sensors Article

Clustering and Beamforming for Efficient Communication in Wireless Sensor Networks Francisco Porcel-Rodríguez 1 , Juan Valenzuela-Valdés 1, *, Pablo Padilla 1 , Francisco Luna-Valero 2 , Rafael Luque-Baena 3 and Miguel Ángel López-Gordo 1 1

2 3

*

Department of Signal Theory, Telematics and Communications—CITIC, University of Granada, 18071 Granada, Spain; [email protected] (F.P.-R.); [email protected] (P.P.); [email protected] (M.Á.L.-G.) Department of Computer Science and Programming Languages, University of Malaga, 29071 Malaga, Spain; [email protected] Department of Computer and Telematics Systems Engineering, University of Extremadura, 06800 Merida, Spain; [email protected] Correspondence: [email protected]; Tel.: +34-958-24-08-49

Academic Editor: Yu Wang Received: 31 May 2016; Accepted: 12 August 2016; Published: 20 August 2016

Abstract: Energy efficiency is a critical issue for wireless sensor networks (WSNs) as sensor nodes have limited power availability. In order to address this issue, this paper tries to maximize the power efficiency in WSNs by means of the evaluation of WSN node networks and their performance when both clustering and antenna beamforming techniques are applied. In this work, four different scenarios are defined, each one considering different numbers of sensors: 50, 20, 10, five, and two nodes per scenario, and each scenario is randomly generated thirty times in order to statistically validate the results. For each experiment, two different target directions for transmission are taken into consideration in the optimization process (φ = 0◦ and θ = 45◦ ; φ = 45◦ , and θ = 45◦ ). Each scenario is evaluated for two different types of antennas, an ideal isotropic antenna and a conventional dipole one. In this set of experiments two types of WSN are evaluated: in the first one, all of the sensors have the same amount of power for communications purposes; in the second one, each sensor has a different amount of power for its communications purposes. The analyzed cases in this document are focused on 2D surface and 3D space for the node location. To the authors’ knowledge, this is the first time that beamforming and clustering are simultaneously applied to increase the network lifetime in WSNs. Keywords: wireless sensors networks; energy efficiency; beamforming; optimization techniques

1. Introduction The next generation 5G wireless communications are currently being developed [1,2]. Indeed, the first deployments of a 5G network are expected to be fully operating in 2020 [3]. The design goals of such systems are shown in Figure 1. These systems are conceived to provide very high data rates (typically Gbps), extremely low latency, manifold increase in base station capacity and significant improvement in users’ perceived quality of service (QoS), compared to current 4G LTE networks [4]. There are many innovative technologies that will be used to satisfy the demands of the massive volume of traffic and various devices: massive MIMO, orthogonal frequency-division multiple access (OFDMA), cloud radio access network (CRAN), software-defined networking (SDN), composite wireless infrastructures, flexible spectrum management, small cells, and heterogeneous network deployment, among others [5].

Sensors 2016, 16, 1334; doi:10.3390/s16081334

www.mdpi.com/journal/sensors

Sensors 2016, 16, 1334 Sensors 2016, 16, 1334

2 of 14 2 of 14

Figure 1. 1. Design goals and requirements forfor 5G5G networks. (adapted from [3]). Figure Design goals and requirements networks. (adapted from [3]).

Theresearch research initiativesbyby industryand andacademia academiahave haveidentified identifiedeight eightmajor majorrequirements requirements[1][1] The initiatives industry of the next generation 5G systems, with “the reduction in energy usage by almost 90%” being one of the next generation 5G systems, with “the reduction in energy usage by almost 90%” being one ofof them. Green technologies are, thus, being considered in the standard bodies. As a consequence, them. Green technologies are, thus, being considered in the standard bodies. As a consequence, energy energy efficiency is, therefore, a key for the of design these networks [6]. Efficiency, in general, efficiency is, therefore, a key issue forissue the design theseofnetworks [6]. Efficiency, in general, and and energy consumption, in particular, clearly involve some sort of optimization. The required energy consumption, in particular, clearly involve some sort of optimization. The required technologies technologies for the 5G as systems, as well scenarios, as the application for the development of development 5G systems, asofwell the application are quite scenarios, numerous.are Onequite of numerous. One of the most potential use cases of 5G [4] is in wireless sensor networks (WSNs). The the most potential use cases of 5G [4] is in wireless sensor networks (WSNs). The expected number number of connected “things” (IoT: will Internet Things) be 7 trillion [3]. Therefore, it is ofexpected connected “things” (IoT: Internet of Things) be 7of trillion [3].will Therefore, it is clear that energy clear that energy preservation forofWSNs is an issue of great concern with respect to network design preservation for WSNs is an issue great concern with respect to network design and deployment, and deployment, protocols, and configurations, trying to maximize the performance of the nodes. protocols, and configurations, trying to maximize the performance of the nodes. Oneoption optionisistotoreduce reducethe theenergy energyconsumption consumptionbybyoptimizing optimizingthe theway waythe thewireless wireless One communications take place. problem to cope theinterface air interface (the signals) radio signals) and communications take place. TheThe problem has has to cope withwith the air (the radio and how how the radiation pattern (beam) is set up to reach the desired QoS with minimal energy the radiation pattern (beam) is set up to reach the desired QoS with minimal energy consumption. consumption. The technique that enables wireless this energy-efficient communication The technique that enables this energy-efficient communicationwireless is beamforming [7], which is beamforming [7],coordinated which consists of several coordinated to generate consists of several antennas radiating jointly to antennas generate aradiating directivejointly beam for covering a directive beam for covering a given area with very accurate precision. The goal is to reduce energy a given area with very accurate precision. The goal is to reduce the energy consumption of thethe sensors of the sensors by performing beamforming. This a very important byconsumption performing collaborative beamforming. This collaborative is a very important problem as the is sensor nodes have problem as the sensor nodes have limited power, and saving energy is critical. In this paper, the limited power, and saving energy is critical. In this paper, the objective is to show that beamforming objective is to show that beamforming can be used in the context of WSNs to perform energy efficient can be used in the context of WSNs to perform energy efficient communications, allowing the lifetime allowing the lifetimeto of kindincreased. of distributed to be steadily ofcommunications, this kind of distributed infrastructure be this steadily To the infrastructure best of our knowledge, this increased. To the best of our knowledge, this is the first approach in which the gain toofthe the is the first approach in which the gain of the beamforming is accurately computed, in relation beamforming is accurately computed, in relation to the number of nodes, organized in different number of nodes, organized in different clusters in the WSN. clusters in theisWSN. The paper organized as follows: Section 2 describes the beamforming model used for a WSN; The paperthe is organized as follows: Section 2 describes the beamforming used forresults, a WSN; Section 3 shows experimental set-up and pattern results; Section 4 provides model the theoretical Section 3 shows experimental and pattern results; Sectionare 4 provides results, and in Section 5 thethe gain results and set-up discussion of the results obtained exposed;the andtheoretical the conclusions and in section 5 the gain results and discussion of the results obtained are exposed; and the are outlined in Section 6. conclusions are outlined in Section 6. 2. Collaborative Beamforming in WSN 2. Collaborative Beamforming in WSN WSNs usually have the nodes arbitrarily located in a defined area. In many cases, this distribution WSNs one. usually theinfluence nodes arbitrarily located in a defined area. In many cases, this is a random Thishave has an in the data transmission when considering collaborative distribution is aeasy random one. This an transmission influence in the data of transmission when considering nodes; not being to maximize thehas global capacity the network. In this context, collaborative nodes; not being easy to maximize the global transmission capacity of the network. beamforming has arisen as a good strategy in WSNs for complete system optimization in terms of In thisefficiency, context, beamforming arisen and as a good strategy WSNs for complete system energy and networkhas capacity reliability. Theinfirst approaches on this are optimization dated only in terms energy efficiency, andprove network capacity and firstconfiguration approaches on are a few yearsofago. In [8], the authors the existence of anreliability. optimum The feeding forthis each dated only a few years ago. In [8], the authors prove the existence of an optimum feeding individual element of the network when trying to maximize the transmission gain in a desired spatial configuration for each individual element ofstated the network when trying to maximize transmission direction. Moreover, in the results in [9] it is that 80% of the network energy the is saved when gain in abeamforming. desired spatialIndirection. in the aresults in [9] proof it is stated 80% of the network applying [10], the Moreover, authors provide theoretical of thethat gain improvement if energy is saved when applying beamforming. In [10], the authors provide a theoretical proof of the

