Optimal Deployment of Sensor Nodes Based on Performance ... - MDPI

sensors Article

Optimal Deployment of Sensor Nodes Based on Performance Surface of Underwater Acoustic Communication Sunhyo Kim

ID

and Jee Woong Choi *

Department of Marine Science and Convergence Engineering, Hanyang University, 55 Hanyangdaehak-ro, Sangnok-gu, Ansan, Gyeonggi-do 15588, Korea; [email protected] * Correspondence: [email protected] Received: 6 August 2017; Accepted: 16 October 2017; Published: 20 October 2017

Abstract: The underwater acoustic sensor network (UWASN) is a system that exchanges data between numerous sensor nodes deployed in the sea. The UWASN uses an underwater acoustic communication technique to exchange data. Therefore, it is important to design a robust system that will function even in severely fluctuating underwater communication conditions, along with variations in the ocean environment. In this paper, a new algorithm to find the optimal deployment positions of underwater sensor nodes is proposed. The algorithm uses the communication performance surface, which is a map showing the underwater acoustic communication performance of a targeted area. A virtual force-particle swarm optimization algorithm is then used as an optimization technique to find the optimal deployment positions of the sensor nodes, using the performance surface information to estimate the communication radii of the sensor nodes in each generation. The algorithm is evaluated by comparing simulation results between two different seasons (summer and winter) for an area located off the eastern coast of Korea as the selected targeted area. Keywords: underwater acoustic sensor network; performance surface; virtual force-particle swarm optimization; optimal deployment

1. Introduction The underwater acoustic sensor network (UWASN) has recently attracted considerable attention because of its potential for use in a variety of military and civilian applications, such as ocean environmental monitoring, ocean exploration, target detection, surveillance systems, and communication among underwater sensor nodes [1–4]. In the UWASN, a number of sensor nodes are deployed in the ocean to collect data, which must be successfully exchanged among adjacent sensor nodes. For this reason, each sensor node must be positioned to be able to perform collaborative communication tasks over a targeted area. The deployment efficiency of sensor nodes in the UWASN is evaluated in terms of their connectivity and the area covered by the limited number of sensor nodes in the targeted area, because underwater sensor devices are more expensive and more difficult to deploy at precise positions compared to sensor nodes on land [5]. Most attempts to find an optimal deployment scheme for sensor nodes have assumed that every sensor node has the same sensing range [6–9], which is reasonable in the case of terrestrial wireless communication systems. However, underwater communication performance is significantly influenced by temporal and spatial variations in the ocean environment [10,11]. Underwater acoustic communication presents a challenging problem in that the underwater acoustic communication

Sensors 2017, 17, 2389; doi:10.3390/s17102389

www.mdpi.com/journal/sensors

Sensors 2017, 17, 2389

2 of 14

channel is a time-varying multipath channel formed by multiple interactions of sound with the sea surface and bottom of the ocean, particularly in shallow water. This multipath channel causes significant delay in spreading, often covering hundreds of symbols, thereby causing inter-symbol interference (ISI). This happens because the underwater sound propagates with relatively low sound speed (mean sound speed in water is around 1500 m/s, which is approximately 200,000 times lower than that of an electromagnetic wave) [12]. The ISI results in a significant degradation in communication performance [13]. In addition, a time-varying channel produces a short coherence time or a large Doppler spread resulting from the temporal and spatial variations in the sea surface [14]. Therefore, if all sensor nodes are deployed at equal distances from each other in the targeted area under the assumption that each sensor node has the same communication radius, performance degradation of the UWASN may occur. It is necessary to develop an optimal deployment algorithm for underwater sensor nodes, which must be able to reflect performance variations in the sensor network due to environmental fluctuations. One of the methods to improve the performance of the UWASN is to use array processing. Array processing techniques using an array of spatially separated receivers have been reported to decrease communication error rates by eliminating the ISI caused by multipath channels in the ocean and to have better communication performance than single receivers that only use the equalizer technique [15–17]. In this paper, a multi-channel combining technique using a vertical line array is applied to the UWASN. The performance of the communication system is then estimated over the targeted area, which is converted to the communication performance surface (PS). The PS is a projection of the geospatial information of the performance of a system. In general, the PS in a sonar system is used to provide insight into the relative performance for the detection range of the system, demonstrating the impact of the ocean environment on the system capability [18]. Here, a communication PS is predicted based on the communication radius estimated on each grid within the targeted area. On the basis of the estimated communication PS, the optimal deployment of sensor nodes is obtained using a hybrid optimization method combining the advantages of a virtual force algorithm [19,20] and a particle swarm optimization [21,22], called virtual force-particle swarm optimization (VFPSO) [23]. The goal of this paper is to achieve the maximum communication coverage maintaining their connectivity with the limited number of underwater sensor nodes, especially in shallow water areas. The paper is organized as follows. Section 2 provides the descriptions of four sub-algorithms for optimal deployment of the sensor nodes: modeling of underwater acoustic channel impulse response; estimation of the communication performance; construction of the communication PS, and; optimal deployment of the sensor nodes. Section 3 presents the results of the optimal deployment positions for sensor nodes simulated for winter and summer seasons at a targeted area located off the east coast of Korea. Finally, Section 4 provides a summary and conclusion. 2. Algorithm for Optimal Deployment The algorithm to obtain the optimal deployment scheme is composed of four sub-categories. First, underwater acoustic channel modeling is carried out using a ray-based acoustic model for the targeted area using the ocean environment database including bathymetry, sound speed profiles, and sediment types. Second, the communication performance is estimated using the array processing technique of a single-input multiple-output (SIMO) system. The signal received at each array channel is simulated via a convolution of the communication source signal with the simulated channel impulse response. Third, the PS for the targeted area is derived using the estimated communication performance. Lastly, the optimal deployment scheme for the sensor nodes is determined using the VFPSO method based on the estimated PS. 2.1. Modeling of the Underwater Acoustic Channel Impulse Response The underwater communication channel is characterized by multipaths caused by the acoustic interactions with the sea surface and bottom interfaces, as well as the water medium itself, which causes

