A Service of

Leibniz-Informationszentrum econstor Wirtschaft Leibniz Information Centre Make Your Publications Visible. zbw for Economics

Anany, M. G.; El Din, Eman Serag; Elmesalawy, Mahmoud M.

Conference Paper Optimal Radio Access Network Selection in Multi- RAT HetNets Using Matching Game Approach

2nd Europe - Middle East - North African Regional Conference of the International Telecommunications Society (ITS): "Leveraging Technologies For Growth", Aswan, Egypt, 18th-21st February, 2019 Provided in Cooperation with: International Telecommunications Society (ITS)

Suggested Citation: Anany, M. G.; El Din, Eman Serag; Elmesalawy, Mahmoud M. (2019) : Optimal Radio Access Network Selection in Multi-RAT HetNets Using Matching Game Approach, 2nd Europe - Middle East - North African Regional Conference of the International Telecommunications Society (ITS): "Leveraging Technologies For Growth", Aswan, Egypt, 18th-21st February, 2019, International Telecommunications Society (ITS), Calgary

This Version is available at: http://hdl.handle.net/10419/201748

Standard-Nutzungsbedingungen: Terms of use:

Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Documents in EconStor may be saved and copied for your Zwecken und zum Privatgebrauch gespeichert und kopiert werden. personal and scholarly purposes.

Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle You are not to copy documents for public or commercial Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich purposes, to exhibit the documents publicly, to make them machen, vertreiben oder anderweitig nutzen. publicly available on the internet, or to distribute or otherwise use the documents in public. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, If the documents have been made available under an Open gelten abweichend von diesen Nutzungsbedingungen die in der dort Content Licence (especially Creative Commons Licences), you genannten Lizenz gewährten Nutzungsrechte. may exercise further usage rights as specified in the indicated licence. www.econstor.eu Optimal Radio Access Network Selection in Multi-RAT HetNets Using Matching Game Approach

M. G. Anany1, Eman Serag El Din2 and Mahmoud M. Elmesalawy3 1Department of Electronics and Communications Canadian International College, CIC, Cairo, Egypt 2Network Planning Department, National Telecommunication Institute, Cairo, Egypt 3Department of Electronics, Communications and Computers Engineering, Faculty of Engineering, Helwan University, Cairo, Egypt. Emails: [email protected], [email protected] and [email protected]

