physical-layer channel authentication for 5g via machine...
TRANSCRIPT
Research ArticlePhysical-Layer Channel Authentication for 5G via MachineLearning Algorithm
Songlin Chen 1 Hong Wen 1 Jinsong Wu2 Jie Chen1 Wenjie Liu1
Lin Hu3 and Yi Chen1
1e National Key Laboratory of Science and Technology on Communications University of Electronic Scienceand Technology of China Chengdu 611731 China2Department of Electrical Engineering Universidad de Chile Santiago 833-0072 Chile3Chongqing Key Laboratory of Mobile Communication Technology Chong Qing University of Post amp Telecommunication of ChinaChongqing China
Correspondence should be addressed to Hong Wen wcdma 2000hotmailcom
Received 26 January 2018 Accepted 19 September 2018 Published 2 October 2018
Academic Editor Vicente Casares-Giner
Copyright copy 2018 Songlin Chen et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
By utilizing the radio channel information to detect spoofing attacks channel based physical layer (PHY-layer) enhancedauthentication can be exploited in light-weight securing 5G wireless communications One major obstacle in the application ofthe PHY-layer authentication is its detection rate In this paper a novel authentication method is developed to detect spoofingattacks without a special test threshold while a trained model is used to determine whether the user is legal or illegal Unlike thethreshold test PHY-layer authentication method the proposed AdaBoost based PHY-layer authentication algorithm increases theauthentication rate with one-dimensional test statistic feature In addition a two-dimensional test statistic features authenticationmodel is presented for further improvement of detection rate To evaluate the feasibility of our algorithm we implement the PHY-layer spoofing detectors in multiple-input multiple-output (MIMO) system over universal software radio peripherals (USRP)Extensive experiences show that the proposed methods yield the high performance without compromising the computingcomplexity
1 Introduction
5G mobile communication system puts forward the require-ments that are high-speed high efficiency and high securityunder three typical application scenarios enhanced MobileBroadband (eMBB) Large-Scale Internet of Things (IoT)and ultraReliableamp Low-latencyConnections (uRLLC) [1 2]The specific application scenarios that enhance the need formobile broadband including high-traffic and high-densitywireless networks are densely used in indoors or urbanareas in which large-area signals of wireless mobile networksare continuously covered in rural areas Meanwhile 5Ginvolves the interconnection and communication betweena large number of machines and equipment which is anecessary condition for the operation of IoT [3]Manymobiledevices access the wireless network at the same time whichresults in heavy burden of authentication computing in the
wireless network Therefore lightweight access methods arerequired for intensive application scenarios of 5G wirelesscommunication networks
In response to this need scholars have successivelycarried out researches on light-weight security measuresbased on computational cryptography [4 5] However it isstill very difficult to use the cipher algorithm that meets theresource-constrained application scenarios such as wirelessmobile terminals IoT and sensor networks Therefore thereis a need to find new technologies to construct the lightweightsecurity scheme In the last decade the research of PHY-layersecurity technology has brought new vitality to the wirelessmobile communication industry [6ndash10] The physical layerof the characteristics is difficult to be counterfeit which canprovide high level security with low cost to overcome thelack of the cipher based security technologies Consequentlyphysical layer characteristics which can be used to improve
HindawiWireless Communications and Mobile ComputingVolume 2018 Article ID 6039878 10 pageshttpsdoiorg10115520186039878
2 Wireless Communications and Mobile Computing
the security of wireless communications have been widelyconcerned for researchers
Several PHY-authentication techniques are proposedIn [11ndash17] the received signal strength (RSS) and channelimpulse response (CIR) as well as channel state information(CSI) are utilized to detect identity-based attacks in wirelessnetworks such as man-in-the middle and denial-of-service(DoS) attacks The work [18] presents a PHY-authenticationframework that can be adapted for multicarrier transmis-sion In order to detect Sybil attacks [19 20] present aPHY-authentication protocol that combines with high-layerauthentication based on the channel response decorrelationsrapidly in space and channel-based detection of Sybil attacksin wireless networks is implemented In [21] Peng Hao et aldeveloped a practical authentication scheme by monitoringand analyzing the packet error rate (PER) and received signalstrength indicator (RSSI) at the same time to enhance thespoofing attack detection capability In [22ndash24] the authorsanalysed the spatial decorrelation property of the channelresponse and validated the efficacy of the channel-basedauthentication for spoofing detection inMIMO system by thecomparison between channel information ldquodifferencerdquo of twoor several frames
However in above-mentioned works artificial thresh-olds are needed to detect spoofing attack In fact thresh-old range cannot be accurately confirmed resulting inspoofing detection with low precision In this paper amachine learning based PHY-layer authentication is devel-oped which provides an intelligent decision method insteadof a one-dimension test threshold Specifically Adaboost[25 26] based algorithm with one-dimensional feature isemployed to detect spoofing attacks To enhance authenti-cation performance the two-dimensional feature is carriedoutThemajor contributions of this paper are summarized asfollows
(1) An AdaBoost based PHY-layer authentication algo-rithm is proposed to increase the authentication rate
(2) The authentication model based on two-dimensionalfeature is established which has a stronger per-formance for cheating detection than the one-dimensional authentication method
(3) The proposed PHY-layer channel authenticationscheme is implemented in a real world environ-ment based on MIMO-OFDM systems The simu-lation results show that the detection rate is greatlyincreased
The rest of this paper is organized as follows Section 2describes system model and problem formulation Our pro-posed algorithm for PHY-layer authentication is presented inSection 3 The system experiment and simulation results arepresented in Section 4 In Section 5 we conclude this paper
2 System Model
In this section we provide a system model of physical layerauthentication and hypothesis testing
Alice
Bob
Scattering clusters
Eve
Scattering clusters
(
(
pilots
pilots
Figure 1 Alice-Bob-Eve model in MIMO system
21 MIMO ree Parts System Model As shown in Figure 1our analysis is based on an Alice-Bob-Eve model in MIMOsystem where Alice and Bob are legitimate users equippedwith N T and N R antennas respectively Eve with 119873119879antennas attempts to spoof Alice by using her identity Theyare assumed to be located in spatially separated positionsIn order to address this spoofing detection Bob tracks theuniqueness of wireless channel responses to discriminatebetween legitimate signals fromAlice and illegitimate signalsfrom Eve That is a physical layer authentication The detailedphysical layer authentication process is as follows Signalswith the pilots which can be used to estimate the channelresponse of the corresponding transmitter are transmittedover the wireless multipath channel to the receiver The 119894-th transmission data contains 119873119891-frames while each frameconsists of119873119904 OFDM symbols
Bob is assumed to obtain the Alice-Bob channel infor-mation for any frame index 119896 gt 1 119860119861119896 and save itwhich extracted by the channel estimation After a whilewhen Bob receives the next data frame the k + 1th dataframe 119860119861119896+1 which is extracted and estimated by Bob theunknown channel response information Bob compares 119860119861119896+1with the channel of Alice 119860119861119896 to determine whether thecorresponding signal is actually send by Alice
If the values of 119860119861119896 and 119860119861119896+1 are approaching Bobconsiders the senderrsquos identity as valid and stores it On thecontrary Bob determines that the senders identity is invalidand directly abandons the data frame
Channel information is detected by the channel estima-tion algorithm denoted by 119860119861119896 and 119860119861119896+1 Each data framecontains 119873s OFDM symbols Thus the channel informationis given by
119860119861119896 = [1198601198611198961 1198601198611198962 119860119861119896119873119904
] (1)
where 119860119861119896119909 (119909 = 1 2 119873119904) denotes the 119909-th OFDMsymbol of channel information
Wireless Communications and Mobile Computing 3
22 Hypothesis Testing A binary hypothesis testing is per-formed to determine the identity authentication in thecontinuous data frames Let the receiver Bob verify thatthe kth data frame originates from the legitimate senderAlice and the extracted channel information is 119867119860119861119896 thesender of the k + 1 th data frame is still unknown and thechannel information is119867119860119861119896+1 the null hypothesis H0 indicatesthat the packet is indeed sent by the Alice The alternativehypothesis H1 is that the real client of the packet is notAliceThe spoofing detection builds the hypothesis test givenby
1198670 119867119860119861119896+1 997888rarr 1198671198601198611198961198671 119867119860119861119896+1 997888rarr 119867119860119861119896
(2)
where all elements of119873119896 and119873119896+1 are iid complex Gaussiannoise samples 119862119873(0 1205752) Therefore if channel informationfor hypothesis testing is directly used the need of consideringthe impact of noise variables will increase the certificationcomplexity To this end since 119873119896 and 119873119896+1 are with thesame statistical characteristics the ldquodifferencerdquo of channelinformation can eliminate the influence of noise variablesThe physical layer authentication translates into the com-parison between the ldquodifferencerdquo of the channel informa-tion and the set threshold Equation (2) can be expressedas
1198670 diff (119867119860119861119896+1 119867119860119861119896 ) lt 120578
1198671 diff (119867119860119861119896+1 119867119860119861119896 ) gt 120578(3)
where diff(119860 119861) denotes the calculating result of the differ-ence between A and B and 120578 is the test threshold
The null hypothesis 1198670 is that the identity is legitimateand Bob accepts this hypothesis if the test statistic hecomputes diff(119860 119861) is below some threshold 120578 OtherwiseBob accepts the alternative hypothesis 1198671 that the identityis illegitimate The channel response ldquodifferencerdquo is recordedas T and (3) can be also written as
119879 = diff (119867119860119861119896+1 119867119860119861119896 )gt 1198671lt 1198670
120578 (4)
As shown in (4) the physical layer authentication isactually a comparison between channel information ldquodif-ferencerdquo and authentication threshold Thus the differencebetween channel information and authentication thresholdis the key of physical layer authentication The test statisticscan measure the similarity of channel information andcalculate the channel information difference In this paperwe use two kinds of test statistic TA and TB respectivelyIn particular assuming Bob obtains two consecutive framechannel response of 119860119861119896minus1119909 and 119860119861119896119909 respectively fromAliceWe build test statistics of 119879119860 and 119879119861 based on the twoframes for the purpose of discrimination identity of Aliceor Eve Subsequently Bob acquires the k+1th frame channelinformation as 119860119861119896+1119909
The test statistics are calculated as
119879119860 (119896) =10038161003816100381610038161003816100381610038161003816100381610038161003816
diff (119860119861119896+1119909 minus 119860119861119896119909)diff (119860119861
119896119909minus 119860119861119896minus1119909
)
10038161003816100381610038161003816100381610038161003816100381610038161003816=1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
sum119873119904119909=1sum119873119898=1sum119873119899=1100381610038161003816100381610038161003816119860119861119896+1119909 (119898 119899) minus 119860119861119896119909 (119898 119899) 119890119895120579(119898119899)
100381610038161003816100381610038161003816sum119873119904119909=1sum119873119898=1sum119873119899=1
10038161003816100381610038161003816119860119861119896119909 (119898 119899) minus 119860119861119896minus1119909 (119898 119899) 119890119895120579(119898119899)10038161003816100381610038161003816
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
gt 1198671lt 1198670
120578119860 (5)
where 120579(119898 119899) is the phase offset and can be denoted by
120579 (119898 119899) = arg (119860119861119896119909 (119898 119899) [119867119883119861119896+1119909 (119898 119899)]lowast) (6)
From (5) 119879119860 can be taken as the difference of thesubcarrier amplitude which avoids the effect of 120579(119898 119899)
Two consecutive data frames 119860119861119896119909 and 119860119861119896+1119909 representmeasurement errors in the phase of the channel responseEach channel response value consists of119873119904 frequency domainchannel matrix which is OFDM symbol of N dimensionalsquare matrix and 119899 denotes the 119898th row and 119899 denotes thecolumn element phase offset
119879119861 (119896) =10038161003816100381610038161003816100381610038161003816100381610038161003816
diff (119860119861119896+1119909 minus 119860119861119896119909)diff (119860119861119896119909 minus 119860119861119896minus1119909)
10038161003816100381610038161003816100381610038161003816100381610038161003816=100381610038161003816100381610038161003816100381610038161003816100381610038161003816
sum119873119904119909=1sum119873119898=1sum119873119899=110038161003816100381610038161003816119883119861119896+1119909 (119898 119899) minus 119860119861119896119909 (119898 119899)
100381610038161003816100381610038162
sum119873119904119909=1sum119873119898=1sum119873119899=110038161003816100381610038161003816119860119861119896119909 (119898 119899) minus 119860119861119896minus1119909 (119898 119899)
100381610038161003816100381610038162
100381610038161003816100381610038161003816100381610038161003816100381610038161003816
gt 1198671lt 1198670
120578119860 (7)
where 119879119861 is the test statistic based on amplitude and phaseinformation We use 119879119860 and 119879119861 as the one-dimensional teststatistic respectively for detecting spoofing attack Unfor-tunately it is hard to find the best threshold for achievinghigh accuracy authentication detection rate To tackle thisproblemwe propose a learning algorithmbased onAdaBoostto achieve physical layer authentication in which 119879119860 and 119879119861are used as training features
3 Physical Authentication withAdaBoost Algorithm
In this section we propose a learning algorithm based onAdaBoost for physical authentication
31 AdaBoost Algorithm AdaBoost is the abbreviation ofadaptive boosting and developed by Yoav Freund [24] and is
4 Wireless Communications and Mobile Computing
71
72
7G
7-
1 (x)
2 (x)
G (x)
- (x)
f(x)=M
summ=1
mGm(x)
Figure 2 AdaBoost algorithm
the most widely used form of boosting algorithm Boostingis a powerful technique combined with base classifiers [25]to produce a form of committee whose performance can besignificantly better than other base classifiers The principalof AdaBoost algorithm is that this algorithm improves itsperformance by the iterative algorithm which is adaptive inthe sense that subsequent weak classifiers called as learnersare adjusted to improve those instances misclassified byprevious classifiers AdaBoost can be seen as a particularmethod of training a boosted classifier A boost classifier isa classifier as follows
119891 (119909) =119872
sum119898=1
120572119898119866119898 (119909) (8)
where each 119866119898(119909) is a weak classifier that takes 119909 as inputand returns a value 119910119898 indicating the class of 119909 The weakclassifiers each of classifiers is trained by using a weightedcoefficient 119908119898119894 from the data set where the weighting coeffi-cient associated depending on the performance of the weakclassifiers such as decision tree (support vector machine)SVM are trained in sequence More specially data pointswhich aremisclassified by one of theweak classifiers are beinggiven greater weight which are used to train the next weakclassifier As illustrated in Figure 2 once all the classifiers have
been trained until there are no misclassified data points thentheir final model is generated via a weight majority votingscheme
32 Physical Authentication with AdaBoost Algorithm Thephysical authentication with AdaBoost algorithm is pro-posed for detection spoofing The performance chart of thealgorithm is illustrated in Figure 3 Bob collects the channelmatrix H1198601198611 which obtained by channel estimation usingthe pilot from Alice and records it When Bob receives thenext data frame from the Alice the Bob collects channelinformation H1198601198612 Similarly Bob collects continuous N-frames channel information from Alice and stores as H119860119861 =[H1198601198611 H1198601198612 H119860119861119873 ] In the same time an Eve sends the dataframes to the Bob and claims that he is Alice In practicalcommunication scenarios we do not know where and whoEves are But in proposed scheme Eves are needed to be testtraining purpose Therefore one or several Eve nodes are setfor this purpose Bob continuously extracts the continuous Nframes channel information from Eve and stores as H119864119861 =[H1198641198611 H1198641198612 H119864119861119873 ]
The data set is preprocessed by Bob Firstly Bob calculatesthe value of data set H119860119861 H119864119861 Secondly Bob calculates thetest statistics based on test statistics 119879119860 119879119861 as
119879119883119861119860 (119896) =10038161003816100381610038161003816100381610038161003816100381610038161003816
diff (119883119861119896+1119909 minus 119883119861119896119909)diff (119883119861
119896119909minus 119883119861119896minus1119909
)
10038161003816100381610038161003816100381610038161003816100381610038161003816=1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
sum119873119904119909=1sum119873119898=1sum119873119899=1100381610038161003816100381610038161003816119883119861119896+1119909 (119898 119899) minus 119883119861119896119909 (119898 119899) 119890119895120579(119898119899)
100381610038161003816100381610038161003816sum119873119904119909=1sum119873119898=1sum119873119899=1
10038161003816100381610038161003816119883119861119896119909 (119898 119899) minus 119883119861119896minus1119909 (119898 119899) 119890119895120579(119898119899)10038161003816100381610038161003816
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
gt 1198671lt 1198670
120578119883119860 (9)
119879119883119861119861 (119896) =10038161003816100381610038161003816100381610038161003816100381610038161003816
diff (119883119861119896+1119909 minus 119883119861119896119909)diff (119883119861
119896119909minus 119883119861119896minus1119909
)
10038161003816100381610038161003816100381610038161003816100381610038161003816=100381610038161003816100381610038161003816100381610038161003816100381610038161003816
sum119873119904119909=1sum119873119898=1sum119873119899=110038161003816100381610038161003816119883119861119896+1119909 (119898 119899) minus 119883119861119896119909 (119898 119899)
100381610038161003816100381610038162
sum119873119904119909=1sum119873119898=1sum119873119899=110038161003816100381610038161003816119883119861119896119909 (119898 119899) minus 119883119861119896minus1119909 (119898 119899)
100381610038161003816100381610038162
100381610038161003816100381610038161003816100381610038161003816100381610038161003816
gt 1198671lt 1198670
120578119883119861 (10)
Finally Bob generates training data set of two categoriesThe first one is
119879119860119861119860 = 1199091 119909119894 119909119873 119910119860 (11a)
119879119860119861119861 = 1199091 119909119894 119909119873 119910119860 (11b)
where 119909119894 isin 119879119860119861119860 (119896) or 119909119894 isin 119879119860119861119861 (119896) 119910119860 = +1 by substitutingH119860119861 into (9) (10) yields 119879119860119861119860 119879119860119861119861 and the value of 119910119860represents that the transmitter is the legal transmitter fromAlice And the second training set is
119879119864119861119860 = 1199091198641 119909119864119894 119909119864119873 119910119864119861 (12a)
Wireless Communications and Mobile Computing 5
Start
Adaboost algorithm is used to generate a strong classifier
Rate of reachingtargeted
The end
Collect the sample of the
legal
Collect the sample of the
illegal
Data preprocessing
Thetesting
set
Thetraining
set
To judge legal or illegal
Collect the sample
Data processing
No Yes
Generate aweak
classifier
Figure 3 Physical authentication with AdaBoost algorithm
119879119864119861119861 = 1199091198641 119909119864119894 119909119864119873 119910119864119861 (12b)
where 119909119864119894 isin 119879119860119864119861(119896) or 119909119864119894 isin 119879119861119864119861(119896) 119910119864119861 = minus1 bysubstituting H119864119861 into (9) and (10) yields 119879119864119861119860 and 119879119864119861119861 andthe value of 119910119894 represents that the transmitter is the illegaltransmitter from Eve Bob uses the two classification trainingdata set 119879119860119861119860 119879119860119861119861 119879119864119861119860 and 119879119864119861119861 as input training set
Spoofing detection is essentially a two-classification prob-lem which is considered to be solved through AdaBoostalgorithm The training data is made up of a bunch of samplepoints Each sample point comprises input sample 119909119894 andlabel 119910119894 where 119910119894 isin minus1 1 Each sample point is given anassociated weight parameter 119908119898119894 119898 means 119898-th trainingand 119894means the number of sample points which is initially set1119894 for all sample pointsWe suppose that we have a procedureavailable for training a weak classifier using weighted samplepoints At each iteration of the training process AdaBoosttrains a new weak classifier by using the sample points inwhich the weighting coefficients are adjusted according to theperformance of the previously trained weak classifier so as togive greater weight to the misclassified data points in whichthe classification error rate 119890119898 is used to evaluate misclassifieddata set119863119898
119890119898 = 119875 (119866119898 (119909119894) = 119910119894) =2119905
sum119894=1
119908119898119894119868 (119866119898 (119909119894) = 119910119894) (13)
Then the coefficient 120572119898 of 119866119898 is calculate as
120572119898 =12 log
1 minus 119890119898119890119898
(14)
Finally we generate a final model that different weight isbeing given to different weak classifiers in (8) The AdaBoostalgorithm is given as in Algorithm 2 in which the point ofthe training data can be doubled by combining with the one-dimension test statistics 119879119860 and 119879119861 together and becomea new two-dimensional features authentication model forspoofing detection Therefore in the AdaBoost algorithmthe input training data set T is following two optionalsets
(1)One-dimension test statistics training data set
119879 = 119879119860119861119860 119879119864119861119860
or 119879 = 119879119860119861119861 119879119864119861119861 (15)
(2) Two-dimension test statistics training data set
119879 = (119879119860119861119860 119879119860119861119861 ) (119879119864119861119860 119879119864119861119861 ) (16)
6 Wireless Communications and Mobile Computing
InputThe channel information of legal transmitter or illgal transmitterProcess1 Bob calculates the value of data set H119860119861 and H119864119861 from Alice and simulated Eve