Sensors 2017, 17, 2389

3 of 14

Sensors 2017, 17, 2389

3 of 14

demodulation of the communication signal complex and difficult. Therefore, it is necessary to determine the optimal deployment scheme based on the communication PS estimated via the ISI, and resulting in serious distortion communication signals. Theassociated ISI makes with demodulation of communication channel modeling. Theofmodel input parameters the ocean the communication signal complex and difficult. Therefore, it is necessary to determine the optimal environmental information, including sound speed profiles, bathymetry, and sediment properties, deployment scheme based on the communication PS estimated communication channel are needed for accurate underwater acoustic channel modeling.via Here, the bathymetry in themodeling. targeted The model input parameters associated with the ocean environmental information, including area is extracted from ETOPO1, which is a global relief model of Earth’s surface integratingsound land speed profiles, bathymetry, and sediment properties, are needed for accurate underwater acoustic topography and ocean bathymetry supplied by NOAA (National Oceanic and Atmospheric channel modeling. Here,with the bathymetry in the targeted is extracted fromprofiles ETOPO1, is a Administration) [24,25], a spatial resolution of ~100 area m. The sound speed arewhich extracted global relief model of Earth’s surface integrating land topography and ocean bathymetry supplied from the GDEM (Generalized Digital Environment Model) with a 1/4° horizontal grid resolution and by NOAA (National Oceanic and Atmospheric Administration) [24,25], with a spatial resolution of a monthly time resolution Theare values of sediment mean grain(Generalized size are obtained by Environment interpolating ~100 m. The sound speed [26]. profiles extracted from the GDEM Digital the mean grain sizes measured from surficial sediment samples taken directly from 14 values positions ◦ Model) with a 1/4 horizontal grid resolution and a monthly time resolution [26]. The of within the targeted area. The mean grain size values of the surficial sediment were converted to the sediment mean grain size are obtained by interpolating the mean grain sizes measured from surficial sediment speed, attenuation coefficient using the empirical formula sediment sound samples takendensity, directlyand from 14 positions within the targeted area. The mean developed grain size by Jackson and Richardson [27]. values of the surficial sediment were converted to the sediment sound speed, density, and attenuation Figureusing 1a shows the targeted area selected in this is 22 km ×[27]. 22 km and its western coefficient the empirical formula developed by paper. JacksonThe andsize Richardson boundary is ~10 km away from the eastern coast of Korea. The targeted area divided 100 Figure 1a shows the targeted area selected in this paper. The size is 22 km × was 22 km and its into western grid points, for km which environmental data were the database for eight azimuthal boundary is ~10 away from the eastern coast of extracted Korea. Thefrom targeted area was divided into 100 grid angles (Figure 1b). Figure 2a,b shows the interpolation maps of bathymetry and mean grain size points, for which environmental data were extracted from the database for eight azimuthal angles (expressed by ϕ, where ϕ= −log ( / ), is the grain diameter in millimeters, and is (Figure 1b). Figure 2a,b shows the interpolation maps of bathymetry and mean grain size (expressedthe by reference length, equal to 1 mm) for the targeted area. Figure 2c shows examples of sound speed φ, where φ= − log2 (d /d0 ), d is the grain diameter in millimeters, and d0 is the reference length, equal profiles for eight azimuthal angles of a grid point speed locatedprofiles at 36° 79′ and 129° 68′ to 1 mm)and forbathymetries the targeted area. Figure 2c shows examples of sound andNbathymetries ◦ 0 ◦ 0 forThe eight azimuthal channel angles of a gridresponses point located at eight 36 79 N and 129 68 E.are The deterministic E. deterministic impulse for the azimuthal directions predicted using channel impulse for theacoustic eight azimuthal directions predicted using BELLHOP, is BELLHOP, whichresponses is a ray-based propagation modelare [28]. The BELLHOP model is which a highly a ray-based propagation modelmodel [28]. The BELLHOP model is a highly and rapid efficient andacoustic rapid acoustic ray tracing suitable for higher frequency andefficient range-dependent acoustic ray tracing model suitable for higher frequency and range-dependent ocean environments. ocean environments.

Figure Figure 1. 1. (a) (a) Location Location of of the the targeted targeted area area and and (b) (b) 100 100 grid grid points points in in the the targeted targeted area. area. The The lines lines in in (b) (b) indicate of eight eight azimuthal azimuthal angles angles at at each each point. point. indicate the the directions directions of

Sensors 2017, 17, 2389

4 of 14

Sensors 2017, 17, 2389

4 of 14

Figure 2. 2. (a) Figure (a) Bathymetry Bathymetry map map and and (b) (b) mean mean grain grain size size distribution distribution of of surficial surficial sediment sediment of of the the targeted targeted area. (c) Sound speed profiles and bathymetries for eight azimuthal angles of a grid point located at at area. (c) Sound speed profiles and bathymetries for eight azimuthal angles of a grid point located ◦ 0 ◦ 0 N and and 129 129° 68 ′ E, which is marked with a green circle circle in in (a). (a). 36° 79 79′ N 36

Figure 3a,b shows the eigenray eigenray tracing tracing results results for for the the azimuthal azimuthal angles angles of of 90 90°◦ and and 270 270°◦, respectively, at a grid point located at 36◦ 790 N and 129◦ 680 E. The eigenray tracing results for respectively, at a grid point located at 36° 79′ N and 129° 68′ E. The eigenray tracing results for the the source-receiver ranges of 100 m, 1 km, and 2 km are shown in the top, middle, and bottom plots, source-receiver ranges of 100 m, 1 km, and 2 km are shown in the top, middle, and bottom plots, respectively. The source and receiver are assumed to be positioned 2 m above the bottom for both respectively. The source and receiver are assumed to be positioned 2 m above the bottom for both azimuthal angles. The main difference between the two different azimuthal angles is that the sound azimuthal angles. The main difference between the two different azimuthal angles is that the sound propagates into the up-slope direction for the azimuthal angle of 90◦ and the down-slope direction for propagates into the of up-slope direction the azimuthal of 90° and theasdown-slope direction the azimuthal angle 270◦ . Figure 3c,dfor shows the channelangle impulse responses a function of arrival ◦ ◦, time for the source-receiver ranges between 100 m and 2.5 km for the azimuthal angles of 90 and 270of for the azimuthal angle of 270°. Figure 3c,d shows the channel impulse responses as a function respectively. The first arrival includes direct (D) and bottom (B) paths. The second arrival includes arrival time for the source-receiver ranges between 100 m and 2.5 km for the azimuthal angles of 90° the paths associated with single sea-surface bounce path, such as sea surface (S), bottom-surface (B-S), 270°, respectively. The firstThe arrival includes direct (D) and bottom (B) paths. Theinteracting second arrival and surface-bottom (S-B) paths. third and fourth arrivals correspond to the paths with the sea surface twice and thrice, respectively. It is interesting to note that spreads of the includes the paths associated with single sea-surface bounce path, such asthe seadelay surface (S), bottomchannel impulse responses for the down-slope larger thanarrivals those for the up-slope direction. surface (B-S), and surface-bottom (S-B) paths.direction The thirdare and fourth correspond to the paths This difference eventually produce different communication performance. Thethe channel interacting withmay the sea surface twice andthe thrice, respectively. It is interesting to note that delay impulse responses simulated as aresponses function of range for the eightthan azimuthal angles spreads of the channel impulse forthe thesource-receiver down-slope direction are larger those for the up-slope direction. This difference may eventually produce the different communication performance. The channel impulse responses simulated as a function of the source-receiver range for

