IAES Inter national J our nal of Artificial Intelligence (IJ-AI) V ol. 9, No. 2, June 2020, pp. 193 202 ISSN: 2252-8938, DOI: 10.11591/ijai.v9i2.pp193-202 r 193 F eatur es detection based blind hando v er using kullback leibler distance f or 5G HetNets systems Adnane El Hanjri 1 , Aawatif Hayar 2 , Abdelkrim Haqiq 3 1,3 IR2M Laboratory , F aculty of Sciences and T echniques, Hassan 1st Uni v ersity , Morocco 2 GREENTIC, ENSEM, Hassan II Uni v ersity , Morocco Article Inf o Article history: Recei v ed No v 03, 2019 Re vised Apr 4, 2020 Accepted Apr 19, 2020 K eyw ords: Akaik e information criterion Akaik e weight Hando v ers K ullback leibler distance Small cells ABSTRA CT The Fifth Generation of Mobile Netw orks (5G) is changing the cellular netw ork infrastructure paradigm, and small cells are a k e y piece of this shift. But the high number of small cell s and their lo w co v erage in v olv e more Hando v ers to pro vide continuous connecti vity , and the selection, quickly and at lo w ener gy cost, of the appropriate one in the vicinity of thousands is also a k e y problem. In this paper , we propose a ne w method, to ha v e an ef ficient, blind and rapid hando v er just by analysing recei v ed signal probability density functi on instead of demodulating and analysing recei v ed signal i tself as in classical hando v er . The proposed method e xploits kullback leibler distance (KLD), akaik e information criterion (AIC) and akaik e weights, in order to decide blindly the best hando v er and the best base station (BS) for each user . This is an open access article under the CC BY -SA license . Corresponding A uthor: EL HANJRI Adnane, IR2M laboratory , F aculty of Sciences and T echniques, Hassan 1st Uni v ersity , Settat, Morocco. Email: adnane.elhanjri@gmail.com 1. INTR ODUCTION Mobile cellular communication [1] has become increasingly one of the most interesting re search area o v er the past fe w years. The e xponentially increasing demand for wireless data services [2] require a massi v e netw ork densification that is neither economically nor ecologically viable with current cellular system architectures. Fifth Generation (5G) [3-4] ha v e recently emer ged to s atisfy the increasing demand for high data bit rates. A crucial requirement for 5G netw orks is the deplo yment of Small Cells (SCs) [5] o v er Macrocells layer whi ch introduces a ne w type of netw orks called Heterogeneous Netw orks (HetNets) [6]. A HetNet is simply the banding together of dif ferent sized cells to pro vide ultra dense co v erage in defined geographic areas. Small Cells (SCs) are lo w-po wered cellular radio access nodes that operate in licensed and unlicens ed spectrum that ha v e a range of 10 meters to a fe w kil ometers. The y will be a crucial component of 5G netw orks, because the y ha v e the a b i lity to significantly increase netw ork capacity , density and co v erage, especially indoors. The y are a relati v ely lo w cost deplo yment option and, because the y are lo w po wer de vices [7], are relati v ely cheap and ef ficient to run to gi v e a lo w total cost of o wnership. Lik e e v ery other technology , SCs ha v e some dra wbacks that gi v e rise to some major concern on part of the end users. In this paper , we are going to study the problem of the management of hando v ers. Hando v er is the practice of retaining a user’ s acti v e connection when a mobile terminal changes its connection point to the access netw ork (called “point of attachment”) [8-9]. Because of the lo w co v erage of SCs, it is essential to J ournal homepage: http://ijai.iaescor e .com Evaluation Warning : The document was created with Spire.PDF for Python.
194 r ISSN: 2252-8938 support seamless hando v ers to pro vide continuous connecti vity wi thin an y wide area netw ork. In addition, due to the high number of SCs, hando v ers increase, and the selection, quickly and at lo w ener gy cost, of the appropriate one in the vicinity of thousands is also a k e y problem. Hence, we propose a ne w method to operate and manage a blind hando v er between a number of users and Base Stations of SCs. Ev ery hando v er process contains three phases logical ly [10]. The first step concerns the mea surement or information g athering phase, where the UE measures the signal strength of e v ery potential neighbor BS and the cur rent serving station. The