Feasibility of Green Network Deployment for Heterogeneous Networks

Green technology is a new term which is used to describe the energy efficient technologies. In the context of mobile communications industry, complying with the green technology strategy is a challenge. This is because of the tradeoff between the Quality of Service (QoS) provided and the total energy used in the transmission. Reducing the transmission energy may cause degradation in the QoS, more distinctively, in highly populated areas. This paper explores the possibility of achieving the green technology goal in planning and deployment of the HetNet mobile network with efficient network QoS. A decoupled two stage multi-objective genetic algorithm is developed to provide the network base station distribution that would satisfy both the network QoS and green network demands. In the first stage the algorithm estimates the base station parameters for more energy efficient HetNet deployment for optimum network coverage. The initial base station candidate locations are provided by a network operator in Kuala Lumpu, Malaysia. The second stage of the developed algorithm selects the number and location of RS associated with each base station optimized in the first stage to improve the network capacity. To optimize the network power, a novel arrival rate based HetNet total power consumption model is derived to investigate the parameters that affect the network power expenditure. Results show that a remarkable energy saving of about 40 % of the operator transmission power could be achieved with full network coverage. The addition of RS associated with each base station would greatly improve network capacity on the expense of its power expenditure. The relative RS to base station capacity plays major rule in reducing HetNet power expenditure.