Sensors 2017, 17, 2389 Sensors 2017, 17, 2389

5 of 14 5 of 14

the eight azimuthal angles are convolved with an original communication sequence to obtain the received communication signals. This processsequence is repeated for 100 points in the targeted area are convolved with an original communication to obtain thegrid received communication signals. shown in Figure 1b. This process is repeated for 100 grid points in the targeted area shown in Figure 1b.

Figure 3. 3. Eigenray Eigenray tracing tracing results results for for the the azimuthal azimuthal angles angles of of (a) (a) 90 90° and(b) (b)270 270°. Channel impulse impulse ◦ and ◦ . Channel Figure ◦ ◦ responses . responses as as aa function function of of arrival arrival time time for for the the azimuthal azimuthal angles angles of of (c) (c) 90 90° and and(d) (d)270 270°.

2.2. Estimate of the Communication Performance Real-time communication with a high data transfer rate and low bit error rate (BER) is required is a islimit to using a single sensor sensor node innode the UWASN because for aa robust robustUWASN UWASNsystem. system.There There a limit to using a single in the UWASN underwater communication is severely affected by the time-varying multipaths because underwater communication is severely affected by the time-varying multipathsininthe the ocean Recently, in in aa number number of of studies studies on on underwater underwater acoustic communication, sensor array environment. Recently, systems have have been been used used to to achieve achieve the the high high data data rate rate with with low low BER BER [15–17]. [15–17]. In general, as the element systems spacing and andthethe number of sensors increase, communication performance spacing number of sensors increase, communication performance becomes becomes enhanced.enhanced. However, However, the spatialaggravates diversity the aggravates the spatial of the UWASN Here, that it is the spatial diversity spatial efficiency of efficiency the UWASN system. Here,system. it is assumed assumed that eachinsensor node inisthe UWASN a vertical line array composed of three hydrophone each sensor node the UWASN a vertical lineisarray composed of three hydrophone receivers with receivers an element of 1.5 m, which to 10 λ (where λ is thebased acoustic an elementwith spacing of 1.5 m,spacing which corresponds to 10 λcorresponds (where λ is the acoustic wavelength) on on a frequency of 10 kHz. awavelength) frequency ofbased 10 kHz. communication sequence used for the communication communication performance simulation is a binary The communication phase-shift keying (BPSK) kHz and a bit raterate of 1ofkbps. To phase-shift (BPSK) sequence sequencewith withaacenter centerfrequency frequencyofof1010 kHz and a bit 1 kbps. simulate thethe communication signals received To simulate communication signals receivedafter afterpropagating propagatingthrough throughthe themultipath multipathocean ocean channel, channel, communication sequence sequence is convolved convolved with the channel impulse responses simulated using the BPSK communication method described described in in Section Section 2.2, 2.2, and and then then isotropic isotropic white white Gaussian Gaussiannoise noisewas wasadded. added. the method

Sensors 2017, 17, 2389 Sensors 2017, 17, 2389

6 of 14 6 of 14

Figure 4 shows shows aa signal signal processing processing block block diagram diagram of the SIMO SIMO system used to decode decode the communication data. data. The Thecommunication communicationdata datareceived receivedbyby each sensor component are multiplied each sensor component are multiplied by − iωt  by to recover baseband waveform, where the angular frequency. that,data the were data e e to recover the the baseband waveform, where ω is ω theisangular frequency. AfterAfter that, the were low-pass filtered passed through adaptive Decisionfeedback feedbackequalizer equalizer(DFE) (DFE) [29] [29] to low-pass filtered and and passed through an an adaptive Decision compensate for the channel distortion by ISI. Recursive Least-Squares (RLS) algorithm was used to ISI. Recursive Least-Squares adaptively update the filter weights of the equalizer, equalizer, and and aa forgetting forgetting factor factor was was 0.995. 0.995. The The number number of feedforward and feedback filter taps in the DFE is related to the delay time spread due to the multipaths underwater communication channel [10,17]. In thisInpaper, tap numbers covering multipaths ininthe the underwater communication channel [10,17]. this the paper, the tap numbers the delay the timedelay spread sufficiently were used were in theused equalizer Then, they are summed to covering time spread sufficiently in theprocess. equalizer process. Then, they are eliminate residual ISI and increase SNR (Signal-to-Noise Ratio) [30,31]. Finally, the communication summed to eliminate residual ISI and increase SNR (Signal-to-Noise Ratio) [30,31]. Finally, the performance is evaluated withisaevaluated BER estimate. communication performance with a BER estimate.