Sensors 2016, 16, 1334

3 of 14

beamforming is applied. In this work, we compute iteratively the beamforming in an array of sensors to calculate the possible gain improvement based on this approach [11,12]. 2.1. Beamforming The beamforming techniques are based on the definition of a highly directive pattern in a desired direction, depending on the network operation conditions and the network transmitting necessities. Based on this, it is necessary to consider the field of each individual element contributing constructively in the desired pattern direction, and destructively in the rest of them, even provoking transmission nulls in some critical directions. The general pattern is made of the aggregation of the elementary ones, and five variables define the antenna array that the nodes form. These variables can be classified into two different categories, which are the radiating element nature and the spatial location: 1. 2. 3.

Radiating element: the amplitude excitation of each unitary radiating element. The phase excitation of each unitary radiating element. The radiation pattern of each element, depending on the kind of antenna considered (dipole, isotropic ideal, etc.). Spatial location:

1. 2.

The separation among elements of the array, in terms of wavelength distance. The antenna array geometry, which may be linear, circular, rectangular, or spherical, among other possibilities.

Considering classical array theory, in terms of computation, the easiest and affordable manner of placing the elements is a 1D line or 2D surface in which the elements are uniformly separated and with the same excitation in both components: amplitude and phase. This simple configuration lets one obtain a variation in the radiation direction when acting progressively over the phase of each element. Unfortunately, this is not the case of WSN networks, whose nodes are distributed in space randomly. However, there is still enough room to achieve benefits from applying beamforming, although it becomes a harder task to find the optimal configuration. 2.2. WSNs Scenario Model The experimental setup considered in this work implies a random node deployment in a squared area or cubic volume with dimensions of 30 on each side. The node location is based on a uniform distribution in both 2D coordinates or 3D coordinates, with no movement once the nodes are settled down. This approach is widely used in the definition of WSN experimental scenarios, as can be found in [7] or [8]. In this scenario, each node has limited power autonomy, which must fulfill the operation requirements for both data transmitting to the High Energy Communication Node (HECN) and environment sensing. The nodes are also equipped with two different antenna types providing different antenna patterns. Although it is typically assumed that the total available power is the same at each sensor, the distribution of power devoted to sensing or to data transmission can be modeled to be diverse. In this work, it is assumed that the node consumption for each task is commanded by random uniformly distributed variable of value [0, 1]. Thus, the available energy at sensor x (x = 1, 2, . . . , max_sensors), Eax , is the following: Eax = Et · Fx (1) where Et is the total available sensor energy and Fx is the uniformly distributed variable described above. Additionally, the energy consumed by each sensor (Ecx ) is: ECx = Ptx_x · t x

(2)

Sensors 2016, 16, 1334

4 of 14

where Ptx_x is the amount of power devoted to transmit the data in sensor x and tx is the transmission time. Notice that the node consumption can be also expressed in terms of power. Thus, the available power, Pax , is provided by: Pax = Ptx_x · Fx (3) As it can be noticed, the maximum lifetime, tlife_x , for the sensors as a whole, is obtained when the energy consumption is equal to the available energy: Et · Fx = Ptx_x · tli f e_x

(4)

As a consequence, the lifetime of each sensor can be written as follow: tli f e_x =

Et · Fx Ptx_x

(5)

At this point, it must be considered that Ptx_x is the power needed to transmit data without considering beamforming. However, when beamforming is applied, Ptx_x , which is related to the excitation in amplitude of each sensor, is multiplied by a gain effect provided by the beamforming. This gain effect depends on the final radiation pattern and may induce gain in some directions and loss in others. Thus, the power to transmit data to the receptor is: Ptx_x_B_A = Ptx_x · GB (θ, φ)

(6)

where Ptx_x_B_A is the excitation in amplitude of each sensor and GB is the global gain value at the different directions of transmission/reception. Thus, Ptx_x considering beamforming is: Ptx_x =

Ptx_x_B_A GB (θ, φ)

(7)

Finally, the lifetime when adding beamforming, tlife_x_B , is: tli f e_x_B =

Et · Fx · GB (θ, φ) Ptx_x_A_B

(8)

The main interest in this work is focused on the maximization of the lifetime of the WSN as a whole, applying beamforming for the data transmission among the sensors. This is translated into a search of phase and amplitude configurations, Ptx_x_B , at each sensor that maximize the final outcome: n   o max tWSN = min tli f e_x_B , xe [1, max_sensors]

(9)

Notice that the sensor that first runs out of battery fixes the lifetime of the WSN. 2.3. Optimization Algorithm In order to address the problem defined above in Equation (9), a genetic algorithm (GA) has been adopted as the optimization algorithm because they have shown good performance over a great variety of optimization problems [8]. GAs manage a pool of candidate solutions (the population), which represent tentative solutions of the target problem. In our case, these solutions are composed of the excitation (module and phase) of the sensors deployed in the WSN. That is, solutions are arrays (vectors) of real-coded values in which the modules of all sensors are arranged first, and their phases afterwards. Given such a solution, the lifetime of all of the sensors is computed with Equation (8) (using the corresponding modules and phases), and the minimum of such lifetimes is the objective to be maximized, as Equation (9) defines. The GA population is iteratively improved by using genetic operators (selection, recombination, and mutation) that follows the idea of “survival of the fittest”, i.e., new solutions are generated in each

Sensors 2016, 16, 1334

5 of 14

generation and replace worse solutions in the population. The process of iterating through successive generations is called evolution, and ends when a termination condition is fulfilled. The GA included in the optimization toolbox of Matlab® (Natick, MA, USA) has been used. The default configuration of this GA and its standard settings are used: the population size has been set to 100; as genetic operators, remainder selection, heuristic crossover (with a probability of 0.8), and uniform mutation (probability of 0.1); finally, the stopping condition is to perform 100 generations. Sensors 2016, 16, 1334 14 These parameter values are widely accepted and they are known to perform well over a5 ofvariety scenarios [13]. These parameter values are widely accepted and they are known to perform well over a variety scenarios [13].