second phase is about the hando v er decision, where the current serving BS decides about initializing the hando v er based on the measured data from the first stage. And the last one, is the cell e xchange, when the UE releases the serving e v olv ed NodeB (eNB) and connects to the ne w one. F or 5G Netw orks, Artificial Intelligence [11] can be broadly applied in the Blind Hando v er techniques. The usage of Artificial Intelligence techniques in the Handof f decision process will reduce the computation comple xity which already e xists in the con v entional methods. The main idea is to operate ef ficient, blind and rapid hando v er just by analysing recei v ed signal probability density function(pdf) instead of demodulating and analysing recei v ed signal itself as in classical hando v er . The goal within our contrib ution is to e xploit kullback leiber distance, akaik e information criterion (AIC) and akaik e weights [12-13] in order to decide blindly the best hando v er and the best BS for each user . The remainder of the paper is or g anized as follo ws. W e be gin by introducing, a brief o v ervie w of related w ork in section 2. In section 3, we re visit KLD and present the formulation of our problem. In section 4 we gi v e a brief re vie w of model selection using AIC: the AIC is presented and the akaik e weights are deri v ed. The approach based on model selection is de v eloped in Section 5. The e v aluation of the result is in section 6. The last section will be de v oted to the conclusion. 2. RELA TED W ORK In mobile telec o m munications systems, there ar e circumstances where it is desirable for a mobile terminal (such as a telephone, portable computer with communications capabilities, etc.), which is operating at a first frequenc y in a first netw ork belonging to a first system to transfer to a second netw ork operating at a second frequenc y (which may belong to a second s ystem), that is, a system using a dif ferent type of technology and defined according to a dif ferent standard. Dif ferent types of hando v er may be en visaged: If the primary netw ork is a time di vision duple x (TDD) netw ork [14] then, e v en while the mobile terminal is transmitting or recei ving data/v oice, there are time slots when it is inacti v e (that is, it is neither sending nor transmitting signals). These time slots can be used to perform measurements on channels operating at other frequencies, thus enabling the terminal to e v aluate the performance of candidate tar get netw orks. Ho we v er , if the primary netw ork is a frequenc y di vision duple x (FDD) netw ork [14], such as a Uni v ersal Mobile T ele communication System (UMTS) FDD netw ork then, when the terminal is acti v e and currently transmitting or recei ving data, there are no inacti v e periods a v ailable for performing measurements at other frequencies. So, in this case, the terminal cannot readily e v aluate the performance of candidate tar get netw orks. V arious techniques ha v e been proposed to enable intra-system inter -frequenc y hando v ers, or inter -system hando v ers, to be performed by terminals operating in primary netw orks using FDD (such as UMTS FDD netw orks). Man y techniques ha v e been propos ed using measurements on the T ar get Netw ork [15]: A first approach which enables measurements to be made on the tar get netw ork is the “dual recei v er” approach which means that mobile station has tw o recei v er branches, one recei ving branch measures the signal strength and quality on the other frequenc y while another recei ving branch are k eeping track on transmitting and recei ving signals of the current frequenc y . This is especially suitable for antenna di v ersity in mobile station. This approach has a number of disadv antages. Firstly , po wer consumption of the terminal is increased. Secondly , if the terminal is adapted to operate both in UMTS FDD netw orks and in GSM 1800 netw orks then a problem ca n arise (due to t h e closeness of the frequencies of the UMTS FDD uplink band and the GSM 1800 do wnlink band) when the contemplated hando v er is from an UMTS FDD netw ork to a GSM 1800 netw ork. More specifically , if the frequencies corresponding to the UMTS FDD uplink band and the GSM 1800 do wnlink band are not perfectly isolated then the dual recei v er terminal may not be able to demodulate them both. In such a