machine learning based antenna design for physical layer...

Research ArticleMachine Learning Based Antenna Design for Physical LayerSecurity in Ambient Backscatter Communications

Tao Hong 1 Cong Liu 1 and Michel Kadoch 2

1School of Electronics and Information Engineering Beihang University China2Department of Electrical Engineering Ecole de Technologie Superieure University of Quebec Canada

Correspondence should be addressed to Tao Hong hongtaobuaaeducn

Received 4 October 2018 Accepted 4 December 2018 Published 1 January 2019

Guest Editor Feng Ye

Copyright copy 2019 TaoHong et alThis is an open access article distributed under theCreative CommonsAttribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Ambient backscatter employs existing radio frequency (RF) signals in the environment to support sustainable and independentcommunications thereby providing a new set of applications that promote the Internet of Things (IoT) However nondirectionalforms of communication are prone to information leakage In order to ensure the security of the IoT communication system inthis paper we propose a machine learning based antenna design schemewhich achieves directional communication from the relaytag to the receiving reader by combining patch antenna with log-periodic dual-dipole antenna (LPDA) A multiobjective geneticalgorithm optimizes the antenna side lobe gain standing wave ratio and return loss with a goal of limiting the number of largeside lobes and reduce the side lobe level (SLL) The simulation results demonstrate that our proposed antenna design is well suitedfor practical applications in physical layer security communication where signal-to-noise ratio of the wiretap channel is reducedcommunication quality of the main channel is ensured and information leakage is prevented

1 Introduction

The Internet of Things (IoT) is a vital component of the fifthgeneration (5G) mobile communications interconnecting alarge number of devices However in traditional backscattercommunication systems radio frequency (RF) power isprovided by the reader and the limited power supply limitsthe widespread use of IoT

In 2013 the proposed ambient backscatter communi-cation technology solved some of the above shortcomings[1] Unlike traditional backscatter communication (eg forpassive sensors and RF identification (RFID) tags) ambientbackscatter does not require specific devices to provideenergy but instead utilizes RF signals in the environmentas both energy resources and signal resources for reflection[2] As a result ambient backscatter provides sustainableand independent communications and the maintenance andimplementation costs of the system can be greatly reduced[3 4] Because the ambient configuration does not requireadditional spectrum resources to operate we chose the 4G5G and Wireless-Fidelity (Wi-Fi) signals with frequencies inthe range of 2GHz - 4GHz as the ambient resources

However several challenges remain The broadcast char-acteristics of wireless signals make it easy for some illegaleavesdroppers to obtain information content and signalsof the same frequency are superimposed at the receiver tocause interference which brings many difficulties to signaldetection [5]

Traditional security techniques which encrypt informa-tion with high computationally complex codec algorithms[6] have gradually failed with the rapid increase in thecomputational power The fundamental principle behindphysical layer security is to exploit the inherent randomnessof noise and communication channels to limit the amountof information that can be extracted at the ldquobitrdquo level by anunauthorized receiver [7] Therefore information-theoreticsecurity is considered to be a key technology to ensure thesecurity of wireless communications

A lot of research has been done on physical layersecurity A cooperative relay scheme was investigated in [8]however it is only applicable to multiantenna and multirelaysystems Artificial noise- (AN-) based methods are alsoinappropriate because of their higher energy expenditure andincreased cochannel interference with any adjacent user [9]

HindawiWireless Communications and Mobile ComputingVolume 2019 Article ID 4870656 10 pageshttpsdoiorg10115520194870656

2 Wireless Communications and Mobile Computing

Ambientsource Legacy

User

Tag

Reader

Backscatter signalsAmbient signals

AntennaEnergy

HarvesterVariableImpedance

Micro-

ℎst

ℎsr

ℎtr

controller

Figure 1 Ambient backscatter system

Comparisons have shown that the method most suited forIoT sensing applications is beamforming which allows thesignal to propagate in a specified direction [10] Throughbeamforming technology the signal-to-noise ratio (SNR)is increased at the legitimate reader whereas it is reducedat the illegal eavesdropper thereby enhancing the securityperformance of the system

In order to achieve similar beamforming functions ina sensor network we focused our research on the antennadesign of the relay tag Existing relay antennas have problemssuch as very large size poor directivity and small transmis-sion gain [11] We designed a dual-antenna system consistingof a patch antenna array and a printed log-periodic dual-dipole antenna (LPDA) with the advantages of orientationand high gain At the same time it has a fairly wide operatingfrequency band which can effectively utilize various types ofsignals in the environment Moreover the small side lobesreduce the SNR received by eavesdroppers from other direc-tions ensuring the security of communication In practicalapplications the patch antenna array receives RF signals fromall directions and then transmits them through the LPDA ina specified direction

Traditional antenna designs are mostly based on expe-rience and simulations to continuously modify the relevantparameters which is time consuming and arduous As amachine learning algorithm genetic algorithms have beenwidely used in antenna design to search for large-scalenonintuitive solution space and find the optimal parametervalue In [12] a genetic algorithm was used to optimize thestructure and length of a wire antenna In [13] an improvedhierarchical Bayesian optimization algorithm was applied tothe optimization of the antenna array feed network Howeverthe single-objective genetic algorithms used in the above-mentioned research may not be suitable for real situations

because it over-emphasizes the importance of one metric Tosolve this problem we use a multiobjective genetic algorithmto optimize the antenna by using gain side lobe returnloss and voltage standing wave ratio (VSWR) as objectivefunctions

The remainder of this paper is organized as followsSection 2 introduces the ambient backscatter communicationmodel and the use of the directional antenna to achievephysical layer security Section 3 illustrates the structureof the proposed antenna and the optimization process ofthe multiobjective genetic algorithm Section 4 presents thesimulation and optimization results of the antenna structurefollowed by Section 5 which concludes the paper

2 System Model

Ambient backscatter has become a promising option for self-sustainable communication systems because of its energy-saving features and has good potential for widespread usein the IoT A typical ambient backscatter system includes anambient source a passive tag and a reader as illustrated inFigure 1

The communication process within the tag is as followsWhen an ambient source broadcasts signals to its legacyusers such as mobile phones and laptops the tag can harvestthe RF energy from the signals and use the collected energyto power the entire system Then the micro-controller in thesystem tunes the variable impedance based on the signalto indicate bit ldquo1rdquo or ldquo0rdquo by backscattering or absorbingthe ambient signals [14 15] Finally the reader decodes thebackscattered signals and recovers the two information bitscompleting the tag-to-reader communication

Evaluating the security of the system and selecting anappropriate solution improves the security of the system A

Wireless Communications and Mobile Computing 3

directional antenna designed for the tag is an effective meansto achieve beamforming in the IoT scenario

21 Ambient Backscatter System and Signal Detection Inambient backscatter systems the detection of the receivedsignal plays a vital role Without loss of generality we denoteℎ119904119905 ℎ119904119903 and ℎ119905119903 as gains of the channels from the source tothe tag from the source to the reader and from the tag to thereader respectively We assume that 119904(119899) represents the RFsource signals with zero mean and unit variance The powerof the ambient source is 119875119904 and is unknown to the receiverThe received signal at the tag is expressed as

119910119905 (119899) = radic119875119904ℎ119904119905119904 (119899) + 119908119905 (119899) (1)

where 119908119905(119899) is the noise inside the tag which can beignored because here and the tag is a passive component ie119908119905(119899) = 0 [16]

The signal backscattered by the tag is

119909119905 (119899) = 120578119909 (119899) 119910119905 (119899) (2)

where 119909(119899) isin 0 1 controls the working condition of thetag antenna The tag reflects the signal when 119909(119899) = 1 andthe tag does not reflect when 119909(119899) = 0 120578 isin [0 1] is theattenuation factor inside the tag The signal received at thereader is

119910119903 (119899) = ℎ119904119903119904 (119899) + ℎ119905119903119909119905 (119899) + 119908 (119899)

= radic119875119904ℎ0119904 (119899) + 119908 (119899) 119909 (119899) = 0radic119875119904ℎ1119904 (119899) + 119908 (119899) 119909 (119899) = 1

(3)

where ℎ0 ≜ ℎ119904119905 ℎ1 ≜ ℎ119904119903 + 120578ℎ119904119905ℎ119905119903 and 119908(119899) is theadditive white Gaussian noise (AWGN) with zero mean and1205902119908 variance

In ambient backscatter systems the amplitude or phaseof the backscattered signals always carries the requiredinformation