3. Experimental Set-up and Pattern Results

3.The Experimental and Pattern radiation Set-up patterns and theirResults gains when applying beamforming are computed with [14], ® . Beamforming requires all of the nodes in the cluster to be synchronized, and a MatlabThe toolbox radiation patterns and their gains when applying beamforming are computed with [14], a ®. Beamforming this Matlab issue takes time: the higher the cluster longer thecluster sync time to be. In this work toolbox requires all size, of thethe nodes in the to be needs synchronized, and this we have fixed five different scenarios regarding the number of sensors implied: 50, 20, 10, five, issue takes time: the higher the cluster size, the longer the sync time needs to be. In this work we haveand five different scenarios regardingeach the number ofis sensors implied: 50, 20, 10,thirty five, and twoinnodes two fixed nodes per scenario. Additionally scenario randomly generated times order to per scenario. Additionally each is randomly thirty in order to statisticallytime statistically validate the results. Forscenario the scenarios of 50, generated 20, 10, five, and times two nodes, a transmission validate results. scenarios 50,synchronization, 20, 10, five, and two nodes, a transmission time of 40% of 40% 30%,the 20%, 15%,For andthe 10% is usedoffor respectively. The time of transmission 30%, 20%, 15%, and 10% is used for synchronization, respectively. The time of transmission the to of the beamforming technique is multiplied by the number of sensors because every sensorofneeds beamforming technique is multiplied by the number of sensors because every sensor needs to transmit the data of all sensors. In each experiment, two different transmission directions are being transmit the data of all sensors. In each experiment, two different transmission directions are being considered for the optimization process (φ = 0◦ and θ = 45◦ ; φ = 45◦ , and θ = 45◦ ). Then, the GA starts considered for the optimization process ( = 0° and θ = 45°;  = 45°, and θ = 45°). Then, the GA starts optimizing the antenna parameters installed in these sensors for performing the desired beamforming optimizing the antenna parameters installed in these sensors for performing the desired (a beam towards (a the HECN, which located at the four different directions previously mentioned). beamforming beam towards the is HECN, which is located at the four different directions previously Eachmentioned). scenario isEach evaluated foristwo different an antennas, ideal isotropic antenna and a conventional scenario evaluated for antennas, two different an ideal isotropic antenna and a dipole one. In this set of experiments all the sensors are considered to be transmitting the same amount conventional dipole one. In this set of experiments all the sensors are considered to be transmitting of power when acting the network as aincollaborative node. the same amount ofin power when acting the network as a collaborative node.

Pattern Results Pattern Results Figures 2 and 3 present theperformance performance results. results. At it should be noticed that,that, Figures 2 and 3 present the Ateach eachsubfigure, subfigure, it should be noticed although the desired transmission direction implies a particular θ and  value, the subfigures provide although the desired transmission direction implies particular θ and φ value, the subfigures provide the radiation pattern resultsfor forthe theparticular particular φ  and and the thisthis θ range, the the the radiation pattern results the entire entireθθrange range(360°). (360◦In ). In θ range, desired θ direction can be easily identified. desired θ direction can be easily identified.

Figure 2. Radiation patterns for two, five, 10, 20 and 50 sensors, different search angles: Figure 2. Radiation patterns for two, five, 10, 20 and 50 sensors, different search angles: isotropic isotropic antenna. antenna.

Figure 2 provides the beamforming results for two, five, 10, 20, and 50 sensors with isotropic antennas for the two different angular configurations. It is observed that although the desired direction is always obtained, for a low number of sensors there is still a high amount of radiated power towards other directions. This limitation is logical regarding the reduced number of sensors considered. As the number of nodes is increased, the radiated power towards directions different from the desired one is reduced, as it is clearly identified in Figure 2. In the case of using a dipole

Sensors 2016, 16, 1334

6 of 14

Figure 2 provides the beamforming results for two, five, 10, 20, and 50 sensors with isotropic antennas for the two different angular configurations. It is observed that although the desired direction is always obtained, for a low number of sensors there is still a high amount of radiated power towards other directions. This limitation is logical regarding the reduced number of sensors considered. As the number of16, nodes one Sensors 2016, 1334 is increased, the radiated power towards directions different from the desired 6 of 14 is reduced, as it is clearly identified in Figure 2. In the case of using a dipole (Figure 3), the dipole radiation adds a limitation in the beam that points towards the desired direction. final (Figure 3),pattern the dipole radiation pattern adds a limitation in the beam that points towards theThe desired beam is, thus, affected byis,the radiation pattern the dipole, reducing global gain achieved, as direction. The final beam thus, affected by theof radiation pattern of thethe dipole, reducing the global expected. When number of nodes is highof (20nodes and 50 nodes), beam more gain achieved, as the expected. When the number is high (20 the anddesired 50 nodes), thebecomes desired beam directive,more the number of the rest of the alsobeams, their gain, is reduced. becomes directive, number ofbeams, the restand of the and also their gain, is reduced.

Figure Radiationpatterns patternsforfor two, five, 10,and 20 50 and 50 sensors, different search dipole angles:antenna. dipole Figure 3. Radiation two, five, 10, 20 sensors, different search angles: antenna.

4. Theoretical Gain Results 4. Theoretical Gain Results In order to analyze the gain results, four different scenarios have been proposed. In the first one In of order analyze thethe gain results, fouravailable differentfor scenarios have been (this proposed. theoffirst one (A) all the to sensors have same power communications is the In case a WSN (A) all of the sensors have the same power available for communications (this is the case of a WSN composed with the same sensor type) and the sensors are placed only in two dimensions. In the second composed the same sensor type)available and thepower sensors placed only in(this twoisdimensions. the scenario (B)with the sensors have random forare communications the case of aInWSN second scenario (B) the sensors power communications (this is the case of composed with different sensor have types)random and theavailable sensors are againfor placed only in two dimensions. In the athird WSN composed with different sensor types) and the sensors are again placed only in two scenario (C) all sensors have the same available power for the communications but the sensors dimensions. thedimensions. third scenario all scenario sensors(D) have the same available power for the are placed in In three In the (C) fourth the sensors have random available power communications but the sensors are placed in three dimensions. In the fourth scenario (D) the for communications and again the sensors are placed in three dimensions. sensors have random available power for communications and again the sensors are placed in 4.1. Theoretical Gain Results three dimensions. In order to analyze the different gain factors implied in the energy efficiency in a WSN, in this 4.1. Theoretical Gain Results subsection, the distance between the nodes and the HECN has not been taken into account. That is, In order to obtained analyze the different gain implied technique. in the energy efficiency in a are WSN, in this the gain factor is only due to thefactors beamforming The gain results analyzed: subsection, distance between thefor nodes anddifferent the HECN has not beenangles, taken all into account.and That is, Figures 4–6 the provide the gain values the two desired search scenarios both the gain factor is 20 only due toand the50 beamforming technique. The gain results analyzed: antennas, with obtained 10 sensors, sensors, sensors, respectively. The gain values areare computed in Figures thefactor gain values for the desired search angles, scenarios terms of4–6 theprovide increment regarding thetwo gaindifferent in the case beamforming is notall used; that is,and the both gain antennas, with 10 sensors, sensors, and 50 sensors,and respectively. Theitsgain values are computed in obtained for a unitary node20for the desired direction provided by antenna pattern. termsEach of the increment is factor regarding the so gain in 30 thedifferent case beamforming is not used; that the gain experiment repeated 30 times that sets of sensor positions are is, defined, in obtained for a unitaryvalidate node forthe theresults. desiredIndirection and provided by its4–6) antenna pattern. have the order to statistically the result figures (Figures the subfigures ◦ ) sensor Each experiment is repeated times so that(φ 30=different positions are left), defined, in following characteristics: isotropic30ideal antenna 0◦ and θsets = 45of in Figure 4a (upper dipole order to statistically validate the results. In the result figures (Figures 4–6) the subfigures have the following characteristics: isotropic ideal antenna ( = 0° and θ = 45°) in Figure 4a (upper left), dipole antenna ( = 0° and θ = 45°) in Figure 4b (upper right), isotropic ideal antenna ( = 45° and θ = 45°) in subfigure c (lower left), and dipole antenna ( = 45° and θ = 45°) in subfigure d (lower right). In all subfigures it is plotted with a red line the mean gain of each scenario. In Figure 4 it is shows the

