operators and assign
them to Ps + 1. Delete the low-ranked
individuals.
END IF
END
Step 2: Save the best evolved Solution and terminate
the procedure.

values retained are then used in testing phase to vali-

date the generalization for real-world applications.

3.1. GA control parameters

The convergence of selective breeding is governed by
an application-oriented predefined fitness function.
Other factors contributing to the decreased time of
convergence are different control parameters custom-
ized for a typical application. The control parameters
in our GA simulations are population and generation
Figure 2. Proposed architecture.
sizes, selection method, cross-over and mutation prob-
abilities, and termination criteria, etc. In addition,
probabilities of genetic operators of replication, cross-
over and mutation are repeatedly used to optimize a
the KPIs are a nonlinear function of a. Furthermore, highly complex cost function.
this relationship is not available in a closed form.
Hence, we have chosen GA to achieve our optimiza- 3.1.1. Population initialization
tion objective. Figure 2 shows the the generalized dia- Initially, the population is initialized by pseudo-ran-
gram for the proposed approach. As shown, the dom values drawn from the standard normal distribu-
proposed optimization method consists of two main tion assuring ramped half and half strategy. A random
steps: the training phase, followed by the testing phase. number is picked between the minimum and maxi-
In the training phase, the a parameters are optimized mum possible values for each decision variable, where
using GA. The dimension of a chromosome depends decision variables are the mean fitness values, obtained
upon the number of genes, where each gene uniquely by simulating concerned chromosome using LTE
represents the a parameter of an individual eNB. The simulator.
total number of genes in a chromosome equals the
number of base stations. The population size deter- 3.1.2. Termination criterion
mines the number of chromosomes in a generation. In GA, termination criterion for simulations can be
These populations are scored and sorted based upon based upon the number of generations or on some
their individual fitness values. The next generation is threshold of fitness level or on some predefined time
evolved by employing genetic operators of cross-over, limit. In our case, we have limited the simulations by
mutation and replication on the best performing indi- specifying maximum numbers of generations to be 30.
viduals in the previous generation (see Table 1 for The best evolved solution is saved.
algorithm details). The process continues till the termi-
nation criterion is reached, which may be based on 3.1.3. Fitness function
time limit, number of generations or a predefined The fitness function score of the chromosomes is used
acceptable fitness value. The termination criteria in to select the best performing chromosomes in a popu-
this case are the number of generations. The optimum lation. We have used mean ABR of all mobiles in the
 AUTOMATIKA 51

Table 2. Genetic algorithm simulation parameters.

Parameters Settings
Fitness function Given as Equation (3)
Selection Generational
Population size 80
Survival mechanism Keep best
Termination 30 generations
Selection method Roulette wheel selection method

network as the fitness function, given as:

X
n
ABRi
U¼ (3)
i¼0
n

where ABRi denotes the average bit rate of all mobiles

in communication with the ith base station at that
instant. The improvement of mean ABR is directly cor-
related with the improvement/reduction of overall
FTTs in the network. Consequently, the accept rate of
Figure 3. The network diagram of the simulated system.
network improves. ABRi is a function of ai:

ABRi ¼ f ðai Þ (4) mobile users is Poisson process with a certain arrival
rate λ (arrivals/sec). The time difference between two
Hence our optimization objective can be given as simulation time steps is 1 sec. Hence, the probability
  Pr of generation of k mobiles during each simulation
ai ¼ argmax U ai 0 (5) time step is given as
0
ai
λk eλ
Pr ðkÞ ¼ (6)
The settings for the control parameters of GA given in k!
Table 2. In Figure 2, it can be seen that the LTE simula- Poisson distribution approximation of binomial distri-
tions run at two times: first, when calculating mean fit- bution is used to calculate k for each simulation time
ness values for each genome in GA during the training step [21]. The network simulation parameters are
phase; second, when the KPIs are averaged for the listed in Table 3.
optimized and non-optimized cases during the testing
phase. The simulations for calculating each chromo- Reference solution
some’s mean fitness value run for 30 sec. Hence, the Reference solution is the default optimal a value for
total time required to obtain optimized a distribution all eNBs. Its value has been chosen as 0.5 as deter-
is 30 £ 80 £ 30 sec = 72,000 sec or 20 hr). While for mined in [22].
averaging KPIs during testing phase, the simulations
run for 2000 sec for both self-optimized and non-
optimized cases. 4.2. Simulation results
Performance obtained using adaption of a-parameters
using GAs is compared with the reference solution.
4. Numerical results
Table 3. The system-level simulation parameters.
4.1. Simulation environment Parameters Settings
System bandwidth 5 MHz
The simulations have been performed using the down- Cell layout 30 eNBs, single sector
link LTE MATLAB simulator as described in [11]. The Maximum eNB transmit 32 dBm
network diagram of the simulated system is shown in power
Inter-site distance 1.5–2 km
Figure 3. Subcarrier spacing 15 kHz
The simulator performs correlated Monte Carlo PRBs per eNB 24 (8 in each sub-band) + 1 for pilot
channel
snapshots with the resolution of 1 sec to account for Path loss L=128.1+37.6 log10(R), R in kilometers
the time evolution of the network. At the end of each Thermal noise density ¡173 dBm/Hz
Shadowing standard 6 dB
time-step mobile positions are updated, HO events are deviation
processed, new mobiles are admitted according to the Traffic model FTP
File size 6300 Kbits
access conditions and some other users leave the net- PRBs assigned per mobile 1–4 (first-come, first-serve basis)
work (end their communications or are dropped). Mobility of mobiles 10%
Traffic model used to simulate the arrival of new Mobile speed 8.33 m/sec
52 A. KASHAF ET AL.

Figure 4. The mean FTT as a function of traffic intensity for the Figure 5. The mean ABR as a function of traffic intensity for the
optimized and non-optimized case. optimized and non-optimized cases.

As mentioned earlier, the network performance is eval-

uated in terms of three KPIs: call accept rate, FTT and
ABR of mobile stations. Figure 4 presents a compari-
son between the FTT of the two systems relative to
traffic arrival rate. It can be observed that for the traffic
value of 1 arrivals/sec, the improvement is only mar-
ginal. This is due to the fact that the number of mobiles
entering the network is not large enough. Conse-
quently, due to low interference and allocation of
PRBs in the edge band, the mobiles quickly download
the file and leave the network. Therefore, the employ-
ment of self-optimization mechanism does not signifi-
cantly improve the mean FTT value. The same holds Figure 6. The accept rate as a function of traffic intensity for
true for the case of the traffic values of 2 and 3 arrivals/- the optimized and non-optimized cases.
sec. However, for traffic value of 4 arrivals/sec, and
onwards, we can see a clear improvement in the FTT. increases accept rate decreases, but at the same time,
As the number of mobiles present in the network the self-optimization algorithm becomes more effective
increases, they stay longer in the network due to con- due to congestion. Hence, for a traffic value of 5 arriv-
gestion and are also allocated PRBs in the centre band. als/sec, we see an 8% increase in the call accept rate.
Hence, adaptation of a leads to greater improvement. Figures 7–12 compare the CDF plots of the ABR
Overall, the FTT increases, but when comparing the of mobiles of two systems for traffic values of
two systems, it can be concluded that the proposed 3–8 arrivals/sec, respectively.
model is better to be used for high traffic rate
situations.
For a traffic value of 5 arrivals/sec we see a maxi-
mum comparative decrease/improvement of 25% in
FTT. Overall, the greater FTT value shows that the net-
work has taken more time for these transfers.
Figure 5 compares mean ABR of mobile stations of
two systems. It is evident that the proposed algorithm
performs better than the non-optimized one at all traf-
fic values considered. Same as other KPI values, it
shows greater improvement at traffic values of 4 and
5 arrivals/sec. For a traffic value of 5, we see a 25%
increase in the ABR. After that, it tends to decrease but
even then the results are better than the non-optimized
case.
From Figure 6, it is seen that the call accept rate for
both the cases is approximately same for traffic values
of 1 and 2 arrivals/sec. This is due to the fact that ini- Figure 7. CDF of individual ABR values for all mobiles in the
tially there are less number of mobiles in network. So, network for traffic intensity of 3 arrivals/sec, with and without
the mobiles are served quickly. While as the traffic optimization.
 AUTOMATIKA 53

Figure 8. CDF of individual ABR values for all mobiles in the

Figure 10. CDF of individual ABR values for all mobiles in the
network for traffic intensity of 4 arrivals/sec, with and without
network for traffic intensity of 6 arrivals/sec, with and without
optimization.
optimization.