case another technique w ould be required in order to enable the terminal to perform measurements on the tar get netw ork. Finally , the mobile terminal comprises tw o recei v ers and, accordingly , requires e xtra circuitry compared to a standard terminal: whi ch increases its size, cost and comple xity . Int J Artif Intell, V ol. 9, No. 2, June 2020 : 193 – 202 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Artif Intell ISSN: 2252-8938 r 195 A second approach which enables the terminal to mak e measurements on the tar get netw ork consists in operating the terminal in “compressed mode’ [16-18]. The compressed mode, often referred to as the slotted mode, is needed when making measurements from another frequenc y in a CDMA system without a f u l l dual recei v er terminal. The compressed mode means that transmission and reception are halted for a short time, in the order of a fe w milliseconds, in order to perform measurements on the other frequencies. The intention is not to lose data b ut to compress the data transmission in the time domain. An e xisting feature in which the netw ork node, e.g. an eNB in case of L TE, may initiate a hando v er procedure for a terminal without doing con v entional measurement configuration and without consideri ng mea- surement reports is Blind Hando v er . This feature may be beneficial when a f ast hando v er is needed and candi- date cell measurements are una v ailable, or w ould impose an unw anted delay . Using the blind hando v er in such case remo v es the time and signaling needed to conduct hando v er measurements, hence gi ving the desired f ast hando v er . Blind Hando v er T echniques [15]: A beacon pilot blind hando v er technique has been proposed in which the tar get netw ork, which normally operates at a frequenc y f. broadcasts a “beacon pilot’ at the same frequenc y f. as the frequenc y of the primary netw ork. This beacon pilot consists of a pilot channel and a synchronisation channel and enables the mobile terminal to e v aluate the propag ation loss between itself and the tar get netw ork. One disadv antage of the “beacon pilot approach is that it requires deplo yment of pi lot antennas, increasing the cost of the system infrastructure. Another disadv antage arises in the case of an intra-system, inter -frequenc y hando v er between primary and tar get netw orks which are UMTS FDD netw orks operating at adjacent frequencies. In this case the pilot transmission can generate i nterference on the tar get netw ork, making its capacity decrease. Another kno wn blind hando v er consists in a “direct blind hando v er in which a look-up table is held, for e xample, in the Radio Netw ork Controller (RNC) of the primary netw ork (assuming an UMTS FDD primary netw ork). This look-up table (or “planning table’) indicates, for each primary cell, which tar get cell should be used in a hando v er . If the hando v er is between systems ha ving co-located cells then this blind hando v er method w orks reasonably well. Ho we v er , in the case where the transfer is an inter -system transfer there is no guarantee that the boundaries of the cells of the tw o systems will be defined in the same locations. If the primary and tar get cells are not co-located then the quality of the connection a v ailable in the tar get cell will v ary depending upon the geographic location of the mobile terminal within the primary c ell. Thus, for mobile terminals at certain locations within the primary cell, the tar get cell specified in t he planning table will not be the best one to use. 3. DESCRIPTION AND FORMULA TION OF THE PR OBLEM The main idea in our contrib ution is to detect the best BS for each user (Best Hando v er) by e xploi ting model selection techniques and especially the AIC. It w as sho wn in [19] that, when signal demodulation cannot be perf o r med, the recei v ed wireless communication signal can be, roughly , modeled using Rayleigh and Rician distrib ution. Therefore, we propose to calculate in blindly process the Recei v ed Signal for each BS and Analyze AIC in order to determine the best hando v er . Figure 1 presents an illustrated model of Small Cells Netw ork. Figure 1. Model of small cells netw ork F eatur es detection based blind hando ver using kullbac k leibler distance for ... (Adnane El Hanjri) Evaluation Warning : The document was created with Spire.PDF for Python.