Sensors 2016, 16, 1334

7 of 14

antenna (φ = 0◦ and θ = 45◦ ) in Figure 4b (upper right), isotropic ideal antenna (φ = 45◦ and θ = 45◦ ) in subfigure c (lower left), and dipole antenna (φ = 45◦ and θ = 45◦ ) in subfigure d (lower right). In all subfigures it is plotted with a red line the mean gain of each scenario. In Figure 4 it is shows the results for 10 nodes and 30 emulations for each scenario (four scenarios for each subfigure). The aspect of all subfigures is very similar; that is, the influence of the directions and the antenna is very limited. Sensors 2016, 16, 1334 7 of 14 For example the mean gain for the scenario A, for Figure 4a is 6.50, for Figure 4b is 6.60, for Figure 4c is 6.51, and 4c Figure is Figure 6.61, which means that the that maximum difference between subfigures is only Figure is 6.51,4d and 4d is 6.61, which means the maximum difference between subfigures 0.11 (1.6%). However, gain factor is different for the different scenarios, around 6.55 for is only 0.11 (1.6%). the However, the gain factor is different for the different scenarios, around 6.55scenario for A; around 9.7A; foraround scenario B; around scenario and around for scenario D. ThisD. means scenario 9.7 for scenario7B;for around 7 forC;scenario C; and10 around 10 for scenario This that means that thefixed scenarios with fixed power gain than the with the scenarios with power have less gain have than less the scenarios withscenarios sensors with withsensors different power. different power. This is because the genetic algorithm tries to assign the node with lower power to This is because the genetic algorithm tries to assign the node with lower power to the communication. the communication. lower the the power to conformthe thehigher beamforming, the higher lifetime the The lower the power to The conform beamforming, the lifetime of thethe WSN. Forof10 nodes WSN. For 10 nodes there exists little difference between 2D and 3D scenarios. there exists little difference between 2D and 3D scenarios.

Figure 4. Gain results forsensors. 10 sensors. Isotropic ideal antenna, 0° and in subfigure Figure 4. Gain results for 10 Isotropic ideal antenna, φ =0◦= and θ =θ45=◦45° in subfigure (a)(a) (upper ◦ ◦  = 0° and θ = 45° in subfigure (b) (upper right), isotropic ideal antenna, (upper left), dipole antenna, left), dipole antenna, φ = 0 and θ = 45 in subfigure (b) (upper right), isotropic ideal antenna, φ = 45◦ 45°◦ in andsubfigure θ = 45° in (c) (lower left) and dipoleφantenna,  =θ45° and θ subfigure = 45° in (d) ◦ in and θ == 45 (c)subfigure (lower left) and dipole antenna, = 45◦ and = 45 subfigure (d) (lower right). (lower right).

Figure 5 shows the results for 20 nodes and 30 emulations for each scenario (four scenarios for

Figure 5 shows the for the 20 nodes 30 of emulations for each scenario (four scenarios each subfigure). For results 20 nodes, aspect and of all the subfigures is also very similar, that is, for influence For of the andaspect the antenna is very limited. Forisexample, thesimilar, mean gain each the subfigure). 20directions nodes, the of all of the subfigures also very thatfor is, the scenario B, for Figure 5a is 16.03, for Figure 5b islimited. 15.91, forFor Figure 5c is 16.41, and for Figure 5d is influence of the directions and the antenna is very example, the mean gain for scenario B, 15.98, which means that the maximum difference between subfigures is only 0.5 (3.0%). However, for Figure 5a is 16.03, for Figure 5b is 15.91, for Figure 5c is 16.41, and for Figure 5d is 15.98, which the gain factor is different for the different scenarios, around 8.4 for scenario A, around 16.2 for means that the maximum difference between subfigures is only 0.5 (3.0%). However, the gain factor scenario B, around 10.7 for scenario C, and around 18.8 for scenario D. Again, the scenarios with fixed is different for the different scenarios, around 8.4 for scenario A, around 16.2 for scenario B, around power have less gain than the scenarios with sensors with different power for the same reasons as 10.7 for scenario C, and around 18.8 for scenario D. Again, the scenarios with fixed power have less that for 10 nodes. For 20 nodes there exists differences between the 2D and the 3D scenarios, and the gain than the scenarios with sensors with different power for ones the same reasons as that 10be nodes. gain factors for 2D scenarios are around 15% lower than the for 3D scenarios. Thisfor may For 20influenced nodes there exists differences between the 2D and the 3D scenarios, and the gain factors for 2D by the major complexity of the calculations in the 3D case which may lead to non-optimal scenarios are around than the ones for 3D scenarios. This may be influenced by the major solutions when the15% GAslower algorithms are applied. Figure 6 show the results for 50 and 30may emulations each scenario (four scenarios for GAs complexity of the calculations in the 3Dnodes case which lead tofor non-optimal solutions when the each subfigure). Again, for 50 nodes, the aspect of all the subfigures is very similar, which means that algorithms are applied. the influence theresults directions andnodes the antenna very limited. example, the(four mean gain for for the each Figure 6 showofthe for 50 and 30isemulations forFor each scenario scenarios scenario D, for Figure 6a is 22.25, for Figure 6b is 21.6, for Figure 6c is 21.66, and Figure 6d is 21.93, subfigure). Again, for 50 nodes, the aspect of all the subfigures is very similar, which means that the which implies that the maximum difference between subfigures is only the 0.65 (2.9%). As it is influence of the directions and the antenna is very limited. For example, the mean gain for the scenario expected, the gain factor is again different for the different scenarios, around 7.75 for scenario A, D, foraround Figure15.8 6a for is 22.25, forB,Figure 21.6, for Figure is 21.66, Figure D. 6dThis is 21.93, which scenario around6b 10.3isfor scenario C, and6c around 21.6and for scenario means, implies that the maximum difference between subfigures is only the 0.65 (2.9%). As it is expected, in a similar way to the Figures 4 and 5, that the scenarios with fixed power have less gain than the the scenarios with sensors with different power. For 50 nodes there exist more differences between 2D and 3D scenarios, the gain factors for 2D scenarios are around a 30% less than for 3D scenarios for fixed power and around 40% less for sensors with random power for communications. It is clear from Figure 6 that the variability of scenarios with 50 sensors is high. For example, in subfigure b in