According to amplitude or phase modulation the maintask of the backscatter reader is to determine the ampli-tude or phase variation In most cases demodulation frombackscattered waves with the binary amplitude modulationrequires envelope detection at the receiver Alternativelyphase demodulation is based on phase detection Commonmethods of phase demodulation include the use of a homo-dyne receiver with an RF in-phasequadrature demodulationand channel estimation [17] Subsequently the demodulatorcan acquire the information bit modulated on the phase byutilizing the channel estimation value

The conventional detection scheme of the reflected sig-nals uses preamble packets as thresholds for detection Inrecent years many other detection schemes have been devel-oped For example a detector based on differential encod-ing can finish the detection without the knowledge of thechannel state information (CSI) [18] A joint-energy detectionscheme requires only the channel variances without requiringthe specific CSI and recently a maximum-likelihood (ML)detector has been commonly used

ℎtr

ℎte

Eve

Ambientsource

Bob

Tag

Figure 2 Wiretap channel model

22 Security Assessment Since the hardware of the tag limitsthe signal processing capability the security of communica-tion becomes a challenge for the IoT Traditional encryptiontechnology that relies on high computational complexitydoes not meet the requirements of the IoT applicationscenario Beamforming AN cooperative interference differ-ential channel estimation and network coding have becomecommon physical layer security solutions Among thembeamforming technology ismost suitable for IoT security andallowswireless signals to propagate only in specific directions

As shown in Figure 2 the data in the tag are modulatedinto the ambient carrier and the information signal is receivedby the legal receiver Bob over the ldquomain channelrdquo whereasit is received by the eavesdropper Eve over an additionalldquowiretap channelrdquo

The secrecy capacity is used to measure the security ofthe system In the wiretap channel the secrecy rate is atransmission rate that can be reliably transmitted on themainchannel but cannot be transmitted on the eavesdroppingchannel In the case of one eavesdropper the secrecy capacityis

119877119904 = max 119877119889 minus 119877119890 (4)

where 119877119889 is the communication rate of the source-destination link and 119877119890 is the communication rate of thesource-eavesdropper link Usually it is calculated as thedifference between the mutual information in the primaryand eavesdropping channels 119868(119860 119861) minus 119868(119860 119864)

In the case ofmultiple eavesdroppers the secrecy capacityis

119877119904 = maxmin119895119877119889 minus 119877119895119890 (5)

The secrecy outage probability is another importantvariable in physical layer security communication It is thelikelihood that the instantaneous secrecy rate 119877119904 is below apredefined threshold 120576 for a particular fading distribution

119875119900119906119905 = 119875 119877119904 lt 120576 120576 gt 0 (6)


Eve

BobBob

Ambientsource

Tag1 Tag2

Figure 3 Influence of antenna side lobes on communication

An analysis from the perspective of information theoryindicates that the mutual information depends on the SNR ofthe received signal which indicates that the secrecy capacityis determined by the SNR of the legal receiver and of theeavesdropper [19]

By using beamforming at the tag we change the directionof the antenna to increase the gain of the main channel andreduce the signal strength of the wiretap channel by reducingthe side lobes In this way the security of the system isenhanced and the secrecy capacity is improved as well

23 Antenna Demand In the IoT application scenario inorder to achieve beamforming and meet the constraints oflimited hardware we designed a directional antenna withhigh gain and small side lobes to be used in the tag

The previous discussion shows that providing differentSNRs for eavesdroppers and readers is a key task to improvecommunication security The directional antenna increasesthe peak gain of the antenna thereby improving spatial reuseand expanding the geographic coverage in a given direction[20] Moreover the use of directional antennas improves thewireless network capacity avoids physical jamming attemptsenhances data availability and suppresses interference fromneighbors In addition the antenna is required to have fewerside lobes

We can observe in Figure 3 that Tag1 has more side lobesthan Tag2 and higher side lobe levels When the side lobesare small the main lobe has a large transmit power whichmaximizes the signal power in the desired direction whilesuppressing signals in undesired directions Thus a goal canbe achieved tomaximize the ratio of the SNRs received by thereader and the eavesdropper

In addition higher frequency signals experience severalorders of magnitude of free space path loss and thereforecommunication coverage is small By using a small sidelobeantenna we can increase the transmission distance in thespecified direction

In order to achieve these goals we designed a dual-antenna system consisting of a patch antenna array and aprinted LPDA The patch antenna receives the RF signalfrom all directions and the LPDA is directed at the reader

Substrate1 Substrate2Ground plane layer

143

66m

m 4 Patchs62mm LPDA

Top layer500mm

Power combiner

LPDA

Bottom layer

Figure 4 Dual-antenna system structure

In addition we use a multiobjective genetic algorithm tooptimize the antenna side lobes reduce the peak value ofthe largest side lobes and enhance the directionality of theantenna

The designed antenna does not require additional powerwhich prevents the disadvantages of traditional physical layersecurity technology The specific structure of the antenna isdescribed in detail in Section 3

3 Antenna Design and Optimization

31 Antenna Model The antenna system structure on thetag is shown in Figure 4 and is located on the XOY planeThe system consists of three components a four-elementpatch antenna array a feeding network and a printed LPDATaking into account the receiving range and antenna gainthe receiving plane uses a simple coaxial probe-fed patchantenna It is located on the top layer The feed networkconsisting of the power combiner and the correspondingsubstrate is located on the bottom layer and shares the groundplane with the patch antenna array The patch antennascoaxial probe is connected to the four input ports of the powercombiner through two layers of substrate and a ground planeTheprinted LPDAacts as a transmit antenna and is connectedto the output port of the combiner The total size of this dual-antenna system is 500 times 14366 times 81751198981198983 It is evidentfrom the reciprocity of the antenna that when the incidentwave is irradiated from the +z direction to the four-elementpatch antenna array the received electromagnetic wave istransmitted to the printed LPDA through the feed networkfor reradiation with polarization transition characteristicsthereby changing the incident wave transmission directionand achieving the function of omnidirectional reception anddirectional transmission

32 Calculation of the Antenna Initial Size

321 Receiving Antenna Because high gain is a priority arectangular patch antenna fed by a coaxial probe is used as


the receiving antenna for the dual-antenna system A RogersTMM4 with a dielectric constant 120576119903 = 45 is chosen asthe substrate with a thickness of 5mm According to theempirical formula provided in [21] the initial length 119882 andwidth 119871 of the patch are calculated as

119871 = 11988821198910

1radic1120576119890 minus 2Δ119871 (7)

119882 = 11988821198910 (

120576119903 + 12 )12 (8)

where 119888 is the speed of light 1198910 is the resonant frequencyℎ is the thickness of the substrate 120576119903 is the dielectric constantof the substrate and 120576119890 is the effective dielectric constant 120576119890and Δ119871 are calculated using the following formula

120576119890 = 120576119903 + 12 + 120576119903 minus 12 (1 + 12ℎ119871)

12

(9)

Δ119871 = 0412ℎ(120576119890 + 03) (119882ℎ + 0264)(120576119890 minus 0258) (119882ℎ + 08) (10)

The position of the feed point can be calculated by

119883119891 = 1198712radic120576119890 (11)

Considering the mutual coupling effect between thepatches the interval between the adjacent units is 05120582119892 sim 120582119892

The four-element rectangular patch antenna array is fedin parallel by a 1-4 power combiner The distances from theinput port to every unit are equal to achieve the same phasefeed

322 Transmitting Antenna An LPDA is a wideband an-tenna In order to make the tag structure more compacta flat printed structure is used to integrate the transceiverantennas into one plane The structure of the printed LPDAis shown in Figure 5 The length of the antenna element isdenoted by 119871119899 and the extension of the end of each antennaelement intersects at a point called a virtual vertex with anopening angle of 120572 The vertical distance from the virtualvertex to each antenna element is represented as 119877119899 thevibrator width is represented as 119908119899 and the adjacent twovibrators are separated by 119889119899

The geometry of the antenna is determined by thegeometric factor 120591 and the spacing factor 120590

120591 = ℎ119899+1ℎ119899 = 119871119899+1

119871119899 = 119889119899+1119889119899 (12)

120590 = 1198891198994ℎ119899 (13)

The number of antenna elements is obtained by thefollowing formula

119873119886 = 1 + 119897119892 (11987021198701)119897119892120591 (14)

where1198701 and 1198702 are the cutoff coefficients

Ln

ℎn

dn

wn

Figure 5 The structure of the printed LPDA

1198701 = 101 minus 0519120591 (15)

1198702 = 7011205913 minus 2131205912 + 2198120591 minus 730+ 120590 (2182 minus 66120591 + 62121205912 minus 18291205913) (16)