196 r ISSN: 2252-8938 In this section, we will gi v e a short re vie w of the basic ideas. In f act, it is assumed that the samples of the Recei v ed Signal for each BS are dis trib uted according to an original probability density function f k where k 2 f 1 ; 2 ; 3 ; 4 ; 5 ; 6 g is the inde x of BS, called the operating model. Since only a finite number of observ ations is a v ailable, the operating model is usually unkno wn. Therefore, approximating model (i.e candidate model) must be specified using the observ ed data, in order to estimate the operating model. The candidate model is denoted as g k , where indicates the U-dimensional parameter v ector , which specifies the probability density function. In information theory [20], the K ullback-Leibler distance describes the discrepanc y between the tw o probability density functions f k and g k and is gi v en by [12]: D ( f k k g k ) = E ( l og ( f k ( x ))) E ( l og ( g k ( x ))) D ( f k k g k ) = h i ( x ) Z f k ( x ) l og ( g k ( x )) dx (1) where h(.) denotes dif ferential entrop y . Since, the original probability density function f k is not kno wn, this distance measure is not directly applicable. It is kno wn, ho we v er , that the K ullback-Leibler distance is nonne g ati v e, this implies that the K ullback- Leibler discrepanc y , Z f k ( x ) l og ( g k ( x )) dx = h i ( x ) + D ( f k k g k ) (2) approaches the dif ferential entrop y of X from abo v e for increasing quality of the model g k . Applying the weak la w of lar ge numbers [21], this e xpression (2) can be approximated by a v eraging the log-lik elihood v alues gi v en the model o v er N independent observ ations x 1 ; x 2 ; :::; x N according to: Z f k ( x ) l og ( g k ( x )) dx 1 N N X n =1 g k ( x n ) (3) The e xpected K ullback-Leibler discrepanc y is gi v en by [17]: E Z f k ( x ) l og ( g k ( x )) dx (4) This e xpression (4) cannot be computed, b ut estimated. 4. MODEL SELECTION USING AKAIKE INFORMA TION CRITERION The information theoretic criteria w as first introduced by Akaik e in [8] for model selection. Assuming a candidate model, the idea is to deci de if the distrib ution of the observ ed signal fits the candi- date model. The AIC criterion is an approximately unbiased estimator for (4) and is gi v en by: AI C k = 2 N X n =1 l og ( g k ( x n )) + 2 U (5) where U indicates the dimension of the parameter v ector . One should select the model that yields the smallest v alue of AIC because t his model i s estimated to be the closest to the unkno wn reality that generated the data, from among the candidate models considered. The parameter v ector for each f amily should be estimated using the minimum discrepanc y estimator b , which minimizes the empirical discrepanc y . This is the discrepanc y between the approximating model and the model obtained by re g arding the observ ations as the whole population. The maximum lik elihood estimator [22] is the minimum discrepanc y estimator for the K ullback-Leibler discrepanc y [12]. Consider a probability distrib ution parameterized by an unkno wn parameter , associated with either a kno wn probability density function or a kno wn probability mass function, denoted as f k . As a function of Int J Artif Intell, V ol. 9, No. 2, June 2020 : 193 – 202 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Artif Intell ISSN: 2252-8938 r 197 with x 1 ; x 2 ; :::; x N fix ed, the lik elihood function is: L k ( ) = f k ( x 1 ; x 2 ; :::; x N ) (6) The method of maximum lik el ihood estimates by finding the v alue of that maximizes L k ( ) . The maximum lik elihood estimator (MLE) [22] of is gi v en by: b = argmax L k ( ) (7) Commonly , one assumes that the data dra wn from a particular distrib ution are i.i.d. with unkno wn parameters. This considerably simplifies the problem because the lik elihood can then be written as a product of N uni v ariate probability densities: L k ( ) = N Y n =1 f k ( x n j ) (8) and since maxima are unaf fected by monotone transformations, one can tak e the log arithm of this e xpression to turn it into a sum: L k ( ) = N X n =1 l og f k ( x n j ) (9) Consequently , the e xpression of the maximum lik elihood in our case is [19]: b = argmax 1 N N X n =1 l og ( g k ( x n )) (10) The maximum of this e xpression can then be found numerically using v arious optimization algorithm s [17]. This contrasts with