Sensors 2016, 16, 1334

8 of 14

gain factor is again different for the different scenarios, around 7.75 for scenario A, around 15.8 for scenario B, around 10.3 for scenario C, and around 21.6 for scenario D. This means, in a similar way to the Figures 4 and 5, that the scenarios with fixed power have less gain than the scenarios with sensors with different power. For 50 nodes there exist more differences between 2D and 3D scenarios, the gain factors for 2D scenarios are around a 30% less than for 3D scenarios for fixed power and around 40% 2016, 88of less forSensors sensors with random power for communications. It is clear from Figure 6 that the variability of Sensors 2016,16, 16,1334 1334 of14 14 scenarios with 50 sensors is high. For example, in subfigure b in scenario D, the emulation with the scenario D, the emulation with the gain aa gain factor of 34 the emulation with less emulation the higher higher gain has haswith gain factor 34 and and thefactor emulation withthat lessis, the higherscenario gain hasD,a the gain factor ofwith 34 and the emulation less gainofhas a gain of 13.4; gain has a gain factor of 13.4; that is, the emulation with less gain is only 39.4% than that of gain has a gain factor of 13.4; that is, the emulation with less gain is only 39.4% than that of the the emulation withwith less more gain gain. is only 39.4% than that of the emulation with more gain. emulation emulation with more gain.

5.5. Gain 20 Isotropic ideal antenna, == ◦45° FigureFigure 5. Gain results for 20for sensors. Isotropic ideal antenna, φ =0=◦= 0° and θ =θθ45 in in subfigure (a)(a) (upper Figure Gain results results for 20 sensors. sensors. Isotropic ideal antenna, 0° and and 45° in subfigure subfigure (a) ◦ ◦  = 0° and θ = 45° in subfigure (b) (upper right), isotropic ideal antenna, (upper left), dipole antenna, left), dipole φ =antenna, 0 and θ==0°45andinθsubfigure (b) (upper right),right), isotropic ideal antenna, φ = 45◦ = 45° in subfigure (b) (upper isotropic ideal antenna, (upperantenna, left), dipole ◦ ◦ ◦  = 45° and θ = 45° in subfigure (c) (lower left) and dipole antenna,  = 45° and θ = 45° in subfigure and θ ==45 in subfigure (c) (lower(c)left) and dipole antenna, φ =45 and insubfigure subfigure (d) 45° and θ = 45° in subfigure (lower left) and dipole antenna, = 45° andθθ== 45 45° in (lower (d) (lowerright). right). (lower(d) right).

6.6. Gain 50 Isotropic ideal antenna, == ◦45° Figure Gain results results for 50 sensors. sensors. Isotropic ideal antenna, 0° and and 45° in subfigure subfigure (a) FigureFigure 6. Gain results for 50for sensors. Isotropic ideal antenna, φ =0=◦= 0° and θ =θθ45 in in subfigure (a)(a) (upper  = 0° and θ = 45° in subfigure (b) (upper right), isotropic ideal antenna, (upper left), dipole antenna, ◦ ◦  = 0° and θ = 45° in subfigure (b) (upper right), isotropic ideal antenna, (upper left), dipole antenna, left), dipole antenna, φ = 0 and θ = 45 in subfigure (b) (upper right), isotropic ideal antenna, φ = 45◦  ==45° and θθ==45° subfigure (c), left) dipole ==45° and ◦θθ==45° 45° and 45°in in (c),(lower (lower left)and and dipoleantenna, antenna, 45° 45°in insubfigure subfigure ◦ in and θ = 45 subfigure (c),subfigure (lower left) and dipole antenna, φ = 45◦ and θ =and 45 in subfigure (d) (lower (d) (lower right). (d) (lower right). right).

Figure Figure77shows showsthat, that,for forthe theresults resultsconsidering consideringtwo, two,five, five,10, 10,20 20and and50 50nodes nodesand and30 30emulations emulations for four scenarios (one scenario for each subfigure), the direction has little influence. Consequently, Figure 7 shows that, the results considering five, 10,has 20 little and 50 nodes Consequently, and 30 emulations for four scenarios (onefor scenario for each subfigure),two, the direction influence. itit isis only for  == 45° == 45°. The the characteristics: isotropic only shown shown forscenario 45° and and 45°.subfigure), The subfigures subfigures have the following following characteristics: isotropic for four scenarios (one forθθeach thehave direction has little influence. Consequently, ideal antenna and scenario B in Figure 7a (upper left), dipole antenna and scenario B in Figure 7b ◦ ◦ ideal antenna and scenario B in Figure 7a (upper left), dipole antenna and scenario B in Figure 7b it is only shown for φ = 45 and θ = 45 . The subfigures have the following characteristics: isotropic (upper right) and isotropic ideal antenna, scenario D in Figure 7c (lower left) and dipole antenna and (upper right) and isotropic ideal antenna, scenario D in Figure 7c (lower left) and dipole antenna and scenario scenario BB in in Figure Figure 7d 7d (lower (lower right). right). Again, Again, in in all all of of the the subfigures, subfigures, the the mean mean gain gain of of each each scenario scenario isis plotted with a red line. It is shown that the variability of the results increases with the plotted with a red line. It is shown that the variability of the results increases with the number number of of

Sensors 2016, 16, 1334

9 of 14

ideal antenna and scenario B in Figure 7a (upper left), dipole antenna and scenario B in Figure 7b (upper right) and isotropic ideal antenna, scenario D in Figure 7c (lower left) and dipole antenna and scenario B in Figure 7d (lower right). Again, in all of the subfigures, the mean gain of each scenario Sensors 2016, 16, 1334 9 of 14 is plotted with a red line. It is shown that the variability of the results increases with the number of Sensors 2016, 16, 1334 9 of 14 sensors. With a low number of sensors (two and five) the standard deviation of results is low, and for a sensors. With a low number of sensors (two and five) the standard deviation of results is low, and for high anumber of sensors (20 and 50) the is very high.high. Additionally, this figure, it is shown sensors. With a of low number sensors (twovariability and five) the standard deviation ofin results is low, and high number sensors (20of and 50)variability the is very Additionally, in this figure, itfor is that the gain increase with the number of elements, except for 50 sensors. More details can be observed a high number of sensors (20 and 50) the variability is very high. Additionally, in this figure, it shown that the gain increase with the number of elements, except for 50 sensors. More details can beis shown that the gain increase with in Table 1 and Figure 8. observed in Table 1 and Figure 8. the number of elements, except for 50 sensors. More details can be observed in Table 1 and Figure 8.