In addition it is necessary to estimate the width of theelements as follows

119885119886 asymp 120 ln(ℎ119886) minus 225 (17)

where ℎ119886 is the half-height-to-radius ratio of the dipoleIn the planar printing structure we use microstrip patchesinstead of cylindrical dipoles Considering the equivalentperimeters of the cylindrical and thin rectangular conduc-tors we used the approximate relationship 119908 asymp 120587119886 where119885119886 represents the average characteristic impedance which is50Ω here 119908 represents the dipole width

The LPDA is an end-fire antenna and the maximum radi-ation direction is from the longest oscillator to the shortestoscillator [22] When the operating frequency changes theradiation area of the antenna moves around the antenna andmaintains similar characteristics therefore the pattern of theantenna changes little with the frequency In general thelarger the value of 120591 the higher the number of oscillatorsin the radiation region the stronger the directivity of theantenna and the smaller the half-power angle of the patternThe lengths of the longest oscillator and the shortest oscillatorof the LPDA determine the operating frequency

The LPDA is a linearly polarized antenna When theLPDArsquos oscillator plane is placed horizontally it radiates orreceives horizontally polarized waves when its oscillatorplane is placed vertically it radiates or receives verticallypolarized waves Circular polarization is easier to achievewith a planar structure

323 Optimization Scheme Since the directional antennahas a larger impact on the physical layer security and thestructure of the LPDA is more complex the transmittingantenna is optimized We select the lengths widths andspacings of the elements as variables for the optimization


Target2

Target1

w1

w2

w3

0

Figure 6 Pareto front

During the design of the antenna the gain bandwidth sidelobes and standing wave ratio (VSWR) of the antenna areimportant indicators of the performance of the antennaTherefore the function corresponding to the abovemen-tioned four indicators is defined as the objective function

In the actual design process of the antenna there areusually multiple targets that need to be optimized Theremay be contradictory relationships between the variousobjectives and it is impossible to achieve optimality at thesame time Therefore a multiobjective genetic algorithm(MOGA) was introduced

In MOGA there exists a set of Pareto-optimal or non-dominated solutions generating a set of Pareto-optimaloutcomesobjective vectors which is called pareto frontExplicitly the pareto front is generated by the specific set ofsolutions for which none of the multiple objectives can beimproved without sacrificing the other objectives as shownin Figure 6

A traditional multiobjective optimization scheme usesa method of assigning weights to convert multiple goalsinto a single goal However due to the nonconcaveness ofmultiobjectives in order to find the Pareto front a three-dimensional search is required for each weight 119908 = [1199081 11990821199083] which is very time-consuming Moreover as the numberof objective functions increases the complexity of theweight-ing method is greatly increased In addition it is challengingto assign weights to each decision variable Therefore a newoptimization solution is needed

In this study amultiobjective genetic algorithm (MOGA)is introduced as an optimization scheme namely the non-sorting genetic algorithm (NSGA)-II The NSGA-II is con-sidered one of the classic MOGAs The algorithm obtains apotential uniformly distributed Pareto optimal solution setby fast nondominated sorting crowded degree calculationand an elitism strategy This is very helpful for improving theexploratory capacity of the NSGA

The specific process is shown in Figure 7The NSGA-II first finds nondominated solutions in the

population and stratifies the population through nondom-inated sorting Subsequently these points are removed andidentified and the nondominated solutions in the remainingpopulation are removed The algorithm updates the current

Start

Initial populationGeneration

Calculate fitness value for every objective

function

Rank every member of population

Check stopping Criteria

Yes

Best individuals

Stop

Crossover

Mutation

Selection

Figure 7 Flowchart of the MOGA

archive by identifying the old archive and all current non-dominated solutions in the aggregateThese layers are used inturn until the maximum archive is reached The point closestto the target value is obtained by considering the crowdingdistance operator

Unlike traditional optimization methodsMOGAs do notconvert multiple targets into a single target for optimizationusing weighting but seek to optimize multiple targets simul-taneously Thereby an optimal solution set can be found thatsatisfies multiple goals

324 Decision Variables A multiobjective optimizationproblem with three decision variables and four objectivefunctions is expressed as

119874119901119905119894119898119886119897119865 (119909) = [1198651 (119909) 1198652 (119909) 1198653 (119909) 1198654 (119909)] (18)

where 119865(119909) is the vector of the objective functions1198651(119909) 1198652(119909) 1198653(119909) 1198654(119909) represent the objective functionswhere 119909 = [1198711 1198712 119871119899 1198891 1198892 119889119899 1199081 1199082 119908119899]

Here 119883 and 119885 represent the search space and the targetspace respectively Thus using the mapping 119865 119883 997888rarr 119885each vector 119909 isin 119883 corresponds to a vector 119911 = 119865(119909) isin 119885

We ensure that the height of the high-order oscillatoris greater than the length of the low-order oscillator Allare within the appropriate range According to the physicalmeaning of the variable its optimization range is given as119871 119894 isin [5 40] 119889119894 isin [1 15] 119908119894 isin [01 3] (unit mm)

33 Objective Function

331 Bandwidth The objective function is designed toincrease the antenna bandwidth so that the transmit antenna


can operate over a wider frequency range The target fre-quency band is determined by a 10119889119861 return loss 11987811 andranges from 2119866119867119911 to 4119866119867119911 Therefore the fitness functionis defined as the average of the return loss of less than minus10119889119861in the frequency band

1198651 (119909) = 1119873119873

sum119894=1

119876 (119891119894) (19)

119876(119891119894) = 10 11987811 lt minus10100381610038161003816100381611987811 (119891119894)1003816100381610038161003816 11987811 ⩾ minus10

(20)

In the above equation 119891119894 is the sampling frequency Ifthe average value of 11987811 at the sampling frequency is lessthan minus10119889119861 it is concluded that the design goal has beenachieved

In the design example we set the parameters as followsThe sampling frequency is 119873 = 5 1198911 = 2119866119867119911 1198912 =25119866119867119911 1198913 = 3119866119867119911 1198914 = 35119866119867119911 and 1198915 = 4119866119867119911 When

1198651(119909) ⩾ 10 the objective function is satisfied

332 VSWR The VSWR is an important indicator to mea-sure the antenna matching state the VSWR is limited to[1 18]

119881119878119882119877 (119891119894) = 119881119878119882119877 119881119878119882119877 ⩽ 1818 119881119878119882119877 lt 18 (21)

1198652 (119909) = 1119873119873

sum119894=1

119881119878119882119877 (119891119894) (22)

where 119873 = 5 is the number of sampling points of 2 sim4119866119867119911 When 1198652(119909) ⩽ 18 the objective function is satisfied

333 Gain The antenna gain is a measure of the ability of anantenna to transmit and receive signals in a specific directionIt is an important indicator used for antenna optimizationThe average gain in the band is used as the objectivefunction

1198653 (119909) = 1119873119873

sum119894=1

119866119886119894119899 (119891119894) (23)

334 Side Lobes Since the antenna has many side lobesthe maximum side lobes tend to have a level that is notmuch different from the maximum gain of the antenna Inthe physical layer security there is a strict requirement forthe orientation of the antenna and it is necessary to reducethe peak value of the highest side lobes as much as possibleThis is required because if the eavesdropper is located in thedirection of the largest side lobe information leakage mayoccur Therefore the optimization goal is the minimizationof the maximum peak of the side lobes

The total radiation pattern factor 119891(120579 120601) of the119872 cells ofthe LPDA shown in Figure 4 is

119891 (120579 120601) = sin 120579119872

sum119901=1

119871119901

sdot exp [119895119896 (119883119901 sin 120579 cos 120601 + 119884119901 sin 120579 sin 120601 + 119885119901 cos 120579)]

sdot119873

sum119899=1

(minus1)119899 times (2119899 minus 1) 119871119899119901 cos (120587119871119901 sdot cos 120579)(2119899 minus 1)2 minus (2119871119901 sdot cos 120579)2

(24)

The optimization goal is min1198654(119909) subject to1198654 (119909) = max

119891119894isin[21198661198671199114119866119867119911](119878119871119871 (119891119894)) (25)

335 Fuzzy Decision Making Fuzzy set theory is a methodto find the optimal compromise solution from the Paretofront Using linear fuzzy membership function modelingthe objective function value is mapped to the satisfactionfunction This defines a linear membership function 119904119891119899

119904119891119899 =

1 119894119891 119911119899 ⩾ 1199111198981198861199091198991 minus 119911119898119886119909119899 minus 119911119899

119911119898119886119909119899 minus 119911119898119894119899119899 119894119891 119911119898119894119899119899 lt 119911119899 lt 1199111198981198861199091198990 119894119891 119911119899 ⩽ 119911119898119894119899119899

(26)

where 119911119898119894119899119899 and 119911119898119886119909119899 are the minimum and maximumvalues of the 119899 minus 119905ℎ objective function respectively Thecanonical membership function of the 119899 minus 119905ℎ nondominatedsolution of the objective function is expressed as