seeking an unbiased estimator of , which may not necessarily yield the MLE b ut which will yield a v alue that (on a v erage) will neither tend to o v er -estimate nor under -estimate the true v alue of . The maximum lik elihood estimator may not be unique, or indeed may not e v en e xist. Because AIC contains v arious constants and is a function of sample size, we routinely recommend computing (and presenting in publications) the AIC dif ferences(in addition to the actual AIC v alues): k = AI C k AI C min (11) where AI C min denotes the minimum AIC v alue o v er all BSs. Akaik e weights can be computed using (5), in order to decide if the distrib ution of the Recei v ed Signal fits the candidate distrib ution or not. The Akaik e weights can be interpreted as estimate for the probabilities that the corresponding candidate distrib ution sho w the best modeling fit. It pro vides another measure of the strength of e vidence for this model, and is gi v en by: W k = e 1 = 2 k P 6 i =1 e 1 = 2 i where k 2 f 1 ; 2 ; 3 ; 4 ; 5 ; 6 g (12) The Akaik e weights allo w us not only to decide if the distri b ut ion of the Recei v ed Signal fits the Gaussian distrib ution, b ut also pro vide information about the relati v e approximation quality of this distrib ution. The maximum Lik elihood estimator is the minimum discrepanc y estimator for the KL discrepanc y [12]. In our problem, we w ant Line Of Sight (LOS) signal between the BS and the users. Consequently , we are going to use the Rice distrib ution [23]. So the probability density function for the Recei v ed Signal for each BS is gi v en by: g k ( x j k ; k ) = x 2 k exp ( x 2 + 2 k ) 2 2 k I 0 ( x k 2 k ) (13) where I 0 ( x k 2 k ) is the modified Bessel function of the first kind with order zero , k is the mean or e xpectation F eatur es detection based blind hando ver using kullbac k leibler distance for ... (Adnane El Hanjri) Evaluation Warning : The document was created with Spire.PDF for Python.
198 r ISSN: 2252-8938 of the distrib ution (and also its median and mode) and k is the standard de viation. The approximated probability density function leads to the follo wing log-lik elihood function : L k ( k ; k ) = l og   Q N i =1 x i 2 N k exp   P N i =1 ( x 2 i + 2 k ) 2 2 k ! N Y i =1 I 0 ( x i k 2 k ) ! (14) P arameters k and k are gi v en by the solution of the follo wing set of equations: 8 < : k 1 N P N i =1 x i I 1 ( x i k 2 k ) I 0 ( x i k 2 k ) = 0 2 k + 2 k 1 N P N i =1 x 2 i = 0 (15) where I 1 ( x i k 2 k ) = I 0 ( x i k 2 k ) + 2 k 2 x k I 0 ( x i k 2 k ) is the modified Bessel function [24] with order one. When x i k 2 k >> 0 : 25 and I 0 ( x i k 2 k ) = exp ( x i k 2 k ) r 2 x i k 2 k , (15) can be e xpressed as: ( 2 k + 1 N P N i =1 x i k 2 k 2 = 0 2 k 1 N P N i =1 x 2 i + 2 2 k = 0 (16) Resolving (16), the MLE for the parameters c k , c k can be e xpressed as: ( c k = 2 P N i =1 x i + p (4( P N i =1 x i ) 2 +5 N P N i =1 x 2 i ) 5 N c k 2 = 1 2 c k 2 + 1 2 N P N i =1 x 2 i (17) And the parameter v ector = ( k ; k ) 5. THE APPR O A CH In this section, we present a ne w approach to detect the best hando v er based on e xploiting model selection techniques and especially AIC introduced by Akaik e in [12, 13]. W e consider that the initial signal can be modeled using Gaussian distrib ution and its norm can be modeled using Rician distrib ution. After the input of the v alues of the Recei v ed Signal for each BS (observ ations), in the first step we compute the parameters c k and c k (MLE parameters), then g k the pdf for the Recei v ed Signal for each BS k . Once we ge t g k , we calculate AI C k and W k for each BS. The Akaik e weights allo w us not only to decide if the distrib ution of the Recei v ed Signal fits the suitable distrib ution, b ut also pro vide information about the best signal (best BS) for each user . If the Akaik e weight of Rician distrib ution of the B S k is higher than the Akaik e weights of other BSs, then there is no Hando v er , and if the Akaik e weight of B S k is lo wer than the Akaik e weight of B S i where i 2 f 1 ; 2 ; 3 ; 4 ; 5 ; 6 g then there is Hando v er from B S k to B S i . thr eshol d ( x n ) = W k W i < thr eshol d Hando v er ( H 0 ) W k W i > thr eshol d No Hando v er ( H 1 ) (18) The decision threshold is determined by using the probability of f alse alarm P F A [25]. The threshold thr eshol d for a gi v en