Introduction
Since the industrial revolution at the beginning of the nineteenth century, the demand of energy resources has been growing up relentlessly.This sudden and huge demand has had its impact on the environment leading to emerging issues of imbalance, and threatening our existence on earth due to global warming.As a result, the need of continuing the current life style has to consider sustainability, which could only be achieved if the environment is considered in the equation.The term green technology refers to all the measures and actions that are conducted in order to efficiently utilize the energy resources, nevertheless, producing the same desired product quality.
Mobile communication networks shares a significant part of the global energy consumption.Nowadays, the booming reliance on mobile services caused the requirements of mobile communication services to grow very rapidly.This increase adds tension on the power resources since it needs wider bandwidth and higher data rates to accommodate the market demands and the expanding number of users and services.Annually, it is estimated that 58 MW is the total power consumption of a network consisting of 20,000 3G base stations [32].Base stations transmission power accounts for 60-80 % of the total power consumed in the network [21].These power consumption figures motivates the consideration of the green network planning as an essential step to reduce the impact of wireless technology on the environment, especially that there is a collective aim globally to decrease carbon footprint and total power consumption up to 20 % by 2020 [7].Green Network planning optimizes the network by spatially distribute the base stations and adjust their transmission power and configuration such that the total power of transmission is minimized.As the QoS is the main concern of all mobile communication companies, green network planning should consider Quality of Service (QoS) in the planning process to ensure the compliance of the green network with the requirements of mobile communication standards.
Long Term Evolution (LTE) is a global term describing what is commercially known as 4G mobile broadband network.The ability of the operators to preserve higher peak throughputs during higher spectrum bandwidth is the key advantage of LTE networks over HSPA?.In order for an LTE network to maintain high peak data rates in high spectrum bandwidth, it uses Orthogonal Frequency Division Multiple Access (OFDMA) for downlink.For uplink, however, it employs an approach known as Single Carrier FDMA (SC-FDMA) [23].Similar to traditional networks, LTE base station power consumption compliance to the green network definition remains an open question.
Varieties of optimization methods are developed for base station location problem in [3,12,14,15,18,[24][25][26][27].The wireless network planning and designing problem is classified as stochastic problem.One of the methods to solve such problem is genetic algorithm, which is used to solve non-polynomial hard problems.The early use of genetic algorithms to solve the problem of wireless network design was in [17], GA was used to maximize the SNR and minimize the transmitter power.Authors in [8] addressed the complexity of planning UMTS.Two algorithms were employed to build two mathematical models: greedy algorithm and genetic algorithm.Their system models showed that both algorithms gave same results in terms of number of base station to cover the same area.
However, the greedy algorithm was faster to reach the solution.Later in [10], the authors used the GA to maximize the coverage and minimize total cost.The results showed that the GA solution covered only 70 % or less of the target area.That is because the GA fitness function only considered the coverage area over the number of selected base stations.The SNR, SI and transmitter power were not included in their fitness function.Authors in [20] faced the same problem as they only considered the coverage in their fitness function.Although they compared three approaches with different population sizes and the random scenarios, the GA fitness function needed to be improved.The multi-objective genetic algorithm obtained a population of solutions by the Pareto fronts to solve the wireless design problem in [22].In their work, GA complexity and the convolution of the problem resulted in a parallel run of the algorithm because of the multi-objective fitness function and multi-level code.A simple genetic algorithm was used in [19] to solve the problem of base stations locations.Although the GA fitness function was based on number of the test points that satisfied the requirements, the study did not consider all planning parameters such as the path loss and delay.
Researchers have done further studies to solve the wireless network radio design and planning problem.In addition to employing genetic algorithm, varieties of heuristic algorithms have been explored.For instance, authors in [9] investigated three heuristic algorithms: greed-like algorithm, Darwinism algorithm and island-based Genetic algorithm.The three algorithms were investigated to examine their capability to solve the optimization radio planning problem.The results showed that the island-based Genetic algorithm gives the best results, however, the computation time of GA is larger than the rest algorithms.The GA was implemented in [25] to solve the UMTS planning problem, the traffic distribution, propagation model and candidate site are inputs of the problem.GA compared to Enumerative method and its showed less competition time in contrast to Enumerative method.
The SINR is used as the fitness function in the GA system developed in [16].The antenna parameters which are Antenna down-tilt, Antenna azimuth, Horizontal beamwidth, and sub-channel power optimized distinctly to improve the average SINR of the wireless network in deployment stage.The achieved SINR improvement was in the maximum up to 45 % by optimizing the sub-channel power and the antenna parameters.However, the single objective fitness function implemented in this work, doesnt consider the network coverage nor the total capacity of the network.The LTE planning and deployment were studied in [29], the Physical Cell ID (PCI) planning was discussed and the interference matrix was build, GA was used to minimize the interference by adjusting the PCI distribution.The main distinction between the different methods is the objectives used which are based on the network requirements and the constraints imposed in the optimization criteria.Particle swarm optimization (PSO) method is developed in [34] to reduce the total system cost.The objective functions used are transmission rate, hand off and number of antennas.For different scenarios, their proposed PSO outperform the uniform planning in terms of number of antennas and total system cost.Tabu search algorithm and GA are compared in [15] for optimization of the number of base stations deployed for a specific area.The main objective of their work is to reduce the number of base station while maintaining maximum number of users served.Both methods showed a satisfactory convergence with better optimization achieved from the Tabu search while GA provided faster solution.The same conclusion was drawn from [27] where GA, Tabu search and simulated annealing algorithms has been compared.In [18], different scenarios of optimization have been compared for optimizing the configuration of base stations for Single Frequency Networks (SFN) using GA.The scenarios focused on the carbon dioxide footprint (CO 2 ), energy consumption, coverage optimization, number of active transmitters, and safety index as the objective functions.The authors compared the results by running each objective distinctively.Energy efficiency objective provided the minimum CO 2 footprint and safety index.Genetic algorithm methods were developed for UMTS networks in [14].Different objective functions have been evaluated to check the distribution of the resulted network.In [24], transmission power is considered part of the objective function to ensure signal strength received by the user to be bounded to provide required signal strength and satisfy the safety limits.More recently, network planning has been addressed for a uniformly distributed random base stations locations for operating network in Kuala Lumpur, Malaysia [3].The total energy optimization was the main objective function.The GA estimates the best position of the base stations, transmission power, and height.The results demonstrated the possibility of deployment of energy efficient network.A similar work is introduced by [33] using single objective genetic algorithm to propose a cost effective HetNet LTE network by minimizing the number of base station deployed.The proposed Genetic algorithm fitness function in [33] does not consider transmission power of the base stations.Though no analysis of coverage or capacity is provided clearly for the genetic algorithm proposed topology, it is claimed that the capacity and coverage are fulfilled.The relation between Energy Efficient (EE) network and user Quality of Experience (QoE) is investigated in [35].They introduced a new QoE measure and highlighted the importance of maintaining the QoE when planning energy efficient networks.
In this paper a green network planning strategy is investigated for HetNet topologies comprising major base stations (BS) each is associated with a set of relay stations (RS).A decoupled two stages genetic algorithm (DGA) is developed to optimize the network topology in order to achieve both network QoS and green network operation demands.The DGA first stage optimizes the distribution and configuration of the BS to provide full coverage of the network whereas the network capacity is the concern of the second stage of the DGA.In the second stage, the number and location of RS associated with each BS proposed in the first stage are optimized.In both stages, the proposed DGA utilizes a multiobjective fitness functions each consists of minimizing and maximizing objectives.In the first stage, the minimizing objectives are the total network power expenditure, number of base stations, and safety index.The maximizing objectives contain coverage probability.The main criteria for the proposed HetNet plan are to comply with the green technology principal while it provides the requirements of the coverage probability and received signal strength.In the second stage, the fitness function maximizing objectives contains the network capacity and the safety index whereas the minimizing objective in the second stage of the DGA is the total power of network.To accurately estimate the total power of the HetNet at any traffic arrival rate, a novel power model is developed considering both the RS and BS power consumption in the network at any given traffic arrival rate.Furthermore, the developed power model considers the spatial distribution of the traffic arrival rate in the region of interest through the RS load factor.
The rest of this paper is organized as follows: Sect. 2 investigates data collection campaign.System model is presented in Sect. 3 followed by a description of DGA and the objective functions in Sect. 4. Fifth Section discusses the results.Finally, the results of this work are concluded and future work is recommended.
Network planning involves distribution of base station antennas in the Region of Interest (RoI) to ensure a full coverage of the area.Base station antenna heights and transmission power should be selected carefully to provide the required QoS and to keep the field exposure and total power consumption within the acceptable range.Therefore, to configure the DGA system to choose the suitable green network plan in a specific area, a digital representation of the area and the base station planned locations along with the base station heights and transmission power are required.Road test data will help to validate the DGA proposed plan.
The RoI is represented by Digital Terrain Map (DTM) with a resolution of 1:5000 m provided by Jabatan Ukur Dan Pemetaan Malaysia (JUPEM).The DTM covers the metropolitan region of Kuala Lumpur city in Malaysia of an area of 54 km 2 .The projection used in the map is WGS84.The details of the latitude and longitude coordination are shown in Table 1.
Network base station distribution and configuration are provided by the Malaysian Communications and Multimedia Commission (MCMC) for Kuala Lumpur region.The planned base station antenna heights are calculated from the ground level and the transmission power is in dBW representing the EIRP power.All the base stations are Kathrin 742 215 model and polarized horizontally with continuously adjustable electrical tilt from 0 to 10 .The locations of the base stations are in decimal latitude/longitude format projected on WGS84.The initial plan contains 160 base stations distributed in the region transmitting at 2112.4 GHz for downlink and 1922.4GHz for uplink.The base stations antenna gain is provided as 18 dBi in the specified range of frequencies.
For the purpose of validating the DGA proposed scenario received signal strength index, estimation of path loss is performed in a route provided by MCMC used for regular roadtest as shown in Fig. 1.The calculation of received signal strength index is accomplished for the DGA proposed plan and for the operator plan.