◦ and ◦ . Isotropic Figure 7. Gain results two, five,10, 10,20 20and and 50 50 sensors sensors with θ =θ45°. ideal ideal Figure 7. Gain results forfor two, five, withφ= =45° 45and = 45Isotropic Figure 7. Gain results for two, five, 10, 20 and 50 sensors with  = 45° and θ = 45°. Isotropic ideal antenna and scenario B in subfigure (a) (upper left), dipole antenna and scenario B in subfigure (b) (b) antenna and scenario B in subfigure (a) (upper left), dipole antenna and scenario B in subfigure antenna and scenario B in subfigure (a) (upper left), dipole antenna and scenario B in subfigure (b) (upper right) isotropic antenna and scenario D in subfigure (c) (lower left) and dipole antenna and (upper right) isotropic antenna and scenario D in subfigure (c) (lower left) and dipole antenna and (upper right) isotropic antenna and scenario D in subfigure (c) (lower left) and dipole antenna and scenario B in subfigure (d) (lower right). scenario B in subfigure (d) (lower right).

scenario B in subfigure (d) (lower right).

Table 1. Mean gain factor for all scenarios with two antenna and  = 45°, θ = 45°. TableTable 1. Mean gaingain factor forfor allall scenarios antennaand and φ = 45θ◦ ,=θ45°. = 45◦ . = 45°, 1. Mean factor scenarioswith with two two antenna (φ = 45° θ = 45°) Scenario A Scenario B Scenario C Scenario D (φN° = 45° = 45°) A A Iso Scenario BB Scenario C Scenario D ◦ θ = 45◦Iso Sensors Dip Dip Iso Iso (φ θ = 45 )Scenario Scenario Scenario Scenario C Dip Scenario D Dip N° Sensors Iso Dip Iso Dip Iso Dip Iso Dip 2 N◦ Sensors 1.83 1.86 1.81 1.79 Iso Dip 1.80Iso Dip Iso Dip1.81 Iso 1.80 Dip 1.80 1.83 1.86 1.80 1.81 1.79 1.81 1.80 1.80 52 4.18 4.19 4.43 4.42 4.20 4.28 4.50 4.52 2 1.83 1.86 1.80 1.81 1.79 1.81 1.80 1.80 5 4.18 4.19 4.43 4.42 4.20 4.28 4.50 10 6.51 6.62 9.63 9.75 6.91 7.14 10.16 10.18 5 4.18 4.19 4.43 4.42 4.20 4.28 4.50 4.52 4.52 10 10 6.51 6.51 6.62 6.62 16.41 9.639.63 15.95 9.75 6.91 7.14 7.1410.16 19.54 10.16 9.75 6.91 10.18 10.18 20 8.42 8.36 10.52 10.57 19.1 16.41 15.16 15.95 10.52 20 20 8.42 8.42 8.36 8.36 16.26 16.41 15.95 10.52 10.57 10.5719.54 21.66 19.5419.1 21.93 19.1 50 8.02 8.06 11.34 11.33 16.26 15.16 15.16 11.34 21.93 21.93 50 50 8.02 8.028.06 8.06 16.26 11.34 11.3311.3321.66 21.66

Figure 8. Results for search angle  = 0°, θ = 45°, all scenarios and both antennas. ◦, θ θ = 45°, and both antennas. Figure 8. Results search angleφ ==00°, Figure 8. Results for for search angle 45◦ ,all allscenarios scenarios and both antennas. Table 1 shows the mean gain factor for all scenarios with two antenna and  = 45° θ = 45°. Table 1 shows the mean gainis factor for1 all two antenna aand  =reference 45° θ = 45°. Figure 8 shows the same data that in Table in ascenarios graphicalwith mode. Moreover, linear for Figure 8 shows the same data that is in Table 1 in a graphical mode. Moreover, a linear reference for

Sensors 2016, 16, 1334

10 of 14

Table 1 shows the mean gain factor for all scenarios with two antenna and φ = 45◦ θ = 45◦ . Figure 8 shows the same data that is in Table 1 in a graphical mode. Moreover, a linear reference for visual performance it is added. In this figure it is possible to observe that, for a low number of sensors (two and five) all scenarios work well; the main gain factor is approximately equal to the number of sensors. For an intermediate number of sensors (10), only scenarios B and D obtain a mean gain factor near the number of sensors. Meanwhile, scenarios A and C lose more than 30% of the main gain factor with respect to Scenarios B and D. No great difference between 2D and 3D position is found. For 20 sensors, the unique scenario that has a mean gain factor similar to the number of sensors is scenario D. With 20 sensors there exist differences between all scenarios; 3D scenarios work better than 2D scenarios, and scenarios with sensors with random power have more possibilities to conform the beamforming than scenarios with sensors with fixed power. Finally, with 50 sensors, only 3D scenarios (C and D) increase the mean gain factor obtained with 20 sensors, and the increase is small and very far from the number of sensors. The main reason that explains these results of the GA has to do with the size of the instances and the stopping condition of the algorithm, which has been kept the same in all the scenarios. As the number of sensors increase, the search space to be explored by the GA becomes much larger, but the sampling size (i.e., the number of function evaluations) is constant: 100 individuals × 100 generations = 10,000 evaluations. This clearly favors the small instances for which the GA is given more chance to find better quality solutions. Either increasing the population size of the GA, its number of generations to stop, or both would be required together with advanced diversity preservation mechanism that avoid the search to get stuck in local optima (premature convergence). In any case, there is room for improvement, especially when large instances are addressed. From the point of view of the node synchronization, the synchronization of 50 sensors can imply a significant problem because all of the nodes must transmit at the same time. Additionally, the calculation of the amplitudes and phases is not a trivial aspect for a genetic algorithm because the number of variables to be optimized is equal to 100. Therefore, a good choice can be the clusterization of the WSN, when the number of sensors is high. This is discussed in the next section. 4.2. Cluster Gain Results In this section, the sensor network has been divided into two clusters to study the behavior of the sensor network when the number of sensors is high. To split into two clusters of sensors, we have used the method of k-means, which is one of the simplest and popular, as explained in [15]: “One of the most popular and simple clustering algorithms, K-means, was first published in 1955. In spite of the fact that K-means was proposed over 50 years ago and thousands of clustering algorithms have been published since then, K-means is still widely used”. The procedure is as follows: first the K-means algorithm is applied to all sensors, obtaining two clusters, then, the beamforming technique is applied to the two clusters. To maintain the computational costs, a unique genetic algorithm has been applied in order to optimize the two clusters simultaneously, so that only one genetic algorithm optimizes the phases and amplitudes of the two clusters. Obviously, it would be a better strategy to apply two different genetic algorithms, one to each of the clusters. Figure 9 shows the results of performing optimization with two clusters, where 10, 20, and 50 sensors are simulated. All of the subfigures are for the Scenario D. The subfigures have the following characteristics: isotropic ideal antenna, φ = 0◦ and θ = 45◦ in subfigure a (upper left), dipole antenna, φ = 0◦ and θ = 45◦ in subfigure b (upper right), isotropic ideal antenna, φ = 45◦ and θ = 45◦ in subfigure c (lower left) and dipole antenna, φ = 45◦ and θ = 45◦ in subfigure D (lower right). Figure 9 shows that the gain factor is reduced (50%) when 10 and 20 sensors are simulated. This result is coherent since two clusters with half of the elements are being combined. As it has been stated in previous sections, the gain for scenario D is approximately equal to the number of elements in the range of 0–20. It is observed that the variability of the results is much higher when considering two clusters than when considering just one. However, with 50 sensors the gain achieved is about 12, not reaching the expected gain of 20. This gain value is expected because,