Figure 4. 4. Block of the the single-input single-input multiple-output multiple-output (SIMO) (SIMO) communication communication receiver receiver system. system. Figure Block diagram diagram of

2.3. Communication Communication Performance Performance Surface Surface Algorithm Algorithm 2.3. As described described in in the the previous previous section, section, the the process process to to estimate estimate the the communication communication performance performance is is As complex and time-consuming. If the process is repeated on candidate positions in the targeted area complex and time-consuming. If the process is repeated on candidate positions in the targeted area in in the computation loop in order to search optimal sensor positions, it will to find the computation loop in order to search for for optimal sensor positions, it will taketake too too longlong to find the the optimal solution. Here, the communication PS for the targeted area is constructed based onBER the optimal solution. Here, the communication PS for the targeted area is constructed based on the BER performance estimated in Section 2.2, and then it is used for rapid computation of the search performance estimated in Section 2.2, and then it is used for rapid computation of the search process process to find the optimal positions. The communication PS represents thedistribution spatial distribution of the to find the optimal positions. The communication PS represents the spatial of the relative relative performance of the underwater communication system, which varies temporally performance of the underwater communication system, which varies temporally and spatially inand the spatially in the ocean [18]. ocean [18]. The BER BER estimates estimates as as aa function function of of the the source-receiver source-receiver range range for for the the eight eight azimuthal azimuthal angles angles of of The each grid point were made to obtain the communication PS of the targeted area. Figure 5a,b show the each grid point were made to obtain the communication PS of the targeted area. Figure 5a,b show interpolation of the BER performances forfor the eight predicted the interpolation of the BER performances the eightazimuthal azimuthalangles anglesofofeach each grid grid point point predicted using the mean sound speed profiles in February and August, respectively. The difference between using the mean sound speed profiles in February and August, respectively. The difference between the the sound speed profiles in the two seasons produces the difference of channel impulse response, sound speed profiles in the two seasons produces the difference of channel impulse response, which which ultimately produces the difference in BER performance. Note vary ultimately produces the difference in BER performance. Note that the that BER the mayBER varymay with the with typesthe of types of equalizer and receiver used in the UWASN. Here, a BER of 2% is chosen as a criterion of equalizer and receiver used in the UWASN. Here, a BER of 2% is chosen as a criterion of tolerance for tolerance for communication, and the range to this as criterion is definedrange. as a communication, and the range corresponding to corresponding this criterion is defined a communication communication range. Figure 6 shows an example of the BER performance estimated as a function of Figure 6 shows an example of the BER performance estimated as a function of range. In this case, range. In this case, the communication range is determined to be 1.8 km. the communication range is determined to be 1.8 km. Now, the communication ranges for the eight azimuthal angles are averaged to represent the communication radius at the grid point. The minimum or maximum communication range can also be chosen optionally by a user as a representation of the communication radius at the grid point.

Sensors 2017, 17, 2389

7 of 14

Sensors 2017, 17, 2389

7 of 14

The communication radii estimated for every grid point are interpolated to create the communication PS of the targeted area. Figure 7 shows an example of the communication PS simulated using a three-channel vertical array for the two different seasons ((a) in February and (b) in August). In our simulation case, the overall communication radii in February are longer than those in August, Sensors 17, Sensors 2017, 17,2389 2389 14 which2017, means that the communication environment in February is better than that in August. 77ofof14

Figure 5. Communication bit error rate (BER) performance fields of 100 grid points obtained by interpolation of the BER estimate as a function of range for the eight azimuthal angles predicted using the mean sound speed profiles in (a) February and (b) August.

Figure 5.5. 5. Communication Communicationbit biterror errorrate rate(BER) (BER)performance performancefields fieldsof of100 100grid gridpoints pointsobtained obtainedby by Figure Figure Communication bit error rate (BER) performance fields of 100 grid points obtained by interpolation of the BER estimate as a function of range for the eight azimuthal angles predicted using interpolation of the BER estimate as a function of range for the eight azimuthal angles predicted using interpolation of the BER estimate as a function of range for the eight azimuthal angles predicted using themean meansound soundspeed speedprofiles profilesin in (a) February and (b) August. the the mean sound speed profiles in(a) (a)February Februaryand and(b) (b)August. August. Figure 6. Example of the BER performance as a function of range. A BER of 2% is used as a criterion of tolerance for communication in this paper.

Now, the communication ranges for the eight azimuthal angles are averaged to represent the communication radius at the grid point. The minimum or maximum communication range can also be chosen optionally by a user as a representation of the communication radius at the grid point. The communication radii estimated for every grid point are interpolated to create the communication PS of the targeted area. Figure 7 shows an example of the communication PS simulated using a three-channel vertical array for the two different seasons ((a) in February and (b) in August). In our simulation case, the Figure 6.Example Exampleof of the BER as function of AA ofof used asas ameans of the Figure 6.6. function of range. BER 2% used acriterion overall communication radii inperformance February are than those in August, which that Figure Example ofthe theBER BER performance asaaalonger function ofrange. range. ABER BER of2% 2%isisis used as acriterion criterion tolerance for communication in this paper. of tolerance for communication in this paper. of tolerance environment for communication in this paper. communication in February is better than that in August. Now, Now, the the communication communication ranges ranges for for the the eight eight azimuthal azimuthal angles angles are are averaged averaged to to represent represent the the communication communication radius radiusat at the the grid grid point. point. The The minimum minimum or or maximum maximum communication communication range range can can also also be be chosen optionally by a user as a representation of the communication radius at the grid point. The chosen optionally by a user as a representation of the communication radius at the grid point. The communication communication radii radii estimated estimated for for every every grid grid point point are are interpolated interpolated to to create create the the communication communication PS PS of of the thetargeted targetedarea. area.Figure Figure77shows showsan anexample exampleof ofthe thecommunication communicationPS PSsimulated simulatedusing usingaathree-channel three-channel vertical verticalarray arrayfor forthe thetwo twodifferent differentseasons seasons((a) ((a)in inFebruary Februaryand and(b) (b)in inAugust). August).In Inour oursimulation simulationcase, case,the the overall overall communication communication radii radii in in February February are are longer longer than than those those in in August, August, which which means means that that the the communication communicationenvironment environmentin inFebruary Februaryisisbetter betterthan thanthat thatin inAugust. August.

Figure 7. forfor thethe targeted areaarea in (a) and Figure 7. Communication Communicationperformance performancesurface surface(PS) (PS)simulated simulated targeted in February (a) February (b) August. and (b) August.