System Model
Assuming a network consisting of k base station antennas distributed in a region of size m Â n km 2 where each base station location is represented by x i latitude and y i longitude forming a set of candidate locations: To allow the DGA to select the active base station locations, the locations where the base stations are installed, each location will be associated with a digital flag defined as follows: The height of each base station antenna is selected from a set of valid heights represented as: where q represents the length of the allowed base station antenna height list.Similarly, the valid transmission power can be defined as a list of allowable transmission power where each base station antenna transmission power is a member in the set as follows: where z is the length of the allowed base station antenna transmission power list.This configuration increases the degree of freedom of the planning process where the DGA is allowed to select the locations, heights and transmission powers independently for the base station antennas to satisfy the required criteria.The proposed solutions of the DGA will be evaluated by calculating the required parameters on a set of regularly distributed test points represented by the set of test location pairs: where u is the number of test points representing users.The proposed network configuration must satisfy two constraints.The first constraint is that all the base station antenna heights in the network are within the range of the defined base station antenna heights.Similarly, the proposed base station antenna transmission power should be in the range specified in the transmission power list.The network capacity is mitigated in the second stage of DGA by introducing a set of RS for each BS optimized in the first stage.Thus, for each BS i ; 1\i\k the set of its associated RS can be defined as: where N i RS is the number of RS associated with BS i , and D i RS is the distance of the RS from BS i .

HetNet Total Power Consumption
The considered HetNet topology consists of a set of k major BS each is associated with a set of RS where the total number of RS in the network is denoted as m as clarified in Fig. 2. In this scenario, the BS and and its associated RS are represented as clusters serving a group of users.The number of BS in the network and the number of RS associated with each BS are a selection parameters based on the coverage and capacity constraints in the network.The RS are assumed to have full BS transmission protocols, i.e. type 3 RS [4], with limited capacity and coverage.The RS and BS power consumption is assumed to vary linearly with the load factor carried by the RS or BS, that is: where P BS and P RS are the power consumed in the antenna power amplifier and cooling system respectively, F BS and F RS are the BS and RS transmitter power amplifier efficiency, and P L BS and P L RS are the power sensitivity to the load carried by the transmitter.The BS and RS power sensitivity to the load P L BS and P L RS represent the variation of the power consumption associated to the percentage of the transmitter bandwidth used in transmission.Denoting the transmitter (BS or RS) load factor as a BS and a RS for BS and RS respectively, the power sensitivity in the BS and RS can be represented as: where a BS and a RS are the load factor of the BS and RS, a BS and a RS 2 ½0; 1.Thus, the total power consumption in the network can be calculated as the aggregation of the power consumed in each BS and RS as follows: However, the RS are deployed to provide higher network capacity and, thus, they carry part of the traffic load.Consequently, the BS total a BSi is related to the total a RSo .This relation can be derived by defining the transmitter normalized load factor as the ratio of the occupied bandwidth R served to the total bandwidth offered R max , hence, for BS and RS respectively, the load factor is: The total normalized arrival rate is defined as the ratio of the total bandwidth occupied in all the BS and RS to the total bandwidth offered by the BS and RS in the network as follows: The arrival rate definition in Eq. 14 can be further simplified by assuming the total maximum bandwidth offered by the BS and RS as follows: By substituting Eqs. ( 15) and ( 16) in Eq. ( 14), the normalized arrival rate is: Rearranging the terms in Eq. ( 18) to have: Calculating for the normalized total load factor of the BS to have: Substituting Eq. ( 19) in Eq. ( 11) and rearranging the terms would yield the total power consumption in the network as a function of the arrival rate and the total load factor in the RS as follows: The total power consumption in HetNets is, obviously, a function of the number of BS and RS deployed in the network, the total load factor shared by the RS and the normalized arrival rate.More importantly, the relative RS to BS capacity influences the total power consumed in the network as shown in Fig. 3.