Abstract— Due to the dramatic growth in Particularly Wireless Local Area Network mobile data traffic, Multiple Radio Access (WLAN) has been introduced by 3GPP Release 8 Technologies (Multi-RAT) heterogeneous [2], as a candidate for interworking with cellular Networks (HetNets) have been proposed as a networks, due its high capacity, low deployment promising solution to cope with the high traffic complexity, and low cost. However, for an operator demand in mobile networks. In this work we adopting Heterogeneous Network (HetNet) with propose a User Equipment (UE) radio access trusted WLAN, some important challenges should be network selection algorithm in a Wireless Local addressed such as seamless authentication, seamless Area Network (WLAN) and LTE Multi-RAT mobility between different RATs, and more HetNet, where matching game approach is importantly the UE association and RAT selection applied. In this algorithm, UEs propose to their decision. In this paper we will use the terms network best candidate based on a utility function that is selection and UE association alternatively. Mobile formulated to maximize their achieved downlink operators require appropriate UE association for data rate. Then base stations accept or reject the efficient utilization of Long Term Evolution (LTE) proposals based on their utility. The performance and WLAN network resources. A UE can associate of the proposed approach is investigated and with WLAN Access Point (WAP) or an LTE base compared to other models, and simulation results station (BS). Thus, a UE association technique that proved its outperformance. optimizes network performance becomes necessary Keywords—Multi-RAT, HetNet, network [3]. selection, matching game. Recently, matching game has emerged as a promising technique for wireless resource allocation, I. INTRODUCTION and user association [4]. It is a Nobel Prize winning Recently it has been noticed a dramatic growth in framework that provides mathematically tractable wireless connectivity by User Equipment (UEs), solutions for the combinatorial problem of matching which in turn led to unprecedented growth in data players in two distinct sets, depending on the traffic. It has been predicted that by 2021 monthly individual information and preference of each player global mobile data traffic will exceed 49 exabytes [4]. It has been used widely for resource allocation in [1]. This puts high pressure and becomes one of the wireless networks, such as in cognitive radio (CR) serious challenges for cellular mobile network networks [5-7], heterogeneous cellular networks [8], operators and their resources. physical layer security systems [9], distributed orthogonal frequency-division multiple access (OFDMA) networks [10], routing, and queuing none of the previous aims to provide an optimum systems [5]. solution, since users individually take selfish RAT In HetNets, UE association is considered a selection decisions. significant challenge that received researchers’ Although matching game has been used widely in attention. In [11] WLAN-first strategy was analyzed, resource allocation for cellular networks, it is still in which UEs should select WLAN whenever it is amateur in the HetNets user association [5]. In [20], available. WLAN-first is considered one of the authors proposed a context-aware user association pioneer works for LTE and WLAN interworking. It approach based on matching theory for small cell is also considered as a baseline access strategy networks, which exploit the information about the among network selection approaches [12]. velocity and trajectory of the users while also taking In [13] the authors proposed a Q-learning into account their quality of service (QoS) algorithm to find the optimal policy that maximizes requirements. Also authors in [8] consider a HetNet a reward parameter. Although the reward parameter and propose a solution that jointly associates UEs to depends on the load of each detected WAP, the the Femto Access Points (FAPs), and allocates the Signal to Interference plus Noise Ratio (SINR), the FAPs to the SPs such that the total