H119860119861 = [H1198601198611 H1198601198612 H119860119861119873 ]H119864119861 = [H1198641198611 H1198641198612 H119864119861119873 ]
2 The data set are preprocessed by Bob3 The data set are divided into two parts and the one is training data set and the other is testing data set4 Use training data set to get the weak classifier5 Use the Adaboost algorithm to generate a strong classifer6 The testing data set is used to verify whether the claasifier can achieve the target detection rate otherwise it will return to
the first step7 The final classifier is the authenticaton decision model which can judge whether the new packets are legitimate or illegal
End
Algorithm 1 Physical authentication
Inputtraining data set 119879Process1 Initialize the weight distribution of the sample points1198631 = (11990811 1199081119894 11990812119905) 1199081119894 =
12119905 119894 = 1 2 2119905
2 for 119898 = 1 to119872 do119898means119898-th training3 Use the training data set of 119863119898 to learn and get the weak classifier
119866119898(119909) 119909119894 997888rarr minus1 +14 Calculate the classification error rate of119863119898 on the training data set
119890119898 = 119875(119866119898(119909119894) = 119910119894) =2119905
sum119894=1
119908119898119894119868 (119866119898 (119909119894) = 119910119894)5 Calculate the coefficient of 119866119898
120572119898 =12 log
1 minus 119890119898119890119898
6 Update the weight distribution of the training data set119863119898+1 = (119908119898+11 119908119898+1119894 119908119898+12119905)119908119898+1119894 =
119908119898119894119885119898
exp (minus120572119898119910119894119866119898 (119909119894)) 119894 = 1 2 2119905
119885119898 =2119905
sum119894=1
119908119898119894 exp (minus120572119898119910119894119866119898 (119909119894))7 Construct a linear combination of weak classifiers
119891(119909) =119872
sum119894=1
120572119898119866119898(119909)End for
return 119866(119909) = sign(119891(119909))
Algorithm 2 AdaBoost
4 Experimental Verification
In this section we will describe the system setup and the testprocess of measuring the Algorithm 1 for detecting Alice andEve
41 System Setup We consider the spoofing detection ofa receiver called Bob the legal transmitter called Aliceand the spoofing node called Eve They were placed inthree separate locations in a room surrounded by manyother devices such as printers desktops and other types of
equipment as shown in Figure 4 There are scattering andrefraction phenomena in the room due to the presence ofobstacles in the wireless channel from Alice to Bob andEve to Bob As shown in Figure 5 we set up experimentalplatform which implemented on USRPs and experimentswere performed in an indoor environment Bob is equippedwith an 8lowast8 MIMO system Alice is equipped with a 2lowast2MIMO system and the spoofing node called Eve is equippedwith a 2lowast2 MIMO system The signals are sent over 2antennas each at center frequency 35GHz with bandwidth2MHz
Wireless Communications and Mobile Computing 7
Bob
Eve
AliceB
devices
devicesOther
Other
5 m1 20 43
1
3
2
m
Figure 4 The experiments consisted of Alice Bob and Eve
Figure 5 Real MIMO communication platform consisted of Alice Bob and Eve
42 Experiment In the experiment the following steps aretaken
Step 1 Bob extracts channel information from Alice and Eveby the existing channel estimation mechanisms respectively
Step 2 Bob preprocesses the dataset according to (5) (7) (9)and (10)while the threshold is between [0 1] (normalization)
Step 3 Bob generates a training data set of two classificationsaccording to (11a) (11b) (12a) and (12b)
Step 4 The two classification training data set T is generatedaccording to (15) or (16)
Step 5 Bob is trained to generate a strong classifier based onthe training data set of two classifications by using AdaBoostalgorithm under Matlab program
Step 6 Bob uses a strong classifier to judge the test set andobtain the authentication detection rate
In the experiment we consider that the collection framesare five hundred frames and the value of test statisticwas normalized between 0 and 1 The test statistic 119879119860 ofchannel information of the Alice and Bob as a function offrames is shown in Figure 6(a) in which the red pointsis 119879119860(119896) in (5) and green points is 119879119864119860(119896) in (9) As canbe seen there is the overlapped area Meanwhile fromFigure 6(b) the overlapped area is large when we chosethe test statistic 119879119861 of channel information in which thered points is 119879119860(119896) in (7) and green points is 119879119864119860(119896) in(10) It is clearly shown that it is difficult to acquire thebest manual test threshold for the accuracy of authentica-tion Moreover we use 119879119860 119879119861 and the number of framesrespectively to draw a three-dimensional plot As shown in
8 Wireless Communications and Mobile Computing
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
EveAlice
Nor
mal
ized
4
Valu
e
(a) Normalized 119879119860 of Alice and Eve
EveAlice
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
Nor
mal
ized
4
Valu
e
(b) Normalized 119879119861 of Alice and Eve
Figure 6 Normalized 119879119860 and 119879119861 value of the legal transmitter Alice and the spoofing node Eve for spoofing detection with center frequency35GHz with bandwidth 2MHz
01
02
08
04
500450
06
06 400350
08
300
Frames04 250
1
20015002 100500 0
EveAlice
Nor
mal
ized
4
Valu
e
Normalized 4 Value
Figure 7 Normalized 119879119860 value and 119879119861 value of the legal transmitterAlice and the spoofing node Eve drawing three dimensional plot
Figure 7 obviously it is hard to use the traditional manualthreshold method to identify the identity of data sets inthe three-dimensional condition Howevermachine learningalgorithm based the authentication model can effectivelysettle this problem and a dividing curved surface can per-form the identification by the AdaBoost adaptive adjustmentalgorithm
43 Simulation Results In this section simulation results areprovided to demonstrate the performance of the proposedauthentication scheme
Test Threshold
0
01
02
03
04
05
06
07
08
09
1
0 01 02 03 04 05 06 07 08 09 1
Det
ectio
n Ra
te
4
4
Figure 8 Correct classification rate of 119879119860 and 119879119861
As a comparison we considered the PHY-layer spoofingdetection [15] with a varied test threshold From the Figure 8we can see that when test threshold equals 04 the bestauthentication detection rate results of using119879119860 or119879119861 reached798 and 654 respectively In addition our proposedmethod which combined two test statistics 119879119860 and 119879119861 as atwo-dimensional feature can improve the accuracy of detec-tion We use 119879119860 119879119861 and the number of frames respectivelyto draw a three-dimensional plot Figure 9 illustrates thecomparison of spoofing detection among the three methodsfrom which we can conclude that manual threshold methodbased on 119879119860 test statistics can achieve 798 detection rate
Wireless Communications and Mobile Computing 9
0
01
02
03
04
05
06
07
08
09
1
Det
ectio
n Ra
te
ML based 4 data setManual Method
MethodManual Method
ML based 4 and4 data sets
Machine Learining based 4 data set Machine Learning based 4 and4 data sets
Figure 9 The simulation result with the different method ofauthentication scheme
while machine learning based authentication method with119879119860 test statistic can acquire 871 detection rate and machinelearning based authentication method with two-dimensionalfeatures 119879119860 and 119879119861 can achieve 913 accuracy rate with anadditional 10 more computation complexity
To sum up the proposed authentication scheme achievesa superior performance over manual threshold strategy [15]Based on the above observation the proposed machinelearning based authentication scheme with tow-dimensionalfeature not only exhibits excellent performance than manualmethod but also has higher authentication rate than that ofthe same algorithm with one-dimensional feature
5 Conclusions
In this paper machine learning algorithm based physical-layer channel authentication for the 5G wireless communica-tion security is proposed A machine learning authenticationmethod could draw a conclusion whether the received pack-ets are from a legitimate transmitter or from a counterfeiterby using one-dimension or two-dimensional joint featuresThe effectiveness of the proposed authentication scheme isvalidated by widely simulations All the data used in thesimulation are derived from real OFDM-MIMO commu-nication platform which provides a real communicationenvironment Moreover the authentication results show thatthe novel methods provide a higher rate in detecting thespoofing attacks than those of the manual threshold basedphysical layer authentication schemes The training of theclassifier can be done offlineTherefore the novel method canperform authentication fast In addition whether we can usemore machine learning algorithms to further optimize ourauthentication model and find a better statistical test of largedifference in channel information is issue that we need to dealwith in the future
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare no conflicts of interest
Acknowledgments
This research was supported by NSFC (no 61572114)Sichuan Sci amp Tech Achievements Transformation Project(no 2016CC003) Sichuan Sci amp Tech Service Develop-ment Project (no 18KJFWSF0368) Hunan Provincial NatureScience Foundation Project 2018JJ2535 Chile CONICYTFONDECYTRegular Project 1181809 andNational Key RampDProgramofChina (2018YFB0904900 and 2018YFB0904905)
References
[1] M Agiwal A Roy andN Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys amp Tutorials vol 18 no 3 pp 1617ndash1655 2016
[2] J Thompson X Ge and H-C Wu ldquo5G wireless communica-tion systems prospects and challenges [Guest Editorial]rdquo IEEECommunications Magazine vol 52 no 2 pp 62ndash64 2014
[3] C Perera C H Liu S Jayawardena and M Chen ldquoA surveyon internet of things from industrial market perspectiverdquo IEEEAccess vol 2 pp 1660ndash1679 2014
[4] A Bogdanov M Knezevic G Leander D Toz K Varıcı andI Verbauwhede ldquoSPONGENT the design space of lightweightcryptographic hashingrdquo IEEE Transactions on Computers vol62 no 10 pp 2041ndash2053 2013
[5] R Zhang L Zhu C Xu and Y Yi ldquoAn Efficient and secureRFID batch authentication protocol with group tags ownershipTransferrdquo IEEE Collaboration and Internet Computing pp 168ndash175 2015
[6] L Hu H Wen B Wu et al ldquoCooperative Jamming forPhysical Layer Security Enhancement in Internet of ThingsrdquoIEEE Internet of ings Journal vol 5 no 1 pp 219ndash228 2018
[7] L Hu H Wen B Wu J Tang F Pan and R-F LiaoldquoCooperative-Jamming-Aided Secrecy Enhancement in Wire-less Networks with Passive Eavesdroppersrdquo IEEE Transactionson Vehicular Technology vol 67 no 3 pp 2108ndash2117 2018
[8] H Wen Physical layer approaches for securing wireless commu-nication systems Springer New York NY USA 2013
[9] LHuHWen BWu J Tang and F Pan ldquoAdaptive Base StationCooperation for Physical Layer Security in Two-Cell WirelessNetworksrdquo IEEE Access vol 4 pp 5607ndash5623 2016
[10] L Hu H Wen B Wu J Tang and F Pan ldquoAdaptive SecureTransmission for Physical Layer Security in Cooperative Wire-less Networksrdquo IEEE Communications Letters vol 21 no 3 pp524ndash527 2017
[11] D B Faria andDRCheriton ldquoDetecting identity-based attacksin wireless networks using signalprintsrdquo in Proceedings of theACM Workshop on Wireless Security pp 43ndash52 Los AngelesCalif USA 2006
[12] M Demirbas and Y Song ldquoAn RSSI-based scheme for sybilattack detection in wireless sensor networksrdquo in Proceedings of
10 Wireless Communications and Mobile Computing
the WoWMoM 2006 2006 International Symposium on aWorldof Wireless Mobile andMultimedia Networks pp 564ndash568 June2006
[13] Y ChenW Trappe and R P Martin ldquoDetecting and localizingwireless spoofing attacksrdquo in Proceedings of the 2007 4th AnnualIEEE Communications Society Conference on Sensor Mesh andAd Hoc Communications and Networks SECON pp 193ndash202San Diego Calif USA June 2007
[14] N Patwari and S K Kasera ldquoRobust location distinction usingtemporal link signaturesrdquo in Proceedings of the ACM Interna-tional Conference on Mobile Computing and NETWORKINGpp 111ndash122 2007
[15] H Wen Y Wang X Zhu J Li and L Zhou ldquoPhysical layerassist authentication technique for smart meter systemrdquo IETCommunications vol 7 no 3 pp 189ndash197 2013
[16] J K Tugnait ldquoWireless user authentication via comparison ofpower spectral densitiesrdquo IEEE Journal on Selected Areas inCommunications vol 31 no 9 pp 1791ndash1802 2013
[17] Z Jiang J Zhao X Li J Han and W Xi ldquoRejecting the attackSource authentication for Wi-Fi management frames using CSIInformationrdquo in Proceedings of the IEEE INFOCOM 2013 -IEEE Conference on Computer Communications pp 2544ndash2552Turin Italy April 2013
[18] X Wu and Z Yang ldquoPhysical-layer authentication for multi-carrier transmissionrdquo IEEE Communications Letters vol 19 no1 pp 74ndash77 2015
[19] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-authentication protocol for spoofing detection in wireless net-worksrdquo IEEE Transactions on Vehicular Technology vol 65 no12 pp 10037ndash10047 2016
[20] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-LayerSpoofing Detection with Reinforcement Learning in WirelessNetworksrdquo IEEE Transactions on Vehicular Technology vol 65no 12 pp 10037ndash10048 2016
[21] P Hao X Wang and A Refaey ldquoAn enhanced cross-layerauthentication mechanism for wireless communications basedon PER and RSSIrdquo in Proceedings of the 2013 13th CanadianWorkshop on Information eory CWIT 2013 pp 44ndash48Canada June 2013
[22] S Chen et al ldquoMachine-to-Machine communications in ultra-dense networksmdashA surveyrdquo IEEE Communications Surveys ampTutorials vol 1 no 1 99 pages 2017
[23] L Xiao L J Greenstein N B Mandayam and W TrappeldquoChannel-based detection of sybil attacks in wireless networksrdquoIEEE Transactions on Information Forensics and Security vol 4no 3 pp 492ndash503 2009
[24] H Wen P-H Ho C Qi and G Gong ldquoPhysical layer assistedauthentication for distributed ad hoc wireless sensor networksrdquoIET Information Security vol 4 no 4 pp 390ndash396 2010
[25] J H Friedman ldquoGreedy function approximation a gradientboosting machinerdquo Annals of Statistics vol 29 no 5 pp 1189ndash1232 2001
[26] J Friedman T Hastie and R Tibshirani ldquoAdditive logisticregression a statistical view of boostingrdquo e Annals of Statis-tics vol 28 no 2 pp 337ndash407 2000
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
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
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
2 Wireless Communications and Mobile Computing
the security of wireless communications have been widelyconcerned for researchers
Several PHY-authentication techniques are proposedIn [11ndash17] the received signal strength (RSS) and channelimpulse response (CIR) as well as channel state information(CSI) are utilized to detect identity-based attacks in wirelessnetworks such as man-in-the middle and denial-of-service(DoS) attacks The work [18] presents a PHY-authenticationframework that can be adapted for multicarrier transmis-sion In order to detect Sybil attacks [19 20] present aPHY-authentication protocol that combines with high-layerauthentication based on the channel response decorrelationsrapidly in space and channel-based detection of Sybil attacksin wireless networks is implemented In [21] Peng Hao et aldeveloped a practical authentication scheme by monitoringand analyzing the packet error rate (PER) and received signalstrength indicator (RSSI) at the same time to enhance thespoofing attack detection capability In [22ndash24] the authorsanalysed the spatial decorrelation property of the channelresponse and validated the efficacy of the channel-basedauthentication for spoofing detection inMIMO system by thecomparison between channel information ldquodifferencerdquo of twoor several frames
However in above-mentioned works artificial thresh-olds are needed to detect spoofing attack In fact thresh-old range cannot be accurately confirmed resulting inspoofing detection with low precision In this paper amachine learning based PHY-layer authentication is devel-oped which provides an intelligent decision method insteadof a one-dimension test threshold Specifically Adaboost[25 26] based algorithm with one-dimensional feature isemployed to detect spoofing attacks To enhance authenti-cation performance the two-dimensional feature is carriedoutThemajor contributions of this paper are summarized asfollows
(1) An AdaBoost based PHY-layer authentication algo-rithm is proposed to increase the authentication rate
(2) The authentication model based on two-dimensionalfeature is established which has a stronger per-formance for cheating detection than the one-dimensional authentication method
(3) The proposed PHY-layer channel authenticationscheme is implemented in a real world environ-ment based on MIMO-OFDM systems The simu-lation results show that the detection rate is greatlyincreased
The rest of this paper is organized as follows Section 2describes system model and problem formulation Our pro-posed algorithm for PHY-layer authentication is presented inSection 3 The system experiment and simulation results arepresented in Section 4 In Section 5 we conclude this paper
2 System Model
In this section we provide a system model of physical layerauthentication and hypothesis testing
Alice
Bob
Scattering clusters
Eve
Scattering clusters
(
(
pilots
pilots
Figure 1 Alice-Bob-Eve model in MIMO system
21 MIMO ree Parts System Model As shown in Figure 1our analysis is based on an Alice-Bob-Eve model in MIMOsystem where Alice and Bob are legitimate users equippedwith N T and N R antennas respectively Eve with 119873119879antennas attempts to spoof Alice by using her identity Theyare assumed to be located in spatially separated positionsIn order to address this spoofing detection Bob tracks theuniqueness of wireless channel responses to discriminatebetween legitimate signals fromAlice and illegitimate signalsfrom Eve That is a physical layer authentication The detailedphysical layer authentication process is as follows Signalswith the pilots which can be used to estimate the channelresponse of the corresponding transmitter are transmittedover the wireless multipath channel to the receiver The 119894-th transmission data contains 119873119891-frames while each frameconsists of119873119904 OFDM symbols
Bob is assumed to obtain the Alice-Bob channel infor-mation for any frame index 119896 gt 1 119860119861119896 and save itwhich extracted by the channel estimation After a whilewhen Bob receives the next data frame the k + 1th dataframe 119860119861119896+1 which is extracted and estimated by Bob theunknown channel response information Bob compares 119860119861119896+1with the channel of Alice 119860119861119896 to determine whether thecorresponding signal is actually send by Alice
If the values of 119860119861119896 and 119860119861119896+1 are approaching Bobconsiders the senderrsquos identity as valid and stores it On thecontrary Bob determines that the senders identity is invalidand directly abandons the data frame
Channel information is detected by the channel estima-tion algorithm denoted by 119860119861119896 and 119860119861119896+1 Each data framecontains 119873s OFDM symbols Thus the channel informationis given by
119860119861119896 = [1198601198611198961 1198601198611198962 119860119861119896119873119904
] (1)
where 119860119861119896119909 (119909 = 1 2 119873119904) denotes the 119909-th OFDMsymbol of channel information
Wireless Communications and Mobile Computing 3
22 Hypothesis Testing A binary hypothesis testing is per-formed to determine the identity authentication in thecontinuous data frames Let the receiver Bob verify thatthe kth data frame originates from the legitimate senderAlice and the extracted channel information is 119867119860119861119896 thesender of the k + 1 th data frame is still unknown and thechannel information is119867119860119861119896+1 the null hypothesis H0 indicatesthat the packet is indeed sent by the Alice The alternativehypothesis H1 is that the real client of the packet is notAliceThe spoofing detection builds the hypothesis test givenby
1198670 119867119860119861119896+1 997888rarr 1198671198601198611198961198671 119867119860119861119896+1 997888rarr 119867119860119861119896
(2)
where all elements of119873119896 and119873119896+1 are iid complex Gaussiannoise samples 119862119873(0 1205752) Therefore if channel informationfor hypothesis testing is directly used the need of consideringthe impact of noise variables will increase the certificationcomplexity To this end since 119873119896 and 119873119896+1 are with thesame statistical characteristics the ldquodifferencerdquo of channelinformation can eliminate the influence of noise variablesThe physical layer authentication translates into the com-parison between the ldquodifferencerdquo of the channel informa-tion and the set threshold Equation (2) can be expressedas
1198670 diff (119867119860119861119896+1 119867119860119861119896 ) lt 120578
1198671 diff (119867119860119861119896+1 119867119860119861119896 ) gt 120578(3)
where diff(119860 119861) denotes the calculating result of the differ-ence between A and B and 120578 is the test threshold
The null hypothesis 1198670 is that the identity is legitimateand Bob accepts this hypothesis if the test statistic hecomputes diff(119860 119861) is below some threshold 120578 OtherwiseBob accepts the alternative hypothesis 1198671 that the identityis illegitimate The channel response ldquodifferencerdquo is recordedas T and (3) can be also written as
119879 = diff (119867119860119861119896+1 119867119860119861119896 )gt 1198671lt 1198670
120578 (4)
As shown in (4) the physical layer authentication isactually a comparison between channel information ldquodif-ferencerdquo and authentication threshold Thus the differencebetween channel information and authentication thresholdis the key of physical layer authentication The test statisticscan measure the similarity of channel information andcalculate the channel information difference In this paperwe use two kinds of test statistic TA and TB respectivelyIn particular assuming Bob obtains two consecutive framechannel response of 119860119861119896minus1119909 and 119860119861119896119909 respectively fromAliceWe build test statistics of 119879119860 and 119879119861 based on the twoframes for the purpose of discrimination identity of Aliceor Eve Subsequently Bob acquires the k+1th frame channelinformation as 119860119861119896+1119909
The test statistics are calculated as
119879119860 (119896) =10038161003816100381610038161003816100381610038161003816100381610038161003816
diff (119860119861119896+1119909 minus 119860119861119896119909)diff (119860119861
119896119909minus 119860119861119896minus1119909
)
10038161003816100381610038161003816100381610038161003816100381610038161003816=1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
sum119873119904119909=1sum119873119898=1sum119873119899=1100381610038161003816100381610038161003816119860119861119896+1119909 (119898 119899) minus 119860119861119896119909 (119898 119899) 119890119895120579(119898119899)
100381610038161003816100381610038161003816sum119873119904119909=1sum119873119898=1sum119873119899=1
10038161003816100381610038161003816119860119861119896119909 (119898 119899) minus 119860119861119896minus1119909 (119898 119899) 119890119895120579(119898119899)10038161003816100381610038161003816
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
gt 1198671lt 1198670
120578119860 (5)
where 120579(119898 119899) is the phase offset and can be denoted by
120579 (119898 119899) = arg (119860119861119896119909 (119898 119899) [119867119883119861119896+1119909 (119898 119899)]lowast) (6)
From (5) 119879119860 can be taken as the difference of thesubcarrier amplitude which avoids the effect of 120579(119898 119899)