Network Capacity Model
The capacity model of the network is represented by defining a set of Users denoted as UE ¼ 1; 2; 3; . ..; U each one is requesting a service of bandwidth denoted as r u in kbps.Thus, the total bandwidth requested from the network is given as: Furthermore, each UE is characterized by its position in the area around the base station.This position is represented by a set of UE loc ¼ fðx u ; y u Þju ¼ 1; 2; . ..; Ug.This location set determines the propagation conditions associated to UE due to its distance from each BS.The set of the transmitters in the network is characterized by BS and RS sets.The set of BS contains the set of locations represented as in Eq. ( 1), set of heights as defined in Eq. ( 3), set of transmission power as denoted in Eq. ( 4), and set of sectors for each BS given as BS sec ¼ fS pj jp j ¼ 1; 2; . ..; P j ; j ¼ 1; 2; . ..; ng as shown in Fig. 4. Similar to the BS, the network contains RS which are distributed around each BS.The RS are used to provide capacity and coverage enhancement to the network.The main difference between the BS and the RS is that RS are considered as omni directional antenna, i.e.RS are not divided into sectors.The height and transmission power of all the RS in the network are considered fixed and denoted as h RS and TP RS respectively.Each BS is considered to be associated with a set of RS distributed around the respective BS.Thus, the list of RS for each BS is characterized as in Eq. ( 6).The grouping between the BS and its set of RS is defined as an association matrix U ¼ ½u io , where u io ¼ 1 if RS o is associated to BS i .Each RS is allowed to be associated with only one BS given that the distance between the RS and its respective BS is bounded within the radius of the coverage area of the BS.This topology is clarified in the schematic diagram shown in Fig. 2 where the BS sectors are not shown for clarity.The path loss at each UE position is calculated by ERICSSON 9999 urban model.The UE direction and antenna tilt effects are considered by defining a variable gain matrix g ¼ ½G lj where G lj ; l ¼ 1; 2; . ..; k; j ¼ 1; 2; . ..; n denotes the BS j gain perceived at UE l location.The UE will be served from the BS or any associated RS based on the signal quality at its location and the total system capacity Rc.Thus, the total number of BS, their location, height, transmission power and the number of associated RS and their position are design parameters to be selected such that the total capacity offered by the network is higher than the demanded capacity in Eq. (21), that is, the network capacity efficiency is defined as: The network capacity efficiency, according to Eq. ( 22), has a lower bound of R c ¼ UE tc .
The main aim of the optimization algorithm is to meet the capacity and QoS demands at each UE position.Furthermore, this has to be achieved with minimum network power consumption.

Distance Calculations
The distance is calculated using the WGS84 reference ellipsoid with the parameters listed in Table 2.

Shadow Fading
In dense areas shadow fading is represented as a summation of lognormal random variables.Each lognormal random variable represents the contribution of an obstacle blocking the line of site between the transmitter and receiver.Estimating the mean and standard deviation for the summation of the lognormal random variables is accomplished using the Wilkinsons method, described in [5,13], assuming that the resulted summand is lognormal distributed random variable.The choice of Wilkinsons method in this work is due to its simplicity and validity in the range of standard deviation and mean used in this work.The number of lognormal variables used in the summation reflects the urbanization of the area under consideration.

Coverage Probability
The cell coverage probability is computed by considering the lognormal shadow fading with standard deviation r.The detailed derivation of the coverage probability is shown in ''Appendix''.
To achieve the required coverage probability, the SINR at the edge of the cell should be maintained greater than or equal to the minimum allowed SINR.For downlink, LTE network minimum SINR is reported to be in the range of 16.5-11 dB for the given transmission power range [6,30].The path loss at the edge of the cell is represented as: where: P t is the transmitting power of every base station.G t is the transmitter gain ¼ 18 dBi.L b body losses ¼ 4 dB.Using these values in Eq. ( 23) and considering the range of transmission power yields the mean path loss at the edge of the cell to be about 150 dB.This value represents the maximum path loss at the edge of the base station.The SFM is, then, calculated as: The proposed system calculates the coverage probability for all the base stations available, and considers the minimum coverage as the objective function to be maximized.