satisfaction of the handover duration, and the achievable rate, it does UEs in an uplink OFDMA network is maximized. not consider the different technologies represented They propose a distributed algorithms to find the by the effect of WLANs MAC protocols on the optimal UE association and FAP allocations based achievable rate calculations. on dynamic matching game theory. However, all the Authors in [14] introduce a practical probabilistic previous consider only the heterogeneity in transmit RAT selection approach in a heterogeneous network power, and they did not consider the heterogeneity in with two throughput classes, these association RATs. probabilities are calculated with the aim of network The rest of the paper is organized as follow: The throughput maximization. However it considers only next section describes the system model and the a general throughput classes, and does not consider problem formulation. Section III presents the data the different network parameter that could affect rates modeling in different RATs. In section IV, the these classes, and consequently the association matching game user association algorithm is probabilities. proposed. Then we evaluate the performance of our Multiple Attribute Decision Making (MADM) algorithm in section V, and the conclusion is methods are widely adopted in HetNet selection as in presented in section VI. [15-19]. Authors in [15] propose a HetNet selection algorithm based on the combination of network II. SYSTEM MODEL AND PROBLEM attributes and user preferences, by the use of a FORMULATION combination of three MADM methods, namely In this work, we consider an integrated cellular Fuzzy Analytic Hierarchy Process (FAHP), Entropy and small cell network, where small cells are WLAN and Technique for Order of preference by Similarity access points (WAP), overlaid under a macro base to Ideal Solution (TOPSIS). Also [16] proposed a station (MBS) as shown in Figure 1. The set of all strategy that depends on Analytic Hierarchy Process base stations (BS) is denoted by ℳ = (AHP) for UEs to select best RAT based on Signal to {1,2, … , 푚, … , 푀}, with cardinality 푀, where 푚 = 1 Noise Ratio (SNR), available bandwidth, delay, and refers to MBS, 푚 = 2,3, … , 푀 implies WAPs jitter. Moreover, authors in [17] proposed a flexible covered by MBS. Moreover, the set of WAPs are hybrid MADM algorithm consisting of FAHP, represented by Β = {2, … , 푀} with cardinality퐵, standard deviation, and Grey Relational Analysis such that Β ⊂ ℳ. A set of UEs are distributed in a (GRA) for the HetNet selection problem. However certain area (i.e. small building, or a mall) under the MBS and WAPs coverage and are denoted by Ι = {1,2, … , 푁} with cardinality 푁. Also, the set of interfering MBSs is denoted by K = {1,2, … , 퐾} with cardinality퐾, where 푘 ∈ K is the index of the interfering MBS. Furthermore, the number of UEs associated with 푀 푊 MBS, and WAPs are denoted by 푁 , and 푁푚 respectively, such that 푚 ∈ Β for WAPs. Moreover a quota for WAPs is represented by the maximum number of UEs that can be associated with a WAP 푁̂푊 ∀푚 ∈ Β , and it can be pre-calculated to achieve 푚 User WAP connection MBS connection Buiilldiing WAP MBS maximum throughput, considering a constant Figure 1. System Model. minimum and maximum contention window sizes (퐶푊 and 퐶푊 respectively) using Giuseppe 푚푖푛 푚푎푥 III. DATA RATE MODELLING Bianchi’s model [9]. Furthermore, a quota for MBS is represented by the maximum number of associated For modelling the UE’s downlink data rate from LTE UEs 푁̂퐿, in which it can be calculated to give a MBS, Shanon’s capacity formula can be used as minimum number of physical resource blocks per follow, 푃푅퐵 ̂푃푅퐵 user 푁푚푖푛 in an LTE system with 푁 physical 푃푅퐵 푃푅퐵 ̂푃푅퐵 푅 = 푁 푊 log (1 + 푆퐼푁푅 ), ̂퐿 푁 푖푚 2 푖푚 resource blocks (i.e. 푁 = 푃푅퐵). 푁푚푖푛 ∀푚 ∈ ℳ\Β (6) Since our aim is to maximize the overall network throughput by maximizing the UEs downlink data where 푊푃푅퐵is the bandwidth per each PRB, and rates, the user association problem can be formulated 푆퐼푁푅푖푚 for UE 푖 when associated with BS 푚 can be as follows, formulated as follows,