Figure 7 shows that for traffic value of 3 arrivals/sec

the optimized system performs better. As shown, more
than 60% mobiles have ABR values greater than
1500 Kbits/sec in the optimized case. While in the
non-optimized case, 60% mobiles have ABR values
greater than 1300 Kbits/sec. We see a greater improve-
ment for traffic value of 4 arrivals/sec, as shown in
Figure 8. Here 70% mobiles have ABR less than
1750 Kbits/sec for optimized case, while for non-
optimized, 70% are with ABR values less than
1500 Kbits/sec. So, for both traffic values, 3 and 4 arri-
vals/sec, optimized case shows greater improvement in
mobile ABR. The same is true for traffic value of 5 as
shown in Figure 9. The results for the traffic values of
6–8 arrivals/sec are shown in Figures 10–12, respec-
tively. However, the improvements tend to decrease.
Figure 11. CDF of individual ABR values for all mobiles in the
This is due to the fact that now the increasing conges-
network for traffic intensity of 7 arrivals/sec, with and without
tion has increased interference to such a point that optimization.
optimizing a has very little effect. For example, for the

Figure 9. CDF of individual ABR values for all mobiles in the Figure 12. CDF of individual ABR values for all mobiles in the
network for traffic intensity of 5 arrivals/sec, with and without Network for traffic intensity of 8 arrivals/sec, with and without
optimization. optimization.
54 A. KASHAF ET AL.