119904119895 = sum119873119900119887119895119899=1 119904119891119895119899sum119872119901119886119903119895=1 sum119873119900119887119895119899=1 119904119891119895119899

(27)

where119873119900119887119895 represents the number of objective functionsand 119872119901119886119903 is the number of nondominated solutions inthe Pareto front We choose the solution vector with themaximum 119904119895 value as the optimal compromise solution

4 Numerical Results

The design examples and results are provided in this sectionand represent the optimal design of the LPDA based on themultiobjective genetic algorithm

The signals in the domestic environment are mainlycomposed of four types WiFi signals terrestrial digital TVbroadcast signals mobile 4G signals and upcoming 5Gsignals which has ultra-high spectrum utilization and ultra-low power consumption and will be widely used in thefuture Considering the requirements of the signal coveragein various environments transmission rate signal stabilitysecurity signal spectrum and transmission power a workingfrequency band of 2119866119867119911 sim 4119866119867119911 is used in order tomeet therequirements of IoT communications The antenna designgoals are shown in Table 1

According to the design optimization scheme 35 vari-ables are selected as the optimization variables including thelength 1198711 sim 11987112 the width 1199081 sim 11990812 and the spacing 1198891 sim11988911 of the elements The optimization ranges 119871 119894 isin [5 40]119889119894 isin [1 15] and 119908119894 isin [01 3] (unit mm) are used


Table 1 Antenna design index

Index ValueWorking frequency 2119866119867119911 sim 4119866119867119911Maximum VSWR in the band lt 2Average VSWR lt 13Minimum gain in the band gt 5119889119861Average gain gt 65119889119861Maximum side lobe level lt minus4119889119861Return loss (11987811) lt minus10119889119861Antenna size lt 30119888119898

Table 2 Simulation parameter optimization (unit mm)

1198711 1198712 1198713 1198714 1198715 1198716 119871736 324 2916 2624 2362 2126 1921198718 1198719 11987110 11987111 11987112 1198891 11988921722 1550 1395 1255 1130 13 1171198893 1198894 1198895 1198896 1198897 1198898 11988991053 948 853 768 691 622 56011988910 11988911 Gain SLL504B 453 70119889119861 minus576119889119861

Based on a large number of simulation calculationswe chose the NSGA-II algorithm to optimize the antennaparameters We set the population size to 100 the maximumnumber of iterations to 250 the crossover probability to 119901119888 =07 and the mutation probability to 119901119898 = 1119899V119886119903 where119899V119886119903 = 35 is the number of decision variables The frequencyis sampled at intervals of 500119872119867119911 in the 2119866119867119911 to 4119866119867119911band After iterative optimization the antenna parameterswere obtained and are shown in Table 2

The criteria for measuring the quality of an algorithmare time complexity and space complexity Time efficiencyrefers to the execution time of the algorithm Regarding thecomputational complexity of genetic algorithms Goldberg etal proposed the concept of takeover time to discuss the timecomplexity of the algorithm [23] In the antenna design ofthis paper the time complexity is defined as the calculationtime ie the number of iterations it takes to find an optimalsolution which is more practical

The genetic algorithm can end with convergence or endwith the number of iterations [24] After several iterationsthe results began to stabilize Therefore this paper selects 250iterations as the end condition At this point each targetmeetsthe design requirements

An excitation source is used to stimulate the receiving andtransmitting antennas at the same time to simulate the overallgains 1198661 and 1198662 A pattern of the LPDA antenna at a centerfrequency of 24119866119867119911 can be obtained as shown in Figure 7It can be seen that the planar LPDA has good directivity andcan be optimized to achieve gains of 70119889119861The reciprocity ofthe antenna indicates that a certain range of electromagneticwaves received by the patch antenna is radiated through theLPDA and used as a relay antenna on the tag

In Figure 8 a comparison of patterns before and afteroptimization is shown Prior to the optimization there are

02468

0

30

60

90

120

150

180

210

240

270

300

330

02468

dB(Gain Total)-OriginaldB(Gain Total)-Optimized Freq=24GHz

600400200000-200

Unit dB

m2

-400 m3

m4minus8

minus6

minus4

minus2

minus8

minus6

minus4

minus2

m1

Name Phi Theta Magm1 90 270 7m2 90 270 571m3 90 45 -576m4 90 90 -4

Figure 8 Gain and side lobes before and after optimization

1 2 3 4 5minus40

minus30

minus20

minus10

0

a4a3

a2

S(1

1) (d

B)

Freq (GHz)OriginalOptimized

a1

Mark Freq(GHz) S(11)(dB)a1 147 -10a2 161 -10a3 396 -10a4 413 -10

Figure 9 Return loss 11987811

1 large side lobes and the maximum side lobe level is minus4119889119861After optimization the maximum side lobes are reduced tominus576119889119861 which effectively enhances the directionality of theantenna

The curves shown in Figures 9 and 10 show the changesin the return loss 11987811 and the VSWR versus the frequencybefore and after optimization respectively It can be seen thatthe optimized 10119889119861 impedance bandwidth is 25119866119867119911 and theaverageVSWR in the band is 13 reaching the expected target

In order to prove that the designed antenna can effectivelyimprove the security of the channel we evaluate the channelsecrecy capacity

Assume that the ambient source power is 119875119905 the trans-mission gain is 119866119905 the distance from the ambient source tothe tag is 1199031 and the distance from the tag to the reader is 1199032


10 15 20 25 30 35 40 45 5010

15

20

25

30

35

40

45

50

VSW

R

Freq (GHz) Original Optimized

Figure 10 VSWR vs frequency

From the radar equation the signal power 119875119903 received by thereader can be obtained as

119875119903 = 119875119905119866119905412058711990321 sdot

1205904120587 sdot

11986011990311990322 (28)

where 120590 is the radar cross section and 119860119903 = 119866119903120582204120587 isthe effective area of the receiving antenna 119866119903 is the gain ofthe reader received signal and 1205820 is the signal wavelength

Assume that the received power of the legitimate user is119875119889 the noise power received by the legitimate user is119873119889 thereceived power of the illegal eavesdropping user is 119875119890 and thereceived noise power is119873119890 119877119889 and 119877119890 represent the primarychannel and eavesdropping channel capacity respectivelyThe secrecy capacity can be calculated as

119877119904 = 119877119889 minus 119877119890 = 12 log(1 +

119875119889119873119889) minus

12 log(1 +

119875119890119873119890) (29)

where 119875119889 and 119875119890 can be obtained from (28) In the samecommunication system each node receives the same noiseie 119873119889 = 119873119890 All simulation data were quantified and theresults of the evaluation are shown in Figure 11 The abscissa119889119905119890 indicates the distance between the eavesdropper and thetag and the ordinate indicates the channel secrecy capacityAfter optimization the channel secrecy capacity is increasedby 05119887119894119905119904 overall

The results indicate that the relay antenna is optimizedby the multiobjective genetic algorithm the gain is 70 dBand the maximum side lobe level is reduced to minus576119889119861which enhances the antennarsquos directionality This makes itmore difficult for the eavesdropper to obtain communicationinformationThe antenna can be safely applied in the ambientbackscatter communication of the IoT

5 Conclusion

In this study we investigated an important communicationform of the IoT ie ambient backscattering and proposed a

01 02 03 04 05 06 07 08 09 1dte

minus3

minus25

minus2

minus15

minus1

minus05

0

05

1

15

Rs (b

its)

OptimizedOriginal

Figure 11 Secrecy capacity assessment

machine learning based antenna design scheme for physicallayer security The directional communication from the relaytag to the reader is achieved by combining a patch antennaand an LPDA In order to reduce antenna side lobes andimprove orientation performance we used a multiobjectivegenetic algorithm to optimize the antenna size and obtain aset of optimal Pareto fronts The simulation results justifiedthat our proposed antenna design has a simple structuresaves energy and can effectively protect the physical layer IoTcommunications

Data Availability

The data used to support the findings of this study areincluded within the article

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

References

[1] V Liu A Parks V Talla S Gollakota D Wetherall and JR Smith ldquoAmbient backscatter wireless communication outof thin airrdquo in Proceedings of the the ACM SIGCOMM 2013conference p 39 Hong Kong China August 2013

[2] D T Hoang D Niyato P Wang D I Kim and Z HanldquoAmbient Backscatter A New Approach to Improve NetworkPerformance for RF-Powered Cognitive Radio Networksrdquo IEEETransactions on Communications vol 65 no 9 pp 3659ndash36742017

