Energy-aware wireless mesh network deployment using optimization mechanism

Wireless mesh networks are widely used to create network infrastructure in rural areas due to their flexible properties, such as self-healing mechanisms and associated redundant paths. For example, a wireless mesh network can suitably operate with limited battery power in wildlife monitoring applications. The locations of mobile routers to support mobile sensor nodes are essential for extending the system lifetime. This study proposes an energy-aware wireless mesh network deployment optimization mechanism. The goal is to determine a suitable location for the mesh routers with the aim of maximizing network lifetime. After the location solution is obtained from the proposed mechanism, it is then evaluated for system lifetime and network performance by the network simulator (ns-3). The proposed method outperforms the brute-force method in terms of computation time for all amounts of mesh clients. For example, for 30 mesh clients, the proposed method uses only a few minutes, while the brute-force mechanism requires more than 200 minutes to complete the process. Furthermore, compared to the brute-force method, it achieves nearly the same system lifetime and other performance parameters, such as throughput, packet delivery ratio, and packet inter-arrival time. In the real implementation, in which the sensor node placements can be changed during the installation period owing to the environmental status or the recommendation of the installers, the results can be recalculated in a short period. This is an open access article under the CC BY-SA license.


1.
INTRODUCTION Currently, Internet access is widely deployed in both urban and rural areas.High access speeds in the range of gigabit per second can be achieved owing to the availability of fibre to the home (FTTH) and asymmetric digital subscriber line (ADSL) in urban areas.The technologies are for fixed line location and too costly for long cable installation that are not suitable for remote scattering locations in rural areas.Conversely, wireless cellular technologies, such as 3G and LTE, are more popular due to their coverage areas and mobility support.High speed Wi-Fi is an another solution that is used in household and public areas.However, its coverage area is limited up to 50 m -100 m.
For wildlife monitoring applications [1], [2], such as tiger monitoring or forest protection [3], wireless communications are required for real-time and near real-time monitoring.In large wildlife sanctuary areas, such as Huai Kha Khaeng located in Uthai Thani and Tak provinces, Thailand, services of cellular technologies are Journal homepage: http://telkomnika.uad.ac.id

TELKOMNIKA Telecommun Comput El Control
❒ 27 only available at the edge of the forest.Unfortunately, at the monitoring site, wildlife animals usually live in deep forest where no wireless signal is available at all.To perform wildlife and forest monitoring, standalone camera traps are installed by wildlife patrollers in the deep forest.After depletion of batteries that last for about 30-45 days, camera traps are brought back to the camp site for retrieving captured pictures or video stored in the SD card.Patrollers must visit each site in harsh walking conditions at least twice.Furthermore, if suspicious or abnormal events occur, it is too slow for officers to adopt a decision and take any action that does not cause severe damage to the natural resources.
New versions of camera traps are equipped with a wireless communication capability for data uploading via 3G, LTE, or Wi-Fi.However, to deploy the camera traps, wireless network infrastructure is required at the monitoring site.Extending the coverage signal of the cellular network is not a feasible solution due to its restricted installation and operation cost for the service providers.
Specifically, Wi-Fi mesh network exhibits ad hoc network topologies and is an interesting solution to create a network infrastructure.Additionally, Wi-Fi mesh network is a self-healing network that is designed to recover communication problems due to broken links or malfunctioned nodes by using any alternative routes.In forest, all nodes are evidently operated with limited battery power that can be recharged by its installed solar power system.However, the solar charger may not always work due to cloudy skies or tree canopy conditions.Furthermore, batteries replacement requires significant time and efforts.Therefore, it is necessary to minimize energy usage to prolong network system up time.
To minimize energy usage, many routing management techniques are proposed, including spectral efficiency routing protocol [4], flow and power allocation routing protocol [5]- [7], load-aware routing algorithm [8], [9], harvesting energy mechanism based on energy-efficient routing protocol [10] and an energyefficient routing protocol with node mobility support [11].Besides, a hybrid (proactive and reactive) routing protocol [12] has also been proposed.Some other mechanisms include software-defined energy-aware routing protocol [13], [14], admission control for preventing fast battery depletion [15], operating certain nodes in sleep mode [16], limiting certain router functions for a period of a time [17], [18], energy aware multi-layer design [19] and topology control mechanism [20], [21].For channel management, a dynamic channel assignment for wireless mesh network [22] or implementing a genetic algorithm and neural network to optimize energy [23] are also presented.
All the aforementioned energy aware techniques are performed during operational period after nodes are installed at the predefined location.Evidently, locations of all nodes are a major factor for system energy usage.Inappropriate node locations can shorten system lifetime.When certain nodes are idle, some nodes can constitute a system bottleneck wherein energy is quickly drained and cannot be alleviated by any routing adjustment techniques.
It is possible to determine optimal locations of all router nodes off-line by using a brute force technique before the installation period.However, the calculation time can be excessively long.In a real deployment situation, locations of all sensor nodes are carefully selected by patrollers.During the installation period, the sensor node locations normally require some adjustments based on the site environment and animal trails.Exact sensor node locations may not be known before-hand, thus the mesh router location findings must be completed in a limited period during the installation phase.
In the study, an Energy-aware wireless mesh network deployment optimization is proposed.The main objective is to determine appropriate locations of all mesh routers in a short period of time by implementing optimization using the R-module.Additionally, selected placement locations should maximize the system lifetime for effectively operating and maintaining the system in the forest.Hence, after obtaining optimization results, the system lifetime and network performance parameters of all sensors and router nodes are evaluated and compared with brute-force mechanism by using an ns-3 simulator.
Related extant studies are presented in the next section.Subsequently, the proposed mechanism, problem formulation and performance evaluation are described in section 3, section 4, and section 5, respectively.The simulation results are presented in section 6.Finally, section 7 concludes the study.