Two consecutive data frames 119860119861119896119909 and 119860119861119896+1119909 representmeasurement errors in the phase of the channel responseEach channel response value consists of119873119904 frequency domainchannel matrix which is OFDM symbol of N dimensionalsquare matrix and 119899 denotes the 119898th row and 119899 denotes thecolumn element phase offset
119879119861 (119896) =10038161003816100381610038161003816100381610038161003816100381610038161003816
diff (119860119861119896+1119909 minus 119860119861119896119909)diff (119860119861119896119909 minus 119860119861119896minus1119909)
10038161003816100381610038161003816100381610038161003816100381610038161003816=100381610038161003816100381610038161003816100381610038161003816100381610038161003816
sum119873119904119909=1sum119873119898=1sum119873119899=110038161003816100381610038161003816119883119861119896+1119909 (119898 119899) minus 119860119861119896119909 (119898 119899)
100381610038161003816100381610038162
sum119873119904119909=1sum119873119898=1sum119873119899=110038161003816100381610038161003816119860119861119896119909 (119898 119899) minus 119860119861119896minus1119909 (119898 119899)
100381610038161003816100381610038162
100381610038161003816100381610038161003816100381610038161003816100381610038161003816
gt 1198671lt 1198670
120578119860 (7)
where 119879119861 is the test statistic based on amplitude and phaseinformation We use 119879119860 and 119879119861 as the one-dimensional teststatistic respectively for detecting spoofing attack Unfor-tunately it is hard to find the best threshold for achievinghigh accuracy authentication detection rate To tackle thisproblemwe propose a learning algorithmbased onAdaBoostto achieve physical layer authentication in which 119879119860 and 119879119861are used as training features
3 Physical Authentication withAdaBoost Algorithm
In this section we propose a learning algorithm based onAdaBoost for physical authentication
31 AdaBoost Algorithm AdaBoost is the abbreviation ofadaptive boosting and developed by Yoav Freund [24] and is
4 Wireless Communications and Mobile Computing
71
72
7G
7-
1 (x)
2 (x)
G (x)
- (x)
f(x)=M
summ=1
mGm(x)
Figure 2 AdaBoost algorithm
the most widely used form of boosting algorithm Boostingis a powerful technique combined with base classifiers [25]to produce a form of committee whose performance can besignificantly better than other base classifiers The principalof AdaBoost algorithm is that this algorithm improves itsperformance by the iterative algorithm which is adaptive inthe sense that subsequent weak classifiers called as learnersare adjusted to improve those instances misclassified byprevious classifiers AdaBoost can be seen as a particularmethod of training a boosted classifier A boost classifier isa classifier as follows
119891 (119909) =119872
sum119898=1
120572119898119866119898 (119909) (8)
where each 119866119898(119909) is a weak classifier that takes 119909 as inputand returns a value 119910119898 indicating the class of 119909 The weakclassifiers each of classifiers is trained by using a weightedcoefficient 119908119898119894 from the data set where the weighting coeffi-cient associated depending on the performance of the weakclassifiers such as decision tree (support vector machine)SVM are trained in sequence More specially data pointswhich aremisclassified by one of theweak classifiers are beinggiven greater weight which are used to train the next weakclassifier As illustrated in Figure 2 once all the classifiers have
been trained until there are no misclassified data points thentheir final model is generated via a weight majority votingscheme
32 Physical Authentication with AdaBoost Algorithm Thephysical authentication with AdaBoost algorithm is pro-posed for detection spoofing The performance chart of thealgorithm is illustrated in Figure 3 Bob collects the channelmatrix H1198601198611 which obtained by channel estimation usingthe pilot from Alice and records it When Bob receives thenext data frame from the Alice the Bob collects channelinformation H1198601198612 Similarly Bob collects continuous N-frames channel information from Alice and stores as H119860119861 =[H1198601198611 H1198601198612 H119860119861119873 ] In the same time an Eve sends the dataframes to the Bob and claims that he is Alice In practicalcommunication scenarios we do not know where and whoEves are But in proposed scheme Eves are needed to be testtraining purpose Therefore one or several Eve nodes are setfor this purpose Bob continuously extracts the continuous Nframes channel information from Eve and stores as H119864119861 =[H1198641198611 H1198641198612 H119864119861119873 ]
The data set is preprocessed by Bob Firstly Bob calculatesthe value of data set H119860119861 H119864119861 Secondly Bob calculates thetest statistics based on test statistics 119879119860 119879119861 as
119879119883119861119860 (119896) =10038161003816100381610038161003816100381610038161003816100381610038161003816
diff (119883119861119896+1119909 minus 119883119861119896119909)diff (119883119861
119896119909minus 119883119861119896minus1119909
)
10038161003816100381610038161003816100381610038161003816100381610038161003816=1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
sum119873119904119909=1sum119873119898=1sum119873119899=1100381610038161003816100381610038161003816119883119861119896+1119909 (119898 119899) minus 119883119861119896119909 (119898 119899) 119890119895120579(119898119899)
100381610038161003816100381610038161003816sum119873119904119909=1sum119873119898=1sum119873119899=1
10038161003816100381610038161003816119883119861119896119909 (119898 119899) minus 119883119861119896minus1119909 (119898 119899) 119890119895120579(119898119899)10038161003816100381610038161003816
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
gt 1198671lt 1198670
120578119883119860 (9)
119879119883119861119861 (119896) =10038161003816100381610038161003816100381610038161003816100381610038161003816
diff (119883119861119896+1119909 minus 119883119861119896119909)diff (119883119861
119896119909minus 119883119861119896minus1119909
)
10038161003816100381610038161003816100381610038161003816100381610038161003816=100381610038161003816100381610038161003816100381610038161003816100381610038161003816
sum119873119904119909=1sum119873119898=1sum119873119899=110038161003816100381610038161003816119883119861119896+1119909 (119898 119899) minus 119883119861119896119909 (119898 119899)
100381610038161003816100381610038162
sum119873119904119909=1sum119873119898=1sum119873119899=110038161003816100381610038161003816119883119861119896119909 (119898 119899) minus 119883119861119896minus1119909 (119898 119899)
100381610038161003816100381610038162
100381610038161003816100381610038161003816100381610038161003816100381610038161003816
gt 1198671lt 1198670
120578119883119861 (10)
Finally Bob generates training data set of two categoriesThe first one is
119879119860119861119860 = 1199091 119909119894 119909119873 119910119860 (11a)
119879119860119861119861 = 1199091 119909119894 119909119873 119910119860 (11b)
where 119909119894 isin 119879119860119861119860 (119896) or 119909119894 isin 119879119860119861119861 (119896) 119910119860 = +1 by substitutingH119860119861 into (9) (10) yields 119879119860119861119860 119879119860119861119861 and the value of 119910119860represents that the transmitter is the legal transmitter fromAlice And the second training set is
119879119864119861119860 = 1199091198641 119909119864119894 119909119864119873 119910119864119861 (12a)
Wireless Communications and Mobile Computing 5
Start
Adaboost algorithm is used to generate a strong classifier
Rate of reachingtargeted
The end
Collect the sample of the
legal
Collect the sample of the
illegal
Data preprocessing
Thetesting
set
Thetraining
set
To judge legal or illegal
Collect the sample
Data processing
No Yes
Generate aweak
classifier
Figure 3 Physical authentication with AdaBoost algorithm
119879119864119861119861 = 1199091198641 119909119864119894 119909119864119873 119910119864119861 (12b)
where 119909119864119894 isin 119879119860119864119861(119896) or 119909119864119894 isin 119879119861119864119861(119896) 119910119864119861 = minus1 bysubstituting H119864119861 into (9) and (10) yields 119879119864119861119860 and 119879119864119861119861 andthe value of 119910119894 represents that the transmitter is the illegaltransmitter from Eve Bob uses the two classification trainingdata set 119879119860119861119860 119879119860119861119861 119879119864119861119860 and 119879119864119861119861 as input training set
Spoofing detection is essentially a two-classification prob-lem which is considered to be solved through AdaBoostalgorithm The training data is made up of a bunch of samplepoints Each sample point comprises input sample 119909119894 andlabel 119910119894 where 119910119894 isin minus1 1 Each sample point is given anassociated weight parameter 119908119898119894 119898 means 119898-th trainingand 119894means the number of sample points which is initially set1119894 for all sample pointsWe suppose that we have a procedureavailable for training a weak classifier using weighted samplepoints At each iteration of the training process AdaBoosttrains a new weak classifier by using the sample points inwhich the weighting coefficients are adjusted according to theperformance of the previously trained weak classifier so as togive greater weight to the misclassified data points in whichthe classification error rate 119890119898 is used to evaluate misclassifieddata set119863119898
119890119898 = 119875 (119866119898 (119909119894) = 119910119894) =2119905
sum119894=1
119908119898119894119868 (119866119898 (119909119894) = 119910119894) (13)
Then the coefficient 120572119898 of 119866119898 is calculate as
120572119898 =12 log
1 minus 119890119898119890119898
(14)
Finally we generate a final model that different weight isbeing given to different weak classifiers in (8) The AdaBoostalgorithm is given as in Algorithm 2 in which the point ofthe training data can be doubled by combining with the one-dimension test statistics 119879119860 and 119879119861 together and becomea new two-dimensional features authentication model forspoofing detection Therefore in the AdaBoost algorithmthe input training data set T is following two optionalsets
(1)One-dimension test statistics training data set
119879 = 119879119860119861119860 119879119864119861119860
or 119879 = 119879119860119861119861 119879119864119861119861 (15)
(2) Two-dimension test statistics training data set
119879 = (119879119860119861119860 119879119860119861119861 ) (119879119864119861119860 119879119864119861119861 ) (16)
6 Wireless Communications and Mobile Computing
InputThe channel information of legal transmitter or illgal transmitterProcess1 Bob calculates the value of data set H119860119861 and H119864119861 from Alice and simulated Eve
H119860119861 = [H1198601198611 H1198601198612 H119860119861119873 ]H119864119861 = [H1198641198611 H1198641198612 H119864119861119873 ]
2 The data set are preprocessed by Bob3 The data set are divided into two parts and the one is training data set and the other is testing data set4 Use training data set to get the weak classifier5 Use the Adaboost algorithm to generate a strong classifer6 The testing data set is used to verify whether the claasifier can achieve the target detection rate otherwise it will return to
the first step7 The final classifier is the authenticaton decision model which can judge whether the new packets are legitimate or illegal
End
Algorithm 1 Physical authentication
Inputtraining data set 119879Process1 Initialize the weight distribution of the sample points1198631 = (11990811 1199081119894 11990812119905) 1199081119894 =
12119905 119894 = 1 2 2119905
2 for 119898 = 1 to119872 do119898means119898-th training3 Use the training data set of 119863119898 to learn and get the weak classifier
119866119898(119909) 119909119894 997888rarr minus1 +14 Calculate the classification error rate of119863119898 on the training data set
119890119898 = 119875(119866119898(119909119894) = 119910119894) =2119905
sum119894=1
119908119898119894119868 (119866119898 (119909119894) = 119910119894)5 Calculate the coefficient of 119866119898
120572119898 =12 log
1 minus 119890119898119890119898
6 Update the weight distribution of the training data set119863119898+1 = (119908119898+11 119908119898+1119894 119908119898+12119905)119908119898+1119894 =
119908119898119894119885119898
exp (minus120572119898119910119894119866119898 (119909119894)) 119894 = 1 2 2119905
119885119898 =2119905
sum119894=1
119908119898119894 exp (minus120572119898119910119894119866119898 (119909119894))7 Construct a linear combination of weak classifiers
119891(119909) =119872
sum119894=1
120572119898119866119898(119909)End for
return 119866(119909) = sign(119891(119909))
Algorithm 2 AdaBoost
4 Experimental Verification
In this section we will describe the system setup and the testprocess of measuring the Algorithm 1 for detecting Alice andEve
41 System Setup We consider the spoofing detection ofa receiver called Bob the legal transmitter called Aliceand the spoofing node called Eve They were placed inthree separate locations in a room surrounded by manyother devices such as printers desktops and other types of
equipment as shown in Figure 4 There are scattering andrefraction phenomena in the room due to the presence ofobstacles in the wireless channel from Alice to Bob andEve to Bob As shown in Figure 5 we set up experimentalplatform which implemented on USRPs and experimentswere performed in an indoor environment Bob is equippedwith an 8lowast8 MIMO system Alice is equipped with a 2lowast2MIMO system and the spoofing node called Eve is equippedwith a 2lowast2 MIMO system The signals are sent over 2antennas each at center frequency 35GHz with bandwidth2MHz
Wireless Communications and Mobile Computing 7
Bob
Eve
AliceB
devices
devicesOther
Other
5 m1 20 43
1
3
2
m
Figure 4 The experiments consisted of Alice Bob and Eve
Figure 5 Real MIMO communication platform consisted of Alice Bob and Eve
42 Experiment In the experiment the following steps aretaken
Step 1 Bob extracts channel information from Alice and Eveby the existing channel estimation mechanisms respectively
Step 2 Bob preprocesses the dataset according to (5) (7) (9)and (10)while the threshold is between [0 1] (normalization)
Step 3 Bob generates a training data set of two classificationsaccording to (11a) (11b) (12a) and (12b)
Step 4 The two classification training data set T is generatedaccording to (15) or (16)
Step 5 Bob is trained to generate a strong classifier based onthe training data set of two classifications by using AdaBoostalgorithm under Matlab program
Step 6 Bob uses a strong classifier to judge the test set andobtain the authentication detection rate
In the experiment we consider that the collection framesare five hundred frames and the value of test statisticwas normalized between 0 and 1 The test statistic 119879119860 ofchannel information of the Alice and Bob as a function offrames is shown in Figure 6(a) in which the red pointsis 119879119860(119896) in (5) and green points is 119879119864119860(119896) in (9) As canbe seen there is the overlapped area Meanwhile fromFigure 6(b) the overlapped area is large when we chosethe test statistic 119879119861 of channel information in which thered points is 119879119860(119896) in (7) and green points is 119879119864119860(119896) in(10) It is clearly shown that it is difficult to acquire thebest manual test threshold for the accuracy of authentica-tion Moreover we use 119879119860 119879119861 and the number of framesrespectively to draw a three-dimensional plot As shown in
8 Wireless Communications and Mobile Computing
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
EveAlice
Nor
mal
ized
4
Valu
e
(a) Normalized 119879119860 of Alice and Eve
EveAlice
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
Nor
mal
ized
4
Valu
e
(b) Normalized 119879119861 of Alice and Eve
Figure 6 Normalized 119879119860 and 119879119861 value of the legal transmitter Alice and the spoofing node Eve for spoofing detection with center frequency35GHz with bandwidth 2MHz
01
02
08
04
500450
06
06 400350
08
300
Frames04 250
1
20015002 100500 0
EveAlice
Nor
mal
ized
4
Valu
e
Normalized 4 Value
Figure 7 Normalized 119879119860 value and 119879119861 value of the legal transmitterAlice and the spoofing node Eve drawing three dimensional plot
Figure 7 obviously it is hard to use the traditional manualthreshold method to identify the identity of data sets inthe three-dimensional condition Howevermachine learningalgorithm based the authentication model can effectivelysettle this problem and a dividing curved surface can per-form the identification by the AdaBoost adaptive adjustmentalgorithm
43 Simulation Results In this section simulation results areprovided to demonstrate the performance of the proposedauthentication scheme
Test Threshold
0
01
02
03
04
05
06
07
08
09
1
0 01 02 03 04 05 06 07 08 09 1
Det
ectio
n Ra
te
4
4
Figure 8 Correct classification rate of 119879119860 and 119879119861
As a comparison we considered the PHY-layer spoofingdetection [15] with a varied test threshold From the Figure 8we can see that when test threshold equals 04 the bestauthentication detection rate results of using119879119860 or119879119861 reached798 and 654 respectively In addition our proposedmethod which combined two test statistics 119879119860 and 119879119861 as atwo-dimensional feature can improve the accuracy of detec-tion We use 119879119860 119879119861 and the number of frames respectivelyto draw a three-dimensional plot Figure 9 illustrates thecomparison of spoofing detection among the three methodsfrom which we can conclude that manual threshold methodbased on 119879119860 test statistics can achieve 798 detection rate
Wireless Communications and Mobile Computing 9
0
01
02
03
04
05
06
07
08
09
1
Det
ectio
n Ra
te
ML based 4 data setManual Method
MethodManual Method
ML based 4 and4 data sets
Machine Learining based 4 data set Machine Learning based 4 and4 data sets
Figure 9 The simulation result with the different method ofauthentication scheme
while machine learning based authentication method with119879119860 test statistic can acquire 871 detection rate and machinelearning based authentication method with two-dimensionalfeatures 119879119860 and 119879119861 can achieve 913 accuracy rate with anadditional 10 more computation complexity
To sum up the proposed authentication scheme achievesa superior performance over manual threshold strategy [15]Based on the above observation the proposed machinelearning based authentication scheme with tow-dimensionalfeature not only exhibits excellent performance than manualmethod but also has higher authentication rate than that ofthe same algorithm with one-dimensional feature
5 Conclusions
In this paper machine learning algorithm based physical-layer channel authentication for the 5G wireless communica-tion security is proposed A machine learning authenticationmethod could draw a conclusion whether the received pack-ets are from a legitimate transmitter or from a counterfeiterby using one-dimension or two-dimensional joint featuresThe effectiveness of the proposed authentication scheme isvalidated by widely simulations All the data used in thesimulation are derived from real OFDM-MIMO commu-nication platform which provides a real communicationenvironment Moreover the authentication results show thatthe novel methods provide a higher rate in detecting thespoofing attacks than those of the manual threshold basedphysical layer authentication schemes The training of theclassifier can be done offlineTherefore the novel method canperform authentication fast In addition whether we can usemore machine learning algorithms to further optimize ourauthentication model and find a better statistical test of largedifference in channel information is issue that we need to dealwith in the future
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare no conflicts of interest
Acknowledgments
This research was supported by NSFC (no 61572114)Sichuan Sci amp Tech Achievements Transformation Project(no 2016CC003) Sichuan Sci amp Tech Service Develop-ment Project (no 18KJFWSF0368) Hunan Provincial NatureScience Foundation Project 2018JJ2535 Chile CONICYTFONDECYTRegular Project 1181809 andNational Key RampDProgramofChina (2018YFB0904900 and 2018YFB0904905)
References
[1] M Agiwal A Roy andN Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys amp Tutorials vol 18 no 3 pp 1617ndash1655 2016
[2] J Thompson X Ge and H-C Wu ldquo5G wireless communica-tion systems prospects and challenges [Guest Editorial]rdquo IEEECommunications Magazine vol 52 no 2 pp 62ndash64 2014
[3] C Perera C H Liu S Jayawardena and M Chen ldquoA surveyon internet of things from industrial market perspectiverdquo IEEEAccess vol 2 pp 1660ndash1679 2014
[4] A Bogdanov M Knezevic G Leander D Toz K Varıcı andI Verbauwhede ldquoSPONGENT the design space of lightweightcryptographic hashingrdquo IEEE Transactions on Computers vol62 no 10 pp 2041ndash2053 2013
[5] R Zhang L Zhu C Xu and Y Yi ldquoAn Efficient and secureRFID batch authentication protocol with group tags ownershipTransferrdquo IEEE Collaboration and Internet Computing pp 168ndash175 2015
[6] L Hu H Wen B Wu et al ldquoCooperative Jamming forPhysical Layer Security Enhancement in Internet of ThingsrdquoIEEE Internet of ings Journal vol 5 no 1 pp 219ndash228 2018
[7] L Hu H Wen B Wu J Tang F Pan and R-F LiaoldquoCooperative-Jamming-Aided Secrecy Enhancement in Wire-less Networks with Passive Eavesdroppersrdquo IEEE Transactionson Vehicular Technology vol 67 no 3 pp 2108ndash2117 2018
[8] H Wen Physical layer approaches for securing wireless commu-nication systems Springer New York NY USA 2013
[9] LHuHWen BWu J Tang and F Pan ldquoAdaptive Base StationCooperation for Physical Layer Security in Two-Cell WirelessNetworksrdquo IEEE Access vol 4 pp 5607ndash5623 2016
[10] L Hu H Wen B Wu J Tang and F Pan ldquoAdaptive SecureTransmission for Physical Layer Security in Cooperative Wire-less Networksrdquo IEEE Communications Letters vol 21 no 3 pp524ndash527 2017
[11] D B Faria andDRCheriton ldquoDetecting identity-based attacksin wireless networks using signalprintsrdquo in Proceedings of theACM Workshop on Wireless Security pp 43ndash52 Los AngelesCalif USA 2006
[12] M Demirbas and Y Song ldquoAn RSSI-based scheme for sybilattack detection in wireless sensor networksrdquo in Proceedings of
10 Wireless Communications and Mobile Computing
the WoWMoM 2006 2006 International Symposium on aWorldof Wireless Mobile andMultimedia Networks pp 564ndash568 June2006
[13] Y ChenW Trappe and R P Martin ldquoDetecting and localizingwireless spoofing attacksrdquo in Proceedings of the 2007 4th AnnualIEEE Communications Society Conference on Sensor Mesh andAd Hoc Communications and Networks SECON pp 193ndash202San Diego Calif USA June 2007
[14] N Patwari and S K Kasera ldquoRobust location distinction usingtemporal link signaturesrdquo in Proceedings of the ACM Interna-tional Conference on Mobile Computing and NETWORKINGpp 111ndash122 2007
[15] H Wen Y Wang X Zhu J Li and L Zhou ldquoPhysical layerassist authentication technique for smart meter systemrdquo IETCommunications vol 7 no 3 pp 189ndash197 2013
[16] J K Tugnait ldquoWireless user authentication via comparison ofpower spectral densitiesrdquo IEEE Journal on Selected Areas inCommunications vol 31 no 9 pp 1791ndash1802 2013
[17] Z Jiang J Zhao X Li J Han and W Xi ldquoRejecting the attackSource authentication for Wi-Fi management frames using CSIInformationrdquo in Proceedings of the IEEE INFOCOM 2013 -IEEE Conference on Computer Communications pp 2544ndash2552Turin Italy April 2013
[18] X Wu and Z Yang ldquoPhysical-layer authentication for multi-carrier transmissionrdquo IEEE Communications Letters vol 19 no1 pp 74ndash77 2015
[19] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-authentication protocol for spoofing detection in wireless net-worksrdquo IEEE Transactions on Vehicular Technology vol 65 no12 pp 10037ndash10047 2016