[3] X LuD NiyatoH Jiang D I Kim Y Xiao and Z Han ldquoAmbi-ent Backscatter Assisted Wireless Powered CommunicationsrdquoIEEEWireless CommunicationsMagazine vol 25 no 2 pp 170ndash177 2018

[4] C Perez-Penichet ldquoPhD Forum Abstract Ambient Backscat-ter Communicationrdquo in Proceedings of the 15th ACMIEEE


International Conference on Information Processing in SensorNetworks IPSN 2016 Austria April 2016

[5] S Han S Xu W Meng and C Li ldquoDense-Device-EnabledCooperative Networks for Efficient and Secure TransmissionrdquoIEEE Network vol 32 no 2 pp 100ndash106 2018

[6] W Zhang W He X Chen Y Cai X Guan and J QuldquoPower allocation for improving physical layer security inD2D communication via stackelberg gamerdquo in Proceedings ofthe 8th International Conference on Wireless Communicationsand Signal Processing WCSP 2016 pp 1ndash5 Yangzhou ChinaOctober 2016

[7] T Q Duong ldquoKeynote talk 1 Trusted communications withphysical layer security for 5G and beyondrdquo in Proceedingsof the International Conference on Advanced Technologies forCommunications (ATC) p xxxiv Quy Nhon Vietnam 2017

[8] P Zhang Y Ma and B Wang ldquoImproving physical layersecurity via multiple-level relay networkrdquo in Proceedings of the2014 12th IEEE International Conference on Signal ProcessingICSP 2014 pp 1851ndash1854 Hangzhou China October 2014

[9] A Mukherjee ldquoPhysical-layer security in the internet of thingssensing and communication confidentiality under resourceconstraintsrdquo Proceedings of the IEEE vol 103 no 10 pp 1747ndash1761 2015

[10] Z Chen H Li G Cui andM Rangaswamy ldquoAdaptive transmitand receive beamforming for interference mitigationrdquo IEEESignal Processing Letters vol 21 no 2 pp 235ndash239 2014

[11] Q Chen S-W Qu J Li L Wang Q Yuan and K SawayaldquoDual-antenna system composed of patch array and planarYagi-Uda arrayrdquo in Proceedings of the 5th European Conferenceon Antennas and Propagation EUCAP 2011 pp 1023ndash1026Rome Italy April 2011

[12] E E Altshuler and D S Linden ldquoWire-antenna designs usinggenetic algorithmsrdquo IEEE Antennas and Propagation Magazinevol 39 no 2 pp 33ndash43 1997

[13] S Santarelli T-L Yu D E Goldberg et al ldquoMilitary antennadesign using simple and competent genetic algorithmsrdquoMathe-matical andComputer Modelling vol 43 no 9-10 pp 990ndash10222006

[14] W Zhao G Wang R Fan L Fan and S Atapattu ldquoAmbientBackscatter Communication Systems Capacity and OutagePerformance Analysisrdquo IEEE Access vol 6 pp 22695ndash227042018

[15] Y Liu G Wang Z Dou and Z Zhong ldquoCoding and DetectionSchemes for Ambient Backscatter Communication SystemsrdquoIEEE Access vol 5 pp 4947ndash4953 2017

[16] GWang F Gao R Fan andC Tellambura ldquoAmbient Backscat-ter Communication Systems Detection andPerformance Anal-ysisrdquo IEEE Transactions on Communications vol 64 no 11 pp4836ndash4846 2016

[17] S J Thomas and M S Reynolds ldquoA 96 Mbitsec 155 pJbit16-QAM modulator for UHF backscatter communicationrdquo inProceedings of the 2012 6th IEEE International Conference onRFID RFID 2012 pp 185ndash190 Orlando FL USA April 2012

[18] D Bharadia K R Joshi M Kotaru and S Katti ldquoBackFi HighThroughput WiFi Backscatterrdquo in Proceedings of the 2015 ACMSIGCOMM London UK 2016

[19] F Zhu and M Yao ldquoImproving Physical-Layer Security forCRNs Using SINR-Based Cooperative Beamformingrdquo IEEETransactions on Vehicular Technology vol 65 no 3 pp 1835ndash1841 2016

[20] W X Liu Y Z Yin W L Xu and S Zuo ldquoCompact open-slot antennawith bandwidth enhancementrdquo IEEEAntennas andWireless Propagation Letters vol 10 pp 850ndash853 2011

[21] K R Carver and JW Mink ldquoMicrostrip AntennaTechnologyrdquoIEEE Transactions on Antennas and Propagation vol 29 no 1pp 2ndash24 1981

[22] R R Pantoja A R Sapienza and F C Medeiros FilholdquoA Microwave Printed Planar Log-Periodic Dipole ArrayAntennardquo IEEE Transactions on Antennas and Propagation vol35 no 10 pp 1176ndash1178 1987

[23] S Han S Xu W Meng and C Li ldquoAn agile confidentialtransmission strategy combining big data driven cluster andOBFrdquo IEEE Transactions on Vehicular Technology no 99 article1 2017

[24] S Qiao X Dai Z Liu J Huang and G Zhu ldquoImprov-ing the optimization performance of NSGA-II algorithm byexperiment design methodsrdquo in Proceedings of the 2012 IEEEInternational Conference on Computational Intelligence for Mea-surement Systems and Applications CIMSA 2012 pp 82ndash85Tianjin China July 2012

International Journal of

AerospaceEngineeringHindawiwwwhindawicom Volume 2018

RoboticsJournal of

Hindawiwwwhindawicom Volume 2018


Active and Passive Electronic Components

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawiwwwhindawicom

Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Control Scienceand Engineering

Journal of



Journal ofEngineeringVolume 2018

SensorsJournal of



RotatingMachinery


Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation


Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia

Submit your manuscripts atwwwhindawicom


Ambientsource Legacy

User

Tag

Reader

Backscatter signalsAmbient signals

AntennaEnergy

HarvesterVariableImpedance

Micro-

ℎst

ℎsr

ℎtr

controller

Figure 1 Ambient backscatter system

Comparisons have shown that the method most suited forIoT sensing applications is beamforming which allows thesignal to propagate in a specified direction [10] Throughbeamforming technology the signal-to-noise ratio (SNR)is increased at the legitimate reader whereas it is reducedat the illegal eavesdropper thereby enhancing the securityperformance of the system

In order to achieve similar beamforming functions ina sensor network we focused our research on the antennadesign of the relay tag Existing relay antennas have problemssuch as very large size poor directivity and small transmis-sion gain [11] We designed a dual-antenna system consistingof a patch antenna array and a printed log-periodic dual-dipole antenna (LPDA) with the advantages of orientationand high gain At the same time it has a fairly wide operatingfrequency band which can effectively utilize various types ofsignals in the environment Moreover the small side lobesreduce the SNR received by eavesdroppers from other direc-tions ensuring the security of communication In practicalapplications the patch antenna array receives RF signals fromall directions and then transmits them through the LPDA ina specified direction

Traditional antenna designs are mostly based on expe-rience and simulations to continuously modify the relevantparameters which is time consuming and arduous As amachine learning algorithm genetic algorithms have beenwidely used in antenna design to search for large-scalenonintuitive solution space and find the optimal parametervalue In [12] a genetic algorithm was used to optimize thestructure and length of a wire antenna In [13] an improvedhierarchical Bayesian optimization algorithm was applied tothe optimization of the antenna array feed network Howeverthe single-objective genetic algorithms used in the above-mentioned research may not be suitable for real situations

because it over-emphasizes the importance of one metric Tosolve this problem we use a multiobjective genetic algorithmto optimize the antenna by using gain side lobe returnloss and voltage standing wave ratio (VSWR) as objectivefunctions

The remainder of this paper is organized as followsSection 2 introduces the ambient backscatter communicationmodel and the use of the directional antenna to achievephysical layer security Section 3 illustrates the structureof the proposed antenna and the optimization process ofthe multiobjective genetic algorithm Section 4 presents thesimulation and optimization results of the antenna structurefollowed by Section 5 which concludes the paper

2 System Model

Ambient backscatter has become a promising option for self-sustainable communication systems because of its energy-saving features and has good potential for widespread usein the IoT A typical ambient backscatter system includes anambient source a passive tag and a reader as illustrated inFigure 1

The communication process within the tag is as followsWhen an ambient source broadcasts signals to its legacyusers such as mobile phones and laptops the tag can harvestthe RF energy from the signals and use the collected energyto power the entire system Then the micro-controller in thesystem tunes the variable impedance based on the signalto indicate bit ldquo1rdquo or ldquo0rdquo by backscattering or absorbingthe ambient signals [14 15] Finally the reader decodes thebackscattered signals and recovers the two information bitscompleting the tag-to-reader communication