RELATED WORK
For the wireless mesh network placement, most extant studies minimize the deployment cost.For example in [24], the mixed-integer linear programming model to reduce the installation cost of the wireless mesh network is proposed by accounting for traffic routing, interference, rate adaptation, and channel assignment.
Energy-aware wireless mesh network deployment using optimization mechanism (Aphirak Jansang) Furthermore, the mechanism can improve the calculation time with the same network performance by using a heuristic algorithm.However, energy awareness is not considered.By extending the algorithm ffrom [24], an enhanced algorithm [25] uses the traffic profiles to turn some nodes on and off to minimize energy usage while the required node transmission is still satisfied.Later, two mathematical models [26], subsequence improvement [24], [25], for energy-aware wireless mesh network management have been proposed to find the balance between the installation cost and energy management operation cost.The first model aims at location findings based on installation cost.In contrast, the second model seeks to minimize energy by turning on-off routers during the operation period based on network demands.
Nevertheless, for rechargeable MR placement with battery limited, two MR association mechanisms for each node have been proposed [27].Furthermore, the energy charging and discharging models are also included in the proposed optimization models to protect router energy from being depleted.Similarly, Wang et al. [28] used the same constraints in the optimization model.However, not only the MR placement but also the Internet gateway have been incorporated to minimize the installation and energy consumption costs by using the time-division multiplexing slotted mechanism for traffic transmission.In [24]- [28], all mechanisms focus on finding the MR location.However, in the case of the installed mesh network, adding more gateway for reducing the energy usage has been proposed [29].The mixed-integer linear programming and a greedy algorithm are presented and compared.The results show that the mixed-integer mechanism gives the optimal result.Meanwhile, only 5% gains the optimal solution for the greedy algorithm with only one percent of the computational time.
A heuristic method [30] is proposed to find the maximum number of noninterfering links for creating concurrent transmission paths in a fixed installed node environment scenario.The energy-saving component was not addressed in this study.Two mixed-integer linear programming and a local search algorithm to determine the optimal gateway candidate location [31] is proposed with the assumption of unlimited power source for mesh router.
The node placement mechanisms by using intelligent hybrid system consisting of particle swarm optimization, hill climbing, and simulated annealing [32] is proposed.By fixing the number of mesh routers, the routers are placed for optimal coverage for all client nodes based on computational time concern.However, performance evaluation is omitted.Another mesh router placement is proposed by implementing the electromagnetism-like mechanism (EM) metaheuristic technique [33].The study focuses on network connectivity of all client nodes.Energy harvest is mentioned in a future study.
The optimizing goal of most previous research initially focuses on minimizing the installation cost.The energy awareness is the next interesting objective.However, the proposed protocol also focuses on maximizing lifetime besides energy awareness.Related research is summarized as shown in Table 1.