[20] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-LayerSpoofing Detection with Reinforcement Learning in WirelessNetworksrdquo IEEE Transactions on Vehicular Technology vol 65no 12 pp 10037ndash10048 2016
[21] P Hao X Wang and A Refaey ldquoAn enhanced cross-layerauthentication mechanism for wireless communications basedon PER and RSSIrdquo in Proceedings of the 2013 13th CanadianWorkshop on Information eory CWIT 2013 pp 44ndash48Canada June 2013
[22] S Chen et al ldquoMachine-to-Machine communications in ultra-dense networksmdashA surveyrdquo IEEE Communications Surveys ampTutorials vol 1 no 1 99 pages 2017
[23] L Xiao L J Greenstein N B Mandayam and W TrappeldquoChannel-based detection of sybil attacks in wireless networksrdquoIEEE Transactions on Information Forensics and Security vol 4no 3 pp 492ndash503 2009
[24] H Wen P-H Ho C Qi and G Gong ldquoPhysical layer assistedauthentication for distributed ad hoc wireless sensor networksrdquoIET Information Security vol 4 no 4 pp 390ndash396 2010
[25] J H Friedman ldquoGreedy function approximation a gradientboosting machinerdquo Annals of Statistics vol 29 no 5 pp 1189ndash1232 2001
[26] J Friedman T Hastie and R Tibshirani ldquoAdditive logisticregression a statistical view of boostingrdquo e Annals of Statis-tics vol 28 no 2 pp 337ndash407 2000
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
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
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 3
22 Hypothesis Testing A binary hypothesis testing is per-formed to determine the identity authentication in thecontinuous data frames Let the receiver Bob verify thatthe kth data frame originates from the legitimate senderAlice and the extracted channel information is 119867119860119861119896 thesender of the k + 1 th data frame is still unknown and thechannel information is119867119860119861119896+1 the null hypothesis H0 indicatesthat the packet is indeed sent by the Alice The alternativehypothesis H1 is that the real client of the packet is notAliceThe spoofing detection builds the hypothesis test givenby
1198670 119867119860119861119896+1 997888rarr 1198671198601198611198961198671 119867119860119861119896+1 997888rarr 119867119860119861119896
(2)
where all elements of119873119896 and119873119896+1 are iid complex Gaussiannoise samples 119862119873(0 1205752) Therefore if channel informationfor hypothesis testing is directly used the need of consideringthe impact of noise variables will increase the certificationcomplexity To this end since 119873119896 and 119873119896+1 are with thesame statistical characteristics the ldquodifferencerdquo of channelinformation can eliminate the influence of noise variablesThe physical layer authentication translates into the com-parison between the ldquodifferencerdquo of the channel informa-tion and the set threshold Equation (2) can be expressedas
1198670 diff (119867119860119861119896+1 119867119860119861119896 ) lt 120578
1198671 diff (119867119860119861119896+1 119867119860119861119896 ) gt 120578(3)
where diff(119860 119861) denotes the calculating result of the differ-ence between A and B and 120578 is the test threshold
The null hypothesis 1198670 is that the identity is legitimateand Bob accepts this hypothesis if the test statistic hecomputes diff(119860 119861) is below some threshold 120578 OtherwiseBob accepts the alternative hypothesis 1198671 that the identityis illegitimate The channel response ldquodifferencerdquo is recordedas T and (3) can be also written as
119879 = diff (119867119860119861119896+1 119867119860119861119896 )gt 1198671lt 1198670
120578 (4)
As shown in (4) the physical layer authentication isactually a comparison between channel information ldquodif-ferencerdquo and authentication threshold Thus the differencebetween channel information and authentication thresholdis the key of physical layer authentication The test statisticscan measure the similarity of channel information andcalculate the channel information difference In this paperwe use two kinds of test statistic TA and TB respectivelyIn particular assuming Bob obtains two consecutive framechannel response of 119860119861119896minus1119909 and 119860119861119896119909 respectively fromAliceWe build test statistics of 119879119860 and 119879119861 based on the twoframes for the purpose of discrimination identity of Aliceor Eve Subsequently Bob acquires the k+1th frame channelinformation as 119860119861119896+1119909
The test statistics are calculated as
119879119860 (119896) =10038161003816100381610038161003816100381610038161003816100381610038161003816
diff (119860119861119896+1119909 minus 119860119861119896119909)diff (119860119861
119896119909minus 119860119861119896minus1119909
)
10038161003816100381610038161003816100381610038161003816100381610038161003816=1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
sum119873119904119909=1sum119873119898=1sum119873119899=1100381610038161003816100381610038161003816119860119861119896+1119909 (119898 119899) minus 119860119861119896119909 (119898 119899) 119890119895120579(119898119899)
100381610038161003816100381610038161003816sum119873119904119909=1sum119873119898=1sum119873119899=1
10038161003816100381610038161003816119860119861119896119909 (119898 119899) minus 119860119861119896minus1119909 (119898 119899) 119890119895120579(119898119899)10038161003816100381610038161003816
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
gt 1198671lt 1198670
120578119860 (5)
where 120579(119898 119899) is the phase offset and can be denoted by
120579 (119898 119899) = arg (119860119861119896119909 (119898 119899) [119867119883119861119896+1119909 (119898 119899)]lowast) (6)
From (5) 119879119860 can be taken as the difference of thesubcarrier amplitude which avoids the effect of 120579(119898 119899)
Two consecutive data frames 119860119861119896119909 and 119860119861119896+1119909 representmeasurement errors in the phase of the channel responseEach channel response value consists of119873119904 frequency domainchannel matrix which is OFDM symbol of N dimensionalsquare matrix and 119899 denotes the 119898th row and 119899 denotes thecolumn element phase offset
119879119861 (119896) =10038161003816100381610038161003816100381610038161003816100381610038161003816
diff (119860119861119896+1119909 minus 119860119861119896119909)diff (119860119861119896119909 minus 119860119861119896minus1119909)
10038161003816100381610038161003816100381610038161003816100381610038161003816=100381610038161003816100381610038161003816100381610038161003816100381610038161003816
sum119873119904119909=1sum119873119898=1sum119873119899=110038161003816100381610038161003816119883119861119896+1119909 (119898 119899) minus 119860119861119896119909 (119898 119899)
100381610038161003816100381610038162
sum119873119904119909=1sum119873119898=1sum119873119899=110038161003816100381610038161003816119860119861119896119909 (119898 119899) minus 119860119861119896minus1119909 (119898 119899)
100381610038161003816100381610038162
100381610038161003816100381610038161003816100381610038161003816100381610038161003816
gt 1198671lt 1198670
120578119860 (7)
where 119879119861 is the test statistic based on amplitude and phaseinformation We use 119879119860 and 119879119861 as the one-dimensional teststatistic respectively for detecting spoofing attack Unfor-tunately it is hard to find the best threshold for achievinghigh accuracy authentication detection rate To tackle thisproblemwe propose a learning algorithmbased onAdaBoostto achieve physical layer authentication in which 119879119860 and 119879119861are used as training features
3 Physical Authentication withAdaBoost Algorithm
In this section we propose a learning algorithm based onAdaBoost for physical authentication
31 AdaBoost Algorithm AdaBoost is the abbreviation ofadaptive boosting and developed by Yoav Freund [24] and is
4 Wireless Communications and Mobile Computing
71
72
7G
7-
1 (x)
2 (x)
G (x)
- (x)
f(x)=M
summ=1
mGm(x)
Figure 2 AdaBoost algorithm
the most widely used form of boosting algorithm Boostingis a powerful technique combined with base classifiers [25]to produce a form of committee whose performance can besignificantly better than other base classifiers The principalof AdaBoost algorithm is that this algorithm improves itsperformance by the iterative algorithm which is adaptive inthe sense that subsequent weak classifiers called as learnersare adjusted to improve those instances misclassified byprevious classifiers AdaBoost can be seen as a particularmethod of training a boosted classifier A boost classifier isa classifier as follows
119891 (119909) =119872
sum119898=1
120572119898119866119898 (119909) (8)
where each 119866119898(119909) is a weak classifier that takes 119909 as inputand returns a value 119910119898 indicating the class of 119909 The weakclassifiers each of classifiers is trained by using a weightedcoefficient 119908119898119894 from the data set where the weighting coeffi-cient associated depending on the performance of the weakclassifiers such as decision tree (support vector machine)SVM are trained in sequence More specially data pointswhich aremisclassified by one of theweak classifiers are beinggiven greater weight which are used to train the next weakclassifier As illustrated in Figure 2 once all the classifiers have
been trained until there are no misclassified data points thentheir final model is generated via a weight majority votingscheme
32 Physical Authentication with AdaBoost Algorithm Thephysical authentication with AdaBoost algorithm is pro-posed for detection spoofing The performance chart of thealgorithm is illustrated in Figure 3 Bob collects the channelmatrix H1198601198611 which obtained by channel estimation usingthe pilot from Alice and records it When Bob receives thenext data frame from the Alice the Bob collects channelinformation H1198601198612 Similarly Bob collects continuous N-frames channel information from Alice and stores as H119860119861 =[H1198601198611 H1198601198612 H119860119861119873 ] In the same time an Eve sends the dataframes to the Bob and claims that he is Alice In practicalcommunication scenarios we do not know where and whoEves are But in proposed scheme Eves are needed to be testtraining purpose Therefore one or several Eve nodes are setfor this purpose Bob continuously extracts the continuous Nframes channel information from Eve and stores as H119864119861 =[H1198641198611 H1198641198612 H119864119861119873 ]
The data set is preprocessed by Bob Firstly Bob calculatesthe value of data set H119860119861 H119864119861 Secondly Bob calculates thetest statistics based on test statistics 119879119860 119879119861 as
119879119883119861119860 (119896) =10038161003816100381610038161003816100381610038161003816100381610038161003816
diff (119883119861119896+1119909 minus 119883119861119896119909)diff (119883119861
119896119909minus 119883119861119896minus1119909
)
10038161003816100381610038161003816100381610038161003816100381610038161003816=1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
sum119873119904119909=1sum119873119898=1sum119873119899=1100381610038161003816100381610038161003816119883119861119896+1119909 (119898 119899) minus 119883119861119896119909 (119898 119899) 119890119895120579(119898119899)
100381610038161003816100381610038161003816sum119873119904119909=1sum119873119898=1sum119873119899=1
10038161003816100381610038161003816119883119861119896119909 (119898 119899) minus 119883119861119896minus1119909 (119898 119899) 119890119895120579(119898119899)10038161003816100381610038161003816
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
gt 1198671lt 1198670
120578119883119860 (9)
119879119883119861119861 (119896) =10038161003816100381610038161003816100381610038161003816100381610038161003816
diff (119883119861119896+1119909 minus 119883119861119896119909)diff (119883119861
119896119909minus 119883119861119896minus1119909
)
10038161003816100381610038161003816100381610038161003816100381610038161003816=100381610038161003816100381610038161003816100381610038161003816100381610038161003816
sum119873119904119909=1sum119873119898=1sum119873119899=110038161003816100381610038161003816119883119861119896+1119909 (119898 119899) minus 119883119861119896119909 (119898 119899)
100381610038161003816100381610038162
sum119873119904119909=1sum119873119898=1sum119873119899=110038161003816100381610038161003816119883119861119896119909 (119898 119899) minus 119883119861119896minus1119909 (119898 119899)
100381610038161003816100381610038162
100381610038161003816100381610038161003816100381610038161003816100381610038161003816
gt 1198671lt 1198670
120578119883119861 (10)
Finally Bob generates training data set of two categoriesThe first one is
119879119860119861119860 = 1199091 119909119894 119909119873 119910119860 (11a)
119879119860119861119861 = 1199091 119909119894 119909119873 119910119860 (11b)
where 119909119894 isin 119879119860119861119860 (119896) or 119909119894 isin 119879119860119861119861 (119896) 119910119860 = +1 by substitutingH119860119861 into (9) (10) yields 119879119860119861119860 119879119860119861119861 and the value of 119910119860represents that the transmitter is the legal transmitter fromAlice And the second training set is
119879119864119861119860 = 1199091198641 119909119864119894 119909119864119873 119910119864119861 (12a)
Wireless Communications and Mobile Computing 5
Start
Adaboost algorithm is used to generate a strong classifier
Rate of reachingtargeted
The end
Collect the sample of the
legal
Collect the sample of the
illegal
Data preprocessing
Thetesting
set
Thetraining
set
To judge legal or illegal
Collect the sample
Data processing
No Yes
Generate aweak
classifier
Figure 3 Physical authentication with AdaBoost algorithm
119879119864119861119861 = 1199091198641 119909119864119894 119909119864119873 119910119864119861 (12b)
where 119909119864119894 isin 119879119860119864119861(119896) or 119909119864119894 isin 119879119861119864119861(119896) 119910119864119861 = minus1 bysubstituting H119864119861 into (9) and (10) yields 119879119864119861119860 and 119879119864119861119861 andthe value of 119910119894 represents that the transmitter is the illegaltransmitter from Eve Bob uses the two classification trainingdata set 119879119860119861119860 119879119860119861119861 119879119864119861119860 and 119879119864119861119861 as input training set
Spoofing detection is essentially a two-classification prob-lem which is considered to be solved through AdaBoostalgorithm The training data is made up of a bunch of samplepoints Each sample point comprises input sample 119909119894 andlabel 119910119894 where 119910119894 isin minus1 1 Each sample point is given anassociated weight parameter 119908119898119894 119898 means 119898-th trainingand 119894means the number of sample points which is initially set1119894 for all sample pointsWe suppose that we have a procedureavailable for training a weak classifier using weighted samplepoints At each iteration of the training process AdaBoosttrains a new weak classifier by using the sample points inwhich the weighting coefficients are adjusted according to theperformance of the previously trained weak classifier so as togive greater weight to the misclassified data points in whichthe classification error rate 119890119898 is used to evaluate misclassifieddata set119863119898
119890119898 = 119875 (119866119898 (119909119894) = 119910119894) =2119905
sum119894=1
119908119898119894119868 (119866119898 (119909119894) = 119910119894) (13)
Then the coefficient 120572119898 of 119866119898 is calculate as
120572119898 =12 log
1 minus 119890119898119890119898
(14)
Finally we generate a final model that different weight isbeing given to different weak classifiers in (8) The AdaBoostalgorithm is given as in Algorithm 2 in which the point ofthe training data can be doubled by combining with the one-dimension test statistics 119879119860 and 119879119861 together and becomea new two-dimensional features authentication model forspoofing detection Therefore in the AdaBoost algorithmthe input training data set T is following two optionalsets
(1)One-dimension test statistics training data set
119879 = 119879119860119861119860 119879119864119861119860
or 119879 = 119879119860119861119861 119879119864119861119861 (15)
(2) Two-dimension test statistics training data set
119879 = (119879119860119861119860 119879119860119861119861 ) (119879119864119861119860 119879119864119861119861 ) (16)
6 Wireless Communications and Mobile Computing
InputThe channel information of legal transmitter or illgal transmitterProcess1 Bob calculates the value of data set H119860119861 and H119864119861 from Alice and simulated Eve
H119860119861 = [H1198601198611 H1198601198612 H119860119861119873 ]H119864119861 = [H1198641198611 H1198641198612 H119864119861119873 ]
2 The data set are preprocessed by Bob3 The data set are divided into two parts and the one is training data set and the other is testing data set4 Use training data set to get the weak classifier5 Use the Adaboost algorithm to generate a strong classifer6 The testing data set is used to verify whether the claasifier can achieve the target detection rate otherwise it will return to
the first step7 The final classifier is the authenticaton decision model which can judge whether the new packets are legitimate or illegal
End
Algorithm 1 Physical authentication
Inputtraining data set 119879Process1 Initialize the weight distribution of the sample points1198631 = (11990811 1199081119894 11990812119905) 1199081119894 =
12119905 119894 = 1 2 2119905
2 for 119898 = 1 to119872 do119898means119898-th training3 Use the training data set of 119863119898 to learn and get the weak classifier
119866119898(119909) 119909119894 997888rarr minus1 +14 Calculate the classification error rate of119863119898 on the training data set
119890119898 = 119875(119866119898(119909119894) = 119910119894) =2119905
sum119894=1
119908119898119894119868 (119866119898 (119909119894) = 119910119894)5 Calculate the coefficient of 119866119898
120572119898 =12 log
1 minus 119890119898119890119898
6 Update the weight distribution of the training data set119863119898+1 = (119908119898+11 119908119898+1119894 119908119898+12119905)119908119898+1119894 =
119908119898119894119885119898
exp (minus120572119898119910119894119866119898 (119909119894)) 119894 = 1 2 2119905
119885119898 =2119905
sum119894=1
119908119898119894 exp (minus120572119898119910119894119866119898 (119909119894))7 Construct a linear combination of weak classifiers
119891(119909) =119872
sum119894=1
120572119898119866119898(119909)End for
return 119866(119909) = sign(119891(119909))
Algorithm 2 AdaBoost
4 Experimental Verification
In this section we will describe the system setup and the testprocess of measuring the Algorithm 1 for detecting Alice andEve
41 System Setup We consider the spoofing detection ofa receiver called Bob the legal transmitter called Aliceand the spoofing node called Eve They were placed inthree separate locations in a room surrounded by manyother devices such as printers desktops and other types of
equipment as shown in Figure 4 There are scattering andrefraction phenomena in the room due to the presence ofobstacles in the wireless channel from Alice to Bob andEve to Bob As shown in Figure 5 we set up experimentalplatform which implemented on USRPs and experimentswere performed in an indoor environment Bob is equippedwith an 8lowast8 MIMO system Alice is equipped with a 2lowast2MIMO system and the spoofing node called Eve is equippedwith a 2lowast2 MIMO system The signals are sent over 2antennas each at center frequency 35GHz with bandwidth2MHz
Wireless Communications and Mobile Computing 7
Bob
Eve
AliceB
devices
devicesOther
Other
5 m1 20 43
1
3
2
m
Figure 4 The experiments consisted of Alice Bob and Eve
Figure 5 Real MIMO communication platform consisted of Alice Bob and Eve
42 Experiment In the experiment the following steps aretaken
Step 1 Bob extracts channel information from Alice and Eveby the existing channel estimation mechanisms respectively
Step 2 Bob preprocesses the dataset according to (5) (7) (9)and (10)while the threshold is between [0 1] (normalization)
Step 3 Bob generates a training data set of two classificationsaccording to (11a) (11b) (12a) and (12b)
Step 4 The two classification training data set T is generatedaccording to (15) or (16)
Step 5 Bob is trained to generate a strong classifier based onthe training data set of two classifications by using AdaBoostalgorithm under Matlab program
Step 6 Bob uses a strong classifier to judge the test set andobtain the authentication detection rate
In the experiment we consider that the collection framesare five hundred frames and the value of test statisticwas normalized between 0 and 1 The test statistic 119879119860 ofchannel information of the Alice and Bob as a function offrames is shown in Figure 6(a) in which the red pointsis 119879119860(119896) in (5) and green points is 119879119864119860(119896) in (9) As canbe seen there is the overlapped area Meanwhile fromFigure 6(b) the overlapped area is large when we chosethe test statistic 119879119861 of channel information in which thered points is 119879119860(119896) in (7) and green points is 119879119864119860(119896) in(10) It is clearly shown that it is difficult to acquire thebest manual test threshold for the accuracy of authentica-tion Moreover we use 119879119860 119879119861 and the number of framesrespectively to draw a three-dimensional plot As shown in
8 Wireless Communications and Mobile Computing
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
EveAlice
Nor
mal
ized
4
Valu
e
(a) Normalized 119879119860 of Alice and Eve
EveAlice
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
Nor
mal
ized
4
Valu
e
(b) Normalized 119879119861 of Alice and Eve
Figure 6 Normalized 119879119860 and 119879119861 value of the legal transmitter Alice and the spoofing node Eve for spoofing detection with center frequency35GHz with bandwidth 2MHz
01
02
08
04
500450
06
06 400350
08
300
Frames04 250
1
20015002 100500 0
EveAlice
Nor
mal
ized
4
Valu
e
Normalized 4 Value
Figure 7 Normalized 119879119860 value and 119879119861 value of the legal transmitterAlice and the spoofing node Eve drawing three dimensional plot
Figure 7 obviously it is hard to use the traditional manualthreshold method to identify the identity of data sets inthe three-dimensional condition Howevermachine learningalgorithm based the authentication model can effectivelysettle this problem and a dividing curved surface can per-form the identification by the AdaBoost adaptive adjustmentalgorithm
43 Simulation Results In this section simulation results areprovided to demonstrate the performance of the proposedauthentication scheme
Test Threshold
0
01
02
03
04
05
06
07
08
09
1
0 01 02 03 04 05 06 07 08 09 1
Det
ectio
n Ra
te
4
4
Figure 8 Correct classification rate of 119879119860 and 119879119861
As a comparison we considered the PHY-layer spoofingdetection [15] with a varied test threshold From the Figure 8we can see that when test threshold equals 04 the bestauthentication detection rate results of using119879119860 or119879119861 reached798 and 654 respectively In addition our proposedmethod which combined two test statistics 119879119860 and 119879119861 as atwo-dimensional feature can improve the accuracy of detec-tion We use 119879119860 119879119861 and the number of frames respectivelyto draw a three-dimensional plot Figure 9 illustrates thecomparison of spoofing detection among the three methodsfrom which we can conclude that manual threshold methodbased on 119879119860 test statistics can achieve 798 detection rate
Wireless Communications and Mobile Computing 9
0
01
02
03
04
05
06
07
08
09
1
Det
ectio
n Ra
te
ML based 4 data setManual Method
MethodManual Method
ML based 4 and4 data sets
Machine Learining based 4 data set Machine Learning based 4 and4 data sets
Figure 9 The simulation result with the different method ofauthentication scheme