(1) 푃푚푔푚푖 푂푃푇 − 푈퐴: max ∑ ∑ 푥푖푚푅푖푚 푆퐼푁푅 = ∀푚 = 1, 푘 ∈ Κ (7) 푥 푖푚 ∑ 푃 푔 + 휎2 푚∈푀 푖∈퐼 푘 푘 푘푖 s.t. (2) where Κ is the set of neighbor interfering MBSs; 푃푚, ∑ 푥푖푚 ≤ 1, ∀푚, 푖 and 푃푘 denote the downlink transmit powers from 푚∈푀 MBS 푚 = 1 and interfering MBS 푘 over one radio 푥 = {0,1}, ∀푚, 푖, (3) 푖푚 channel respectively; 푔 , and 푔 represent the ̂퐿 (4) 푚푖 푘푖 ∑ 푥푖푚 ≤ 푁 , ∀푚 ∈ ℳ\Β, 푖, average channel gains of the link between MBS 푚 =

푖∈퐼 1 and UE 푖, and between interfering MBSs 푘 and UE ̂푊 (5) ∑ 푥푖푚 ≤ 푁푚 , ∀푚 ∈ 퐵, 푖, 푖, respectively; and 휎2 denotes the additive noise 푖∈퐼 power over each channel.

Furthermore, the downlink data rates achieved by where objective function in (1) aims to maximize the UE 푖 from BS 푚 ≥ 2 can be calculated as follows, total system throughput as a summation of UEs downlink data rates. Constraint (2) ensures that each 푅 = 푅푀퐴퐶(푛) 푊푊퐿퐴푁 log (1 + 푆퐼푁푅 ), UE is associated to only one BS, constraint (3) 푖푚 푚 푚 2 푖푚 ∀푚 ∈ Β (8) ensures that the decision variable 푥 is a binary 푀퐴퐶 푖푚 Here 푅 (푛) is the UE’s normalized throughput decision variable, constraint (4) and (5) ensures that 푚 that depends on associated users 푛 as in [21], and each BS will not exceed its quota. 푊퐿퐴푁 푊푚 is the WLAN channel bandwidth. The UE’s 푀퐴퐶 IV. MATCHING GAME BASED USER normalized throughput 푅푚 (푛) can be reformulated as follows, ASSOCIATION In order to develop a distributed algorithm for 푀퐴퐶 푅푚 푊 user association, the one-to-many matching game is 휏(1 − 휏)푁푚 퐷/(푊푊퐿퐴푁 푙표푔 (1 + 푆퐼푁푅 )) = 푚 2 푖푚 used as it can capture an optimal solution for the 푊 푁푊 1 푇 + 퐷휏(푁푚 + 1)(1 − 휏) 푚 ( 푊퐿퐴푁 ) OPT-UA problem. Under this design, each UE will 푊푚 log2(1 + 푆퐼푁푅푖푚) ∀푚 ∈ Β (9) be matched to at most one BS, while each BS 푚 can ̂퐿 ̂푊 be assigned to at most 푁 UEs, ∀푚 = 1, and 푁푚 푊 UEs, ∀푚 > 1. Here 푁푚 represents the number of users associated with WAP 푚 ∈ Β; 휏 denotes the channel In the matching game, we use the UE’s contention probability; 퐷 is the maximum allowed modulation order and code rate (modulation 푊퐿퐴푁 efficiency 퐸 ), to represent the UE’s utility function size of user packets; and 푊푚 is the bandwidth of 푖푚 each WAP for 푚 ∈ Β. For clarification, the when associated with BS 푚. It can be expressed as numerator represents the average data transferred in follows, a time slot, and the denominator is the average length 푈푖(푚) = 퐸푖푚 (11) of a time slot [21,22]. Moreover 푇 can be calculated ( ) by, Here 푈푖 푚 represents UE 푖 utility function when connecting to different BS 푚, 퐸푖푚 denotes the 푊 modulation efficiency and it can be formulated as 푁푚 +1 푇 = (1 − 휏) 푒 + (1 − (1 − 푊 푁푚 +1 푊 퐿푀퐶푆 퐿푀퐶푆 휏) )(푇푅푇푆 + 푇퐷퐼퐹푆) + (푁푚 + 1)휏(1 − 퐸푖푚 = 푁푖푚 . 퐶푅푖푚 (12) 푊 휏)푁푚 (푇 + 푇 + 3푇 ) (10) 퐶푇푆 퐴퐶퐾 푆퐼퐹푆 where 푁퐿푀퐶푆 is the number of bits in one symbol, 푖 and 퐶푅퐿푀퐶푆 is the coding rate. Particularly 푁푀퐶푆and where 푒 is the duration of an empty slot time; 푖 푖 퐶푅푀퐶푆are mapped to the Channel Quality Indicator 푇 , 푇 , 푇 , 푇 , and 푇 are the durations of 푖 푅푇푆 퐷퐼퐹푆 퐶푇푆 퐴퐶퐾 푆퐼퐹푆 (CQI) index which can be determined from the SINR the Request to Send (RTS) short frame, DCF values measured at the UE. This directly affects the Interframe Space, Clear to Send (CTS) short frame, achievable downlink data rate calculation for each Acknowledgment short frame, and Short Interframe UE when associated with BS 푚. Space respectively. In order to maximize objective function (1), BS Furthermore the 푆퐼푁푅 in (8) for UE 푖 푖푚 utility function must be efficiently designed in user associated with WAP 푚 ∈ Β can be calculated by association process. Therefore, the BS 푚 utility

function that sorts its preference list for biding UEs 푖 푃푑푔 푆퐼푁푅 = 푚 푚푖 can be expressed as follows, 푖푚 푑 2 ∑푚′ 푃푚′푔푚′푖 + 휎 ′ ∀푚 ∈ Β, 푚 ∈ Β\{m} (11) 푈푚(푖) = 푆퐼푁푅푖푚 (13)