f alse alarm probability [25] is determined by solving the equation P F A = P ( thr eshol d ( x ) < thr eshol d j H 1 ) (19) The flo w chart of the proposed algorithm is sho wn in Figure 2, which can be implemented in four steps: Int J Artif Intell, V ol. 9, No. 2, June 2020 : 193 – 202 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Artif Intell ISSN: 2252-8938 r 199 Compu t e par amet er s σ an d   μ Compu t e  and   Com p u t e   the   p d f   of   the    Rece ived Signal  f o r   each   B S In p u t Rece ived Signal Thr es hold Thr es hold No  H and o v er H and o v er Figure 2. Flo wchart of Algorithm of Blind Hando v er based on distrib ution analysis 6. RESUL T AND AN AL YSIS The proposed Blind Detection approach is e v aluated using the softw are package Matlab R2016a. The Figure 3, sho ws the v alues of Akaik e W eights of the six BSs in a time t. W e apply the approach in Figure 2 and we compute the Akaik e W eights for the BSs in terms to choose the best BS for the us er . Figure 3 depicts the Akaik e W eights with Gaussian distrib ution obtained from the 6 BSs. It is clearly sho wn that the BS which has the Maximum Akaik e weight is the first BS, so the best BS for the user is the B S 1 . In Figure 4 we can see the dif ference between Rice and Rayleigh Distrib ution of the Recei v ed Signal. When the Signal between the BS and the UE is suf fering from shado wing by a high b uilding o v er the sensing channel, it definitely can decrease the Recei v ed Signal due to the lo w recei v ed SNR. When the SNR is lo w , the noise distrib ution will dominate in the con v olution and the resulting distrib ution will tend to become close to Gaussian e v en if the signal has an arbitrary non Gaussian distrib ution, and the en v elope (norm) distrib ution of the signal is close to Rayleigh distrib ution. Another important property is the contri b ut ion of the dominant propag at ion paths on the dis trib ution of the Recei v ed Signal. The en v elope distrib ution of the Recei v ed Signal tend to become close to R ician e v en if the input has a non Rician distrib ution . The Akaik e weight of Rician distrib ution is higher than Akaik e weight of Rayleigh distrib ution that mean that BS with Rician Distrib ution is the best for the UE. 1 2 3 4 5 6 7 BS Index 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Akaike weight Figure 3. Akaik e weights of the six BSs at time t F eatur es detection based blind hando ver using kullbac k leibler distance for ... (Adnane El Hanjri) Evaluation Warning : The document was created with Spire.PDF for Python.
200 r ISSN: 2252-8938 1 2 BS Index 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Akaike weight The BS with Rice Distribution The BS with Rayleigh Distribution Figure 4. The comparison of akaik e weight of tw o BSs with dif ferent distrib ution 7. CONCLUSION In this w ork, we studied a ne w method to manage the hando v ers between a number of users and Base Stations of Small Cells. Our idea has been based on analysing the probability density function of the Recei v ed Signal for each BS, to pro vide an indication of the intensity of the Recei v ed Signal, and e xploit KL Di v er gence, Akaik e Information Criterion and Akaik e W eight in order to decide the best hando v er and the best BS for each user . The proposed Blind Detection Approach is e v aluated using the softw are package Matlab R2016a. REFERENCES [1] D. Pinedal, C. Hernandez, “Cogniti v e radio for TVWS usage, ” TELK OMNIKA T elecommunication, Computing, Electronics and Control , v ol. 17, no. 6, pp. 2735-2746, December 2019. [2] E. Halepo vic, C. W illiamson, and M. Ghaderi , “W ireless Data T raf fic: A Decade of Change, ” IEEE Netw ork, pp. 20-26, March/April 2009. [3] D. T . Do, C. B. Le, H. N. Nguyen, T . N. Kieu, S. P . Le, N. L. Nguyen, N. T . Nguyen, M. V oznak, “W ireless po wer transfer enabled NOMA relay systems: tw o SIC modes and performance e v aluation, ” TELK OMNIKA T elecommunication, Comput ing, Electronics and Control , v ol. 17, no. 6, pp. 2697-2703, December 2019. [4] A. Gupta, R.K. Jha, “ A Surv e y o f 5G Netw orks: Architecture and Emer ging T echnologies, ” IEEE Access , v ol. 3, pp. 1206-1232, 2015. [5] J. Ho ydisMari, M. K obayashi and M. Debbah, “Green Small-cell Netw orks, ” IEEE V ehicular T echnology Mag azine, v ol. 6, no. 1, pp. 37-43, April 2011. [6] A. Damnjano