Safety Index
Safety index is defined as the percentage of the amount of power received per unit area of the receiver to the allowable exposure limit.To avoid any health hazard from the propagating microwave, standards identified the limits of exposure allowed for human body [11].In multi-antenna network, safety index is calculated as the total sum of the fractions of power densities received at the user site for every frequency (in case of multi frequency transmission).Power density at a specific distance r (m) from an base station antenna transmitting at power P (W) can be calculated as: For n base stations, the total power density is: According to the IRPA (International Radiation Protection Association) Radio frequency safety guidelines and Standards, the general public exposure limits to time varying electromagnetic field at 2000 MHz is 10 w/m2 [28].Thus the safety index can be defined as:

Signal to Noise Ratio
To account for the interference from the neighboring BS, Signal to Noise Ratio (SINR) is calculated as: The developed DGA consists of two stages to optimize the HetNet topology to meet the green network concept with efficient QoS.The first is optimizing the base stations distribution and configuration to provide the topology that would have full coverage in the RoI with minimum number of BS deployed.The second stage optimizes the number and location of the RS associated with each BS optimized in the first stage so that the network capacity meets the demand.The two stages of the developed DGA are functioning distinctively and sequentially.Thus, each stage implements its distinctive fitness objectives and penalty, however, the output of the first stage is the input to the second stage.

First Stage DGA
The main objects in the DGA are the genes which form the chromosomes.Each genome represents the configuration of a base station and, hence, each chromosome describes the total proposed network configuration.In this work the chromosomes are constructed as binary strings of g Ã k digits where g is the number of digits for each genome and k is the total number of genomes (base stations) in the topology.The number of digits on each genome g can be computed from Eqs. ( 2), ( 3) and ( 4) as follows: The operator : d e represents the ceiling function.Where the first digit is assigned for the flag, followed by the digits of the heights and then finally the digits of the transmission power, the construction of each genome and the corresponding chromosome are demonstrated in Fig. 5.
Initially the first stage DGA proposes a uniformly distributed set of chromosomes called the population.Based on a set of objective functions, the DGA will narrow down the search domain to the best population.The following subsections discuss the objective functions and the steps considered in the first stage of the DGA to converge to the solution.

First Stage CMOGANO Fitness Function
Each chromosome in the generated population will be evaluated based on a set of network model parameters and accordingly it will be assigned a fitness value indicating its suitability.The main aim of this work is to check the ability of the LTE network to comply with the green technology concept.This compliance suggests the reduction of total power consumed by the network while maintaining the QoS.Hence, the objective functions are designed in two groups.The first group consists of minimizing functions while the second group contains the maximizing functions.The following subsection will explain the minimizing group of functions followed by the maximizing functions The minimizing objectives used are:

• Number of active base stations
The first strategy to reduce the total power consumed is by reducing the number of base stations used in the network.Thus, number of active base stations proposed is impeded in the objective functions as follows: where k is the total number of initial transmitters • Total power As each base station transmission power is considered a decision parameter in the DGA, the total power consumed in all the BS should be mitigated as a minimizing objective function.The total power consumed in the HetNet BS is calculated for the first stage DGA by Eq. (7).By controlling the transmission power of each BS, the total power expenditure in the network could be minimized.

• Safety Index
Minimizing the max SI received by the users from all the available base stations ensures the compliance of the proposed network configurations to the IRPA limits.The SI can be represented by: where SI is the safety index calculated using Eq. ( 27) for u users.
Similarly, the maximizing objective is -Coverage probability Reducing the total transmission power of the network and the number of active base stations has a negative effect on the QoS.Quality of Service can be maintained by maximizing the coverage probability of the base stations.The cell coverage probability Pcell is calculated using Eq.(41).Maximizing the minimum coverage ensures the agreement of the proposed network configuration with the requirements of QoS.Thus, the coverage probability objective function is written as: Thus, the first stage DGA fitness function is defined as the summation of the weighted objective functions as follows: subject to the restriction: This fitness function will be evaluated for every chromosome to gauge its suitability.The chromosomes which produce the minimum fitness will be selected to undergo the crossover and mutation to generate the next offspring generation.