′ Such that 푚 is the interfering WAPs to the 푚 WAP. where 푆퐼푁푅푖푚 is the average signal to interference Moreover equation (9) captures the MAC protocol plus noise ratio measured by UE 푖 when associated effect on WLAN data rate calculation and reflects the with BS 푚. difference in downlink data rate calculations The proposed one-to-many matching game for between LTE and WLAN due to different radio access network selection is described in detail technologies. by Algorithm 1. After initialization, each UE Algorithm 1 Matching Game for User Association Table I Simulation Parameters. Initialization: 푀, 퐼, 푁. LTE parameters Values Discovery and utility computation: ( ) LTE system bandwidth 20 MHz 1: Every UEi construct ≻푖 using 푈푖 푚 Path loss model for MBS (d: 128.1+37.6 * Find stable Matching: distance in Km) log10(푑) 2: While ∑∀푖,푚 푏푖→푚 ≠ 0 do: Wi-Fi parameters Values 3: For each unassociated UE: WLAN system Bandwidth 20 MHz WLAN technology 802.11n 4: Find 푚 = arg max 푈푖(푚). 푚∈≻푖 Minimum contention window 16 5: Send a request 푏푖→푚 = 1 to BS 푚. (W) 6: For all BS 푚: Maximum number 6 7: Update 퐼푟푒푞 ← {푖 ∶ 푏 = 1, 푖 ∈ 퐼}. of retransmissions (휇) 푚 푖→푚 Slot time 9 휇푠 8: Construct ≻푚 based on 푈푚(푖). DIFS 50 휇푠 9: repeat SIFS 10 휇푠 10: Accept 푖 = arg max 푈푚(푖). ACK 160 bits 푖∈≻ 푚 RTS 208 bits 11: Update 퐼푚 ← 퐼푚 ∪ 푖. CTS 160 bits ̂퐿 12: until 퐼푚 = 푁푚, 푚 ∈ ℳ\Β ̂푊 13: or 퐼푚 = 푁푚 , 푚 ∈ B the 3 WAPs, and a 10 MHz LTE system bandwidth 푟푒푗 푟푒푞 14: Update 퐼푚 ← {퐼푚 \퐼푚}. of 50 PRBs for an MBS are considered. The MBS 15: Remove MBS or has a coverage of radius 1000m, and the WAPs has 푟푒푗 WAP 푚 ∈ ≻푖, ∀푖 ∈ 퐼푚 radii of 50m each. The UEs are uniformly distributed 16: end while in a 200m2 area that is 700 m far away from the MBS. ∗ 17: Results: A stable matching 휇푈퐴 The transmit power of the MBS is considered to be 46dBm, while the transmit power of each WAP is constructs its preference relations ≻푖 using (7) and considered to be 200mW. A constant maximum sends a biding request 푏푖→푚 = 1 to BS 푚 with the payload 퐷 of 1500 bytes is also considered. The rest highest utility (lines 1-5). In order to find a stable of the parameters are summarized in Table I, and the matching 휇푈퐴, each BS insert all requesting UEs into results are obtained based on averaging out 500 푟푒푞 the set 퐼푚 , and construct its preference relations for simulation runs. 푟푒푞 퐼푚 based on (8) (lines 6-8), then it accepts biding For performance evaluation, we compare our users and updates its matched list 퐼푚under the proposed association algorithm with the well-known 퐿 matching 휇푈퐴(푚) until reaching its quota (푁̂푚 for WLAN-first algorithm, in which UEs are associated 푊 LTE-BSs, and 푁̂푚 for WAPs), and rejects the rest of with WAPs whenever they are in coverage, and 푟푒푗 푟푒푞 associate with the WAP that have the best SINR if the biding users such that 퐼푚 = {퐼푚 \퐼푚} (lines 9- 푟푒푗 more than WAP are in coverage. 14). Each UE in the rejected list 퐼푚 removes BS 푚 In Figure 2, the performance of our proposed from its preference relation ≻푖 (line 15). This process algorithm is compared to WLAN-first strategy, such is repeated until there are no biding UEs. that they are evaluated based on total system throughput. System throughput is defined as the sum V. PERFORMANCE EVALUATION of the achieved rate for each UE. It can be noticed For performance evaluation, we consider a Multi- that the proposed matching algorithm outperforms RAT HetNet comprises of two tiers, an MBS with the WLAN-first at different number of UEs, this is three WAPs under its coverage, which represents a because WLAN-first algorithm doesn’t consider wireless network in a building (e.g. mall, campus, associating with other RATs while WLAN is bank, etc.). A 20 MHz WLAN system for each of available, although in many cases associating with