Sensors 2016, 16, 1334

11 of 14

with two clusters with more than 20 sensors, the gain at each cluster must be equal to the number of sensors in the cluster. The same explanation as in the previous section holds here. More sensors enlarge the search space to be explored by the GA, whilst is configuration remains the same concerning the number function evaluations performed (sampling size). Sensors 2016, 16,of1334 11 of 14

◦ Figure 9. 9. Gain Gain results results with with two two clusters, clusters, 10, 10, 20 20 and and 50 50 sensors, sensors, Scenario Scenario D. D. Isotropic Isotropic antenna antenna φ == 00° Figure ◦ ◦ ◦  = 0° and θ = 45° in subfigure (b) (upper and θ = 45° in subfigure (a) (upper left), dipole antenna, and θ = 45 in subfigure (a) (upper left), dipole antenna, φ = 0 and θ = 45 in subfigure (b) (upper right), isotropic isotropicantenna, antenna,φ = 45 = ◦45° and 45°subfigure in subfigure (c) (lower left) and antenna, dipole antenna, right), and θ =θ45=◦ in (c) (lower left) and dipole φ = 45◦ ◦  = 45° 45° in subfigure (d) right). (lower right). and θ =and 45 θ in=subfigure (d) (lower

5. Gain 5. GainResults Results Up to to now, we have have taken taken into into account the gain in aa certain Up now, we account the gain with with beamforming beamforming in certain pointing pointing direction direction and compared it with the directivity of the sensor in that particular direction. In this section, section, the the and compared it with the directivity of the sensor in that particular direction. In this received power power calculation calculation based based on on the the distance distance to to HECN HECN applying applying the the Friis Friis formula formula is is applied applied to to received calculate the propagation loss. The phase center of the array and the exact angle that is the HECN are calculate the propagation loss. The phase center of the array and the exact angle that is the HECN are available, as as well well as asthe thedistance distancetotothe thecentral centralnode nodeand andthe theangles angles each nodes transmit available, ofof each of of thethe nodes to to transmit to to the HECN. Simulations have been performed with 10 and 20 sensors, and with one and two the HECN. Simulations have been performed with 10 and 20 sensors, and with one and two clusters. clusters. Simulations with 50 sensors have notbecause, been done because, as depicted in the previous Simulations with 50 sensors have not been done as depicted in the previous sections, genetic sections, genetic algorithms with the standard parameters are not able to find a solution when the algorithms with the standard parameters are not able to find a solution when the number of sensors number of sensors increases. 10 shows the resultsoptimization of performingwith optimization withwith one cluster, increases. Figure 10 shows Figure the results of performing one cluster, 10 and with 10 and 20 sensors. This figure provides that the performance of the cases with perfect isotropic 20 sensors. This figure provides that the performance of the cases with perfect isotropic antenna is antenna is very similar to those obtained in the previous section. However, results with an antenna, very similar to those obtained in the previous section. However, results with an antenna, such as a such asare a dipole, are significantly better forThis 3D is cases. This due to the radiation pattern forwith the dipole, significantly better for 3D cases. due to theisradiation pattern for the sensors sensors with dipole antennas. As the dipole can have very low directivity in certain directions, if any dipole antennas. As the dipole can have very low directivity in certain directions, if any of these low of these low directivity directions correspond to thetoexact direction to the HECN station,value a low directivity directions correspond to the exact direction the HECN station, a low directivity is directivity value is obtained. This makes the lifetime of those sensors very low. This problem obtained. This makes the lifetime of those sensors very low. This problem does not occur whendoes the not occur when the beamforming technique the is applied. Therefore, gain factor, in beamforming technique is applied. Therefore, gain factor, in somethe cases, increases upsome to 45. cases, increases up11toshows 45. the results of performing optimization with two clusters, with 10 and 20 sensors. Figure Figure 11 shows the results of performing optimization with two clusters, and 20 sensors. The results with two clusters for ideal isotropic antenna are very similar to thosewith of the10previous section. The results with two clusters for ideal isotropic antenna are very similar to those of the previous However, as it occurs in the case of one cluster, the case of two clusters, dipole antenna, and 3D, the section. However, asin it the occurs thecompared case of onetocluster, thecase. caseAs of two clusters,the dipole antenna, and gain factor increases realincase the ideal an example, mean gain factor 3D, the gain factor increases in the real case compared to the ideal case. As an example, the mean gain obtained in the Scenario D for two clusters is 12 for 20 nodes, while for the same case in Section 4.2 the factor gain obtained Scenario mean factorinisthe equal to 9. D for two clusters is 12 for 20 nodes, while for the same case in Section 4.2 the mean gain factor is equal to 9.

Sensors 2016, 2016, 16, 16, 1334 1334 Sensors Sensors 2016, 16, 1334

12 of of 14 14 12 12 of 14

Figure 10. Gain results with one cluster, 10 and 20 sensors, all scenarios. Isotropic antenna and 10 Figure Gain results results with withone onecluster, cluster,10 10and and20 20sensors, sensors,allallscenarios. scenarios.Isotropic Isotropic antenna and Figure 10. 10. Gain 10 sensors in subfigure (a) (upper left), dipole antenna and 10 sensors in subfigure (b)antenna (upper and right), 10 sensors in subfigure (a) (upper left), dipole antenna and 10 sensors in subfigure (b) (upper right), sensors inantenna subfigure dipole antenna andleft), 10 sensors in subfigure (b) (upper right), isotropic and(a) 20(upper sensorsleft), in subfigure (c) (lower and dipole antenna and 20 sensors in isotropic antenna and 20 sensors in subfigure (c) (lower left), and dipole antenna and 20 sensors in isotropic antenna and 20 sensors in subfigure (c) (lower left), and dipole antenna and 20 sensors in subfigure (d) (lower right). subfigure subfigure (d) (d) (lower (lower right). right).

Figure Gain results results with Figure 11. 11. Gain with two two clusters, clusters, 10 10 and and 20 20 sensors, sensors, all all scenarios. scenarios. Isotropic Isotropic antenna antenna and and Figure 11. Gain results with two clusters, 10 and 20 sensors, all scenarios. Isotropic antenna and 10 sensors in subfigure (a) (upper left), dipole antenna and 10 Sensors in subfigure (b) (upper right), 10 sensors in subfigure (a) (upper left), dipole antenna and 10 Sensors in subfigure (b) (upper right), 10 sensors in subfigure (a) (upper left), dipole antenna and 10 Sensors in subfigure (b) (upper right), isotropic antenna and 20 sensors in subfigure (c) (lower left) and dipole antenna and 20 sensors in isotropic antenna and 20 sensors in subfigure (c) (lower left) and dipole antenna and 20 sensors subfigure (d) (lower right). isotropic antenna and 20 sensors in subfigure (c) (lower left) and dipole antenna and 20 sensors in subfigure (d) (lower right). in subfigure (d) (lower right).