First Stage CMOGANO Penalty
Due to the random generation of population in the GA, the chromosomes might include base stations with heights or transmission powers which do not comply with the height or power limits or produce coverage probability not in compliance to the requirements.To reduce the possibility of the participation of those chromosomes in the next generation, a static penalty is applied on their fitness function as follows: where Ft p is the penalized fitness value, Ft is the fitness value, c 1 is a constant assigned the value 100.The coverage probability penalty is applied when the coverage probability x covp is less than a threshold: where covp min is the minimum coverage probability.The threshold of the coverage probability is assumed to be 0.95.

Second Stage DGA
The first stage of CMOGANO optimizes the location and configuration of the main BS such that the coverage and power efficiency demands are met.However, the network capacity is not considered in the optimization process in the first stage.This is because the capacity of the network is enhanced by adding the RS distributed around the selected BS.Thus, the aim of the second stage of the CMOGANO is to optimize the configuration of the RS in the network such that the capacity of the network meets the requirements of the network operator.In this stage of CMOGANO, the selection parameters are, simply, the number of RS per BS and the distance separating the RS from the respective BS as shown in Fig. 2. Accordingly, the chromosome structure, in the second stage, and the fitness penalty function are constructed for the purpose of RS number and position optimization.Similar to the first stage, CMOGANO second stage gene is a binary string representing the number of RS per BS and their distance as shown on Fig. 6.

CMOGANO Second Stage Fitness Function
The second stage fitness function is constructed to optimize the network capacity.Thus, network coverage probability is replaced by the network capacity computed in Eq. 22.Moreover, the second stage fitness function includes the SI and total power of the network as minimization objective parameters whereas the network capacity is the maximization objective.The second stage fitness function is computed as: where w i ; i ¼ 1; . ..; 6 are weights satisfying

Second Stage CMOGANO Fitness Penalty
The network coverage is optimized in the first stage by selecting the appropriate base station location and configuration.Thus, the addition of the RS in the second stage of the DGA would not reduce the coverage in the network.However, the SINR perceived by UE might be affected by the addition of the RS due to the RS to RS interference and RS to BS interference.Therefore, the penalty in the second stage is applied when the SINR received by the UE is less than a threshold.
where Ft p is the penalized fitness value, Ft is the fitness value, C 2 is a penalty constant equal to 100, minSINR ij is the minimum SINR perceived by the user from his respective transmitter (RS or BS), and SINR th is the SINR threshold assigned the value of -20 dB.
5 Decoupled GA Computational Complexity Analysis GA computational complexity analysis has been investigated in [1,31] by introducing the domino convergence and drifting rates.Domino convergence is the sequential of bits in the genome starting from the most salient to the least salient genes.For small population size, least salient genes may be lost by mutation probability prior to the effect of domino convergence.This problem gives rise to the random genetic drifting phenomena [1].The combination of domino falling and random genetic drifting characterizes the GA computational complexity.
In [1], domino falling convergence time is modeled mathematically as a linear function of the population size as follows: where k is the block size (bits), I is the selection intensity, and k t is the number of converged blocks.
Similarly the random genetic drifting is shown to behave according to the following model: where N is the population size.Both Eqs. ( 40) and ( 41) can be combined to predict where in the string is the drift likely to start as follows [1]: Equation (40) shows that the number of converged blocks k is a linear function of the population size N and the linearity factor is a function of the block size k.Thus, for a given population size N, the size of the block in the GA is the main contributor in the speed of domino convergence.For a GA operating at a population of size N and strings consisting of blocks of k À n bits length, where n ¼ 0; 1; 2; . ..; k À 1, the drift is likely to start at: The ratio k k2 yields: The ratio represented in Eq. ( 44) shows the effect of reducing the block size by n bits on the GA domino convergence rate as compared to k block size strings.This ratio is shown in Fig. 7 for different values of block size k and n ¼ 1.It can be deduced that a reduction of one bit of the block size may enhance the domino convergence rate by more than five times.This result emphasizes that breaking up the heterogeneous network planning problem into sub-problems operating smaller block size would enhance the theoretical computational complexity of the system.Block size (k), n=1 GA algorithm performance ratio ( Fig. 7 GA performance ratio for different block size k, and n ¼ 1 The operator topology used in the results includes the network distribution and the base stations configuration in excel sheet format.The data provided includes the configuration a total of 160 base stations candidate locations covering 54 km 2 area.The topology proposed by the DGA is compared with the operator and with a random plans in terms of coverage probability, the exposure safety index, total power of transmission, and number of base stations used.All the experiments are conducted on Dell station model Precision T3610, intel Xeon CPU 3GHz with 8GB RAM running on Windows 7, 64 bits OS.The proposed system is implemented on MATLAB.The DGA algorithm parameters used for both stages are listed in Table 3 and the network parameters are shown in Table 4 for the first stage.
The second stage DGA parameters are given in Table 5 for optimizing the RS per BS.
The following subsection starts by discussing investigating the proposed topology QoS in contrast with the operator and the random plans followed by the network capacity discussion.The last subsection elaborates on the validity of the proposed network in terms of path loss and its compliance with the green network definition.

QoS of the Proposed Scheme
The scope of this discussion is to examine the behavior of the objective functions during the optimization and the validity of the final solution proposed by the GA in terms of total transmission power, number of base stations proposed, the distribution of the network, the coverage probability, and the exposure safety index.Figures 8, 9, 10 and 11 shows the progress of the objective functions during the optimization.The network coverage probability shows a significant improvement and stability as shown in Fig. 8, reaching approximately 98 % which is well above the standard of 95 % required by the operator.This high coverage probability resulted from a network transmitting at only 41 % of the total power assumed and only 54 % of the total number of base stations initially planned by the operator as it is clear from Figs. 9 and 10 noting that the operator network was transmitting at 67 % of the total transmission power assumed with 70 % of the total base stations are activated.This reduction in total power motivates the Green network planning strategy in HetNet topologies.
Moreover, comparing Figs. 12 and 13 shows that the DGA first stage produced similar BS distribution increased BS density in the high population density areas.This distribution helps improving the coverage probability utilizing less number of base stations and lower levels of total transmission power.The path loss map of the operator network and the path loss map of the GA proposed network topolgy are shown in Figs. 14 and 15 respectively.Comparing both path loss maps, it is clear that the GA proposed network managed to maintain coverage levels in the range of the operator coverage with minimized number of BS.Finally, the safety index progress in Fig. 11 shows stability with values in compliance to the standard.

Network Capacity
The network capacity is considered in the second stage of the DGA by adding a set of RS to each BS proposed in the first stage.For capacity calculation, the actual population size in the region is considered.The user capacity demand, r u , is selected from [64, 128, 256, 512] kbps distributed normally over the users.The total number of users considered in this area is the population density at rush hours multiplied by the area of the RoI.Based on the data provided by JUPEM, the population density is 6696 inhabitants per square kilometer.Thus, in a region of 54 km   RS allocated per BS is 0 and maximum is 32 in steps of 2. Similarly, the RS distances from the respective BS is selected from a list of distances ranging from 400 to 1000 m in steps of 100 m.The optimization of the network capacity in the second stage of DGA is depicted in Fig. 16.This capacity is achieved with the number of RS per BS is distributed as shown in Fig. 17 with average number of RS per BS in the range of 6.The distribution of the distances of the RS from their respective BS for the proposed DGA topology is shown in Fig. 18.The total network power is calculated using Eq. ( 13) and is shown in Fig. 19 for the operator network, the first stage of DGA optimization, and the second stage of