while machine learning based authentication method with119879119860 test statistic can acquire 871 detection rate and machinelearning based authentication method with two-dimensionalfeatures 119879119860 and 119879119861 can achieve 913 accuracy rate with anadditional 10 more computation complexity
To sum up the proposed authentication scheme achievesa superior performance over manual threshold strategy [15]Based on the above observation the proposed machinelearning based authentication scheme with tow-dimensionalfeature not only exhibits excellent performance than manualmethod but also has higher authentication rate than that ofthe same algorithm with one-dimensional feature
5 Conclusions
In this paper machine learning algorithm based physical-layer channel authentication for the 5G wireless communica-tion security is proposed A machine learning authenticationmethod could draw a conclusion whether the received pack-ets are from a legitimate transmitter or from a counterfeiterby using one-dimension or two-dimensional joint featuresThe effectiveness of the proposed authentication scheme isvalidated by widely simulations All the data used in thesimulation are derived from real OFDM-MIMO commu-nication platform which provides a real communicationenvironment Moreover the authentication results show thatthe novel methods provide a higher rate in detecting thespoofing attacks than those of the manual threshold basedphysical layer authentication schemes The training of theclassifier can be done offlineTherefore the novel method canperform authentication fast In addition whether we can usemore machine learning algorithms to further optimize ourauthentication model and find a better statistical test of largedifference in channel information is issue that we need to dealwith in the future
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare no conflicts of interest
Acknowledgments
This research was supported by NSFC (no 61572114)Sichuan Sci amp Tech Achievements Transformation Project(no 2016CC003) Sichuan Sci amp Tech Service Develop-ment Project (no 18KJFWSF0368) Hunan Provincial NatureScience Foundation Project 2018JJ2535 Chile CONICYTFONDECYTRegular Project 1181809 andNational Key RampDProgramofChina (2018YFB0904900 and 2018YFB0904905)
References
[1] M Agiwal A Roy andN Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys amp Tutorials vol 18 no 3 pp 1617ndash1655 2016
[2] J Thompson X Ge and H-C Wu ldquo5G wireless communica-tion systems prospects and challenges [Guest Editorial]rdquo IEEECommunications Magazine vol 52 no 2 pp 62ndash64 2014
[3] C Perera C H Liu S Jayawardena and M Chen ldquoA surveyon internet of things from industrial market perspectiverdquo IEEEAccess vol 2 pp 1660ndash1679 2014
[4] A Bogdanov M Knezevic G Leander D Toz K Varıcı andI Verbauwhede ldquoSPONGENT the design space of lightweightcryptographic hashingrdquo IEEE Transactions on Computers vol62 no 10 pp 2041ndash2053 2013
[5] R Zhang L Zhu C Xu and Y Yi ldquoAn Efficient and secureRFID batch authentication protocol with group tags ownershipTransferrdquo IEEE Collaboration and Internet Computing pp 168ndash175 2015
[6] L Hu H Wen B Wu et al ldquoCooperative Jamming forPhysical Layer Security Enhancement in Internet of ThingsrdquoIEEE Internet of ings Journal vol 5 no 1 pp 219ndash228 2018
[7] L Hu H Wen B Wu J Tang F Pan and R-F LiaoldquoCooperative-Jamming-Aided Secrecy Enhancement in Wire-less Networks with Passive Eavesdroppersrdquo IEEE Transactionson Vehicular Technology vol 67 no 3 pp 2108ndash2117 2018
[8] H Wen Physical layer approaches for securing wireless commu-nication systems Springer New York NY USA 2013
[9] LHuHWen BWu J Tang and F Pan ldquoAdaptive Base StationCooperation for Physical Layer Security in Two-Cell WirelessNetworksrdquo IEEE Access vol 4 pp 5607ndash5623 2016
[10] L Hu H Wen B Wu J Tang and F Pan ldquoAdaptive SecureTransmission for Physical Layer Security in Cooperative Wire-less Networksrdquo IEEE Communications Letters vol 21 no 3 pp524ndash527 2017
[11] D B Faria andDRCheriton ldquoDetecting identity-based attacksin wireless networks using signalprintsrdquo in Proceedings of theACM Workshop on Wireless Security pp 43ndash52 Los AngelesCalif USA 2006
[12] M Demirbas and Y Song ldquoAn RSSI-based scheme for sybilattack detection in wireless sensor networksrdquo in Proceedings of
10 Wireless Communications and Mobile Computing
the WoWMoM 2006 2006 International Symposium on aWorldof Wireless Mobile andMultimedia Networks pp 564ndash568 June2006
[13] Y ChenW Trappe and R P Martin ldquoDetecting and localizingwireless spoofing attacksrdquo in Proceedings of the 2007 4th AnnualIEEE Communications Society Conference on Sensor Mesh andAd Hoc Communications and Networks SECON pp 193ndash202San Diego Calif USA June 2007
[14] N Patwari and S K Kasera ldquoRobust location distinction usingtemporal link signaturesrdquo in Proceedings of the ACM Interna-tional Conference on Mobile Computing and NETWORKINGpp 111ndash122 2007
[15] H Wen Y Wang X Zhu J Li and L Zhou ldquoPhysical layerassist authentication technique for smart meter systemrdquo IETCommunications vol 7 no 3 pp 189ndash197 2013
[16] J K Tugnait ldquoWireless user authentication via comparison ofpower spectral densitiesrdquo IEEE Journal on Selected Areas inCommunications vol 31 no 9 pp 1791ndash1802 2013
[17] Z Jiang J Zhao X Li J Han and W Xi ldquoRejecting the attackSource authentication for Wi-Fi management frames using CSIInformationrdquo in Proceedings of the IEEE INFOCOM 2013 -IEEE Conference on Computer Communications pp 2544ndash2552Turin Italy April 2013
[18] X Wu and Z Yang ldquoPhysical-layer authentication for multi-carrier transmissionrdquo IEEE Communications Letters vol 19 no1 pp 74ndash77 2015
[19] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-authentication protocol for spoofing detection in wireless net-worksrdquo IEEE Transactions on Vehicular Technology vol 65 no12 pp 10037ndash10047 2016
[20] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-LayerSpoofing Detection with Reinforcement Learning in WirelessNetworksrdquo IEEE Transactions on Vehicular Technology vol 65no 12 pp 10037ndash10048 2016
[21] P Hao X Wang and A Refaey ldquoAn enhanced cross-layerauthentication mechanism for wireless communications basedon PER and RSSIrdquo in Proceedings of the 2013 13th CanadianWorkshop on Information eory CWIT 2013 pp 44ndash48Canada June 2013
[22] S Chen et al ldquoMachine-to-Machine communications in ultra-dense networksmdashA surveyrdquo IEEE Communications Surveys ampTutorials vol 1 no 1 99 pages 2017
[23] L Xiao L J Greenstein N B Mandayam and W TrappeldquoChannel-based detection of sybil attacks in wireless networksrdquoIEEE Transactions on Information Forensics and Security vol 4no 3 pp 492ndash503 2009
[24] H Wen P-H Ho C Qi and G Gong ldquoPhysical layer assistedauthentication for distributed ad hoc wireless sensor networksrdquoIET Information Security vol 4 no 4 pp 390ndash396 2010
[25] J H Friedman ldquoGreedy function approximation a gradientboosting machinerdquo Annals of Statistics vol 29 no 5 pp 1189ndash1232 2001
[26] J Friedman T Hastie and R Tibshirani ldquoAdditive logisticregression a statistical view of boostingrdquo e Annals of Statis-tics vol 28 no 2 pp 337ndash407 2000
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
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
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
4 Wireless Communications and Mobile Computing
71
72
7G
7-
1 (x)
2 (x)
G (x)
- (x)
f(x)=M
summ=1
mGm(x)
Figure 2 AdaBoost algorithm
the most widely used form of boosting algorithm Boostingis a powerful technique combined with base classifiers [25]to produce a form of committee whose performance can besignificantly better than other base classifiers The principalof AdaBoost algorithm is that this algorithm improves itsperformance by the iterative algorithm which is adaptive inthe sense that subsequent weak classifiers called as learnersare adjusted to improve those instances misclassified byprevious classifiers AdaBoost can be seen as a particularmethod of training a boosted classifier A boost classifier isa classifier as follows
119891 (119909) =119872
sum119898=1
120572119898119866119898 (119909) (8)
where each 119866119898(119909) is a weak classifier that takes 119909 as inputand returns a value 119910119898 indicating the class of 119909 The weakclassifiers each of classifiers is trained by using a weightedcoefficient 119908119898119894 from the data set where the weighting coeffi-cient associated depending on the performance of the weakclassifiers such as decision tree (support vector machine)SVM are trained in sequence More specially data pointswhich aremisclassified by one of theweak classifiers are beinggiven greater weight which are used to train the next weakclassifier As illustrated in Figure 2 once all the classifiers have
been trained until there are no misclassified data points thentheir final model is generated via a weight majority votingscheme
32 Physical Authentication with AdaBoost Algorithm Thephysical authentication with AdaBoost algorithm is pro-posed for detection spoofing The performance chart of thealgorithm is illustrated in Figure 3 Bob collects the channelmatrix H1198601198611 which obtained by channel estimation usingthe pilot from Alice and records it When Bob receives thenext data frame from the Alice the Bob collects channelinformation H1198601198612 Similarly Bob collects continuous N-frames channel information from Alice and stores as H119860119861 =[H1198601198611 H1198601198612 H119860119861119873 ] In the same time an Eve sends the dataframes to the Bob and claims that he is Alice In practicalcommunication scenarios we do not know where and whoEves are But in proposed scheme Eves are needed to be testtraining purpose Therefore one or several Eve nodes are setfor this purpose Bob continuously extracts the continuous Nframes channel information from Eve and stores as H119864119861 =[H1198641198611 H1198641198612 H119864119861119873 ]
The data set is preprocessed by Bob Firstly Bob calculatesthe value of data set H119860119861 H119864119861 Secondly Bob calculates thetest statistics based on test statistics 119879119860 119879119861 as
119879119883119861119860 (119896) =10038161003816100381610038161003816100381610038161003816100381610038161003816
diff (119883119861119896+1119909 minus 119883119861119896119909)diff (119883119861
119896119909minus 119883119861119896minus1119909
)
10038161003816100381610038161003816100381610038161003816100381610038161003816=1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
sum119873119904119909=1sum119873119898=1sum119873119899=1100381610038161003816100381610038161003816119883119861119896+1119909 (119898 119899) minus 119883119861119896119909 (119898 119899) 119890119895120579(119898119899)
100381610038161003816100381610038161003816sum119873119904119909=1sum119873119898=1sum119873119899=1
10038161003816100381610038161003816119883119861119896119909 (119898 119899) minus 119883119861119896minus1119909 (119898 119899) 119890119895120579(119898119899)10038161003816100381610038161003816
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
gt 1198671lt 1198670
120578119883119860 (9)
119879119883119861119861 (119896) =10038161003816100381610038161003816100381610038161003816100381610038161003816
diff (119883119861119896+1119909 minus 119883119861119896119909)diff (119883119861
119896119909minus 119883119861119896minus1119909
)
10038161003816100381610038161003816100381610038161003816100381610038161003816=100381610038161003816100381610038161003816100381610038161003816100381610038161003816
sum119873119904119909=1sum119873119898=1sum119873119899=110038161003816100381610038161003816119883119861119896+1119909 (119898 119899) minus 119883119861119896119909 (119898 119899)
100381610038161003816100381610038162
sum119873119904119909=1sum119873119898=1sum119873119899=110038161003816100381610038161003816119883119861119896119909 (119898 119899) minus 119883119861119896minus1119909 (119898 119899)
100381610038161003816100381610038162
100381610038161003816100381610038161003816100381610038161003816100381610038161003816
gt 1198671lt 1198670
120578119883119861 (10)
Finally Bob generates training data set of two categoriesThe first one is
119879119860119861119860 = 1199091 119909119894 119909119873 119910119860 (11a)
119879119860119861119861 = 1199091 119909119894 119909119873 119910119860 (11b)
where 119909119894 isin 119879119860119861119860 (119896) or 119909119894 isin 119879119860119861119861 (119896) 119910119860 = +1 by substitutingH119860119861 into (9) (10) yields 119879119860119861119860 119879119860119861119861 and the value of 119910119860represents that the transmitter is the legal transmitter fromAlice And the second training set is
119879119864119861119860 = 1199091198641 119909119864119894 119909119864119873 119910119864119861 (12a)
Wireless Communications and Mobile Computing 5
Start
Adaboost algorithm is used to generate a strong classifier
Rate of reachingtargeted
The end
Collect the sample of the
legal
Collect the sample of the
illegal
Data preprocessing
Thetesting
set
Thetraining
set
To judge legal or illegal
Collect the sample
Data processing
No Yes
Generate aweak
classifier
Figure 3 Physical authentication with AdaBoost algorithm
119879119864119861119861 = 1199091198641 119909119864119894 119909119864119873 119910119864119861 (12b)
where 119909119864119894 isin 119879119860119864119861(119896) or 119909119864119894 isin 119879119861119864119861(119896) 119910119864119861 = minus1 bysubstituting H119864119861 into (9) and (10) yields 119879119864119861119860 and 119879119864119861119861 andthe value of 119910119894 represents that the transmitter is the illegaltransmitter from Eve Bob uses the two classification trainingdata set 119879119860119861119860 119879119860119861119861 119879119864119861119860 and 119879119864119861119861 as input training set
Spoofing detection is essentially a two-classification prob-lem which is considered to be solved through AdaBoostalgorithm The training data is made up of a bunch of samplepoints Each sample point comprises input sample 119909119894 andlabel 119910119894 where 119910119894 isin minus1 1 Each sample point is given anassociated weight parameter 119908119898119894 119898 means 119898-th trainingand 119894means the number of sample points which is initially set1119894 for all sample pointsWe suppose that we have a procedureavailable for training a weak classifier using weighted samplepoints At each iteration of the training process AdaBoosttrains a new weak classifier by using the sample points inwhich the weighting coefficients are adjusted according to theperformance of the previously trained weak classifier so as togive greater weight to the misclassified data points in whichthe classification error rate 119890119898 is used to evaluate misclassifieddata set119863119898
119890119898 = 119875 (119866119898 (119909119894) = 119910119894) =2119905
sum119894=1
119908119898119894119868 (119866119898 (119909119894) = 119910119894) (13)
Then the coefficient 120572119898 of 119866119898 is calculate as
120572119898 =12 log
1 minus 119890119898119890119898
(14)
Finally we generate a final model that different weight isbeing given to different weak classifiers in (8) The AdaBoostalgorithm is given as in Algorithm 2 in which the point ofthe training data can be doubled by combining with the one-dimension test statistics 119879119860 and 119879119861 together and becomea new two-dimensional features authentication model forspoofing detection Therefore in the AdaBoost algorithmthe input training data set T is following two optionalsets
(1)One-dimension test statistics training data set
119879 = 119879119860119861119860 119879119864119861119860
or 119879 = 119879119860119861119861 119879119864119861119861 (15)
(2) Two-dimension test statistics training data set
119879 = (119879119860119861119860 119879119860119861119861 ) (119879119864119861119860 119879119864119861119861 ) (16)
6 Wireless Communications and Mobile Computing
InputThe channel information of legal transmitter or illgal transmitterProcess1 Bob calculates the value of data set H119860119861 and H119864119861 from Alice and simulated Eve
H119860119861 = [H1198601198611 H1198601198612 H119860119861119873 ]H119864119861 = [H1198641198611 H1198641198612 H119864119861119873 ]
2 The data set are preprocessed by Bob3 The data set are divided into two parts and the one is training data set and the other is testing data set4 Use training data set to get the weak classifier5 Use the Adaboost algorithm to generate a strong classifer6 The testing data set is used to verify whether the claasifier can achieve the target detection rate otherwise it will return to
the first step7 The final classifier is the authenticaton decision model which can judge whether the new packets are legitimate or illegal
End
Algorithm 1 Physical authentication
Inputtraining data set 119879Process1 Initialize the weight distribution of the sample points1198631 = (11990811 1199081119894 11990812119905) 1199081119894 =
12119905 119894 = 1 2 2119905
2 for 119898 = 1 to119872 do119898means119898-th training3 Use the training data set of 119863119898 to learn and get the weak classifier
119866119898(119909) 119909119894 997888rarr minus1 +14 Calculate the classification error rate of119863119898 on the training data set
119890119898 = 119875(119866119898(119909119894) = 119910119894) =2119905
sum119894=1
119908119898119894119868 (119866119898 (119909119894) = 119910119894)5 Calculate the coefficient of 119866119898
120572119898 =12 log
1 minus 119890119898119890119898
6 Update the weight distribution of the training data set119863119898+1 = (119908119898+11 119908119898+1119894 119908119898+12119905)119908119898+1119894 =
119908119898119894119885119898
exp (minus120572119898119910119894119866119898 (119909119894)) 119894 = 1 2 2119905
119885119898 =2119905
sum119894=1
119908119898119894 exp (minus120572119898119910119894119866119898 (119909119894))7 Construct a linear combination of weak classifiers
119891(119909) =119872
sum119894=1
120572119898119866119898(119909)End for
return 119866(119909) = sign(119891(119909))
Algorithm 2 AdaBoost
4 Experimental Verification
In this section we will describe the system setup and the testprocess of measuring the Algorithm 1 for detecting Alice andEve
41 System Setup We consider the spoofing detection ofa receiver called Bob the legal transmitter called Aliceand the spoofing node called Eve They were placed inthree separate locations in a room surrounded by manyother devices such as printers desktops and other types of
equipment as shown in Figure 4 There are scattering andrefraction phenomena in the room due to the presence ofobstacles in the wireless channel from Alice to Bob andEve to Bob As shown in Figure 5 we set up experimentalplatform which implemented on USRPs and experimentswere performed in an indoor environment Bob is equippedwith an 8lowast8 MIMO system Alice is equipped with a 2lowast2MIMO system and the spoofing node called Eve is equippedwith a 2lowast2 MIMO system The signals are sent over 2antennas each at center frequency 35GHz with bandwidth2MHz
Wireless Communications and Mobile Computing 7
Bob
Eve
AliceB
devices
devicesOther
Other
5 m1 20 43
1
3
2
m
Figure 4 The experiments consisted of Alice Bob and Eve
Figure 5 Real MIMO communication platform consisted of Alice Bob and Eve
42 Experiment In the experiment the following steps aretaken
Step 1 Bob extracts channel information from Alice and Eveby the existing channel estimation mechanisms respectively
Step 2 Bob preprocesses the dataset according to (5) (7) (9)and (10)while the threshold is between [0 1] (normalization)
Step 3 Bob generates a training data set of two classificationsaccording to (11a) (11b) (12a) and (12b)
Step 4 The two classification training data set T is generatedaccording to (15) or (16)
Step 5 Bob is trained to generate a strong classifier based onthe training data set of two classifications by using AdaBoostalgorithm under Matlab program
Step 6 Bob uses a strong classifier to judge the test set andobtain the authentication detection rate
In the experiment we consider that the collection framesare five hundred frames and the value of test statisticwas normalized between 0 and 1 The test statistic 119879119860 ofchannel information of the Alice and Bob as a function offrames is shown in Figure 6(a) in which the red pointsis 119879119860(119896) in (5) and green points is 119879119864119860(119896) in (9) As canbe seen there is the overlapped area Meanwhile fromFigure 6(b) the overlapped area is large when we chosethe test statistic 119879119861 of channel information in which thered points is 119879119860(119896) in (7) and green points is 119879119864119860(119896) in(10) It is clearly shown that it is difficult to acquire thebest manual test threshold for the accuracy of authentica-tion Moreover we use 119879119860 119879119861 and the number of framesrespectively to draw a three-dimensional plot As shown in
8 Wireless Communications and Mobile Computing
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
EveAlice
Nor
mal
ized
4
Valu
e
(a) Normalized 119879119860 of Alice and Eve
EveAlice
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
Nor
mal
ized
4
Valu
e
(b) Normalized 119879119861 of Alice and Eve
Figure 6 Normalized 119879119860 and 119879119861 value of the legal transmitter Alice and the spoofing node Eve for spoofing detection with center frequency35GHz with bandwidth 2MHz
01
02
08
04
500450
06
06 400350
08
300
Frames04 250
1
20015002 100500 0
EveAlice
Nor
mal
ized
4
Valu
e
Normalized 4 Value
Figure 7 Normalized 119879119860 value and 119879119861 value of the legal transmitterAlice and the spoofing node Eve drawing three dimensional plot
Figure 7 obviously it is hard to use the traditional manualthreshold method to identify the identity of data sets inthe three-dimensional condition Howevermachine learningalgorithm based the authentication model can effectivelysettle this problem and a dividing curved surface can per-form the identification by the AdaBoost adaptive adjustmentalgorithm
43 Simulation Results In this section simulation results areprovided to demonstrate the performance of the proposedauthentication scheme
Test Threshold
0
01
02
03
04
05
06
07
08
09
1
0 01 02 03 04 05 06 07 08 09 1
Det
ectio
n Ra
te
4
4
Figure 8 Correct classification rate of 119879119860 and 119879119861
As a comparison we considered the PHY-layer spoofingdetection [15] with a varied test threshold From the Figure 8we can see that when test threshold equals 04 the bestauthentication detection rate results of using119879119860 or119879119861 reached798 and 654 respectively In addition our proposedmethod which combined two test statistics 119879119860 and 119879119861 as atwo-dimensional feature can improve the accuracy of detec-tion We use 119879119860 119879119861 and the number of frames respectivelyto draw a three-dimensional plot Figure 9 illustrates thecomparison of spoofing detection among the three methodsfrom which we can conclude that manual threshold methodbased on 119879119860 test statistics can achieve 798 detection rate
Wireless Communications and Mobile Computing 9
0
01
02
03
04
05
06
07
08
09
1
Det
ectio
n Ra
te
ML based 4 data setManual Method
MethodManual Method
ML based 4 and4 data sets
Machine Learining based 4 data set Machine Learning based 4 and4 data sets
Figure 9 The simulation result with the different method ofauthentication scheme
while machine learning based authentication method with119879119860 test statistic can acquire 871 detection rate and machinelearning based authentication method with two-dimensionalfeatures 119879119860 and 119879119861 can achieve 913 accuracy rate with anadditional 10 more computation complexity
To sum up the proposed authentication scheme achievesa superior performance over manual threshold strategy [15]Based on the above observation the proposed machinelearning based authentication scheme with tow-dimensionalfeature not only exhibits excellent performance than manualmethod but also has higher authentication rate than that ofthe same algorithm with one-dimensional feature