6. Conclusions 6. Conclusions 6. Conclusions This This work work provides provides the the evaluation evaluation of of WSN WSN node node networks networks and and their their performance performance when when both both This work provides the evaluation of WSN node networks and their performance both clustering and antenna beamforming are applied. In this work we have fixed four different scenarios clustering and antenna beamforming are applied. In this work we have fixed four differentwhen scenarios clustering and antenna beamforming are applied. In this work we have fixed four different scenarios and with different number of sensors implied: 50, 20, five, nodes and each eachscenario scenarioisissimulated simulated with different number of sensors implied: 50,10,20, 10,and five,two and two and each scenario is simulated with different number of sensors implied: 50, 20, 10, five, and per scenario, where each scenario is randomly generated thirty times in order to validate the results nodes per scenario, where each scenario is randomly generated thirty times in order to validate two the nodes per scenario, where scenario randomly generated thirty times in order to validate are the and their repeatability and each reliability. Foriseach experiment, two different transmission directions results and their repeatability and reliability. For each experiment, two different transmission ◦ ◦ ◦ ◦ results and their repeatability and reliability. For each experiment, two different transmission considered (φ considered = 0 and θ = ; φand = 45 45 )and for the process. Each scenario is directions are (45 = 0° θ =, and 45°; θ  == 45°, θ = optimization 45°) for the optimization process. Each directions are considered (  = 0° and θ = 45°;  = 45°, and θ = 45°) for the optimization process. Each evaluated for two different antennas, an ideal isotropic antenna and a conventional dipole one. In this scenario is evaluated for two different antennas, an ideal isotropic antenna and a conventional dipole scenario is evaluated fortypes two different antennas, an are ideal isotropic antenna andsensors a conventional dipole set ofIn experiments, two of two WSNtypes are considered: inconsidered: the first one, the same one. this set of experiments, of WSN inall theoffirst one, all have of thethe sensors one. In this set of experiments, two types of WSN are considered: in the first one, all of the sensors amount of power for communication purposes; in the second one, each sensor has a different amount have the same amount of power for communication purposes; in the second one, each sensor has a have the amount same amount of power for The communication purposes; in the second one, each has a different of available power. analyzed cases in this document are focused onsensor static nodes different amount of available power. The analyzed cases in this document are focused on static nodes

Sensors 2016, 16, 1334

13 of 14

of available power. The analyzed cases in this document are focused on static nodes (no movement after the random scenario generation) and 2D surface and 3D space for the node location. In the results, when the desired direction is obtained with beamforming increasing the array directivity, the gain factor for a low number of nodes is equal to the number of sensors, which is an important conclusion of this paper. For a medium number of sensors, the gain factor is equal to the number of sensors only for Scenario D (sensors with different available power for communications and 3D distribution), that is, the scenario with the best performance is the scenario D, which is another important conclusion of this paper. Regarding the number of clusters, it is observed that clusterization is a good strategy for a number of nodes higher than 20. The gain in the two-cluster case is equal to the number of nodes of the cluster. Moreover, in a real environment, with real antennas (dipole), the gain obtained with beamforming is higher for both cases (one and two clusters). To the authors’ knowledge, this is the first time that beamforming and clustering have been simultaneously applied to increase the network performance in WSNs. Future work is related to an improvement in the optimization algorithms in order to solve adequately scenarios with a high number of sensors. Also, the optimization of the number of clusters and the amplitude and phase of the sensors simultaneously may become another interesting work. Finally, it is possible to develop scenarios with multiple HECN (multiple directions) and multiple directions from which the WSN is being attacked (multiple directions towards which decrease the transmission power). Acknowledgments: This work has been partially funded by the Government of Extremadura and the European Regional Development Fund (FEDER) under the IB13113 project. Author Contributions: All authors contributed equally to this work. Conflicts of Interest: The authors declare no conflicts of interest.

References 1. 2. 3. 4. 5. 6. 7. 8.

9.

10. 11.

Agiwal, M.; Roym, A.; Saxena, N. Next Generation 5G Wireless Networks: A Comprehensive Survey. IEEE Commun. Surv. Tutor. 2016. [CrossRef] Boccardi, F.; Heath, R.W.; Lozano, A.; Marzetta, T.L.; Popovski, P. Five disruptive technology directions for 5G. IEEE Commun. Mag. 2014, 52, 74–80. [CrossRef] Community Research and Development Information Service (CORDIS), European Comission. Research*eu, Focus Magazine 5G; Brussels, Belgium, 2015. Bogale, T.E.; Le, L.B. Massive MIMO and mmWave for 5G Wireless HetNet: Potential Benefits and Challenges. IEEE Veh. Technol. Mag. 2016, 11, 64–75. [CrossRef] Tsai, C.W.; Cho, H.H.; Shih, T.K.; Pan, J.S.; Rodrigues, J.J.P.C. Metaheuristics for the deployment of 5G. IEEE Wirel. Commun. 2015, 22, 40–46. [CrossRef] Buzzi, S.; Chih-Lin, I.; Klein, T.E.; Poor, V.; Yang, C.; Zappone, A. A Survey of Energy-Efficient Techniques for 5G Networks and Challenges Ahead. IEEE J. Sel. Areas Commun. 2016. in press. [CrossRef] Haro, B.B.; Zazo, S.; Palomar, D.P. Energy Efficient Collaborative Beamforming in Wireless Sensor Networks. IEEE Trans. Signal Process. 2014, 62, 496–510. [CrossRef] Padilla, J.L.; Padilla, P.; Valenzuela-Valdés, J.F.; Serrán-González, J.; López-Gordo, M.A. Performance Analysis of Different Link Layer Protocols in Wireless Sensor Networks (WSN). Wirel. Pers. Commun. 2015, 84, 3075–3089. [CrossRef] Feng, J.; Lu, Y.; Jung, B.; Peroulis, D. Energy efficient collaborative beamforming in wireless sensor networks. In Proceedings of the IEEE International Symposium on Circuits and Systems, ISCAS, Taipei, Taiwan, 24–27 May 2009. Ahmed, M.F.A.; Vorobyov, S.A. Collaborative beamforming for wireless sensor networks with Gaussian distributed sensor nodes. IEEE Trans. Wirel. Commun. 2009, 8, 638–643. [CrossRef] Valenzuela-Valdés, J.F.; Luna, F.; Luque, R.M.; Padilla, P. Saving energy in WSNs with beamforming. In Proceedings of the 2014 IEEE 3rd International Conference on Cloud Networking (CloudNet), Luxembourg, Luxembourg, 8–10 October 2014; pp. 255–260.

Sensors 2016, 16, 1334

12.

13. 14. 15.

14 of 14

Agudo, J.E.; Valenzuela-Valdés, J.F.; Luna, F.; Luque-Baena, R.M.; Padilla, P. Analysis of beamforming for improving the energy efficiency in wireless sensor networks with metaheuristics. Prog. Artif. Intell. 2016, 5. [CrossRef] Bäck, T.; Fogel, D.; Michalewicz, Z. (Eds.) Handbook of Evolutionary Computation; Oxford University Press: Oxford, UK, 1997. Matlab Toolbox. Available online: http://www.mathworks.es/es/help/phased/beamforming.html (accessed on 18 August 2016). Jain, A.K. Data clustering: 50 Years beyond K-means. Pattern Recognit. Lett. 2010, 31, 651–666. [CrossRef] © 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).