Evaluating the security of the system and selecting anappropriate solution improves the security of the system A




119910119905 (119899) = radic119875119904ℎ119904119905119904 (119899) + 119908119905 (119899) (1)



119909119905 (119899) = 120578119909 (119899) 119910119905 (119899) (2)


119910119903 (119899) = ℎ119904119903119904 (119899) + ℎ119905119903119909119905 (119899) + 119908 (119899)

= radic119875119904ℎ0119904 (119899) + 119908 (119899) 119909 (119899) = 0radic119875119904ℎ1119904 (119899) + 119908 (119899) 119909 (119899) = 1

(3)





ℎtr

ℎte

Eve

Ambientsource

Bob

Tag





119877119904 = max 119877119889 minus 119877119890 (4)



119877119904 = maxmin119895119877119889 minus 119877119895119890 (5)


119875119900119906119905 = 119875 119877119904 lt 120576 120576 gt 0 (6)


Eve

BobBob

Ambientsource

Tag1 Tag2










143

66m

m 4 Patchs62mm LPDA

Top layer500mm

Power combiner

LPDA

Bottom layer










119871 = 11988821198910

1radic1120576119890 minus 2Δ119871 (7)

119882 = 11988821198910 (

120576119903 + 12 )12 (8)


120576119890 = 120576119903 + 12 + 120576119903 minus 12 (1 + 12ℎ119871)

12

(9)

Δ119871 = 0412ℎ(120576119890 + 03) (119882ℎ + 0264)(120576119890 minus 0258) (119882ℎ + 08) (10)


119883119891 = 1198712radic120576119890 (11)





120591 = ℎ119899+1ℎ119899 = 119871119899+1

119871119899 = 119889119899+1119889119899 (12)

120590 = 1198891198994ℎ119899 (13)


119873119886 = 1 + 119897119892 (11987021198701)119897119892120591 (14)


Ln

ℎn

dn

wn


1198701 = 101 minus 0519120591 (15)



119885119886 asymp 120 ln(ℎ119886) minus 225 (17)






Target2

Target1

w1

w2

w3

0









Start



function



Yes

Best individuals

Stop

Crossover

Mutation

Selection





119874119901119905119894119898119886119897119865 (119909) = [1198651 (119909) 1198652 (119909) 1198653 (119909) 1198654 (119909)] (18)








1198651 (119909) = 1119873119873

sum119894=1

119876 (119891119894) (19)

119876(119891119894) = 10 11987811 lt minus10100381610038161003816100381611987811 (119891119894)1003816100381610038161003816 11987811 ⩾ minus10

(20)





119881119878119882119877 (119891119894) = 119881119878119882119877 119881119878119882119877 ⩽ 1818 119881119878119882119877 lt 18 (21)

1198652 (119909) = 1119873119873

sum119894=1

119881119878119882119877 (119891119894) (22)



1198653 (119909) = 1119873119873

sum119894=1

119866119886119894119899 (119891119894) (23)



119891 (120579 120601) = sin 120579119872

sum119901=1

119871119901


sdot119873

sum119899=1


(24)


119891119894isin[21198661198671199114119866119867119911](119878119871119871 (119891119894)) (25)


119904119891119899 =

1 119894119891 119911119899 ⩾ 1199111198981198861199091198991 minus 119911119898119886119909119899 minus 119911119899

119911119898119886119909119899 minus 119911119898119894119899119899 119894119891 119911119898119894119899119899 lt 119911119899 lt 1199111198981198861199091198990 119894119891 119911119899 ⩽ 119911119898119894119899119899

(26)


119904119895 = sum119873119900119887119895119899=1 119904119891119895119899sum119872119901119886119903119895=1 sum119873119900119887119895119899=1 119904119891119895119899

(27)


4 Numerical Results








1198711 1198712 1198713 1198714 1198715 1198716 119871736 324 2916 2624 2362 2126 1921198718 1198719 11987110 11987111 11987112 1198891 11988921722 1550 1395 1255 1130 13 1171198893 1198894 1198895 1198896 1198897 1198898 11988991053 948 853 768 691 622 56011988910 11988911 Gain SLL504B 453 70119889119861 minus576119889119861






02468

0

30

60

90

120

150

180

210

240

270

300

330

02468


600400200000-200

Unit dB

m2

-400 m3

m4minus8

minus6

minus4

minus2

minus8

minus6

minus4

minus2

m1



1 2 3 4 5minus40

minus30

minus20

minus10

0

a4a3

a2

S(1

1) (d

B)


a1








10 15 20 25 30 35 40 45 5010

15

20

25

30

35

40

45

50

VSW

R




119875119903 = 119875119905119866119905412058711990321 sdot

1205904120587 sdot

11986011990311990322 (28)



119877119904 = 119877119889 minus 119877119890 = 12 log(1 +

119875119889119873119889) minus

12 log(1 +

119875119890119873119890) (29)



5 Conclusion


01 02 03 04 05 06 07 08 09 1dte

minus3

minus25

minus2

minus15

minus1

minus05

0

05

1

15

Rs (b

its)

OptimizedOriginal



Data Availability




References





























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia





119910119905 (119899) = radic119875119904ℎ119904119905119904 (119899) + 119908119905 (119899) (1)



119909119905 (119899) = 120578119909 (119899) 119910119905 (119899) (2)


119910119903 (119899) = ℎ119904119903119904 (119899) + ℎ119905119903119909119905 (119899) + 119908 (119899)

= radic119875119904ℎ0119904 (119899) + 119908 (119899) 119909 (119899) = 0radic119875119904ℎ1119904 (119899) + 119908 (119899) 119909 (119899) = 1

(3)





ℎtr

ℎte

Eve

Ambientsource

Bob

Tag





119877119904 = max 119877119889 minus 119877119890 (4)



119877119904 = maxmin119895119877119889 minus 119877119895119890 (5)


119875119900119906119905 = 119875 119877119904 lt 120576 120576 gt 0 (6)


Eve

BobBob

Ambientsource

Tag1 Tag2










143

66m

m 4 Patchs62mm LPDA

Top layer500mm

Power combiner

LPDA

Bottom layer










119871 = 11988821198910

1radic1120576119890 minus 2Δ119871 (7)

119882 = 11988821198910 (

120576119903 + 12 )12 (8)


120576119890 = 120576119903 + 12 + 120576119903 minus 12 (1 + 12ℎ119871)

12

(9)

Δ119871 = 0412ℎ(120576119890 + 03) (119882ℎ + 0264)(120576119890 minus 0258) (119882ℎ + 08) (10)


119883119891 = 1198712radic120576119890 (11)





120591 = ℎ119899+1ℎ119899 = 119871119899+1

119871119899 = 119889119899+1119889119899 (12)

120590 = 1198891198994ℎ119899 (13)


119873119886 = 1 + 119897119892 (11987021198701)119897119892120591 (14)


Ln

ℎn

dn

wn


1198701 = 101 minus 0519120591 (15)



119885119886 asymp 120 ln(ℎ119886) minus 225 (17)






Target2

Target1

w1

w2

w3

0









Start



function



Yes

Best individuals

Stop

Crossover

Mutation

Selection





119874119901119905119894119898119886119897119865 (119909) = [1198651 (119909) 1198652 (119909) 1198653 (119909) 1198654 (119909)] (18)








1198651 (119909) = 1119873119873

sum119894=1

119876 (119891119894) (19)

119876(119891119894) = 10 11987811 lt minus10100381610038161003816100381611987811 (119891119894)1003816100381610038161003816 11987811 ⩾ minus10

(20)





119881119878119882119877 (119891119894) = 119881119878119882119877 119881119878119882119877 ⩽ 1818 119881119878119882119877 lt 18 (21)

1198652 (119909) = 1119873119873

sum119894=1

119881119878119882119877 (119891119894) (22)



1198653 (119909) = 1119873119873

sum119894=1

119866119886119894119899 (119891119894) (23)



119891 (120579 120601) = sin 120579119872

sum119901=1

119871119901


sdot119873

sum119899=1


(24)


119891119894isin[21198661198671199114119866119867119911](119878119871119871 (119891119894)) (25)


119904119891119899 =

1 119894119891 119911119899 ⩾ 1199111198981198861199091198991 minus 119911119898119886119909119899 minus 119911119899

119911119898119886119909119899 minus 119911119898119894119899119899 119894119891 119911119898119894119899119899 lt 119911119899 lt 1199111198981198861199091198990 119894119891 119911119899 ⩽ 119911119898119894119899119899

(26)


119904119895 = sum119873119900119887119895119899=1 119904119891119895119899sum119872119901119886119903119895=1 sum119873119900119887119895119899=1 119904119891119895119899

(27)


4 Numerical Results








1198711 1198712 1198713 1198714 1198715 1198716 119871736 324 2916 2624 2362 2126 1921198718 1198719 11987110 11987111 11987112 1198891 11988921722 1550 1395 1255 1130 13 1171198893 1198894 1198895 1198896 1198897 1198898 11988991053 948 853 768 691 622 56011988910 11988911 Gain SLL504B 453 70119889119861 minus576119889119861






02468

0

30

60

90

120

150

180

210

240

270

300

330

02468


600400200000-200

Unit dB

m2

-400 m3

m4minus8

minus6

minus4

minus2

minus8

minus6

minus4

minus2

m1



1 2 3 4 5minus40

minus30

minus20

minus10

0

a4a3

a2

S(1

1) (d

B)


a1








10 15 20 25 30 35 40 45 5010

15

20

25

30

35

40

45

50

VSW

R




119875119903 = 119875119905119866119905412058711990321 sdot

1205904120587 sdot

11986011990311990322 (28)



119877119904 = 119877119889 minus 119877119890 = 12 log(1 +

119875119889119873119889) minus

12 log(1 +

119875119890119873119890) (29)



5 Conclusion


01 02 03 04 05 06 07 08 09 1dte

minus3

minus25

minus2

minus15

minus1

minus05

0

05

1

15

Rs (b

its)

OptimizedOriginal



Data Availability




References





























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



Eve

BobBob

Ambientsource

Tag1 Tag2










143

66m

m 4 Patchs62mm LPDA

Top layer500mm

Power combiner

LPDA

Bottom layer










119871 = 11988821198910

1radic1120576119890 minus 2Δ119871 (7)

119882 = 11988821198910 (

120576119903 + 12 )12 (8)


120576119890 = 120576119903 + 12 + 120576119903 minus 12 (1 + 12ℎ119871)

12

(9)

Δ119871 = 0412ℎ(120576119890 + 03) (119882ℎ + 0264)(120576119890 minus 0258) (119882ℎ + 08) (10)


119883119891 = 1198712radic120576119890 (11)





120591 = ℎ119899+1ℎ119899 = 119871119899+1

119871119899 = 119889119899+1119889119899 (12)

120590 = 1198891198994ℎ119899 (13)


119873119886 = 1 + 119897119892 (11987021198701)119897119892120591 (14)


Ln

ℎn

dn

wn


1198701 = 101 minus 0519120591 (15)



119885119886 asymp 120 ln(ℎ119886) minus 225 (17)






Target2

Target1

w1

w2

w3

0









Start



function



Yes

Best individuals

Stop

Crossover

Mutation

Selection





119874119901119905119894119898119886119897119865 (119909) = [1198651 (119909) 1198652 (119909) 1198653 (119909) 1198654 (119909)] (18)








1198651 (119909) = 1119873119873

sum119894=1

119876 (119891119894) (19)

119876(119891119894) = 10 11987811 lt minus10100381610038161003816100381611987811 (119891119894)1003816100381610038161003816 11987811 ⩾ minus10

(20)





119881119878119882119877 (119891119894) = 119881119878119882119877 119881119878119882119877 ⩽ 1818 119881119878119882119877 lt 18 (21)

1198652 (119909) = 1119873119873

sum119894=1

119881119878119882119877 (119891119894) (22)



1198653 (119909) = 1119873119873

sum119894=1

119866119886119894119899 (119891119894) (23)



119891 (120579 120601) = sin 120579119872

sum119901=1

119871119901


sdot119873

sum119899=1


(24)


119891119894isin[21198661198671199114119866119867119911](119878119871119871 (119891119894)) (25)


119904119891119899 =

1 119894119891 119911119899 ⩾ 1199111198981198861199091198991 minus 119911119898119886119909119899 minus 119911119899

119911119898119886119909119899 minus 119911119898119894119899119899 119894119891 119911119898119894119899119899 lt 119911119899 lt 1199111198981198861199091198990 119894119891 119911119899 ⩽ 119911119898119894119899119899

(26)


119904119895 = sum119873119900119887119895119899=1 119904119891119895119899sum119872119901119886119903119895=1 sum119873119900119887119895119899=1 119904119891119895119899

(27)


4 Numerical Results








1198711 1198712 1198713 1198714 1198715 1198716 119871736 324 2916 2624 2362 2126 1921198718 1198719 11987110 11987111 11987112 1198891 11988921722 1550 1395 1255 1130 13 1171198893 1198894 1198895 1198896 1198897 1198898 11988991053 948 853 768 691 622 56011988910 11988911 Gain SLL504B 453 70119889119861 minus576119889119861






02468

0

30

60

90

120

150

180

210

240

270

300

330

02468


600400200000-200

Unit dB

m2

-400 m3

m4minus8

minus6

minus4

minus2

minus8

minus6

minus4

minus2

m1



1 2 3 4 5minus40

minus30

minus20

minus10

0

a4a3

a2

S(1

1) (d

B)


a1








10 15 20 25 30 35 40 45 5010

15

20

25

30

35

40

45

50

VSW

R




119875119903 = 119875119905119866119905412058711990321 sdot

1205904120587 sdot

11986011990311990322 (28)



119877119904 = 119877119889 minus 119877119890 = 12 log(1 +

119875119889119873119889) minus

12 log(1 +

119875119890119873119890) (29)



5 Conclusion


01 02 03 04 05 06 07 08 09 1dte

minus3

minus25

minus2

minus15

minus1

minus05

0

05

1

15

Rs (b

its)

OptimizedOriginal



Data Availability




References





























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia




119871 = 11988821198910

1radic1120576119890 minus 2Δ119871 (7)

119882 = 11988821198910 (

120576119903 + 12 )12 (8)


120576119890 = 120576119903 + 12 + 120576119903 minus 12 (1 + 12ℎ119871)

12

(9)

Δ119871 = 0412ℎ(120576119890 + 03) (119882ℎ + 0264)(120576119890 minus 0258) (119882ℎ + 08) (10)


119883119891 = 1198712radic120576119890 (11)





120591 = ℎ119899+1ℎ119899 = 119871119899+1

119871119899 = 119889119899+1119889119899 (12)

120590 = 1198891198994ℎ119899 (13)


119873119886 = 1 + 119897119892 (11987021198701)119897119892120591 (14)


Ln

ℎn

dn

wn


1198701 = 101 minus 0519120591 (15)



119885119886 asymp 120 ln(ℎ119886) minus 225 (17)






Target2

Target1

w1

w2

w3

0









Start



function



Yes

Best individuals

Stop

Crossover

Mutation

Selection





119874119901119905119894119898119886119897119865 (119909) = [1198651 (119909) 1198652 (119909) 1198653 (119909) 1198654 (119909)] (18)








1198651 (119909) = 1119873119873

sum119894=1

119876 (119891119894) (19)

119876(119891119894) = 10 11987811 lt minus10100381610038161003816100381611987811 (119891119894)1003816100381610038161003816 11987811 ⩾ minus10

(20)





119881119878119882119877 (119891119894) = 119881119878119882119877 119881119878119882119877 ⩽ 1818 119881119878119882119877 lt 18 (21)

1198652 (119909) = 1119873119873

sum119894=1

119881119878119882119877 (119891119894) (22)



1198653 (119909) = 1119873119873

sum119894=1

119866119886119894119899 (119891119894) (23)



119891 (120579 120601) = sin 120579119872

sum119901=1

119871119901


sdot119873

sum119899=1


(24)


119891119894isin[21198661198671199114119866119867119911](119878119871119871 (119891119894)) (25)


119904119891119899 =

1 119894119891 119911119899 ⩾ 1199111198981198861199091198991 minus 119911119898119886119909119899 minus 119911119899

119911119898119886119909119899 minus 119911119898119894119899119899 119894119891 119911119898119894119899119899 lt 119911119899 lt 1199111198981198861199091198990 119894119891 119911119899 ⩽ 119911119898119894119899119899

(26)


119904119895 = sum119873119900119887119895119899=1 119904119891119895119899sum119872119901119886119903119895=1 sum119873119900119887119895119899=1 119904119891119895119899

(27)


4 Numerical Results








1198711 1198712 1198713 1198714 1198715 1198716 119871736 324 2916 2624 2362 2126 1921198718 1198719 11987110 11987111 11987112 1198891 11988921722 1550 1395 1255 1130 13 1171198893 1198894 1198895 1198896 1198897 1198898 11988991053 948 853 768 691 622 56011988910 11988911 Gain SLL504B 453 70119889119861 minus576119889119861