5 Conclusions
In this paper machine learning algorithm based physical-layer channel authentication for the 5G wireless communica-tion security is proposed A machine learning authenticationmethod could draw a conclusion whether the received pack-ets are from a legitimate transmitter or from a counterfeiterby using one-dimension or two-dimensional joint featuresThe effectiveness of the proposed authentication scheme isvalidated by widely simulations All the data used in thesimulation are derived from real OFDM-MIMO commu-nication platform which provides a real communicationenvironment Moreover the authentication results show thatthe novel methods provide a higher rate in detecting thespoofing attacks than those of the manual threshold basedphysical layer authentication schemes The training of theclassifier can be done offlineTherefore the novel method canperform authentication fast In addition whether we can usemore machine learning algorithms to further optimize ourauthentication model and find a better statistical test of largedifference in channel information is issue that we need to dealwith in the future
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare no conflicts of interest
Acknowledgments
This research was supported by NSFC (no 61572114)Sichuan Sci amp Tech Achievements Transformation Project(no 2016CC003) Sichuan Sci amp Tech Service Develop-ment Project (no 18KJFWSF0368) Hunan Provincial NatureScience Foundation Project 2018JJ2535 Chile CONICYTFONDECYTRegular Project 1181809 andNational Key RampDProgramofChina (2018YFB0904900 and 2018YFB0904905)
References
[1] M Agiwal A Roy andN Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys amp Tutorials vol 18 no 3 pp 1617ndash1655 2016
[2] J Thompson X Ge and H-C Wu ldquo5G wireless communica-tion systems prospects and challenges [Guest Editorial]rdquo IEEECommunications Magazine vol 52 no 2 pp 62ndash64 2014
[3] C Perera C H Liu S Jayawardena and M Chen ldquoA surveyon internet of things from industrial market perspectiverdquo IEEEAccess vol 2 pp 1660ndash1679 2014
[4] A Bogdanov M Knezevic G Leander D Toz K Varıcı andI Verbauwhede ldquoSPONGENT the design space of lightweightcryptographic hashingrdquo IEEE Transactions on Computers vol62 no 10 pp 2041ndash2053 2013
[5] R Zhang L Zhu C Xu and Y Yi ldquoAn Efficient and secureRFID batch authentication protocol with group tags ownershipTransferrdquo IEEE Collaboration and Internet Computing pp 168ndash175 2015
[6] L Hu H Wen B Wu et al ldquoCooperative Jamming forPhysical Layer Security Enhancement in Internet of ThingsrdquoIEEE Internet of ings Journal vol 5 no 1 pp 219ndash228 2018
[7] L Hu H Wen B Wu J Tang F Pan and R-F LiaoldquoCooperative-Jamming-Aided Secrecy Enhancement in Wire-less Networks with Passive Eavesdroppersrdquo IEEE Transactionson Vehicular Technology vol 67 no 3 pp 2108ndash2117 2018
[8] H Wen Physical layer approaches for securing wireless commu-nication systems Springer New York NY USA 2013
[9] LHuHWen BWu J Tang and F Pan ldquoAdaptive Base StationCooperation for Physical Layer Security in Two-Cell WirelessNetworksrdquo IEEE Access vol 4 pp 5607ndash5623 2016
[10] L Hu H Wen B Wu J Tang and F Pan ldquoAdaptive SecureTransmission for Physical Layer Security in Cooperative Wire-less Networksrdquo IEEE Communications Letters vol 21 no 3 pp524ndash527 2017
[11] D B Faria andDRCheriton ldquoDetecting identity-based attacksin wireless networks using signalprintsrdquo in Proceedings of theACM Workshop on Wireless Security pp 43ndash52 Los AngelesCalif USA 2006
[12] M Demirbas and Y Song ldquoAn RSSI-based scheme for sybilattack detection in wireless sensor networksrdquo in Proceedings of
10 Wireless Communications and Mobile Computing
the WoWMoM 2006 2006 International Symposium on aWorldof Wireless Mobile andMultimedia Networks pp 564ndash568 June2006
[13] Y ChenW Trappe and R P Martin ldquoDetecting and localizingwireless spoofing attacksrdquo in Proceedings of the 2007 4th AnnualIEEE Communications Society Conference on Sensor Mesh andAd Hoc Communications and Networks SECON pp 193ndash202San Diego Calif USA June 2007
[14] N Patwari and S K Kasera ldquoRobust location distinction usingtemporal link signaturesrdquo in Proceedings of the ACM Interna-tional Conference on Mobile Computing and NETWORKINGpp 111ndash122 2007
[15] H Wen Y Wang X Zhu J Li and L Zhou ldquoPhysical layerassist authentication technique for smart meter systemrdquo IETCommunications vol 7 no 3 pp 189ndash197 2013
[16] J K Tugnait ldquoWireless user authentication via comparison ofpower spectral densitiesrdquo IEEE Journal on Selected Areas inCommunications vol 31 no 9 pp 1791ndash1802 2013
[17] Z Jiang J Zhao X Li J Han and W Xi ldquoRejecting the attackSource authentication for Wi-Fi management frames using CSIInformationrdquo in Proceedings of the IEEE INFOCOM 2013 -IEEE Conference on Computer Communications pp 2544ndash2552Turin Italy April 2013
[18] X Wu and Z Yang ldquoPhysical-layer authentication for multi-carrier transmissionrdquo IEEE Communications Letters vol 19 no1 pp 74ndash77 2015
[19] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-authentication protocol for spoofing detection in wireless net-worksrdquo IEEE Transactions on Vehicular Technology vol 65 no12 pp 10037ndash10047 2016
[20] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-LayerSpoofing Detection with Reinforcement Learning in WirelessNetworksrdquo IEEE Transactions on Vehicular Technology vol 65no 12 pp 10037ndash10048 2016
[21] P Hao X Wang and A Refaey ldquoAn enhanced cross-layerauthentication mechanism for wireless communications basedon PER and RSSIrdquo in Proceedings of the 2013 13th CanadianWorkshop on Information eory CWIT 2013 pp 44ndash48Canada June 2013
[22] S Chen et al ldquoMachine-to-Machine communications in ultra-dense networksmdashA surveyrdquo IEEE Communications Surveys ampTutorials vol 1 no 1 99 pages 2017
[23] L Xiao L J Greenstein N B Mandayam and W TrappeldquoChannel-based detection of sybil attacks in wireless networksrdquoIEEE Transactions on Information Forensics and Security vol 4no 3 pp 492ndash503 2009
[24] H Wen P-H Ho C Qi and G Gong ldquoPhysical layer assistedauthentication for distributed ad hoc wireless sensor networksrdquoIET Information Security vol 4 no 4 pp 390ndash396 2010
[25] J H Friedman ldquoGreedy function approximation a gradientboosting machinerdquo Annals of Statistics vol 29 no 5 pp 1189ndash1232 2001
[26] J Friedman T Hastie and R Tibshirani ldquoAdditive logisticregression a statistical view of boostingrdquo e Annals of Statis-tics vol 28 no 2 pp 337ndash407 2000
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
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
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 5
Start
Adaboost algorithm is used to generate a strong classifier
Rate of reachingtargeted
The end
Collect the sample of the
legal
Collect the sample of the
illegal
Data preprocessing
Thetesting
set
Thetraining
set
To judge legal or illegal
Collect the sample
Data processing
No Yes
Generate aweak
classifier
Figure 3 Physical authentication with AdaBoost algorithm
119879119864119861119861 = 1199091198641 119909119864119894 119909119864119873 119910119864119861 (12b)
where 119909119864119894 isin 119879119860119864119861(119896) or 119909119864119894 isin 119879119861119864119861(119896) 119910119864119861 = minus1 bysubstituting H119864119861 into (9) and (10) yields 119879119864119861119860 and 119879119864119861119861 andthe value of 119910119894 represents that the transmitter is the illegaltransmitter from Eve Bob uses the two classification trainingdata set 119879119860119861119860 119879119860119861119861 119879119864119861119860 and 119879119864119861119861 as input training set
Spoofing detection is essentially a two-classification prob-lem which is considered to be solved through AdaBoostalgorithm The training data is made up of a bunch of samplepoints Each sample point comprises input sample 119909119894 andlabel 119910119894 where 119910119894 isin minus1 1 Each sample point is given anassociated weight parameter 119908119898119894 119898 means 119898-th trainingand 119894means the number of sample points which is initially set1119894 for all sample pointsWe suppose that we have a procedureavailable for training a weak classifier using weighted samplepoints At each iteration of the training process AdaBoosttrains a new weak classifier by using the sample points inwhich the weighting coefficients are adjusted according to theperformance of the previously trained weak classifier so as togive greater weight to the misclassified data points in whichthe classification error rate 119890119898 is used to evaluate misclassifieddata set119863119898
119890119898 = 119875 (119866119898 (119909119894) = 119910119894) =2119905
sum119894=1
119908119898119894119868 (119866119898 (119909119894) = 119910119894) (13)
Then the coefficient 120572119898 of 119866119898 is calculate as
120572119898 =12 log
1 minus 119890119898119890119898
(14)
Finally we generate a final model that different weight isbeing given to different weak classifiers in (8) The AdaBoostalgorithm is given as in Algorithm 2 in which the point ofthe training data can be doubled by combining with the one-dimension test statistics 119879119860 and 119879119861 together and becomea new two-dimensional features authentication model forspoofing detection Therefore in the AdaBoost algorithmthe input training data set T is following two optionalsets
(1)One-dimension test statistics training data set
119879 = 119879119860119861119860 119879119864119861119860
or 119879 = 119879119860119861119861 119879119864119861119861 (15)
(2) Two-dimension test statistics training data set
119879 = (119879119860119861119860 119879119860119861119861 ) (119879119864119861119860 119879119864119861119861 ) (16)
6 Wireless Communications and Mobile Computing
InputThe channel information of legal transmitter or illgal transmitterProcess1 Bob calculates the value of data set H119860119861 and H119864119861 from Alice and simulated Eve
H119860119861 = [H1198601198611 H1198601198612 H119860119861119873 ]H119864119861 = [H1198641198611 H1198641198612 H119864119861119873 ]
2 The data set are preprocessed by Bob3 The data set are divided into two parts and the one is training data set and the other is testing data set4 Use training data set to get the weak classifier5 Use the Adaboost algorithm to generate a strong classifer6 The testing data set is used to verify whether the claasifier can achieve the target detection rate otherwise it will return to
the first step7 The final classifier is the authenticaton decision model which can judge whether the new packets are legitimate or illegal
End
Algorithm 1 Physical authentication
Inputtraining data set 119879Process1 Initialize the weight distribution of the sample points1198631 = (11990811 1199081119894 11990812119905) 1199081119894 =
12119905 119894 = 1 2 2119905
2 for 119898 = 1 to119872 do119898means119898-th training3 Use the training data set of 119863119898 to learn and get the weak classifier
119866119898(119909) 119909119894 997888rarr minus1 +14 Calculate the classification error rate of119863119898 on the training data set
119890119898 = 119875(119866119898(119909119894) = 119910119894) =2119905
sum119894=1
119908119898119894119868 (119866119898 (119909119894) = 119910119894)5 Calculate the coefficient of 119866119898
120572119898 =12 log
1 minus 119890119898119890119898
6 Update the weight distribution of the training data set119863119898+1 = (119908119898+11 119908119898+1119894 119908119898+12119905)119908119898+1119894 =
119908119898119894119885119898
exp (minus120572119898119910119894119866119898 (119909119894)) 119894 = 1 2 2119905
119885119898 =2119905
sum119894=1
119908119898119894 exp (minus120572119898119910119894119866119898 (119909119894))7 Construct a linear combination of weak classifiers
119891(119909) =119872
sum119894=1
120572119898119866119898(119909)End for
return 119866(119909) = sign(119891(119909))
Algorithm 2 AdaBoost
4 Experimental Verification
In this section we will describe the system setup and the testprocess of measuring the Algorithm 1 for detecting Alice andEve
41 System Setup We consider the spoofing detection ofa receiver called Bob the legal transmitter called Aliceand the spoofing node called Eve They were placed inthree separate locations in a room surrounded by manyother devices such as printers desktops and other types of
equipment as shown in Figure 4 There are scattering andrefraction phenomena in the room due to the presence ofobstacles in the wireless channel from Alice to Bob andEve to Bob As shown in Figure 5 we set up experimentalplatform which implemented on USRPs and experimentswere performed in an indoor environment Bob is equippedwith an 8lowast8 MIMO system Alice is equipped with a 2lowast2MIMO system and the spoofing node called Eve is equippedwith a 2lowast2 MIMO system The signals are sent over 2antennas each at center frequency 35GHz with bandwidth2MHz
Wireless Communications and Mobile Computing 7
Bob
Eve
AliceB
devices
devicesOther
Other
5 m1 20 43
1
3
2
m
Figure 4 The experiments consisted of Alice Bob and Eve
Figure 5 Real MIMO communication platform consisted of Alice Bob and Eve
42 Experiment In the experiment the following steps aretaken
Step 1 Bob extracts channel information from Alice and Eveby the existing channel estimation mechanisms respectively
Step 2 Bob preprocesses the dataset according to (5) (7) (9)and (10)while the threshold is between [0 1] (normalization)
Step 3 Bob generates a training data set of two classificationsaccording to (11a) (11b) (12a) and (12b)
Step 4 The two classification training data set T is generatedaccording to (15) or (16)
Step 5 Bob is trained to generate a strong classifier based onthe training data set of two classifications by using AdaBoostalgorithm under Matlab program
Step 6 Bob uses a strong classifier to judge the test set andobtain the authentication detection rate
In the experiment we consider that the collection framesare five hundred frames and the value of test statisticwas normalized between 0 and 1 The test statistic 119879119860 ofchannel information of the Alice and Bob as a function offrames is shown in Figure 6(a) in which the red pointsis 119879119860(119896) in (5) and green points is 119879119864119860(119896) in (9) As canbe seen there is the overlapped area Meanwhile fromFigure 6(b) the overlapped area is large when we chosethe test statistic 119879119861 of channel information in which thered points is 119879119860(119896) in (7) and green points is 119879119864119860(119896) in(10) It is clearly shown that it is difficult to acquire thebest manual test threshold for the accuracy of authentica-tion Moreover we use 119879119860 119879119861 and the number of framesrespectively to draw a three-dimensional plot As shown in
8 Wireless Communications and Mobile Computing
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
EveAlice
Nor
mal
ized
4
Valu
e
(a) Normalized 119879119860 of Alice and Eve
EveAlice
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
Nor
mal
ized
4
Valu
e
(b) Normalized 119879119861 of Alice and Eve
Figure 6 Normalized 119879119860 and 119879119861 value of the legal transmitter Alice and the spoofing node Eve for spoofing detection with center frequency35GHz with bandwidth 2MHz
01
02
08
04
500450
06
06 400350
08
300
Frames04 250
1
20015002 100500 0
EveAlice
Nor
mal
ized
4
Valu
e
Normalized 4 Value
Figure 7 Normalized 119879119860 value and 119879119861 value of the legal transmitterAlice and the spoofing node Eve drawing three dimensional plot
Figure 7 obviously it is hard to use the traditional manualthreshold method to identify the identity of data sets inthe three-dimensional condition Howevermachine learningalgorithm based the authentication model can effectivelysettle this problem and a dividing curved surface can per-form the identification by the AdaBoost adaptive adjustmentalgorithm
43 Simulation Results In this section simulation results areprovided to demonstrate the performance of the proposedauthentication scheme
Test Threshold
0
01
02
03
04
05
06
07
08
09
1
0 01 02 03 04 05 06 07 08 09 1
Det
ectio
n Ra
te
4
4
Figure 8 Correct classification rate of 119879119860 and 119879119861
As a comparison we considered the PHY-layer spoofingdetection [15] with a varied test threshold From the Figure 8we can see that when test threshold equals 04 the bestauthentication detection rate results of using119879119860 or119879119861 reached798 and 654 respectively In addition our proposedmethod which combined two test statistics 119879119860 and 119879119861 as atwo-dimensional feature can improve the accuracy of detec-tion We use 119879119860 119879119861 and the number of frames respectivelyto draw a three-dimensional plot Figure 9 illustrates thecomparison of spoofing detection among the three methodsfrom which we can conclude that manual threshold methodbased on 119879119860 test statistics can achieve 798 detection rate
Wireless Communications and Mobile Computing 9
0
01
02
03
04
05
06
07
08
09
1
Det
ectio
n Ra
te
ML based 4 data setManual Method
MethodManual Method
ML based 4 and4 data sets
Machine Learining based 4 data set Machine Learning based 4 and4 data sets
Figure 9 The simulation result with the different method ofauthentication scheme
while machine learning based authentication method with119879119860 test statistic can acquire 871 detection rate and machinelearning based authentication method with two-dimensionalfeatures 119879119860 and 119879119861 can achieve 913 accuracy rate with anadditional 10 more computation complexity
To sum up the proposed authentication scheme achievesa superior performance over manual threshold strategy [15]Based on the above observation the proposed machinelearning based authentication scheme with tow-dimensionalfeature not only exhibits excellent performance than manualmethod but also has higher authentication rate than that ofthe same algorithm with one-dimensional feature
5 Conclusions
In this paper machine learning algorithm based physical-layer channel authentication for the 5G wireless communica-tion security is proposed A machine learning authenticationmethod could draw a conclusion whether the received pack-ets are from a legitimate transmitter or from a counterfeiterby using one-dimension or two-dimensional joint featuresThe effectiveness of the proposed authentication scheme isvalidated by widely simulations All the data used in thesimulation are derived from real OFDM-MIMO commu-nication platform which provides a real communicationenvironment Moreover the authentication results show thatthe novel methods provide a higher rate in detecting thespoofing attacks than those of the manual threshold basedphysical layer authentication schemes The training of theclassifier can be done offlineTherefore the novel method canperform authentication fast In addition whether we can usemore machine learning algorithms to further optimize ourauthentication model and find a better statistical test of largedifference in channel information is issue that we need to dealwith in the future
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare no conflicts of interest
Acknowledgments
This research was supported by NSFC (no 61572114)Sichuan Sci amp Tech Achievements Transformation Project(no 2016CC003) Sichuan Sci amp Tech Service Develop-ment Project (no 18KJFWSF0368) Hunan Provincial NatureScience Foundation Project 2018JJ2535 Chile CONICYTFONDECYTRegular Project 1181809 andNational Key RampDProgramofChina (2018YFB0904900 and 2018YFB0904905)
References
[1] M Agiwal A Roy andN Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys amp Tutorials vol 18 no 3 pp 1617ndash1655 2016
[2] J Thompson X Ge and H-C Wu ldquo5G wireless communica-tion systems prospects and challenges [Guest Editorial]rdquo IEEECommunications Magazine vol 52 no 2 pp 62ndash64 2014
[3] C Perera C H Liu S Jayawardena and M Chen ldquoA surveyon internet of things from industrial market perspectiverdquo IEEEAccess vol 2 pp 1660ndash1679 2014
[4] A Bogdanov M Knezevic G Leander D Toz K Varıcı andI Verbauwhede ldquoSPONGENT the design space of lightweightcryptographic hashingrdquo IEEE Transactions on Computers vol62 no 10 pp 2041ndash2053 2013
[5] R Zhang L Zhu C Xu and Y Yi ldquoAn Efficient and secureRFID batch authentication protocol with group tags ownershipTransferrdquo IEEE Collaboration and Internet Computing pp 168ndash175 2015
[6] L Hu H Wen B Wu et al ldquoCooperative Jamming forPhysical Layer Security Enhancement in Internet of ThingsrdquoIEEE Internet of ings Journal vol 5 no 1 pp 219ndash228 2018
[7] L Hu H Wen B Wu J Tang F Pan and R-F LiaoldquoCooperative-Jamming-Aided Secrecy Enhancement in Wire-less Networks with Passive Eavesdroppersrdquo IEEE Transactionson Vehicular Technology vol 67 no 3 pp 2108ndash2117 2018
[8] H Wen Physical layer approaches for securing wireless commu-nication systems Springer New York NY USA 2013
[9] LHuHWen BWu J Tang and F Pan ldquoAdaptive Base StationCooperation for Physical Layer Security in Two-Cell WirelessNetworksrdquo IEEE Access vol 4 pp 5607ndash5623 2016
[10] L Hu H Wen B Wu J Tang and F Pan ldquoAdaptive SecureTransmission for Physical Layer Security in Cooperative Wire-less Networksrdquo IEEE Communications Letters vol 21 no 3 pp524ndash527 2017
[11] D B Faria andDRCheriton ldquoDetecting identity-based attacksin wireless networks using signalprintsrdquo in Proceedings of theACM Workshop on Wireless Security pp 43ndash52 Los AngelesCalif USA 2006
[12] M Demirbas and Y Song ldquoAn RSSI-based scheme for sybilattack detection in wireless sensor networksrdquo in Proceedings of
10 Wireless Communications and Mobile Computing
the WoWMoM 2006 2006 International Symposium on aWorldof Wireless Mobile andMultimedia Networks pp 564ndash568 June2006
[13] Y ChenW Trappe and R P Martin ldquoDetecting and localizingwireless spoofing attacksrdquo in Proceedings of the 2007 4th AnnualIEEE Communications Society Conference on Sensor Mesh andAd Hoc Communications and Networks SECON pp 193ndash202San Diego Calif USA June 2007
[14] N Patwari and S K Kasera ldquoRobust location distinction usingtemporal link signaturesrdquo in Proceedings of the ACM Interna-tional Conference on Mobile Computing and NETWORKINGpp 111ndash122 2007
[15] H Wen Y Wang X Zhu J Li and L Zhou ldquoPhysical layerassist authentication technique for smart meter systemrdquo IETCommunications vol 7 no 3 pp 189ndash197 2013
[16] J K Tugnait ldquoWireless user authentication via comparison ofpower spectral densitiesrdquo IEEE Journal on Selected Areas inCommunications vol 31 no 9 pp 1791ndash1802 2013
[17] Z Jiang J Zhao X Li J Han and W Xi ldquoRejecting the attackSource authentication for Wi-Fi management frames using CSIInformationrdquo in Proceedings of the IEEE INFOCOM 2013 -IEEE Conference on Computer Communications pp 2544ndash2552Turin Italy April 2013
[18] X Wu and Z Yang ldquoPhysical-layer authentication for multi-carrier transmissionrdquo IEEE Communications Letters vol 19 no1 pp 74ndash77 2015
[19] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-authentication protocol for spoofing detection in wireless net-worksrdquo IEEE Transactions on Vehicular Technology vol 65 no12 pp 10037ndash10047 2016
[20] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-LayerSpoofing Detection with Reinforcement Learning in WirelessNetworksrdquo IEEE Transactions on Vehicular Technology vol 65no 12 pp 10037ndash10048 2016
[21] P Hao X Wang and A Refaey ldquoAn enhanced cross-layerauthentication mechanism for wireless communications basedon PER and RSSIrdquo in Proceedings of the 2013 13th CanadianWorkshop on Information eory CWIT 2013 pp 44ndash48Canada June 2013
[22] S Chen et al ldquoMachine-to-Machine communications in ultra-dense networksmdashA surveyrdquo IEEE Communications Surveys ampTutorials vol 1 no 1 99 pages 2017
[23] L Xiao L J Greenstein N B Mandayam and W TrappeldquoChannel-based detection of sybil attacks in wireless networksrdquoIEEE Transactions on Information Forensics and Security vol 4no 3 pp 492ndash503 2009