Sensors 2017, 17, 2389

8 of 14

2.4. Optimal Deployment Algorithm for the Sensor Nodes In general, the performance of the algorithm for optimal deployment of the sensor nodes is evaluated in terms of maximizing the communication coverage rate and maintaining connectivity among the sensor nodes with a given number of nodes in the targeted area. Here, a hybrid method combining the advantages of the virtual force algorithm (VFA) [19,20] and the particle swarm optimization (PSO) [21,22] is used as a search algorithm, which is referred to as the VFPSO [23]. The VFA is a self-organizing search algorithm to determine the optimal distances among sensor nodes, which attempts to maximize the communication coverage using a combination of attractive and repulsive forces. As an initial step, the sensors are randomly placed in the targeted area. The distances between one sensor node and its neighboring nodes are compared to a communication threshold. In this work, the communication threshold was defined to be the smaller of the communication radii at two sensor node positions for two-way communication. The communication radius at a certain node position can be extracted from the communication PS map. If the distance is smaller than the threshold, a repulsive force arises between the two sensor nodes. In contrast, if the distance is too far apart, an attractive force arises. This process is performed for all neighboring nodes based on one sensor node, and a weighted sum is calculated to obtain the total force exerted on the node. This is repeated for each node. As the iteration of the algorithm continues, all sensors nodes are positioned, maintaining optimal distances between the sensor nodes. The PSO is a stochastic search algorithm that mimics social behavior of animals moving in flocks in order to find optimal positions. It has been widely used because it is an efficient optimization algorithm for solving dynamic optimization problems having fast convergence and robustness [21,22]. The PSO uses particles, which correspond to the individual sensor nodes in this study, and the set of all sensor nodes is referred to as the swarm in the PSO algorithm. After being initialized with random positions and velocities, the particles estimate their communication radii at their positions and memorize this information. At each generation, particles evaluate the best positions by comparison with the positions achieved at previous generations. The velocity of each particle is adjusted based on the experiences of the particle and its companions. The position in the next generation is then updated by the sum of the present position and the adjusted velocity [23]. The VFA has outstanding performance in adjusting the distance among the sensor nodes. However, there may be a limit because the strong attractive and repulsive forces among the sensor nodes may hinder the nodes from settling at optimal positions. In contrast, the PSO has the ability to find a global optimal solution. However, it does not have the ability to adjust the distances among the sensor nodes, which hinders the maximization of the communication coverage rate. Here, the VFPSO algorithm is applied to find the optimal positions of the sensor nodes in the UWASN, which is expressed by [23]. → Fi

k

=

∑

→

Fi, j

(1)

j = 1, j 6= i

→

Fi, j =

vi, d (n + 1)

  



h   w R


 w A di, j − dth , αi, j i f di, j > dth 0 i f di, j = dthi  1

di,

j

−

1 dth

, αi, j + π

i f di, j < dth

  

h i → = vi, d (n) + c1 r1 [ pi, d (n) − xi, d (n)]+c2 r2 p g, d (n) − xi, d (n) + Fi xi, d (n + 1) = xi, d (n) + vi, d (n + 1),

(2)

  (3) (4)

→

where Fi, j is the resultant force of the attractive force and repulsive force between the i-th and j-th →

sensor nodes. Fi is the total sum of the forces between the i-th sensor node and others. di, j is the Euclidean distance between the i-th and j-th sensor nodes, and dth is the communication threshold, which was defined as the smaller value of the communication radii for the i-th and j-th sensor nodes

Sensors 2017, 17, 2389

9 of 14

as mentioned earlier. αi, j indicates the direction from the i-th to j-th sensor nodes. w A and w R are the Sensors 2017, 17, 2389 9 of 14 weighting values for the attractive and repulsive forces, respectively. xi, d (n) and vi, d (n) represent the and velocity of the i-th sensor node in the d-th dimension at iteration n, respectively. andposition , ( ) represent the position and velocity of the -th sensor node in the d-th dimension at p n is the optimal position of the i-th sensor node until iteration n, and p until is the optimal ( ) i,d d ( n )iteration iteration n, respectively. , and , ( ) is the optimal position of the -th sensor node g, position obtained from the experiences of all sensor nodes until iteration n, which is called the global ( ) is the optimal position obtained from the experiences of all sensor nodes until iteration , , optimal position. The constants c and c are acceleration weight constants, and r and r are random 2 2 1 1 which is called the global optimal position. The constants and are acceleration weight numbers 0and and 1. are In this study,numbers the effective communication coverage area the was effective used as constants,between and random between 0 and 1. In this study, an objection function. The pseudocode theobjection VFPSO used to find optimal positions for the communication coverage area was used for as an function. Thethe pseudocode for the VFPSO sensor nodes is illustrated in Figure 8 and its flow chart is shown in Figure 9. The loop finishes used to find the optimal positions for the sensor nodes is illustrated in Figure 8 and its flow chart is when coverage sufficiently converges and its standard deviation (s.d.) over shownthe in communication Figure 9. The loop finishesrate when the communication coverage rate sufficiently converges 10 less10than 0.5%. Figure 10 illustrates showing andconsecutive its standarditerations deviationbecomes (s.d.) over consecutive iterations becomes an lessexample than 0.5%. Figurethe 10 convergence progress of the sensor nodes. Through iteration of the VFPSO simulation, the sensor illustrates an example showing the convergence progress of the sensor nodes. Through iteration of nodes randomly distributed in the targeted area at the initial stage toward theat optimal positions the VFPSO simulation, the sensor nodes randomly distributed inmoved the targeted area the initial stage where network connectivity was well maintained. moved toward the optimal positions where network connectivity was well maintained.

Figure 8. Pseudocode of the virtual force-particle swarm optimization (VFPSO) algorithm for the Figure 8. Pseudocode of the virtual force-particle swarm optimization (VFPSO) algorithm for the search of the optimal deployment positions for the sensor nodes. search of the optimal deployment positions for the sensor nodes.

Sensors 2017, 17, 2389 Sensors 2017, 2017, 17, 17, 2389 2389 Sensors

10 of 14 10 of of 14 14 10

Figure 9. 9. Flow Flow chart chart of of the the VFPSO algorithm for for the the search search of positions for for optimal deployment Figure VFPSO algorithm of the the optimal optimal deployment positions the sensor nodes. the sensor nodes. nodes. the sensor

(a) (a)

(b) (b)

Figure 10. 10. (a) (a) Example Example showing showing the the convergence convergence progress progress of of the the sensor sensor nodes. nodes. (b) (b) Sensor Sensor positions positions Figure Figure 10. (a) Example showing the convergence progress of the sensor nodes. (b) Sensor positions after after optimal optimal deployment deployment (blue (blue dots). dots). Green Green circles circles indicate indicate the the communication communication radii radii of of the the sensor sensor after optimal deployment (blue dots). Green circles indicate the communication radii of the sensor nodes. nodes. nodes.