Figure 2 System throughput vs. number of users for different Figure 3 Outage probability vs. number of users for different algorithms. algorithms. LTE can provide better achievable downlink data rates, thus our proposed algorithm which aims to achieve an optimal association that maximizes the total system throughput has a significant system throughput gain over WLAN-first. Moreover, it is also notable that the two algorithms has a concave behavior with the increase of UEs numbers, this is because of the WAPs MAC protocol effect from (9), where the rate of increasing the downlink throughput decreases with the increase of number of UEs until it reaches the maximum saturation throughput. In Figure 3, the performance of our proposed algorithm is compared to WLAN-first algorithm in Figure 4 System throughput vs. number of users at different terms of outage probability. Here we define outage WLAN system bandwidths. probability as the probability of failure to achieve a probability, because it aims to the find best RAT reference rate for each UE, which can be set by match for UEs which provides higher achieved data operator to measure quality of service (QoS). It can rates, also because it provides a quota 푁̂푊 for each be noticed that the outage probability of the two 푚 WAP that achieves the maximum saturation system algorithms is almost equal at small number of UEs, throughput, this WAP quota maintains better and our proposed algorithm has always a lower downlink data rates for associated UEs compared to outage probability than WLAN-first algorithm with WLAN-first when the load on the system increases. the increase of number of UEs. This is because when The effect of using different WLAN system the number of UEs is small, the load on the system is bandwidths is illustrated in Figure 4. It evaluates the low, and thus UEs can achieve high downlink data performance of matching game algorithm using rates using any association algorithm. With the WAPs with different system bandwidths (i.e. 20 increase of number of UEs in the system, the load MHz, 40 MHz, and 80 MHz). It can be noticed that increases, and the number of UEs associated with increasing the system bandwidth will lead to increase each WAP increases, which decreases the achieved in the total system throughput, however the downlink data rates for each UE as shown in (9). Our difference between the three curves increases with proposed matching game algorithm has lower outage the increase of number of UEs. This is because system throughput is the sum of UEs throughput, and “Integrating 3G/4G and Wi-Fi Architectures for increasing the number of UEs increases the total gain Diverse Offloading Capabilities”. of using larger system bandwidths for each UE. Moreover, it is notable that the difference between REFERENCES the 20 MHz system bandwidth and the 40 MHz system bandwidth is greater than the difference [1] T. Cisco, “Global mobile data traffic forecast between the 40 MHz and the 80 MHz system update, 2014–2019 white paper,” Cisco Public bandwidths. This is because of the MAC protocol Information, Feb. 2015. effect in (9), in which increasing the system [2] 3GPP, Overview of 3GPP Release 8 V0.3.3, bandwidth results in an exponential decrease in the September 2008. normalized system throughput, while when [3] A. Roy and A. Karandikar, “Optimal radio multiplied with the physical rate as in (8), it gives a access technology selection policy for lte-wifi concave increase in the total system throughput. This network,” 2015 13th International Symposium implies that the system throughput gain from on Modeling and Optimization in Mobile, Ad increasing the system bandwidth decreases with the Hoc, and Wireless Networks (WiOpt), vol. 1, pp. increase of system bandwidth. 