The Validity of the Proposed Scheme
To examine the validity of the DGA proposed network, the path loss map is calculated for the operator network plan and for the DGA proposed network.The DGA proposed network path loss map, shown in Fig. 15, demonstrates the validity of the proposed network.The path loss provided by the proposed network ranges between 139.364 and 139.53 dBW as compared with the operator network in Fig. 14 which is in the range 140,883 and 140.052 dBW and the random plan which is in the range of.Furthermore, the proposed network parameters in terms of coverage probability, the received RSSI level, SINR, and SI distributions are compared with the operator network and the random generated plan in Figs. 20, 21, 22 and 23 respectively.The proposed network parameters show a good proximity to the operator network parameters, however, the random generated network resulted in the lowest RSSI and coverage probabilities.The received power estimation for the operator plan in the route provided by the MCMC are compared with the calculated received power for the proposed network in Figs.24 and 25 shows the CDF of the operator plan received power and the received power calculations of the proposed network respectively.
The provided results illustrate the applicability of the green network notion to the HetNet networks without a significant loss of QoS.Moreover, the introduction of RS technology in the HetNet provided great insights in the improving the total network Fig. 14 Operator network EIRP path loss map capacity.It is worth to mention that HetNet topologies with small BS (RS) can be optimized for energy efficient network topology.The introduction of RS in HetNet may cause a remarkable increase in the network total power expenses.This motivates the adoption of the green network deployment strategy even for HetNet networks.To further enhance the power efficiency of HetNet topologies, traffic aware RS switching algorithm can be implemented which switch off the RS during low traffic rates.However, due to the complex relation resulted from the capacity differences between the BS and RS and their power consumption a robust mathematical model is introduced in [2] to optimize the switching parameters and provide efficient traffic aware RS switching algorithm.Needless to mention the encouraging effects of such optimized network deployment strategies in reducing the degradation of the environment quality on the long run.This paper covers the first of green wireless network optimization for LTE project.The first stage is the GA optimization of the major base station locations, height and transmission power.the second stage of the project concerns the network capacity, traffic, and user demand.In this paper Green network planning is investigated for LTE network.A new multi-objective decoupled genetic algorithm system is developed to propose the optimum HetNet configuration.The DGA first stage proposes the availability of the base stations and their locations from a list of candidate sites proposed by the operator along with the transmission power setting for each base station and its height.The system selects the best network based on defined weighted objective functions.The RS are configured in the second stage DGA to optimize the network capacity.In this stage, the DGA selects the  number of RS per BS and the distance of the RS the BS to provide the required capacity demands.For the purpose of comparison, the operator network is assumed to install base stations in all the operator candidate sites with the antenna heights and transmission power proposed by the operator.Moreover, the generality of the proposed DGA plan is highlighted by comparing its performance metrics with a random generated plan with the same number of BS proposed by the DGA is performance metrics.To calculate the total power consumption in HetNet, a novel arrival rate based power consumption model is developed.This total power model shows the significance of the RS to BS capacity ratio in HetNet total power consumption.The DGA proposed network plan coverage probability, received signal strength and interference noise are calculated and compared with the operator network plan.The results show that a reduction of up to 40 % of transmission power utilized by the operator network can be achieved.This reduction   resulted from the utilization of less than 60 % of the total base stations to cover the same area with optimized transmission power for BS.The coverage probability and safety index of the proposed network are in close matching to the operator network plan and the random generated plan.Furthermore, road test path loss estimation in the operator plan is compared with the proposed network plan estimated path loss, and it demonstrated the validity of the proposed network plan.In the second stage the DGS added 516 RS in total with and average of 6 RS per BS to improve the network capacity.This addition of RS boosted the network capacity to the range of 98 %, however, it significantly increased the total power consumption in the network, particularly, during high traffic rate periods of time.In general, this work verified the feasibility of Green network planning in mobile communication systems.Aside to the energy efficiency achieved, Green network concept could reduce the installation and operation costs of the network while providing comparable area coverage, received signal strength, capacity and signal to noise ratio.Moreover,  CDF CDF for GA using Operator Plan CDF for Operator Plan CDF for GA using random Plan Fig. 22 CDF of perceived SINR in the area of interest the results of this study show that compliance of operating network to the green network concept is attainable by optimizing the locations of the BS and RS and their configuration in the future versions of HetNet.The objective functions used in this are effective in guiding the DGA to propose a balanced network plan with acceptable QoS measures.However, other objective functions could be utilized based on the requirements of the system.To provide more emphasis on the users spatial distribution, frequency reuse and user density are recommended to be included in the objective function calculation in the future work.
where r is the user distance from the base station and R is the radius of the base station.Using Eq. (47) P cov ¼ PrðX [ A À Blog 10 ðRÞ À ðA þ Blog 10 À SFMÞ ð 51Þ From Eq. ( 45), the coverage probability can be written as: Defining t ¼ x=r and substituting in Eq. ( 53) Equation (55) defines the coverage probability at a distance r from the base station.For cell coverage probability (Pcell), it is defined as the average of the coverage probability at all possible locations covered by the cell.Mathematically Pcell can be written as: where Pðr; uÞ is the distribution function of the mobile users.
For uniform users distribution, the probability of a user to be at distance r with angle u from the base station can be defined as follows: Pðr; uÞ ¼ r pR 2 ; 0\r\R; and 0\u\2p From Eqs. (53), ( 56) and (57), the cell probability can be written as: where the erf is defined as: Let define Calculating for r, yields Equation (66) will be: Changing the limits of the integration and substitute in Eq. ( 58), yields: By integration Eq. (71) by parts, the cell coverage probability is represented as:

Fig. 3
Fig. 3 Total power consumed in the network for different C RS2BS

Fig. 10
Fig. 9 total power of transmission percentage

Fig. 16
Fig. 15 GA proposed network EIRP path loss map Fig.17CDF of the number of RS per BS in the optimized network

Fig. 18
Fig.17CDF of the number of RS per BS in the optimized network

Fig. 20
Fig. 20 CDF of network coverage in the area of interest using Operator Plan CDF for Operator Plan CDF for GA using random Plan

Fig. 21
Fig. 21 CDF of received RSSI in the area of interest Fig. 23 CDF of perceived SI in the area of interest

Fig. 24
Fig. 24 CDF of road test

Table 2
r is the receiver antenna gain ¼ 0 dBi.L r is the receiver losses ¼ 0.5 dB.N r receiver noise ¼ -132.28 dB. G

Table 3
Decoupled genetic algorithm parameters 2, the total number of users is 361,584.Modulation technique considered, in this optimization, is 16-QAM with code rate 7/8 to provide the satisfactory bit rate.The DGA second stage optimization is by selecting the number of RS per BS and their distance from the BS to provide the required user capacity demand.The number of RS per BS is selected from a list where the minimum number of

Table 4
Total network power consumed for the operator network, first stage optimization, and second stage optimization with 516 RS added