[24] H Wen P-H Ho C Qi and G Gong ldquoPhysical layer assistedauthentication for distributed ad hoc wireless sensor networksrdquoIET Information Security vol 4 no 4 pp 390ndash396 2010
[25] J H Friedman ldquoGreedy function approximation a gradientboosting machinerdquo Annals of Statistics vol 29 no 5 pp 1189ndash1232 2001
[26] J Friedman T Hastie and R Tibshirani ldquoAdditive logisticregression a statistical view of boostingrdquo e Annals of Statis-tics vol 28 no 2 pp 337ndash407 2000
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
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
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
6 Wireless Communications and Mobile Computing
InputThe channel information of legal transmitter or illgal transmitterProcess1 Bob calculates the value of data set H119860119861 and H119864119861 from Alice and simulated Eve
H119860119861 = [H1198601198611 H1198601198612 H119860119861119873 ]H119864119861 = [H1198641198611 H1198641198612 H119864119861119873 ]
2 The data set are preprocessed by Bob3 The data set are divided into two parts and the one is training data set and the other is testing data set4 Use training data set to get the weak classifier5 Use the Adaboost algorithm to generate a strong classifer6 The testing data set is used to verify whether the claasifier can achieve the target detection rate otherwise it will return to
the first step7 The final classifier is the authenticaton decision model which can judge whether the new packets are legitimate or illegal
End
Algorithm 1 Physical authentication
Inputtraining data set 119879Process1 Initialize the weight distribution of the sample points1198631 = (11990811 1199081119894 11990812119905) 1199081119894 =
12119905 119894 = 1 2 2119905
2 for 119898 = 1 to119872 do119898means119898-th training3 Use the training data set of 119863119898 to learn and get the weak classifier
119866119898(119909) 119909119894 997888rarr minus1 +14 Calculate the classification error rate of119863119898 on the training data set
119890119898 = 119875(119866119898(119909119894) = 119910119894) =2119905
sum119894=1
119908119898119894119868 (119866119898 (119909119894) = 119910119894)5 Calculate the coefficient of 119866119898
120572119898 =12 log
1 minus 119890119898119890119898
6 Update the weight distribution of the training data set119863119898+1 = (119908119898+11 119908119898+1119894 119908119898+12119905)119908119898+1119894 =
119908119898119894119885119898
exp (minus120572119898119910119894119866119898 (119909119894)) 119894 = 1 2 2119905
119885119898 =2119905
sum119894=1
119908119898119894 exp (minus120572119898119910119894119866119898 (119909119894))7 Construct a linear combination of weak classifiers
119891(119909) =119872
sum119894=1
120572119898119866119898(119909)End for
return 119866(119909) = sign(119891(119909))
Algorithm 2 AdaBoost
4 Experimental Verification
In this section we will describe the system setup and the testprocess of measuring the Algorithm 1 for detecting Alice andEve
41 System Setup We consider the spoofing detection ofa receiver called Bob the legal transmitter called Aliceand the spoofing node called Eve They were placed inthree separate locations in a room surrounded by manyother devices such as printers desktops and other types of
equipment as shown in Figure 4 There are scattering andrefraction phenomena in the room due to the presence ofobstacles in the wireless channel from Alice to Bob andEve to Bob As shown in Figure 5 we set up experimentalplatform which implemented on USRPs and experimentswere performed in an indoor environment Bob is equippedwith an 8lowast8 MIMO system Alice is equipped with a 2lowast2MIMO system and the spoofing node called Eve is equippedwith a 2lowast2 MIMO system The signals are sent over 2antennas each at center frequency 35GHz with bandwidth2MHz
Wireless Communications and Mobile Computing 7
Bob
Eve
AliceB
devices
devicesOther
Other
5 m1 20 43
1
3
2
m
Figure 4 The experiments consisted of Alice Bob and Eve
Figure 5 Real MIMO communication platform consisted of Alice Bob and Eve
42 Experiment In the experiment the following steps aretaken
Step 1 Bob extracts channel information from Alice and Eveby the existing channel estimation mechanisms respectively
Step 2 Bob preprocesses the dataset according to (5) (7) (9)and (10)while the threshold is between [0 1] (normalization)
Step 3 Bob generates a training data set of two classificationsaccording to (11a) (11b) (12a) and (12b)
Step 4 The two classification training data set T is generatedaccording to (15) or (16)
Step 5 Bob is trained to generate a strong classifier based onthe training data set of two classifications by using AdaBoostalgorithm under Matlab program
Step 6 Bob uses a strong classifier to judge the test set andobtain the authentication detection rate
In the experiment we consider that the collection framesare five hundred frames and the value of test statisticwas normalized between 0 and 1 The test statistic 119879119860 ofchannel information of the Alice and Bob as a function offrames is shown in Figure 6(a) in which the red pointsis 119879119860(119896) in (5) and green points is 119879119864119860(119896) in (9) As canbe seen there is the overlapped area Meanwhile fromFigure 6(b) the overlapped area is large when we chosethe test statistic 119879119861 of channel information in which thered points is 119879119860(119896) in (7) and green points is 119879119864119860(119896) in(10) It is clearly shown that it is difficult to acquire thebest manual test threshold for the accuracy of authentica-tion Moreover we use 119879119860 119879119861 and the number of framesrespectively to draw a three-dimensional plot As shown in
8 Wireless Communications and Mobile Computing
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
EveAlice
Nor
mal
ized
4
Valu
e
(a) Normalized 119879119860 of Alice and Eve
EveAlice
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
Nor
mal
ized
4
Valu
e
(b) Normalized 119879119861 of Alice and Eve
Figure 6 Normalized 119879119860 and 119879119861 value of the legal transmitter Alice and the spoofing node Eve for spoofing detection with center frequency35GHz with bandwidth 2MHz
01
02
08
04
500450
06
06 400350
08
300
Frames04 250
1
20015002 100500 0
EveAlice
Nor
mal
ized
4
Valu
e
Normalized 4 Value
Figure 7 Normalized 119879119860 value and 119879119861 value of the legal transmitterAlice and the spoofing node Eve drawing three dimensional plot
Figure 7 obviously it is hard to use the traditional manualthreshold method to identify the identity of data sets inthe three-dimensional condition Howevermachine learningalgorithm based the authentication model can effectivelysettle this problem and a dividing curved surface can per-form the identification by the AdaBoost adaptive adjustmentalgorithm
43 Simulation Results In this section simulation results areprovided to demonstrate the performance of the proposedauthentication scheme
Test Threshold
0
01
02
03
04
05
06
07
08
09
1
0 01 02 03 04 05 06 07 08 09 1
Det
ectio
n Ra
te
4
4
Figure 8 Correct classification rate of 119879119860 and 119879119861
As a comparison we considered the PHY-layer spoofingdetection [15] with a varied test threshold From the Figure 8we can see that when test threshold equals 04 the bestauthentication detection rate results of using119879119860 or119879119861 reached798 and 654 respectively In addition our proposedmethod which combined two test statistics 119879119860 and 119879119861 as atwo-dimensional feature can improve the accuracy of detec-tion We use 119879119860 119879119861 and the number of frames respectivelyto draw a three-dimensional plot Figure 9 illustrates thecomparison of spoofing detection among the three methodsfrom which we can conclude that manual threshold methodbased on 119879119860 test statistics can achieve 798 detection rate
Wireless Communications and Mobile Computing 9
0
01
02
03
04
05
06
07
08
09
1
Det
ectio
n Ra
te
ML based 4 data setManual Method
MethodManual Method
ML based 4 and4 data sets
Machine Learining based 4 data set Machine Learning based 4 and4 data sets
Figure 9 The simulation result with the different method ofauthentication scheme
while machine learning based authentication method with119879119860 test statistic can acquire 871 detection rate and machinelearning based authentication method with two-dimensionalfeatures 119879119860 and 119879119861 can achieve 913 accuracy rate with anadditional 10 more computation complexity
To sum up the proposed authentication scheme achievesa superior performance over manual threshold strategy [15]Based on the above observation the proposed machinelearning based authentication scheme with tow-dimensionalfeature not only exhibits excellent performance than manualmethod but also has higher authentication rate than that ofthe same algorithm with one-dimensional feature
5 Conclusions
In this paper machine learning algorithm based physical-layer channel authentication for the 5G wireless communica-tion security is proposed A machine learning authenticationmethod could draw a conclusion whether the received pack-ets are from a legitimate transmitter or from a counterfeiterby using one-dimension or two-dimensional joint featuresThe effectiveness of the proposed authentication scheme isvalidated by widely simulations All the data used in thesimulation are derived from real OFDM-MIMO commu-nication platform which provides a real communicationenvironment Moreover the authentication results show thatthe novel methods provide a higher rate in detecting thespoofing attacks than those of the manual threshold basedphysical layer authentication schemes The training of theclassifier can be done offlineTherefore the novel method canperform authentication fast In addition whether we can usemore machine learning algorithms to further optimize ourauthentication model and find a better statistical test of largedifference in channel information is issue that we need to dealwith in the future
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare no conflicts of interest
Acknowledgments
This research was supported by NSFC (no 61572114)Sichuan Sci amp Tech Achievements Transformation Project(no 2016CC003) Sichuan Sci amp Tech Service Develop-ment Project (no 18KJFWSF0368) Hunan Provincial NatureScience Foundation Project 2018JJ2535 Chile CONICYTFONDECYTRegular Project 1181809 andNational Key RampDProgramofChina (2018YFB0904900 and 2018YFB0904905)
References
[1] M Agiwal A Roy andN Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys amp Tutorials vol 18 no 3 pp 1617ndash1655 2016
[2] J Thompson X Ge and H-C Wu ldquo5G wireless communica-tion systems prospects and challenges [Guest Editorial]rdquo IEEECommunications Magazine vol 52 no 2 pp 62ndash64 2014
[3] C Perera C H Liu S Jayawardena and M Chen ldquoA surveyon internet of things from industrial market perspectiverdquo IEEEAccess vol 2 pp 1660ndash1679 2014
[4] A Bogdanov M Knezevic G Leander D Toz K Varıcı andI Verbauwhede ldquoSPONGENT the design space of lightweightcryptographic hashingrdquo IEEE Transactions on Computers vol62 no 10 pp 2041ndash2053 2013
[5] R Zhang L Zhu C Xu and Y Yi ldquoAn Efficient and secureRFID batch authentication protocol with group tags ownershipTransferrdquo IEEE Collaboration and Internet Computing pp 168ndash175 2015
[6] L Hu H Wen B Wu et al ldquoCooperative Jamming forPhysical Layer Security Enhancement in Internet of ThingsrdquoIEEE Internet of ings Journal vol 5 no 1 pp 219ndash228 2018
[7] L Hu H Wen B Wu J Tang F Pan and R-F LiaoldquoCooperative-Jamming-Aided Secrecy Enhancement in Wire-less Networks with Passive Eavesdroppersrdquo IEEE Transactionson Vehicular Technology vol 67 no 3 pp 2108ndash2117 2018
[8] H Wen Physical layer approaches for securing wireless commu-nication systems Springer New York NY USA 2013
[9] LHuHWen BWu J Tang and F Pan ldquoAdaptive Base StationCooperation for Physical Layer Security in Two-Cell WirelessNetworksrdquo IEEE Access vol 4 pp 5607ndash5623 2016
[10] L Hu H Wen B Wu J Tang and F Pan ldquoAdaptive SecureTransmission for Physical Layer Security in Cooperative Wire-less Networksrdquo IEEE Communications Letters vol 21 no 3 pp524ndash527 2017
[11] D B Faria andDRCheriton ldquoDetecting identity-based attacksin wireless networks using signalprintsrdquo in Proceedings of theACM Workshop on Wireless Security pp 43ndash52 Los AngelesCalif USA 2006
[12] M Demirbas and Y Song ldquoAn RSSI-based scheme for sybilattack detection in wireless sensor networksrdquo in Proceedings of
10 Wireless Communications and Mobile Computing
the WoWMoM 2006 2006 International Symposium on aWorldof Wireless Mobile andMultimedia Networks pp 564ndash568 June2006
[13] Y ChenW Trappe and R P Martin ldquoDetecting and localizingwireless spoofing attacksrdquo in Proceedings of the 2007 4th AnnualIEEE Communications Society Conference on Sensor Mesh andAd Hoc Communications and Networks SECON pp 193ndash202San Diego Calif USA June 2007
[14] N Patwari and S K Kasera ldquoRobust location distinction usingtemporal link signaturesrdquo in Proceedings of the ACM Interna-tional Conference on Mobile Computing and NETWORKINGpp 111ndash122 2007
[15] H Wen Y Wang X Zhu J Li and L Zhou ldquoPhysical layerassist authentication technique for smart meter systemrdquo IETCommunications vol 7 no 3 pp 189ndash197 2013
[16] J K Tugnait ldquoWireless user authentication via comparison ofpower spectral densitiesrdquo IEEE Journal on Selected Areas inCommunications vol 31 no 9 pp 1791ndash1802 2013
[17] Z Jiang J Zhao X Li J Han and W Xi ldquoRejecting the attackSource authentication for Wi-Fi management frames using CSIInformationrdquo in Proceedings of the IEEE INFOCOM 2013 -IEEE Conference on Computer Communications pp 2544ndash2552Turin Italy April 2013
[18] X Wu and Z Yang ldquoPhysical-layer authentication for multi-carrier transmissionrdquo IEEE Communications Letters vol 19 no1 pp 74ndash77 2015
[19] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-authentication protocol for spoofing detection in wireless net-worksrdquo IEEE Transactions on Vehicular Technology vol 65 no12 pp 10037ndash10047 2016
[20] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-LayerSpoofing Detection with Reinforcement Learning in WirelessNetworksrdquo IEEE Transactions on Vehicular Technology vol 65no 12 pp 10037ndash10048 2016
[21] P Hao X Wang and A Refaey ldquoAn enhanced cross-layerauthentication mechanism for wireless communications basedon PER and RSSIrdquo in Proceedings of the 2013 13th CanadianWorkshop on Information eory CWIT 2013 pp 44ndash48Canada June 2013
[22] S Chen et al ldquoMachine-to-Machine communications in ultra-dense networksmdashA surveyrdquo IEEE Communications Surveys ampTutorials vol 1 no 1 99 pages 2017
[23] L Xiao L J Greenstein N B Mandayam and W TrappeldquoChannel-based detection of sybil attacks in wireless networksrdquoIEEE Transactions on Information Forensics and Security vol 4no 3 pp 492ndash503 2009
[24] H Wen P-H Ho C Qi and G Gong ldquoPhysical layer assistedauthentication for distributed ad hoc wireless sensor networksrdquoIET Information Security vol 4 no 4 pp 390ndash396 2010
[25] J H Friedman ldquoGreedy function approximation a gradientboosting machinerdquo Annals of Statistics vol 29 no 5 pp 1189ndash1232 2001
[26] J Friedman T Hastie and R Tibshirani ldquoAdditive logisticregression a statistical view of boostingrdquo e Annals of Statis-tics vol 28 no 2 pp 337ndash407 2000
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
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
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 7
Bob
Eve
AliceB
devices
devicesOther
Other
5 m1 20 43
1
3
2
m
Figure 4 The experiments consisted of Alice Bob and Eve
Figure 5 Real MIMO communication platform consisted of Alice Bob and Eve
42 Experiment In the experiment the following steps aretaken
Step 1 Bob extracts channel information from Alice and Eveby the existing channel estimation mechanisms respectively
Step 2 Bob preprocesses the dataset according to (5) (7) (9)and (10)while the threshold is between [0 1] (normalization)
Step 3 Bob generates a training data set of two classificationsaccording to (11a) (11b) (12a) and (12b)
Step 4 The two classification training data set T is generatedaccording to (15) or (16)
Step 5 Bob is trained to generate a strong classifier based onthe training data set of two classifications by using AdaBoostalgorithm under Matlab program
Step 6 Bob uses a strong classifier to judge the test set andobtain the authentication detection rate
In the experiment we consider that the collection framesare five hundred frames and the value of test statisticwas normalized between 0 and 1 The test statistic 119879119860 ofchannel information of the Alice and Bob as a function offrames is shown in Figure 6(a) in which the red pointsis 119879119860(119896) in (5) and green points is 119879119864119860(119896) in (9) As canbe seen there is the overlapped area Meanwhile fromFigure 6(b) the overlapped area is large when we chosethe test statistic 119879119861 of channel information in which thered points is 119879119860(119896) in (7) and green points is 119879119864119860(119896) in(10) It is clearly shown that it is difficult to acquire thebest manual test threshold for the accuracy of authentica-tion Moreover we use 119879119860 119879119861 and the number of framesrespectively to draw a three-dimensional plot As shown in
8 Wireless Communications and Mobile Computing
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
EveAlice
Nor
mal
ized
4
Valu
e
(a) Normalized 119879119860 of Alice and Eve
EveAlice
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
Nor
mal
ized
4
Valu
e
(b) Normalized 119879119861 of Alice and Eve
Figure 6 Normalized 119879119860 and 119879119861 value of the legal transmitter Alice and the spoofing node Eve for spoofing detection with center frequency35GHz with bandwidth 2MHz
01
02
08
04
500450
06
06 400350
08
300
Frames04 250
1
20015002 100500 0
EveAlice
Nor
mal
ized
4
Valu
e
Normalized 4 Value
Figure 7 Normalized 119879119860 value and 119879119861 value of the legal transmitterAlice and the spoofing node Eve drawing three dimensional plot
Figure 7 obviously it is hard to use the traditional manualthreshold method to identify the identity of data sets inthe three-dimensional condition Howevermachine learningalgorithm based the authentication model can effectivelysettle this problem and a dividing curved surface can per-form the identification by the AdaBoost adaptive adjustmentalgorithm
43 Simulation Results In this section simulation results areprovided to demonstrate the performance of the proposedauthentication scheme
Test Threshold
0
01
02
03
04
05
06
07
08
09
1
0 01 02 03 04 05 06 07 08 09 1
Det
ectio
n Ra
te
4
4
Figure 8 Correct classification rate of 119879119860 and 119879119861
As a comparison we considered the PHY-layer spoofingdetection [15] with a varied test threshold From the Figure 8we can see that when test threshold equals 04 the bestauthentication detection rate results of using119879119860 or119879119861 reached798 and 654 respectively In addition our proposedmethod which combined two test statistics 119879119860 and 119879119861 as atwo-dimensional feature can improve the accuracy of detec-tion We use 119879119860 119879119861 and the number of frames respectivelyto draw a three-dimensional plot Figure 9 illustrates thecomparison of spoofing detection among the three methodsfrom which we can conclude that manual threshold methodbased on 119879119860 test statistics can achieve 798 detection rate
Wireless Communications and Mobile Computing 9
0
01
02
03
04
05
06
07
08
09
1
Det
ectio
n Ra
te
ML based 4 data setManual Method
MethodManual Method
ML based 4 and4 data sets
Machine Learining based 4 data set Machine Learning based 4 and4 data sets
Figure 9 The simulation result with the different method ofauthentication scheme
while machine learning based authentication method with119879119860 test statistic can acquire 871 detection rate and machinelearning based authentication method with two-dimensionalfeatures 119879119860 and 119879119861 can achieve 913 accuracy rate with anadditional 10 more computation complexity
To sum up the proposed authentication scheme achievesa superior performance over manual threshold strategy [15]Based on the above observation the proposed machinelearning based authentication scheme with tow-dimensionalfeature not only exhibits excellent performance than manualmethod but also has higher authentication rate than that ofthe same algorithm with one-dimensional feature
5 Conclusions
In this paper machine learning algorithm based physical-layer channel authentication for the 5G wireless communica-tion security is proposed A machine learning authenticationmethod could draw a conclusion whether the received pack-ets are from a legitimate transmitter or from a counterfeiterby using one-dimension or two-dimensional joint featuresThe effectiveness of the proposed authentication scheme isvalidated by widely simulations All the data used in thesimulation are derived from real OFDM-MIMO commu-nication platform which provides a real communicationenvironment Moreover the authentication results show thatthe novel methods provide a higher rate in detecting thespoofing attacks than those of the manual threshold basedphysical layer authentication schemes The training of theclassifier can be done offlineTherefore the novel method canperform authentication fast In addition whether we can usemore machine learning algorithms to further optimize ourauthentication model and find a better statistical test of largedifference in channel information is issue that we need to dealwith in the future
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare no conflicts of interest
Acknowledgments
This research was supported by NSFC (no 61572114)Sichuan Sci amp Tech Achievements Transformation Project(no 2016CC003) Sichuan Sci amp Tech Service Develop-ment Project (no 18KJFWSF0368) Hunan Provincial NatureScience Foundation Project 2018JJ2535 Chile CONICYTFONDECYTRegular Project 1181809 andNational Key RampDProgramofChina (2018YFB0904900 and 2018YFB0904905)
References
[1] M Agiwal A Roy andN Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys amp Tutorials vol 18 no 3 pp 1617ndash1655 2016
[2] J Thompson X Ge and H-C Wu ldquo5G wireless communica-tion systems prospects and challenges [Guest Editorial]rdquo IEEECommunications Magazine vol 52 no 2 pp 62ndash64 2014
[3] C Perera C H Liu S Jayawardena and M Chen ldquoA surveyon internet of things from industrial market perspectiverdquo IEEEAccess vol 2 pp 1660ndash1679 2014
[4] A Bogdanov M Knezevic G Leander D Toz K Varıcı andI Verbauwhede ldquoSPONGENT the design space of lightweightcryptographic hashingrdquo IEEE Transactions on Computers vol62 no 10 pp 2041ndash2053 2013
[5] R Zhang L Zhu C Xu and Y Yi ldquoAn Efficient and secureRFID batch authentication protocol with group tags ownershipTransferrdquo IEEE Collaboration and Internet Computing pp 168ndash175 2015
[6] L Hu H Wen B Wu et al ldquoCooperative Jamming forPhysical Layer Security Enhancement in Internet of ThingsrdquoIEEE Internet of ings Journal vol 5 no 1 pp 219ndash228 2018
[7] L Hu H Wen B Wu J Tang F Pan and R-F LiaoldquoCooperative-Jamming-Aided Secrecy Enhancement in Wire-less Networks with Passive Eavesdroppersrdquo IEEE Transactionson Vehicular Technology vol 67 no 3 pp 2108ndash2117 2018
[8] H Wen Physical layer approaches for securing wireless commu-nication systems Springer New York NY USA 2013
[9] LHuHWen BWu J Tang and F Pan ldquoAdaptive Base StationCooperation for Physical Layer Security in Two-Cell WirelessNetworksrdquo IEEE Access vol 4 pp 5607ndash5623 2016
[10] L Hu H Wen B Wu J Tang and F Pan ldquoAdaptive SecureTransmission for Physical Layer Security in Cooperative Wire-less Networksrdquo IEEE Communications Letters vol 21 no 3 pp524ndash527 2017
[11] D B Faria andDRCheriton ldquoDetecting identity-based attacksin wireless networks using signalprintsrdquo in Proceedings of theACM Workshop on Wireless Security pp 43ndash52 Los AngelesCalif USA 2006