3. Simulation Simulation Results Results 3. 3. Simulation Results The optimal optimal deployment deploymentperformance performanceofofthe the sensor nodes based on the communication PS The sensor nodes based on the the communication PS was was The optimal deployment performance of the sensor nodes based on communication PS was simulated for different two different seasons, winter (February) and summer (August), the targeted simulated for two two seasons, winter (February) and summer summer (August), in the theintargeted targeted area simulated for different seasons, winter (February) and (August), in area area located off the eastern coast of Korea. Optimal deployment in our simulation means the sensor located off off the the eastern eastern coast coast of of Korea. Korea. Optimal Optimal deployment deployment in in our our simulation simulation means means the the sensor sensor located deployment having a amaximized communication coverage while maintaining connectivity of the sensor deployment having maximized communication coverage while maintaining connectivity of the the deployment having a maximized communication coverage while maintaining connectivity of nodes. Here, the communication coverage rate is defined as the ratio of the area of communication sensor nodes. nodes. Here, Here, the the communication communication coverage coverage rate rate is is defined defined as as the the ratio ratio of of the the area area of of sensor coverage by the coverage sensor nodes to that of the targeted area. communication by the sensor nodes to that of the targeted area. communication coverage by the sensor nodes to that of the targeted area. Figure 11 shows examples of of the the simulation simulation results results of of optimal optimal deployment deployment for for the the two two different different Figure 11 shows shows examples examples Figure 11 of the simulation results of optimal deployment for the two different seasons. The Theparameters parametersused usedininthe the simulation given in Table 1. The parameters belong in seasons. simulation areare given in Table Table 1. The The parameters belong in four four seasons. The parameters used in the simulation are given in 1. parameters belong in sub-categories: environmental environmental parameters; parameters; channel channel modelling modelling parameters; parameters; communication communication sub-categories:

Sensors 2017, 17, 2389

11 of 14

Sensors 2017, 17, 2389

11 of 14

four sub-categories: environmental parameters; channel modelling parameters; communication parameters, and; and; optimal optimal deployment deployment parameters. parameters, parameters. For For the the simulation, simulation, 100 100 sensor sensor nodes nodes were were used. used. As an initial step, 100 sensor nodes were randomly distributed in the targeted area, as shown in As an initial step, 100 sensor nodes were randomly distributed in the targeted area, as shown in Figure 11a. The VFPSO search algorithm was then applied. The deployment results for February and Figure 11a. The VFPSO search algorithm was then applied. The deployment results for February and August are are displayed displayed on August on their their communication communication PS, PS, which which are are illustrated illustrated in in Figure Figure 11b,c, 11b,c,respectively. respectively. The communication coverage rate for February was estimated to be 85.2%, which is higher than The communication coverage rate for February was estimated to be 85.2%, which is higher than thatthat for for August (estimated to be 80.6%). This result is reasonable because the ocean condition of the August (estimated to be 80.6%). This result is reasonable because the ocean condition of the targeted targeted area in February a better underwater communication environment that in area in February provides provides a better underwater communication environment than thatthan in August, August, as mentioned in Section 2.3. as mentioned in Section 2.3.

Figure 11. Examples sensor nodes. nodes. (a) Randomly Examples of the optimal deployment simulation using 100 sensor deployed initial initialpositions. positions. (b,c) areoptimal the optimal deployment in and February August, deployed (b,c) are the deployment results inresults February August,and respectively. respectively. Table 1. The parameters used in the optimal deployment simulation. Table 1. The parameters used in the optimal deployment simulation. Environmental Parameters Environmental Month Parameters Longitude direction distance Latitude direction distance Month Wind speed Azimuth angle interval Longitude direction Grid points distance Communication Parameters Latitude direction distance Symbol number Wind speed Symbol rate Pulse shaping Azimuth Equalizer angle interval BER criterion Grid points

Value

Channel Modeling Parameters

2, 8 Value 22 km 22 km 2,108 m/s 45◦ 100 22 km

Frequency Channel Modeling Parameters Source level Source depth Frequency Receiver depth (Three vertical receiver array) Element spacing Source level

Value 22 km 3500 10 m/ssps 1000 Root Raised Cosine filter 45° Adaptive DFE(RLS) 2% 100

Optimal Deployment Parameters Source depth Loop number Sensor node Receiver depthnumber Weight value of attractive force (Three vertical receiver array) Weight value of repulsive force Acceleration weight Element spacing

Value 10 kHz Value 140 dB 2 m above the bottom 10 kHz 0.5~3.5 m above the bottom 1.5 m140 (10 dB λ) Value 2 m above the bottom 50 100 0.5~3.5 m above the 0.01 bottom 0.5 1 1.5 m (10 )

Communication Another important Parameters

Valueto be examined Optimal Deployment Parameters Value nodes, which point is the connectivity among the sensor is crucial for effective data communication of the UWASN system. The connectivity among the Symbol number 3500 Loop number 50 sensor nodes was estimated for the final deployment scheme to verify the performance of the VFPSO algorithm. andnode its number neighboring sensor nodes Symbol rateThe distance between 1000 spseach sensor node Sensor 100was compared to the communication threshold defined in Section 2.4, and the results for the two seasons are shown Root Raised Cosine shaping Weightindicate value of attractive force two-way 0.01 communication inPulse Figure 12. The solid lines between sensor nodes that effective filter is possible. For both cases, 100% connectivity was achieved. Consequentially, the sensor nodes were Equalizer Adaptive DFE(RLS) Weight value of repulsive force 0.5 positioned within the high communication performance area to achieve the maximal communication BER criterion 2%varying between 1.5 Acceleration weight coverage rate, with intervals and 2.0 km for both seasons. 1

Another important point to be examined is the connectivity among the sensor nodes, which is crucial for effective data communication of the UWASN system. The connectivity among the sensor nodes was estimated for the final deployment scheme to verify the performance of the VFPSO algorithm. The distance between each sensor node and its neighboring sensor nodes was compared to the communication threshold defined in Section 2.4, and the results for the two seasons are shown

Sensors 2017, 17, 2389

12 of 14

in Figure 12. The solid lines between sensor nodes indicate that effective two-way communication is possible. For both cases, 100% connectivity was achieved. Consequentially, the sensor nodes were positioned within Sensors 2017, 17, 2389 the high communication performance area to achieve the maximal communication 12 of 14 coverage rate, with intervals varying between 1.5 and 2.0 km for both seasons.