Related work
Energy-ware design Rechargable enenrgy Optimization method Minimize installation cost Maximize lifetime Amaldi et al. [24] x

PROPOSED MECHANISM
The scope of the study is to determine appropriate mesh router (MR) locations in a forest in a short period of time.All sensor nodes are termed as mesh client (MC) including camera traps that are assumed to be placed in the forest for only uploading information to MR, which is eventually forwarded to the gateway.The locations of sensing devices, MCs, are determined by the patrol officer in a range of few kilometers.

Proposed system model
In the study, each MR is operated only by a standalone battery without solar system.Hence, the MR lifetime depends upon its operation and transmission activities.We assume that all MRs are installed with same battery capacity.In order to evaluate the scenario, Wi-Fi communication (IEEE 802.11g, 2.4 GHz, 54 Mbps) technology is selected.
The monitoring area is defined as M × M meters, which is divided into small grid cells based on its Wi-Fi communication capabilities.In the study, the placement scenario example is shown in Figure 2(a) where M equals 600 m.The MC can be placed anywhere in the monitoring area that is then mapped into the center of a MC grid cell, 50m × 50m in size, as shown in Figure 2(b).After performing the optimization, all MR locations can be found and placed in the center of a MR grid cell, 100m × 100m in size, as shown in Figure 2(c).It is noted that to decrease the MR location search space in our proposed optimization model, the MR grid cell is twice the MC grid cell.
By assuming that the transmission range for each sensing node to MR and vice versa is set to 250 m from the center, it approximately equals to the next 4 MC grid cells.Conversely, the transmission range between MRs is set to 550 m, which approximately corresponds to the next 5 MR grid cells.Hence, with respect to 600m × 600m, four MRs can extend coverage in all the areas.However, the addition of one more MR increases flexibility for MR location adjustments.
The constraint for the MR placement is that each MC must be covered by at least one MR to ensure connectivity.We assume that all sensing devices are equipped with an SD card and are used to periodically upload information at a constant bit rate (CBR).Furthermore, physical data transmission rate is considered as a fixed value independent of the communication distance.
The proposed model is formulated as a mixed integer linear programming.In order to determine the optimized solution, R program with the linprog package [34] is used.From a network viewpoint, the derived locations from the model are evaluated with ns-3 (network simulator 3) [35] for network performance and battery utilization that are subsequently compared with the brute-force mechanism.
Energy-aware wireless mesh network deployment using optimization mechanism (Aphirak Jansang) The locations of all MCs and gateway are used as the input of the proposed model that is subsequently mapped to grid cells in the optimization model parameters, as shown in Figure 2(a).Let C = {1, ..., c}, R = {1, ..., r}, and G = {1, ..., g} be the set of MC, MR, and gateway grid cells, respectively.Each MC location is assigned such that it is a member of a corresponding MC grid cell.In MC grid cell i point of view, the reachable parameter a ij of each MR grid cell j that communicates with (from each MC location to reachable MR location) each i ∈ C, j ∈ R is: All the reachable cells between each candidate location are calculated and stored as MR range parameter b jl j, l ∈ R: All the reachable cells from gateway are calculated and stored as gateway range parameter w jg j ∈ R, g ∈ G: 1, if MR in location j and gateway in location g are in each communication range Next, we define a decision variable.The MR assignment variable z j for each j ∈ R: The MC assignment variable x ij for each i ∈ C, j ∈ R: The MR connection variable y jl for each j, l ∈ R: The gateway assignment variable Y jg for each j ∈ R, g ∈ G: Y jg = 1, if MR at location j connect with gateway at location g 0, otherwise