02468

0

30

60

90

120

150

180

210

240

270

300

330

02468


600400200000-200

Unit dB

m2

-400 m3

m4minus8

minus6

minus4

minus2

minus8

minus6

minus4

minus2

m1



1 2 3 4 5minus40

minus30

minus20

minus10

0

a4a3

a2

S(1

1) (d

B)


a1








10 15 20 25 30 35 40 45 5010

15

20

25

30

35

40

45

50

VSW

R




119875119903 = 119875119905119866119905412058711990321 sdot

1205904120587 sdot

11986011990311990322 (28)



119877119904 = 119877119889 minus 119877119890 = 12 log(1 +

119875119889119873119889) minus

12 log(1 +

119875119890119873119890) (29)



5 Conclusion


01 02 03 04 05 06 07 08 09 1dte

minus3

minus25

minus2

minus15

minus1

minus05

0

05

1

15

Rs (b

its)

OptimizedOriginal



Data Availability




References





























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



Target2

Target1

w1

w2

w3

0









Start



function



Yes

Best individuals

Stop

Crossover

Mutation

Selection





119874119901119905119894119898119886119897119865 (119909) = [1198651 (119909) 1198652 (119909) 1198653 (119909) 1198654 (119909)] (18)








1198651 (119909) = 1119873119873

sum119894=1

119876 (119891119894) (19)

119876(119891119894) = 10 11987811 lt minus10100381610038161003816100381611987811 (119891119894)1003816100381610038161003816 11987811 ⩾ minus10

(20)





119881119878119882119877 (119891119894) = 119881119878119882119877 119881119878119882119877 ⩽ 1818 119881119878119882119877 lt 18 (21)

1198652 (119909) = 1119873119873

sum119894=1

119881119878119882119877 (119891119894) (22)



1198653 (119909) = 1119873119873

sum119894=1

119866119886119894119899 (119891119894) (23)



119891 (120579 120601) = sin 120579119872

sum119901=1

119871119901


sdot119873

sum119899=1


(24)


119891119894isin[21198661198671199114119866119867119911](119878119871119871 (119891119894)) (25)


119904119891119899 =

1 119894119891 119911119899 ⩾ 1199111198981198861199091198991 minus 119911119898119886119909119899 minus 119911119899

119911119898119886119909119899 minus 119911119898119894119899119899 119894119891 119911119898119894119899119899 lt 119911119899 lt 1199111198981198861199091198990 119894119891 119911119899 ⩽ 119911119898119894119899119899

(26)


119904119895 = sum119873119900119887119895119899=1 119904119891119895119899sum119872119901119886119903119895=1 sum119873119900119887119895119899=1 119904119891119895119899

(27)


4 Numerical Results








1198711 1198712 1198713 1198714 1198715 1198716 119871736 324 2916 2624 2362 2126 1921198718 1198719 11987110 11987111 11987112 1198891 11988921722 1550 1395 1255 1130 13 1171198893 1198894 1198895 1198896 1198897 1198898 11988991053 948 853 768 691 622 56011988910 11988911 Gain SLL504B 453 70119889119861 minus576119889119861






02468

0

30

60

90

120

150

180

210

240

270

300

330

02468


600400200000-200

Unit dB

m2

-400 m3

m4minus8

minus6

minus4

minus2

minus8

minus6

minus4

minus2

m1



1 2 3 4 5minus40

minus30

minus20

minus10

0

a4a3

a2

S(1

1) (d

B)


a1








10 15 20 25 30 35 40 45 5010

15

20

25

30

35

40

45

50

VSW

R




119875119903 = 119875119905119866119905412058711990321 sdot

1205904120587 sdot

11986011990311990322 (28)



119877119904 = 119877119889 minus 119877119890 = 12 log(1 +

119875119889119873119889) minus

12 log(1 +

119875119890119873119890) (29)



5 Conclusion


01 02 03 04 05 06 07 08 09 1dte

minus3

minus25

minus2

minus15

minus1

minus05

0

05

1

15

Rs (b

its)

OptimizedOriginal



Data Availability




References





























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia




1198651 (119909) = 1119873119873

sum119894=1

119876 (119891119894) (19)

119876(119891119894) = 10 11987811 lt minus10100381610038161003816100381611987811 (119891119894)1003816100381610038161003816 11987811 ⩾ minus10

(20)





119881119878119882119877 (119891119894) = 119881119878119882119877 119881119878119882119877 ⩽ 1818 119881119878119882119877 lt 18 (21)

1198652 (119909) = 1119873119873

sum119894=1

119881119878119882119877 (119891119894) (22)



1198653 (119909) = 1119873119873

sum119894=1

119866119886119894119899 (119891119894) (23)



119891 (120579 120601) = sin 120579119872

sum119901=1

119871119901


sdot119873

sum119899=1


(24)


119891119894isin[21198661198671199114119866119867119911](119878119871119871 (119891119894)) (25)


119904119891119899 =

1 119894119891 119911119899 ⩾ 1199111198981198861199091198991 minus 119911119898119886119909119899 minus 119911119899

119911119898119886119909119899 minus 119911119898119894119899119899 119894119891 119911119898119894119899119899 lt 119911119899 lt 1199111198981198861199091198990 119894119891 119911119899 ⩽ 119911119898119894119899119899

(26)


119904119895 = sum119873119900119887119895119899=1 119904119891119895119899sum119872119901119886119903119895=1 sum119873119900119887119895119899=1 119904119891119895119899

(27)


4 Numerical Results








1198711 1198712 1198713 1198714 1198715 1198716 119871736 324 2916 2624 2362 2126 1921198718 1198719 11987110 11987111 11987112 1198891 11988921722 1550 1395 1255 1130 13 1171198893 1198894 1198895 1198896 1198897 1198898 11988991053 948 853 768 691 622 56011988910 11988911 Gain SLL504B 453 70119889119861 minus576119889119861






02468

0

30

60

90

120

150

180

210

240

270

300

330

02468


600400200000-200

Unit dB

m2

-400 m3

m4minus8

minus6

minus4

minus2

minus8

minus6

minus4

minus2

m1



1 2 3 4 5minus40

minus30

minus20

minus10

0

a4a3

a2

S(1

1) (d

B)


a1








10 15 20 25 30 35 40 45 5010

15

20

25

30

35

40

45

50

VSW

R




119875119903 = 119875119905119866119905412058711990321 sdot

1205904120587 sdot

11986011990311990322 (28)



119877119904 = 119877119889 minus 119877119890 = 12 log(1 +

119875119889119873119889) minus

12 log(1 +

119875119890119873119890) (29)



5 Conclusion


01 02 03 04 05 06 07 08 09 1dte

minus3

minus25

minus2

minus15

minus1

minus05

0

05

1

15

Rs (b

its)

OptimizedOriginal



Data Availability




References





























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia






1198711 1198712 1198713 1198714 1198715 1198716 119871736 324 2916 2624 2362 2126 1921198718 1198719 11987110 11987111 11987112 1198891 11988921722 1550 1395 1255 1130 13 1171198893 1198894 1198895 1198896 1198897 1198898 11988991053 948 853 768 691 622 56011988910 11988911 Gain SLL504B 453 70119889119861 minus576119889119861






02468

0

30

60

90

120

150

180

210

240

270

300

330

02468


600400200000-200

Unit dB

m2

-400 m3

m4minus8

minus6

minus4

minus2

minus8

minus6

minus4

minus2

m1



1 2 3 4 5minus40

minus30

minus20

minus10

0

a4a3

a2

S(1

1) (d

B)


a1








10 15 20 25 30 35 40 45 5010

15

20

25

30

35

40

45

50

VSW

R




119875119903 = 119875119905119866119905412058711990321 sdot

1205904120587 sdot

11986011990311990322 (28)



119877119904 = 119877119889 minus 119877119890 = 12 log(1 +

119875119889119873119889) minus

12 log(1 +

119875119890119873119890) (29)



5 Conclusion


01 02 03 04 05 06 07 08 09 1dte

minus3

minus25

minus2

minus15

minus1

minus05

0

05

1

15

Rs (b

its)

OptimizedOriginal



Data Availability




References





























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



10 15 20 25 30 35 40 45 5010

15

20

25

30

35

40

45

50

VSW

R




119875119903 = 119875119905119866119905412058711990321 sdot

1205904120587 sdot

11986011990311990322 (28)



119877119904 = 119877119889 minus 119877119890 = 12 log(1 +

119875119889119873119889) minus

12 log(1 +

119875119890119873119890) (29)



5 Conclusion


01 02 03 04 05 06 07 08 09 1dte

minus3

minus25

minus2

minus15

minus1

minus05

0

05

1

15

Rs (b

its)

OptimizedOriginal



Data Availability




References





























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia


























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia


machine learning based antenna design for physical layer...

Documents