Figure nodes after optimal deployment in (a) February andand (b) Figure 12. 12. Network Networkconnections connectionsofofthe thesensor sensor nodes after optimal deployment in (a) February August. (b) August.

4. 4. Summary Summary and and Conclusions Conclusions A new method A new method for for the the optimal optimal deployment deployment of of underwater underwater sensor sensor nodes, nodes, which which can can be be applied applied to proposed in this paper. The algorithm includes four sub-algorithms, which to the theUWASN, UWASN,has hasbeen been proposed in this paper. The algorithm includes four sub-algorithms, are a routine for underwater acoustic channel modeling in temporally and spatially varying ocean which are a routine for underwater acoustic channel modeling in temporally and spatially varying environments, a routine for estimating the the communication performance ocean environments, a routine for estimating communication performanceusing usingthe the simulated simulated underwater acoustic channels, a routine for constructing the communication PS, and underwater acoustic channels, a routine for constructing the communication PS, and aa routine routine for for searching searching for for the the optimal optimal deployment deployment of of the the sensor sensor nodes nodes on on the the PS. PS. Most Most previous previous studies studies for for the the optimal optimal deployment deployment of of sensor sensor nodes nodes have have been been performed performed under under the the assumption assumption that that the the sensing sensing range of every sensor node is the same without consideration of the spatial and temporal variations range of every sensor node is the same without consideration of the spatial and temporal variations of of the ocean environment. In contrast, the proposed method is an algorithm finding the optimal the ocean environment. In contrast, the proposed method is an algorithm finding the optimal positions positions of sensor nodes using the communication radiifrom extracted from the communication of sensor nodes using the communication radii extracted the communication PS, which is PS, the which is the main difference from the previous algorithms. Moreover, it uses the VFPSO main difference from the previous algorithms. Moreover, it uses the VFPSO algorithm, algorithm, which is a which a hybrid optimization method combining the advantages thethe VFA andTherefore, the PSO. Therefore, hybridisoptimization method combining the advantages of the VFAofand PSO. the sensor the sensor nodes are not placed in equidistant intervals. The optimal deployment based has nodes are not placed in equidistant intervals. The optimal deployment based on theon PSthe hasPSbeen been simulated for two different seasons for an area located off the eastern coast of Korea as the simulated for two different seasons for an area located off the eastern coast of Korea as the selected selected targeted result, the communication rate in winter (February)better was targeted area. As aarea. result,As thea communication coverage rate coverage in winter (February) was somewhat somewhat better than that in summer (August), implying that the optimal deployment scheme may than that in summer (August), implying that the optimal deployment scheme may vary with variations vary with variations of the ocean environment. The algorithm does not propose the optimal number of the ocean environment. The algorithm does not propose the optimal number of sensor node, but it of sensor node, but it finds the optimal positions of sensor nodes for a given number of sensor nodes. finds the optimal positions of sensor nodes for a given number of sensor nodes. In this paper, it was assumed that the communication channel was time invariant. If the system In this paper, it was assumed that the communication channel was time invariant. If the system is used for a short time period using portable sensors such as sonobuoy-type devices, the ocean is used for a short time period using portable sensors such as sonobuoy-type devices, the ocean parameters corresponding to that time should be used as input parameters. However, if it is for the parameters corresponding to that time should be used as input parameters. However, if it is for the long-term ocean surveillance monitoring for a certain area, it is highly recommended to use the ocean long-term ocean surveillance monitoring for a certain area, it is highly recommended to use the ocean environmental parameters producing the worst communication radii to obtain the robust environmental parameters producing the worst communication radii to obtain the robust deployment deployment scheme and, in this case, the sensor nodes would be closer to each other. Also, it was scheme and, in this case, the sensor nodes would be closer to each other. Also, it was assumed that assumed that the BPSK signal with a center frequency of 10 kHz was transmitted as a communication the BPSK signal with a center frequency of 10 kHz was transmitted as a communication sequence sequence and each sensor node was composed of a three-receiver array with an element spacing of and each sensor node was composed of a three-receiver array with an element spacing of 1.5 m. 1.5 m. Other communication sequences [such as OFDM (Orthogonal Frequency Division Other communication sequences [such as OFDM (Orthogonal Frequency Division Multiplexing)] or a Multiplexing)] or a different type of receiver system, which might be more robust to (frequency/timedifferent type of receiver system, which might be more robust to (frequency/time-selective)] fading selective)] fading channels, could provide better communication performance and produce a channels, could provide better communication performance and produce a different deployment scheme; however, these are beyond the scope of this paper. Nevertheless, our algorithm can be applied to other types of communication sequences and receivers.

Sensors 2017, 17, 2389

13 of 14

In this paper, it was assumed that all sensor nodes were positioned near the seafloor. The depth change of the sensor nodes may yield a better communication performance, and in addition the combinations of different types of sensors, including AUV (Autonomous Underwater Vehicle), may improve the performance. The algorithm suggested herein can be extended to these cases, which will be our future work. Acknowledgments: This work was supported by the Agency for Defense Development, Korea (UD160006DD), the National Research Foundation of Korea (NRF-2016R1D1A1B03930983) and the Ministry of Oceans and Fisheries (the project titled : Development of Distributed Underwater Monitoring & Control Networks). The authors would like to thank Joung-Soo Park of Agency for Defense Development, Korea for helpful discussions concerning algorithm for optimal deployment of underwater sensor nodes. Author Contributions: Jee Woong Choi conceived and designed the research; Sunhyo Kim applied the algorithm to UWASN; Both authors performed research and wrote the paper. Conflicts of Interest: The authors declare no conflict of interest.

References 1. 2. 3. 4. 5.

6. 7. 8. 9. 10. 11. 12. 13.

14. 15. 16. 17. 18.