Optimization model
From the objective function, the optimal location for all specific number of MRs that makes the longest operation lifetime is calculated.The system lifetime that is defined as one of MRs is depleted.Given certain constraints, the system lifetime T is maximized (or minimizing the inverse of T ) (8).
The x ij is a binary operator that controls the association between MR and MC.If MC i can connect with MR j, the sum of x ij for each MC i should be at least 1.Therefore, constraints (9) ensures that MC i can associate with only one MR j.While constraints (10) are coherence conditions to guarantee that MC i can be associated with MR j only if there exists MR at j is within communication range of MC at i. Constraints ( 11) is defined the flow balance at MR j, which is the inbound data rate that needs to be equal to the outbound data rate.Let d i be a demanded data rate that MC i to send to any MRs.Let F r lj be the data rate that MR l requires to send to MR j and let F g jg be the data rate that MR j requires to send to gateway g.The term i d i • x ij is the total needed data rate to transmit to associated MR j.
l F r lj is the total data rate that MR i needed to send to MR j.It is the total inbound traffic for MR j.In contrast, l F r jl and g F g jg are the total traffic of MR j sent to other MRs and gateways, respectively, which is the outbound direction.
Let Cap jl be the capacity of the wireless link between MR j and MR l.Constraints (12) ensure that the link between MR j and MR l exists and the sum of the outbound data rate and inbound data rate between MR j and MR l must not exceed the link's capacity.Constraints ( 13)-( 15) define the existence of a wireless link between MR j and MR l, which depends on the existence of MR at location i, location l and communication range variable b jl .Constraints (16) ensure that the number of links between MR to MR and MR to gateways does not exceed the link's degree.Thus, the sum of any MR j's links (MR to MR, MR to gateways) must not surpass its Deg j , where Deg j be the maximum degree of MR j.
Since the connection from any MRs to the gateway is only uplink direction.Thus, constraints (17) ensures that link from MR j to gateway g must exist and limit by Cap jg , and only outbound traffic from MR j to the gateway g is considered.Let Cap jg be the capacity of the wireless link between MR j to the gateway g.Constraints (18) define the existence of a wireless link between MR j and gateway g, which depends Energy-aware wireless mesh network deployment using optimization mechanism (Aphirak Jansang)

❒
ISSN: 1693-6930 on the existence of MR at location i, the gateway at location g, and the communication range variable w jg .Constraints (19) ensure that the total number of links from any possible MRs to gateway g must not exceed its degree, as denoted by Deg g .Let Ener j be the initial energy of MR j.Let T denote the system lifetime that starts from the startup time until the first MR runs out of battery.The number of required MRs is defined by variable numr.Given that energy consumption on each MR depends on its data rate and system lifetime T .The outbound data rate for MR j is summing up the data rate from other MRs to MR j (defined as l F r lj ) and its required data rate to gateway g (defined as g F g gj ).Then, the constraint ( 20) is used to ensure that the energy usage of each MR j ( l F r lj + g F g gj multiplied by time T ) will not exceed its Ener j .Constraints (21) ensure that the total number of selected MRs ( j z j ) must not exceed numr.
From the equation shown above, there are many possible sets of MR locations (z j ) that are qualified by all constrains.However, the objective of the proposed mechanism is to find for the best outcome of z j which yields the longest system lifetime.Therefore, the optimization solver is required for obtaining z j , which delivers the best result. Maximize: Subject to:

PERFORMANCE EVALUATION
In the study, five cases of various MC nodes (i.e. 15, 20, 25, 30, and 45) are evaluated.In each case, three replications of the MC locations are randomly generated with uniform distribution.The gateway location is assumed as fixed for all cases.For system lifetime evaluation, each brute-force solution (from all possible MR locations) for each case is simulated and compared with the corresponding optimization solution.

Extracting optimization results
Before the proposed optimization model can determine the results of potential MR locations on the map, the basic information from the actual map, such as the MCs' locations and the location of the gateway used to relay data, must first be gathered.The collected data will then be converted into the model's initial vector variables, such as matrix a ij for MC locations, w jg for gateway locations, which is described in section 4.1.Next, initialize the model's variables by calculating all possible MR location pairs in the deployment area from vector b jl and generating them into the grid cell as matrix w jg .At the same time, the matrix Y jg is generated from vector w jg to keep all possible MR locations that can communicate with gateway g.The other required vector variables (e.g.z j , x ij , and y lj ) and all constraints are set up before running the linear optimization solver.Subsequently, the program for solving linear equations will then start to search for the possible solutions that make the most extended system lifetime while satisfying all constraints.The software is written in R and solves linear programming tasks with the linprog package.After obtaining the optimization result for the MR's locations in grid cell format, it will convert them back to their actual map locations before being used in the real implementation.The flowchart in Figure 3 illustrates the process of extracting the results from the proposed optimization model.