291–298, 2017. [4] Y. Gu, W. Saad, M. Bennis, M. Debbah, and Z. Han, “Matching Theory for Future Wireless VI. CONCLUSION Networks: Fundamentals and Applications,” In this paper, a distributed algorithm is proposed IEEE Communications Magazine, Vol. 53, No. to optimize radio access network selection in Multi- 5, pp. 52-59, May 2015. RAT HetNet. In this Multi-RAT HetNet, a set of [5] S. Bayat, Y. Li, L. Song, and Z. Han, “Matching WAPs are deployed under an MBS, where the WAPs theory: Applications in wireless commun.” represent a WLAN in a building (i.e. mall, bank, IEEE Signal Process. Mag., vol. 33, no. 6, pp. etc.). The Multi-RAT HetNet user association has 103–122, Nov 2016. been formulated as an optimization problem, which [6] Y. Leshem and E. Zehavi, “Stable matching for aims to maximize the total system throughput while channel access control in cognitive radio guaranteeing a defined quota for the MBS and the systems,” in Proc. 2nd Int. Workshop Cognitive WAPs. To solve this problem, a distributed Information Processing (CIP), Elba Island, algorithm based on matching game theory has been Italy, 2010, pp. 470–475. proposed, where the one-to-many matching game is [7] S. Bayat, R. H. Y. Louie, B. Vucetic, and Y. Li, used. In this game, users are matched to the base “Dynamic decentralized algorithms for station that serves them with the highest downlink cognitive radio relay networks with multiple data rates, while considering a predefined quota to primary and secondary users utilizing matching achieve maximum saturation throughput for WAPs. theory,” Trans. Emerg. Telecomms. Techs., vol. Simulation results have shown that our proposed 24, no. 5, pp. 486– 502, Aug. 2013. algorithm outperforms the WLAN-first algorithm in [8] S. Bayat, R. Louie, Z. Han, B. Vucetic, and Y. terms of system throughput, and in terms of outage Li, “Distributed user association and femtocell probability as well. Simulation results also have allocation in heterogeneous wireless networks,” shown that when using matching game, increasing IEEE Trans. Commun., vol. 62, no. 8, pp. 3027– the WLAN’s system bandwidth increases the system 3043, Aug. 2014. throughput but with a decreased gain. [9] S. Bayat, R. Louie, Z. Han, B. Vucetic, and Y. Li, “Physical-layer security in distributed wireless networks using matching theory,” ACKNOWLEDGMENT IEEE Trans. Inf. Forensics Security, vol. 8, no. This work was supported by the National 5, pp. 717–732, May 2013. Telecommunications Regulatory Authority (NTRA), [10] E. Jorswieck, “Stable matchings for resource Egypt through the funded research project entitled allocation in wireless networks,” in Proc. 17th Int. Conf. Digital Signal Processing (DSP), networks,” in Proc. 2014 IEEE Int. Conf. on Corfu, Greece, 2011, pp. 1–8. Acoustics, Speech and Signal Processing [11] W. Song, H. Jiang, and W. Zhuang, (ICASSP), May 2014, pp. 439–443. “Performance analysis of the WLANfirst [21] Y. Zhou, J. Chen, Y. Kuo, "Cooperative cross- scheme in cellular/WLAN interworking,” layer resource allocation for self-healing in IEEE Trans. Wireless Commun., vol. 6, no. 5, interworking of WLAN and femtocell pp. 1932-1952, May 2007. systems", IEEE Commun. Lett., vol. 21, no. 1, [12] S. Andreev, M. Gerasimenko, O. Galinina, Y. pp. 136-139, Jan. 2017. Koucheryavy, N. Himayat, S. P. Yeh, and S. [22] G. Bianchi, “Performance analysis of the IEEE Talwar, “Intelligent access network selection 802.11 distributed coordination function,” in converged multi-radio heterogeneous IEEE J. Select. Areas of Commun., vol. 18, pp. networks,” IEEE Wireless Commun., vol. 21, 535–547, March 2000. no. 6, pp. 86-96, Dec. 2014. [13] Fakhfakh E, Hamouda S, “Optimised Q- learning for WiFi offloading in dense cellular networks,” IET Commun 11, (15):2380–2385. [14] B. Soleyman, A. Zaman, S. H. Rastegar, and V. Shah-Mansour, “RAT Selection Based on Association Probability in 5G Heterogeneous Networks,” in IEEE Symposium on Communications and Vehicular Technology (SCVT), pp. 1–6, Nov 2017. [15] Yu HW, Zhang B, “A heterogeneous network selection algorithm based on network attribute and user preference,” Ad Hoc Networks, pp. 68-80, Apr 2018. [16] J. Bukhari, N. Akkari, “QoS based approach for LTE-WiFi handover,” Computer Science and Information Technology (CSIT), 2016 7th International Conference. [17] Yu HW, Zhang B, “A Hybrid MADM Algorithm Based on Attribute Weight and Utility Value for Heterogeneous Network Selection” Journal of Network and Systems Management, pp. 1-28, 2018. [18] F. Bouali, K. Moessner, and M. Fitch, “A Context-aware User-driven Framework for Network Selection in 5G multi-RAT Environments,” in IEEE 84th Vehicular Technology Conference (VTC Fall), September 2016. [19] Orimolade J, Ventura N, Falowo O, “ANDSF- based WLAN offloading in the Evolved Packet System (EPS),” In Electrotechnical Conference (MELECON), pp. 1-6, Apr 2016. [20] N. Namvar, W. Saad, B. Maham, and S. Valentin, “A context-aware matching game for user association in wireless small cell