vic, J. Montojo, Y .W ei, T . J. Luo, M. V ajape yam, T . Y oo, O. Song and D. Malladi, ”A Surv e y on 3GPP Heterogeneous Netw orks, ” IEEE W ireless Communications, June 2011. [7] D. Imededdin, A. Salih, H. Medk our , “Design and implementation of lo w po wer consumption wireless sensor node, ” TELK OMNIKA T elecommunication, Computing, Electronics and Control , v ol. 17, no. 6, pp.2729-2734, December 2019. [8] A. Ahamad, A. Sunda v arajan, M. Ismail, “Hando v er in L TE- adv anced wireless netw orks: state of art and surv e y of decision algorithm, ” Springer Science + Business Media , Ne w Y ork, 2017. [9] I. F . Ak yildiz, J. McNair , J. S. Ho, H. Uzunalioglu, and W . W ang, “Mobility management in ne xt- generation wireless systems, ” Proceedings of the IEEE, v ol. 87, no. 8, pp. 1347-1384, 1999. [10] L. K. Johal, A. S. Sandhu, “ An Ov ervie w of V ertical Hando v er Process and T echniques, ” Indian Journal of Science and T echnology , v ol. 9, no. 14, April 2016. [11] J. Mata, I. de Miguel, R .J . Dur ´ an, N . Merayo, S.cK. Singh, A. Jukan, M. Chamania, ”Artifi- Int J Artif Intell, V ol. 9, No. 2, June 2020 : 193 – 202 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Artif Intell ISSN: 2252-8938 r 201 cial intelligence (AI) methods in optical netw orks: A comprehensi v e surv e y , ” Optical Switching and Netw orking, 2018. [12] H. Akaik e, “Information theory and an e xtension of the maximum lik elihood principle, ” 2nd International Symposium on Information Theory , pp. 267-281, 1973. [13] H. Akaik e, “On the lik elihood of a time series model, ” The S tatistician, v ol. 27, no. 3/4, pp. 217-235, December 1978. [14] T . Ojanpera, R. Prasad, “WCDMA: T o w ards IP Mobility and Mobile Internet, ” Library of Congress Catatoging in Publication Data, 2001. [15] M. Pischella, “ Blind Hando v er T echnique, ” P atent NO: US 7, 224, 972 B2, 29 May , 2007. [16] M.Gustafsson, K. Jamal, E. Dahlman, “Compressed Mode techniques for Inter Frenquenc y Measurements in a wide bands DS-CDMA System, ” Proc. of PIMRC 97, pp. 23-35, Sep 1997. [17] H. Holma, A. T oskala, “WCDMA for UMTS, Radio Access for Third Generation Mobile Communica- tion, ” John W ile y and Sons, Inc, 2000. [18] M. Ridouani, A. Hayar and A.Haqiq, ”Perform sensing and transmission in parallel in cogniti v e radio systems: Spectrum and ener gy ef ficienc y , ” Digit. Signal Process, 2016. [19] B. Zayen, A. Hayar and D. Nussbaum, “ Blind Spectrum Sensing for Cogniti v e Radio Based on Model Selection, ” Cro wnCom08, 3rd International Conference on Cogniti v e Radio Oriented W ireless Netw orks and Communications, Mai 2008. [20] E. T . Jaynes, “ Information Theory and Statistical Mechanics, ” Ph ysical Re vie w , v ol. 106, no. 4, 1957. [21] L. V . Dung, T . C. Son and N. T . H. Y en, “ W eak La ws of Lar ge Numbers for Sequences of Random V ariables with Infinite rth Moments, ” Acta Math, 2018. [22] J. Jiao, K. V enkat, Y . Han and T . W eissman, “ Maximum Lik elihood Estimation of Functionals of Discrete Distrib utions, ” IEEE T ransactions on Information Theory , 2017. [23] J. Sijbers, Arnold J. den Dekk er , P . Scheunders, and D. V an Dyck, “ Maximum-Lik elihood Estimation of Rician Distrib ution P arameters, ” IEEE T ranssactions Ransactions on Medical Imaging, v ol. 17, no. 3, June 1998. [24] C. Robert, “ M odified Bessel functions and their applications in probability and statistics, ” Statistics and Probability Letters, v ol. 9, no. 2, pp. 155–161, 1990. [25] B. Zayen, A. Hayar and K. Kansanen, “ B lind Spectrum Sensing for Cogniti v e Radio Based on Signal Space Dimension Estimation, ” ICC’09, IEEE International Conference on Communications, June 2009. BIOGRAPHIES OF A UTHORS Adnane El Hanjri recei v ed his Bachelor’ s de gree in Applied Mathematics at the F aculty of Sciences, Ibn Zohr Uni v ersity , Ag adir , Morocco in 2013. In 2016, he obtained his Masters de gree in Mathe- matics and Applications from Hassan 1st Uni v ersity , Settat, Morocco. He is currently a Ph.D. student in Applied Mathematics and Computer Science at Computer , Netw orks, Mobility and Modeling lab- oratory , F aculty of Sciences and T echniques, Hassan 1st Uni v ersity , Settat, Morocco. His research interests include Information theory , stochastic processes, Mark o v chains and their applications for modeling wireless netw orks. F eatur es detection based blind hando ver using kullbac k leibler distance for ... (Adnane El Hanjri) Evaluation Warning : The document was created with Spire.PDF for Python.