Simulation setup
To compare the network performance between the proposed and brute-force mechanisms, the optimization results from the R-program are simulated by ns-3 simulator.The simulation field area is defined as 600 × 600 m 2 .The MC grid cell are defined as 50 × 50 m 2 for placing each MC.Conversely, the MR grid cell of 100 × 100 m 2 is used for MR placement.The MC traffic patterns for uploading data to the gateway are defined at a constant bit rate of 1 Mbps.All MRs are deployed with the same initial capacities and energy discharge model.Specifically, IEEE 802.11g with fixed propagation delay with the default range propagation loss model is used throughout the simulation.We modified the optimized link state routing protocol (OLSR) to support an energy-aware routing protocol by considering the remaining energy in each path before transmitting the data.The total simulation time corresponds to 80 s.Each MC automatically starts uploading at the 40th second after the simulation is started.The simulation parameters are listed in Table 2. Optimization/brute-force

Energy Usage Model
The default ns-3 energy model is used throughout the simulation.The energy usages are based on its activities, namely transmit, receive, or idle, as shown in Table 3.Each MR stops working if it runs out of energy.

SIMULATION RESULTS
The proposed mechanism is evaluated for the computation time and network performance by comparing it with the brute-force method.As shown in Figure 4, the computational times (including the simulation times) for various MCs of the proposed optimization are almost constant.Furthermore, the brute-force mechanism computation time is significantly varied and significantly high when compared to the proposed mechanism.With respect to network performance, the system life times, as shown in Figure 5, for both mechanisms are almost identical, and they both run out of battery at approximately 65th second.For the average throughput and packet delivery ratio, shown in Figures 6 and 7, for the low number of clients, the predefined condition of the optimization might restrict the allowable degree of MC connecting for each MR.Therefore, once the increasing number of clients reaches 30, both brute-force and optimization mechanisms perform the same due to the limited choices of MR location for covering all MCs.However, in Figure 8 of the brute-force mechanism for many connecting clients shows a significant variation due to the unbalanced connecting degree for each MR.
From Figure 9, an example of 25 clients randomly scattered on the map is presented.The results of the different methods for determining the placement of 5 MR locations are shown in Figure 10.The optimization solution (shown in Figure 10(a) and Figure 10(b)) yields more scatter MR locations than the brute-force mechanism (shown in Figure 10(c) and Figure 10(d)) because the optimization mechanism tried to limit the degree of connecting MC for each MR, which is more apparent in the case of the higher number of required MRs.The inter-arrival time of the case of 25 MCs (from Figure 10) is shown in Figure 11.Given the cumulative distribution function (CDF) value of 0.80, the performance of the optimization methods exceeds that of the brute-force method.Additionally, with respect to different number of MC nodes, the average of inter-arrival time at 80 th percentile is shown in Figure 12.The performance of the proposed method slightly exceeds that of the brute-force method for a high number of MC nodes.When compared to the brute-force mechanism, the network performance obtained by our proposed mechanism are significantly identical while the computation time significantly outperforms the brute-force mechanism.

CONCLUSIONS
In the study, we proposed the optimized wireless router placement method with energy aware for wireless mesh network in a rural area.Our optimization goal involves achieving the longest lifetime of the system compared with brute-force method.The simulation results indicate that the optimization method yields significantly less computation time while obtaining the same results in term of network parameters.This is significantly crucial in the real deployment situation where sensor node (client) locations are mandatory adjusted based on the environmental status or the recommendation of the patrollers.a new calculation.Therefore, the proposed mechanism becomes more desirable due to its short computation time.Furthermore, the heuristic approach can be implemented for better performance.

29 Figure 1 .
Figure 1.Human habitat and forest area

Table 2 .
Parameter of simulation