traffic of 7 arrivals/sec, 30% of mobiles have ABR val- [9] Ghimire B, Auer G, Haas H. Busy burst for trading-off
ues less than 350 Kbits/sec for the self-optimized case. throughput and fairness in cellular OFDMA-TDD.
While for non-optimized case, 38% of mobiles have EURASIP J Wirel Commun Net. 2009;2009(10).
[10] Dirani M, Altman Z. A cooperative reinforcement
ABR values less than 350 Kbits/sec. Similarly, for traffic learning approach for inter-cell interference coordina-
value of 8 arrivals/sec, 60% of mobiles have ABR values tion in OFDMA cellular networks. Proceedings of the
greater than 400 Kbits/sec. While for the non- 8th International Symposium on Modeling and Opti-
optimized case, 35% of mobiles have an ABR value mization in Mobile, Ad Hoc and Wireless Networks
greater than 400 Kbits/sec. (WiOpt), Avignion; 2010.
[11] Nasri R, Altman Z. Handover adaptation for dynamic
load balancing in 3GPP long term evolution systems.
5. Conclusion 5th International Conference on Advanced in Mobile
Computing & Multimedia (MoMM2007), Jakarta;
Interference mitigation is a challenging issue in mod- 2007.
ern wireless communication systems. Our main focus [12] Lobinger A, Stefanski S, Jansen T, et al. Load Balancing
in this proposed work is on the self-optimization part in Downlink LTE Self Optimizing Networks. In Pro-
ceedings of IEEE 71st Vehicular Technology Confer-
of SONs. The objective was to reduce interference in ence (VTC); 2010.
LTE networks and simultaneously improve network [13] Hasselbach PP, Klein A, Gaspard I. Dynamic resource
performance by self-optimization of related RRM assignment (DRA) with minimum outage in cellular
parameters. In turn, this automated management will mobile radio networks. In: IEEE Vehicular Technology
reduce OPEX of the network. By comparing all the Conference; 2008 May 11–14; Singapore. p. 1811–1815.
[14] Mu~ noz P, Barco R, de la Bandera I, et al. Optimization
results for optimized and non-optimized cases, a quite
of a fuzzy logic controller for handover-based load bal-
significant improvement in network KPIs is observed. ancing. International Workshop on Self-Organizing
The proposed self-optimization model can easily be Networks (IWSON), IEEE Vehicular Technology Con-
extended to other RRM parameters like mobility load ference (VTC), Budapest; 2011 May.
balancing and packet scheduling. [15] Rodriguez J, de la Bandera I, Munoz P, et al. Load bal-
ancing in a realistic urban scenario for LTE networks.
IEEE Vehicular Technology Conference (VTC Spring),
Disclosure statement Budapest; 2011 May.
[16] Islam NU, Mitschele-Thiel MA. Reinforcement learn-
No potential conflict of interest was reported by the authors. ing strategies for self-organized coverage and capacity
optimization. In: IEEE Wireless Communications and
Networking Conference (WCNC); 2012; Shanghai,
References China. p. 2818–2823.
[17] Razavi R, Klein S, Claussen H. Self-optimization of
[1] Kato T. Next-generation mobile network. FUJITSU Sci
capacity and coverage in LTE networks using a fuzzy
Tech J. 2012;48(1):11–16.
reinforcement learning approach. In: 2010 IEEE 21st
[2] 3GPP TS 36.300: Evolved Universal Terrestrial Radio
Int Symposium on Personal Indoor and Mobile Radio
Access (E-UTRA) and Evolved Universal Terrestrial
Communications (PIMRC); 2010 Sep 26–30; Instan-
Radio Access Network (E-UTRAN): overall descrip-
bul, Turkey. p. 1865–1870.
tion; Stage 2, version 11.2; 2012.
[18] Li J, Zeng J, Su X, et al. Self-Optimization of Coverage
[3] NGMN, Recommendation on SON and O&M
and Capacity in LTE Networks Based on Central Con-
Requirements, version 1.1; 2008 Dec.
trol and Decentralized Fuzzy Q-Learning. Int J Distrib-
[4] Sallent O, Perez-Romero J, Sanchez-Gonzalez J, et al. A
uted Sensor Net. 2012;2012, Article ID 878595, 10 doi:
roadmap from UMTS optimization to LTE Self-Opti-
10.1155/2012/878595.
mization. IEEE Commun Mag. 2011;49(6):172–182.
[19] Yen K, Hanzo L. Genetic algorithm assisted joint mul-
[5] Hu H, Zhang J, Zheng X, et al. Self-configuration and
tiuser symbol detection and fading channel estimation
self-optimization for LTE networks. IEEE Commun
for synchronous CDMA systems. IEEE J Select Areas
Mag. 2010;48(2):94–100.
Commun. Aug. 2001;19:985–998.
[6] Magnusson P, Oom J. An architecture for self-tuning
[20] Hassan A, Doucet A, Amshah DI. GSR: a new genetic
cellular systems. In: Proceedings of the 2001 IEEE/IFIP
algorithm for improving source and channel estimates.
International Symposium on Integrated Network Man-
IEEE Trans Circuits Sys I: Regular Papers. 2007;54
agement; 2001; Seattle, WA. p. 231–245.
(5):1088–1098.
[7] H€oglund, Valkealahti K. Automated optimization of
[21] Hodges JL, Cam LL. The Poisson approximation to the
key WCDMA parameters. Wirel Commun Mobile
Poisson binomial distribution. Ann Math Stat. 1960;31
Comput. 2005;5(3):257–271 doi: 10.1002/wcm.212.
(3):737–740.
[8] Nasri R, Samhat A, Altman Z. A new approach of
[22] Tiwana MI, Sayrac B, Altman Z. Statistical learning in
UMTS-WLAN load balancing; algorithm and its
automated troubleshooting: Application to LTE inter-
dynamic optimization. 1st IEEE WoWMoM Work-
ference mitigation. IEEE Trans Vehicular Technol.
shop on Autonomic Wireless Access 2007(IWAS07),
2010;59(7):3651–3656.
Helsinki; 2007.