202 r ISSN: 2252-8938 Aa w atif Hayar recei v ed, with honors as the First Moroccan, the de gree of ”Agre g ation Genie Elec- trique” from Ecole Normale Superieure de Cachan in 1992. She recei v ed the ”Diplome d’Etudes Ap- profondies” in Signal processing Image and Communications and the de gree of Engineer in T elecom- munications Systems and Netw orks from ENSEEIHT de T oulouse in 1997. She recei v ed with honors the Ph.D. de gree in Signal Processing and T elecommunications from Institut National Polytechnique in T oulouse in 2001. She w as research and teaching associate at EURECOM’ s Mobile Communica- tion Department from 2001 to 2010 in Sophia Antipolis-Fra nce. Aa w atif Hayar has an HDR (Habil- itation a Diriger la Recherche) from Uni v ersity Sud T oulon V ar from France on Cogniti v e W ideband W ireless Systems on 2010 and an HDR on Green T elecommunication from Uni v ersity Hassan II Casablanca (UH2C) on 2013. She has joined in 2011 the engineering school ENSEM-UH2C. Her research interests includes fields such as cogniti v e green communications systems, UWB systems, smart grids, smart cities, ICT for s ocial eco-friendly smart socio-economic de v elopment. Aa w atif Hayar w as a Guest Editor of Else vier Ph ycom Journal Special i ssue on Cogniti v e Radio Algorithms and System Design in 2009 and General Co-chair of Cro wncom2010 (France) , IW2GN2011, IEEE DL T Chair for EMEA re gion since 2014. General co-chair of ICT 2013 Conference, A w ards Chair for ICUWB2014 conference and T echnical Program Committee co-chair for Ne xt-Gwin W orkshop in 2014. She recei v ed with one of her PhD students the ”best student paper” a w ard at CogArt2010 and has a patent on sub space based blind sensing for cogniti v e radio. Aa w atif Hayar is currently leading or in v olv ed in a couple of R&D projects on Social Smart home, smart grids and frug al smart cities. Pr . Aa w atif Hayar is currently leading the Casablanca IEEE Core Smart city project, and the Hassan II Uni v ersity President.   Abdelkrim Haqiq has a High Study De gree and a PhD , both in the field of modeling and performance e v aluation of computer communication netw orks, from the Uni v ersity of Mohammed V , Agdal, F ac- ulty of Sciences, Rabat, Morocco. Since September 1995 he has been w orking as a Professor at the department of Mathematics and Computer at the F aculty of Sciences and T echniques, Settat, Mo- rocco. He is the Director of Computer , Netw orks, Mobility and Modeling laboratory . He is also the General Secretary of the electronic Ne xt Generation Netw orks (e -NGN) Research Group, Moroccan section. He is an IEEE Senior member and an IEEE Communications Society member . He is also a member of Machi ne Intelligence Research Labs (MIR Labs), W ashington, USA. He w as a co-director of a N A T O Multi-Y ear project entitled ”Cyber Security Analysis and Assurance using Cloud-Based Security Measurement system”, ha ving the code: SPS-984425. Dr . Abdelkrim HA QIQ’ s interests lie in the areas of modeling and performanc e e v aluation of communication netw orks, mobile com- munications netw orks, cloud computing and security , queueing theory and g ame theory . He is the author and co-author of more than 160 papers (international journals and conferences/w orkshops). He is also a member of the board of the International Journal of Intelligent Engineering Informat- ics. He is an associate editor of the Inte rnational Journal of Computer International Systems and Industrial Management Applications (IJCISM), an editorial board member of the International Jour - nal of Intelligent Engineering Informatics (IJIEI) and of the International Journal of Blockchains and Cryptocurrencies (IJBC), an international advisory board member of the International Journal of Smart Security T echnologies (IJSST) and of the International Journal of Applied Research on Smart Surv eillance T echnologies and Society (IJ ARSSTS). He is also an editorial re vie w board of the In- ternational Journal of F og Computing (IJFC) and of the International Journal of Digital Crime and F orensics (IJDCF). Int J Artif Intell, V ol. 9, No. 2, June 2020 : 193 – 202 Evaluation Warning : The document was created with Spire.PDF for Python.