Akyildiz, I.F.; Pompili, D.; Melodia, T. Underwater acoustic sensor networks: Research challenges. Ad Hoc Netw. 2005, 3, 257–279. [CrossRef] Lloret, J. Underwater sensor nodes and networks. Sensors 2013, 13, 11782–11796. [CrossRef] [PubMed] Akkaya, K.; Newell, A. Self-deployment of sensors for maximized coverage in underwater acoustic sensor networks. Comput. Commun. 2009, 32, 1233–1244. Garcia-Pineda, M.; Sendra, S.; Atenas, M.; Lloret, J. Underwater wireless ad-hoc networks: A survey. In Mobile Ad Hoc Networks: Current Status and Future Trends; CRC Press: Boca Raton, FL, USA, 2011; pp. 379–411. Senel, F.; Akkaya, K.; Yilmaz, T. Autonomous deployment of sensors for maximized coverage and guaranteed connectivity in underwater acoustic sensor networks. In Proceedings of the 38th IEEE Conference on Local Computer Networks (LCN), Sydney, Australia, 21–24 October 2013; pp. 211–218. Jiang, P.; Liu, J.; Ruan, B.; Jiang, L.; Wu, F. A new node deployment and location dispatch algorithm for underwater sensor networks. Sensors 2016, 16, 82. [CrossRef] [PubMed] Du, H.; Xia, N.; Zheng, R. Particle swarm inspired underwater sensor self-deployment. Sensors 2014, 14, 15262–15281. [CrossRef] [PubMed] Luo, H.; Guo, Z.; Dong, W.; Hong, F.; Zhao, Y. LDB: Localization with directional beacons for sparse 3D underwater acoustic sensor networks. J. Netw. 2010, 5, 28–38. [CrossRef] Pompili, D.; Melodia, T.; Akyildiz, I.F. Three-dimensional and two-dimensional deployment analysis for underwater acoustic sensor networks. Ad Hoc Netw. 2009, 7, 778–790. [CrossRef] Son, S.U.; Kim, H.; Joo, J.; Choi, J.W. Multipath Effects on High-Frequency Coherent Acoustic Communications in Shallow Water. Jpn. J. Appl. Phys. 2013, 52, 07HG03. [CrossRef] Emokpae, L.; Younis, M. Surface Based Underwater Communications. In Proceedings of the IEEE Global Telecommunications Conference (GLOBECOM 2010), Miami, FL, USA, 6–10 December 2010. Walree, P.A. Propagation and scattering effects in underwater acoustic communication channels. IEEE J. Ocean. Eng. 2013, 38, 614–631. [CrossRef] Rouseff, D.; Badiey, M.; Song, A. Effect of reflected and refracted signals on coherent underwater acoustic communication: Results from the Kauai experiment (KauaiEx 2003). J. Acoust. Soc. Am. 2009, 126, 1359–2366. [CrossRef] [PubMed] Yang, T.C. Properties of underwater acoustic communication channels in shallow water. J. Acoust. Soc. Am. 2012, 131, 129–145. [CrossRef] [PubMed] Hung, H.S.; Chang, S.H.; Chen, S.L.; Chang, C.W. Real time implementation of frost beamforming for underwater communications. J. Mar. Sci. Eng. 1999, 7, 1–7. Song, H.C.; Hodgkiss, W.S.; Kuperman, W.A. High-rate synthetic aperture communications in shallow water. J. Acoust. Soc. Am. 2009, 126, 3057–3061. [CrossRef] [PubMed] Kilfoyle, D.B.; Baggeroer, A.B. The State of the Art in Underwater Acoustic Telemetry. IEEE J. Ocean. Eng. 2000, 25, 4–27. [CrossRef] McDowell, P. Environmental and Statistical Performance Mapping Model for Underwater Acoustic Detection Systems. Ph.D. Thesis, University of New Orleans, New Orleans, LA, USA, 2010.

Sensors 2017, 17, 2389

19. 20.

21.

22. 23. 24. 25. 26. 27. 28. 29. 30. 31.

14 of 14

Chen, J.; Li, S.; Sun, Y. Novel Deployment Schemes for Mobile Sensor Networks. Sensors 2007, 7, 2907–2919. [CrossRef] [PubMed] Zou, Y.; Chakrabarty, K. Sensor deployment and target localization based on virtual forces. In Proceedings of the Twenty-Second Annual Joint Conference of the IEEE Computer and Communications (INFOCOM 2003), San Francisco, CA, USA, 30 March–3 April 2003. Majid, A.S.; Joelianto, E. Optimal Sensor Deployment in Non-Convex Region using Discrete Particle Swarm Optimization Algorithm. In Proceedings of the IEEE Conference on Control, Systems & Industrial Informatics, Bandung, Indonesia, 23–26 September 2012; pp. 23–26. Wang, X.; Ma, J.; Wang, S.; Bi, D.W. Distributed particle swarm optimization and simulated annealing for energy-efficient coverage in wireless sensor networks. Sensors 2007, 7, 628–648. [CrossRef] Wang, X.; Wang, S.; Ma, J.J. An improved co-evolutionary particle swarm optimization for wireless sensor networks with dynamic deployment. Sensors 2007, 7, 354–370. [CrossRef] Amante, C.; Earkins, B.W. ETOPO1 1 Arc-Minute Global Relief Model: Procedures, Data Sources and Analysis; NOAA Technical Memorandum; National Geophysical Data Center: Boulder, CO, USA, 2009. Ryan, W.B.F.; Carbotte, S.M.; Coplan, J.O. Global Multi-Resolution Topography Synthesis. Geochem. Geophys. Geosyst. 2009, 10, 1–9. [CrossRef] Carnes, M.R. Description and Evaluation of GDEM-V 3.0; Memorandum Report; Naval Research Laboratory: Hancock County, MS, USA, 2009. Jackson, D.R.; Richardson, M.D. High-Frequency Seafloor Acoustics; Springer: New York, NY, USA, 2006; pp. 313–314. Porter, M.B.; Bucker, H.P. Gaussian beam tracing for computing ocean acoustic fields. J. Acoust. Soc. Am. 1987, 82, 1349–1359. [CrossRef] Proakis, J.G.; Salehi, M. Digital Communications; McGraw-Hill: New York, NY, USA, 2008; pp. 710–721. Stojanovic, M.; Catipovic, J.A.; Proakis, J.G. Phase-Coherent Digital Communications for Underwater Acoustic Channels. IEEE J. Ocean. Eng. 1993, 19, 100–111. [CrossRef] Abdi, A.; Guo, H. A New Compact Multichannel Receiver for Underwater Wireless Communication Networks. IEEE Trans. Commun. 2009, 8, 3326–3329. [CrossRef] © 2017 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/).