[12] M Demirbas and Y Song ldquoAn RSSI-based scheme for sybilattack detection in wireless sensor networksrdquo in Proceedings of
10 Wireless Communications and Mobile Computing
the WoWMoM 2006 2006 International Symposium on aWorldof Wireless Mobile andMultimedia Networks pp 564ndash568 June2006
[13] Y ChenW Trappe and R P Martin ldquoDetecting and localizingwireless spoofing attacksrdquo in Proceedings of the 2007 4th AnnualIEEE Communications Society Conference on Sensor Mesh andAd Hoc Communications and Networks SECON pp 193ndash202San Diego Calif USA June 2007
[14] N Patwari and S K Kasera ldquoRobust location distinction usingtemporal link signaturesrdquo in Proceedings of the ACM Interna-tional Conference on Mobile Computing and NETWORKINGpp 111ndash122 2007
[15] H Wen Y Wang X Zhu J Li and L Zhou ldquoPhysical layerassist authentication technique for smart meter systemrdquo IETCommunications vol 7 no 3 pp 189ndash197 2013
[16] J K Tugnait ldquoWireless user authentication via comparison ofpower spectral densitiesrdquo IEEE Journal on Selected Areas inCommunications vol 31 no 9 pp 1791ndash1802 2013
[17] Z Jiang J Zhao X Li J Han and W Xi ldquoRejecting the attackSource authentication for Wi-Fi management frames using CSIInformationrdquo in Proceedings of the IEEE INFOCOM 2013 -IEEE Conference on Computer Communications pp 2544ndash2552Turin Italy April 2013
[18] X Wu and Z Yang ldquoPhysical-layer authentication for multi-carrier transmissionrdquo IEEE Communications Letters vol 19 no1 pp 74ndash77 2015
[19] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-authentication protocol for spoofing detection in wireless net-worksrdquo IEEE Transactions on Vehicular Technology vol 65 no12 pp 10037ndash10047 2016
[20] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-LayerSpoofing Detection with Reinforcement Learning in WirelessNetworksrdquo IEEE Transactions on Vehicular Technology vol 65no 12 pp 10037ndash10048 2016
[21] P Hao X Wang and A Refaey ldquoAn enhanced cross-layerauthentication mechanism for wireless communications basedon PER and RSSIrdquo in Proceedings of the 2013 13th CanadianWorkshop on Information eory CWIT 2013 pp 44ndash48Canada June 2013
[22] S Chen et al ldquoMachine-to-Machine communications in ultra-dense networksmdashA surveyrdquo IEEE Communications Surveys ampTutorials vol 1 no 1 99 pages 2017
[23] L Xiao L J Greenstein N B Mandayam and W TrappeldquoChannel-based detection of sybil attacks in wireless networksrdquoIEEE Transactions on Information Forensics and Security vol 4no 3 pp 492ndash503 2009
[24] H Wen P-H Ho C Qi and G Gong ldquoPhysical layer assistedauthentication for distributed ad hoc wireless sensor networksrdquoIET Information Security vol 4 no 4 pp 390ndash396 2010
[25] J H Friedman ldquoGreedy function approximation a gradientboosting machinerdquo Annals of Statistics vol 29 no 5 pp 1189ndash1232 2001
[26] J Friedman T Hastie and R Tibshirani ldquoAdditive logisticregression a statistical view of boostingrdquo e Annals of Statis-tics vol 28 no 2 pp 337ndash407 2000
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
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
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
8 Wireless Communications and Mobile Computing
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
EveAlice
Nor
mal
ized
4
Valu
e
(a) Normalized 119879119860 of Alice and Eve
EveAlice
0 50 100 150 200 250 300 350 400 450 500Frames
0
01
02
03
04
05
06
07
08
09
1
Nor
mal
ized
4
Valu
e
(b) Normalized 119879119861 of Alice and Eve
Figure 6 Normalized 119879119860 and 119879119861 value of the legal transmitter Alice and the spoofing node Eve for spoofing detection with center frequency35GHz with bandwidth 2MHz
01
02
08
04
500450
06
06 400350
08
300
Frames04 250
1
20015002 100500 0
EveAlice
Nor
mal
ized
4
Valu
e
Normalized 4 Value
Figure 7 Normalized 119879119860 value and 119879119861 value of the legal transmitterAlice and the spoofing node Eve drawing three dimensional plot
Figure 7 obviously it is hard to use the traditional manualthreshold method to identify the identity of data sets inthe three-dimensional condition Howevermachine learningalgorithm based the authentication model can effectivelysettle this problem and a dividing curved surface can per-form the identification by the AdaBoost adaptive adjustmentalgorithm
43 Simulation Results In this section simulation results areprovided to demonstrate the performance of the proposedauthentication scheme
Test Threshold
0
01
02
03
04
05
06
07
08
09
1
0 01 02 03 04 05 06 07 08 09 1
Det
ectio
n Ra
te
4
4
Figure 8 Correct classification rate of 119879119860 and 119879119861
As a comparison we considered the PHY-layer spoofingdetection [15] with a varied test threshold From the Figure 8we can see that when test threshold equals 04 the bestauthentication detection rate results of using119879119860 or119879119861 reached798 and 654 respectively In addition our proposedmethod which combined two test statistics 119879119860 and 119879119861 as atwo-dimensional feature can improve the accuracy of detec-tion We use 119879119860 119879119861 and the number of frames respectivelyto draw a three-dimensional plot Figure 9 illustrates thecomparison of spoofing detection among the three methodsfrom which we can conclude that manual threshold methodbased on 119879119860 test statistics can achieve 798 detection rate
Wireless Communications and Mobile Computing 9
0
01
02
03
04
05
06
07
08
09
1
Det
ectio
n Ra
te
ML based 4 data setManual Method
MethodManual Method
ML based 4 and4 data sets
Machine Learining based 4 data set Machine Learning based 4 and4 data sets
Figure 9 The simulation result with the different method ofauthentication scheme
while machine learning based authentication method with119879119860 test statistic can acquire 871 detection rate and machinelearning based authentication method with two-dimensionalfeatures 119879119860 and 119879119861 can achieve 913 accuracy rate with anadditional 10 more computation complexity
To sum up the proposed authentication scheme achievesa superior performance over manual threshold strategy [15]Based on the above observation the proposed machinelearning based authentication scheme with tow-dimensionalfeature not only exhibits excellent performance than manualmethod but also has higher authentication rate than that ofthe same algorithm with one-dimensional feature
5 Conclusions
In this paper machine learning algorithm based physical-layer channel authentication for the 5G wireless communica-tion security is proposed A machine learning authenticationmethod could draw a conclusion whether the received pack-ets are from a legitimate transmitter or from a counterfeiterby using one-dimension or two-dimensional joint featuresThe effectiveness of the proposed authentication scheme isvalidated by widely simulations All the data used in thesimulation are derived from real OFDM-MIMO commu-nication platform which provides a real communicationenvironment Moreover the authentication results show thatthe novel methods provide a higher rate in detecting thespoofing attacks than those of the manual threshold basedphysical layer authentication schemes The training of theclassifier can be done offlineTherefore the novel method canperform authentication fast In addition whether we can usemore machine learning algorithms to further optimize ourauthentication model and find a better statistical test of largedifference in channel information is issue that we need to dealwith in the future
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare no conflicts of interest
Acknowledgments
This research was supported by NSFC (no 61572114)Sichuan Sci amp Tech Achievements Transformation Project(no 2016CC003) Sichuan Sci amp Tech Service Develop-ment Project (no 18KJFWSF0368) Hunan Provincial NatureScience Foundation Project 2018JJ2535 Chile CONICYTFONDECYTRegular Project 1181809 andNational Key RampDProgramofChina (2018YFB0904900 and 2018YFB0904905)
References
[1] M Agiwal A Roy andN Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys amp Tutorials vol 18 no 3 pp 1617ndash1655 2016
[2] J Thompson X Ge and H-C Wu ldquo5G wireless communica-tion systems prospects and challenges [Guest Editorial]rdquo IEEECommunications Magazine vol 52 no 2 pp 62ndash64 2014
[3] C Perera C H Liu S Jayawardena and M Chen ldquoA surveyon internet of things from industrial market perspectiverdquo IEEEAccess vol 2 pp 1660ndash1679 2014
[4] A Bogdanov M Knezevic G Leander D Toz K Varıcı andI Verbauwhede ldquoSPONGENT the design space of lightweightcryptographic hashingrdquo IEEE Transactions on Computers vol62 no 10 pp 2041ndash2053 2013
[5] R Zhang L Zhu C Xu and Y Yi ldquoAn Efficient and secureRFID batch authentication protocol with group tags ownershipTransferrdquo IEEE Collaboration and Internet Computing pp 168ndash175 2015
[6] L Hu H Wen B Wu et al ldquoCooperative Jamming forPhysical Layer Security Enhancement in Internet of ThingsrdquoIEEE Internet of ings Journal vol 5 no 1 pp 219ndash228 2018
[7] L Hu H Wen B Wu J Tang F Pan and R-F LiaoldquoCooperative-Jamming-Aided Secrecy Enhancement in Wire-less Networks with Passive Eavesdroppersrdquo IEEE Transactionson Vehicular Technology vol 67 no 3 pp 2108ndash2117 2018
[8] H Wen Physical layer approaches for securing wireless commu-nication systems Springer New York NY USA 2013
[9] LHuHWen BWu J Tang and F Pan ldquoAdaptive Base StationCooperation for Physical Layer Security in Two-Cell WirelessNetworksrdquo IEEE Access vol 4 pp 5607ndash5623 2016
[10] L Hu H Wen B Wu J Tang and F Pan ldquoAdaptive SecureTransmission for Physical Layer Security in Cooperative Wire-less Networksrdquo IEEE Communications Letters vol 21 no 3 pp524ndash527 2017
[11] D B Faria andDRCheriton ldquoDetecting identity-based attacksin wireless networks using signalprintsrdquo in Proceedings of theACM Workshop on Wireless Security pp 43ndash52 Los AngelesCalif USA 2006
[12] M Demirbas and Y Song ldquoAn RSSI-based scheme for sybilattack detection in wireless sensor networksrdquo in Proceedings of
10 Wireless Communications and Mobile Computing
the WoWMoM 2006 2006 International Symposium on aWorldof Wireless Mobile andMultimedia Networks pp 564ndash568 June2006
[13] Y ChenW Trappe and R P Martin ldquoDetecting and localizingwireless spoofing attacksrdquo in Proceedings of the 2007 4th AnnualIEEE Communications Society Conference on Sensor Mesh andAd Hoc Communications and Networks SECON pp 193ndash202San Diego Calif USA June 2007
[14] N Patwari and S K Kasera ldquoRobust location distinction usingtemporal link signaturesrdquo in Proceedings of the ACM Interna-tional Conference on Mobile Computing and NETWORKINGpp 111ndash122 2007
[15] H Wen Y Wang X Zhu J Li and L Zhou ldquoPhysical layerassist authentication technique for smart meter systemrdquo IETCommunications vol 7 no 3 pp 189ndash197 2013
[16] J K Tugnait ldquoWireless user authentication via comparison ofpower spectral densitiesrdquo IEEE Journal on Selected Areas inCommunications vol 31 no 9 pp 1791ndash1802 2013
[17] Z Jiang J Zhao X Li J Han and W Xi ldquoRejecting the attackSource authentication for Wi-Fi management frames using CSIInformationrdquo in Proceedings of the IEEE INFOCOM 2013 -IEEE Conference on Computer Communications pp 2544ndash2552Turin Italy April 2013
[18] X Wu and Z Yang ldquoPhysical-layer authentication for multi-carrier transmissionrdquo IEEE Communications Letters vol 19 no1 pp 74ndash77 2015
[19] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-authentication protocol for spoofing detection in wireless net-worksrdquo IEEE Transactions on Vehicular Technology vol 65 no12 pp 10037ndash10047 2016
[20] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-LayerSpoofing Detection with Reinforcement Learning in WirelessNetworksrdquo IEEE Transactions on Vehicular Technology vol 65no 12 pp 10037ndash10048 2016
[21] P Hao X Wang and A Refaey ldquoAn enhanced cross-layerauthentication mechanism for wireless communications basedon PER and RSSIrdquo in Proceedings of the 2013 13th CanadianWorkshop on Information eory CWIT 2013 pp 44ndash48Canada June 2013
[22] S Chen et al ldquoMachine-to-Machine communications in ultra-dense networksmdashA surveyrdquo IEEE Communications Surveys ampTutorials vol 1 no 1 99 pages 2017
[23] L Xiao L J Greenstein N B Mandayam and W TrappeldquoChannel-based detection of sybil attacks in wireless networksrdquoIEEE Transactions on Information Forensics and Security vol 4no 3 pp 492ndash503 2009
[24] H Wen P-H Ho C Qi and G Gong ldquoPhysical layer assistedauthentication for distributed ad hoc wireless sensor networksrdquoIET Information Security vol 4 no 4 pp 390ndash396 2010
[25] J H Friedman ldquoGreedy function approximation a gradientboosting machinerdquo Annals of Statistics vol 29 no 5 pp 1189ndash1232 2001
[26] J Friedman T Hastie and R Tibshirani ldquoAdditive logisticregression a statistical view of boostingrdquo e Annals of Statis-tics vol 28 no 2 pp 337ndash407 2000
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
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
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 9
0
01
02
03
04
05
06
07
08
09
1
Det
ectio
n Ra
te
ML based 4 data setManual Method
MethodManual Method
ML based 4 and4 data sets
Machine Learining based 4 data set Machine Learning based 4 and4 data sets
Figure 9 The simulation result with the different method ofauthentication scheme
while machine learning based authentication method with119879119860 test statistic can acquire 871 detection rate and machinelearning based authentication method with two-dimensionalfeatures 119879119860 and 119879119861 can achieve 913 accuracy rate with anadditional 10 more computation complexity
To sum up the proposed authentication scheme achievesa superior performance over manual threshold strategy [15]Based on the above observation the proposed machinelearning based authentication scheme with tow-dimensionalfeature not only exhibits excellent performance than manualmethod but also has higher authentication rate than that ofthe same algorithm with one-dimensional feature
5 Conclusions
In this paper machine learning algorithm based physical-layer channel authentication for the 5G wireless communica-tion security is proposed A machine learning authenticationmethod could draw a conclusion whether the received pack-ets are from a legitimate transmitter or from a counterfeiterby using one-dimension or two-dimensional joint featuresThe effectiveness of the proposed authentication scheme isvalidated by widely simulations All the data used in thesimulation are derived from real OFDM-MIMO commu-nication platform which provides a real communicationenvironment Moreover the authentication results show thatthe novel methods provide a higher rate in detecting thespoofing attacks than those of the manual threshold basedphysical layer authentication schemes The training of theclassifier can be done offlineTherefore the novel method canperform authentication fast In addition whether we can usemore machine learning algorithms to further optimize ourauthentication model and find a better statistical test of largedifference in channel information is issue that we need to dealwith in the future
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare no conflicts of interest
Acknowledgments
This research was supported by NSFC (no 61572114)Sichuan Sci amp Tech Achievements Transformation Project(no 2016CC003) Sichuan Sci amp Tech Service Develop-ment Project (no 18KJFWSF0368) Hunan Provincial NatureScience Foundation Project 2018JJ2535 Chile CONICYTFONDECYTRegular Project 1181809 andNational Key RampDProgramofChina (2018YFB0904900 and 2018YFB0904905)
References
[1] M Agiwal A Roy andN Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys amp Tutorials vol 18 no 3 pp 1617ndash1655 2016
[2] J Thompson X Ge and H-C Wu ldquo5G wireless communica-tion systems prospects and challenges [Guest Editorial]rdquo IEEECommunications Magazine vol 52 no 2 pp 62ndash64 2014
[3] C Perera C H Liu S Jayawardena and M Chen ldquoA surveyon internet of things from industrial market perspectiverdquo IEEEAccess vol 2 pp 1660ndash1679 2014
[4] A Bogdanov M Knezevic G Leander D Toz K Varıcı andI Verbauwhede ldquoSPONGENT the design space of lightweightcryptographic hashingrdquo IEEE Transactions on Computers vol62 no 10 pp 2041ndash2053 2013
[5] R Zhang L Zhu C Xu and Y Yi ldquoAn Efficient and secureRFID batch authentication protocol with group tags ownershipTransferrdquo IEEE Collaboration and Internet Computing pp 168ndash175 2015
[6] L Hu H Wen B Wu et al ldquoCooperative Jamming forPhysical Layer Security Enhancement in Internet of ThingsrdquoIEEE Internet of ings Journal vol 5 no 1 pp 219ndash228 2018
[7] L Hu H Wen B Wu J Tang F Pan and R-F LiaoldquoCooperative-Jamming-Aided Secrecy Enhancement in Wire-less Networks with Passive Eavesdroppersrdquo IEEE Transactionson Vehicular Technology vol 67 no 3 pp 2108ndash2117 2018
[8] H Wen Physical layer approaches for securing wireless commu-nication systems Springer New York NY USA 2013
[9] LHuHWen BWu J Tang and F Pan ldquoAdaptive Base StationCooperation for Physical Layer Security in Two-Cell WirelessNetworksrdquo IEEE Access vol 4 pp 5607ndash5623 2016
[10] L Hu H Wen B Wu J Tang and F Pan ldquoAdaptive SecureTransmission for Physical Layer Security in Cooperative Wire-less Networksrdquo IEEE Communications Letters vol 21 no 3 pp524ndash527 2017
[11] D B Faria andDRCheriton ldquoDetecting identity-based attacksin wireless networks using signalprintsrdquo in Proceedings of theACM Workshop on Wireless Security pp 43ndash52 Los AngelesCalif USA 2006
[12] M Demirbas and Y Song ldquoAn RSSI-based scheme for sybilattack detection in wireless sensor networksrdquo in Proceedings of
10 Wireless Communications and Mobile Computing
the WoWMoM 2006 2006 International Symposium on aWorldof Wireless Mobile andMultimedia Networks pp 564ndash568 June2006
[13] Y ChenW Trappe and R P Martin ldquoDetecting and localizingwireless spoofing attacksrdquo in Proceedings of the 2007 4th AnnualIEEE Communications Society Conference on Sensor Mesh andAd Hoc Communications and Networks SECON pp 193ndash202San Diego Calif USA June 2007
[14] N Patwari and S K Kasera ldquoRobust location distinction usingtemporal link signaturesrdquo in Proceedings of the ACM Interna-tional Conference on Mobile Computing and NETWORKINGpp 111ndash122 2007
[15] H Wen Y Wang X Zhu J Li and L Zhou ldquoPhysical layerassist authentication technique for smart meter systemrdquo IETCommunications vol 7 no 3 pp 189ndash197 2013
[16] J K Tugnait ldquoWireless user authentication via comparison ofpower spectral densitiesrdquo IEEE Journal on Selected Areas inCommunications vol 31 no 9 pp 1791ndash1802 2013
[17] Z Jiang J Zhao X Li J Han and W Xi ldquoRejecting the attackSource authentication for Wi-Fi management frames using CSIInformationrdquo in Proceedings of the IEEE INFOCOM 2013 -IEEE Conference on Computer Communications pp 2544ndash2552Turin Italy April 2013
[18] X Wu and Z Yang ldquoPhysical-layer authentication for multi-carrier transmissionrdquo IEEE Communications Letters vol 19 no1 pp 74ndash77 2015
[19] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-authentication protocol for spoofing detection in wireless net-worksrdquo IEEE Transactions on Vehicular Technology vol 65 no12 pp 10037ndash10047 2016
[20] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-LayerSpoofing Detection with Reinforcement Learning in WirelessNetworksrdquo IEEE Transactions on Vehicular Technology vol 65no 12 pp 10037ndash10048 2016
[21] P Hao X Wang and A Refaey ldquoAn enhanced cross-layerauthentication mechanism for wireless communications basedon PER and RSSIrdquo in Proceedings of the 2013 13th CanadianWorkshop on Information eory CWIT 2013 pp 44ndash48Canada June 2013
[22] S Chen et al ldquoMachine-to-Machine communications in ultra-dense networksmdashA surveyrdquo IEEE Communications Surveys ampTutorials vol 1 no 1 99 pages 2017
[23] L Xiao L J Greenstein N B Mandayam and W TrappeldquoChannel-based detection of sybil attacks in wireless networksrdquoIEEE Transactions on Information Forensics and Security vol 4no 3 pp 492ndash503 2009
[24] H Wen P-H Ho C Qi and G Gong ldquoPhysical layer assistedauthentication for distributed ad hoc wireless sensor networksrdquoIET Information Security vol 4 no 4 pp 390ndash396 2010
[25] J H Friedman ldquoGreedy function approximation a gradientboosting machinerdquo Annals of Statistics vol 29 no 5 pp 1189ndash1232 2001
[26] J Friedman T Hastie and R Tibshirani ldquoAdditive logisticregression a statistical view of boostingrdquo e Annals of Statis-tics vol 28 no 2 pp 337ndash407 2000
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
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
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
10 Wireless Communications and Mobile Computing
the WoWMoM 2006 2006 International Symposium on aWorldof Wireless Mobile andMultimedia Networks pp 564ndash568 June2006
[13] Y ChenW Trappe and R P Martin ldquoDetecting and localizingwireless spoofing attacksrdquo in Proceedings of the 2007 4th AnnualIEEE Communications Society Conference on Sensor Mesh andAd Hoc Communications and Networks SECON pp 193ndash202San Diego Calif USA June 2007
[14] N Patwari and S K Kasera ldquoRobust location distinction usingtemporal link signaturesrdquo in Proceedings of the ACM Interna-tional Conference on Mobile Computing and NETWORKINGpp 111ndash122 2007
[15] H Wen Y Wang X Zhu J Li and L Zhou ldquoPhysical layerassist authentication technique for smart meter systemrdquo IETCommunications vol 7 no 3 pp 189ndash197 2013
[16] J K Tugnait ldquoWireless user authentication via comparison ofpower spectral densitiesrdquo IEEE Journal on Selected Areas inCommunications vol 31 no 9 pp 1791ndash1802 2013
[17] Z Jiang J Zhao X Li J Han and W Xi ldquoRejecting the attackSource authentication for Wi-Fi management frames using CSIInformationrdquo in Proceedings of the IEEE INFOCOM 2013 -IEEE Conference on Computer Communications pp 2544ndash2552Turin Italy April 2013
[18] X Wu and Z Yang ldquoPhysical-layer authentication for multi-carrier transmissionrdquo IEEE Communications Letters vol 19 no1 pp 74ndash77 2015
[19] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-authentication protocol for spoofing detection in wireless net-worksrdquo IEEE Transactions on Vehicular Technology vol 65 no12 pp 10037ndash10047 2016
[20] L Xiao Y Li G Han G Liu and W Zhuang ldquoPHY-LayerSpoofing Detection with Reinforcement Learning in WirelessNetworksrdquo IEEE Transactions on Vehicular Technology vol 65no 12 pp 10037ndash10048 2016
[21] P Hao X Wang and A Refaey ldquoAn enhanced cross-layerauthentication mechanism for wireless communications basedon PER and RSSIrdquo in Proceedings of the 2013 13th CanadianWorkshop on Information eory CWIT 2013 pp 44ndash48Canada June 2013
[22] S Chen et al ldquoMachine-to-Machine communications in ultra-dense networksmdashA surveyrdquo IEEE Communications Surveys ampTutorials vol 1 no 1 99 pages 2017
[23] L Xiao L J Greenstein N B Mandayam and W TrappeldquoChannel-based detection of sybil attacks in wireless networksrdquoIEEE Transactions on Information Forensics and Security vol 4no 3 pp 492ndash503 2009
[24] H Wen P-H Ho C Qi and G Gong ldquoPhysical layer assistedauthentication for distributed ad hoc wireless sensor networksrdquoIET Information Security vol 4 no 4 pp 390ndash396 2010
[25] J H Friedman ldquoGreedy function approximation a gradientboosting machinerdquo Annals of Statistics vol 29 no 5 pp 1189ndash1232 2001
[26] J Friedman T Hastie and R Tibshirani ldquoAdditive logisticregression a statistical view of boostingrdquo e Annals of Statis-tics vol 28 no 2 pp 337ndash407 2000
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
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
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
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
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom