Access Point Selection for Hybrid Li-Fi and Wi-Fi Networks

Hybrid light fidelity (Li-Fi) and wireless fidelity (Wi-Fi) networks are an emerging technology for future indoor wireless communications. This hybrid network combines the high-speed data transmission offered by visible light communication and the ubiquitous coverage of radio-frequency techniques. While a hybrid network can improve the system throughput and users’ experience, it also challenges the process of access point selection (APS) due to the mixture of heterogeneous access points. In this paper, the differences between homogeneous and heterogeneous networks regarding APS are discussed, and a two-stage APS method is proposed for hybrid Li-Fi/Wi-Fi networks. In the first stage, a fuzzy logic system is developed to determine the users that should be connected to Wi-Fi. In the second stage, the remaining users are assigned in the environment of a homogeneous Li-Fi network. Compared with the optimisation method, the proposed method achieves a close-to-optimal throughput at significantly reduced complexity. Simulation results also show that our method greatly improves the system throughput over the conventional methods, such as the signal strength strategy and load balancing, at slightly increased complexity.


Introduction
Mobile communication has been techni cally challenged by exponentially increasing demands for data traffi c.The Cisco visual networking index (VNI) [2] reports that global mobile data traffi c reached 2.5 exabytes per month at the end of 2014, which is 69% more than the traffi c at the end o f 2013.During the same period, the average cellular network connection speed increas ed by 20% only.One solution to relieve the pressure on existing bas e stations is offloading traffic to wireless fidelity (Wi-Fi), based on the fact that over 80% of mobile data traffic comes from indoor locations.However, the dense deployment of Wi-Fi hotspots becomes the bottleneck of improving the system capacity.An alternative short-range wireless communication technology is visible light communication (VLC) [3] and its networking variant, light fidelity (Li-Fi) [4].In Li-Fi, light-emitting diode (LED) lamps act as access points (APs), and light is used as a medium to carry information bits via intensity modulation and direct detection (IM/DD).At the receiver, a photon diode (PD) is employed to collect photons and convert them into electri c current.Unlike the radio-frequency (RF) techniques including Wi-Fi, Li-Fi does not experience interference from other sources becaus e it is contained within a speci fic area, and light is not trans ferred through opaque objects such walls.In addition, Li-Fi offers a much wider spectrum than RF, and is licence-free.Furthermore, Li-Fi can be used in RF-restri cted areas such as hospitals and underwat er.Recent research shows that by using a single LED, Li-Fi is capable of offering high-speed data transmission in the Gbps range [5].

Literature Review
A Li-Fi AP has a smaller coverage area than Wi-Fi, of approximately 2-3 m diameter [6].In order to provide enhanced coverage, a hybrid Li-Fi and Wi-Fi network, which combines the high-speed data transmission of Li-Fi and the relatively large coverage of Wi-Fi, is envisioned for indoor wireless communications [7].In [8], it was shown that such a hybrid network can achieve a great er throughput than stand-alone Wi-Fi or Li -Fi networks.In that study, all of the users are first connect ed to the Li-Fi network, and then those of low achievable data rates are switched to Wi-Fi.This access point selection (APS) method fails to take into account the fact that the required data rates might vary with users.Also, due to the limited Wi-Fi resource, switching a Li-Fi user that achieves a low data rate to Wi-Fi does not necess arily benefit the overall network perform ance.An apparent example is that user receives a very weak Wi-Fi signal and could drain the Wi-Fi resource to meet its demand for data rate.
Motivated by this, we propose a novel APS method based on fuzzy logic for a hybrid Li-Fi and Wi-Fi network.Fuzzy logic, whi ch was first introduced by Zadeh [13] in 1965, is an approach to computing based on " degrees of truth" rather than the usual "true or false" Boolean logic.This approach can readily handle a complicated problem by transforming it into a checklist of rules, and thus has been widely used in control.
A Li-Fi AP has a smaller coverage area than Wi-Fi, of approximat ely 2-3 m diamet er [6].In order to provide enhanced coverage, a hybrid Li-Fi and Wi-Fi network, which combines the high-speed dat a transmission of Li-Fi and the relatively large coverage of Wi-Fi, is envisioned for indoor wireless communications [7].In [8], it was shown that such a hybrid network can achi eve a greater throughput than stand-alone Wi-Fi or Li-Fi networks.In that study, all of the users are fi rst connected to the Li-Fi network, and then those of low achievable dat a rat es are switched to Wi-Fi.This access point selection (APS) method fails to take into account the fact that the required dat a rates might vary with users.Also, due to the limited Wi-Fi resource, switching a Li-Fi user that achi eves a low data rate to Wi-Fi does not necess arily benefit the overall network performance.
An apparent example is that user receives a very weak Wi-Fi signal and could drain the Wi-Fi resource to meet its demand for data rate.

S ystem Model A. Hybrid Li-Fi and Wi-Fi Network
Consider a generalised hybrid Li-Fi and Wi-Fi network for indoor downlink communications, where a number of rooms or compartments are t aken into account, as shown in Fig. 1.Each room has a number of ceiling LED lamps, and each lamp is enabled as a Li-Fi AP covering a confined area.Also, a Wi-Fi AP is fitted in each room, providing coverage for the entire room.Though the APs might be irregularly placed in practice, we assume Li-Fi APs to be arranged in a rectangular shape and Wi-Fi APs in the room centres for the purpose of simplicity.Carrier sense multiple access with collision avoidance (CSMA/CA) is used in the Wi-Fi system [15], and therefore no interference occurs among Wi-Fi APs.
Regarding the Li -Fi system, all of the Li-Fi APs reuse the sam e bandwidth.Since light does not penetrat e walls, interference only exists between those Li-Fi APs in the same room.At each Li-Fi AP, time-division multiple access (TDMA) is adopted to serve multiple users.

B. Li-Fi Channel Model
A VLC channel is comprised of the line of sight (LOS) and non line of sight (NLOS) paths.The geometry of indoor VLC propagation is present ed in Fig. 2. It is assumed that each user device is fitted with a PD vertically facing upwards.The LOS path of Li-Fi AP i and user u is the straight line between them, and the corresponding Euclidean distance is denoted by di,u .The angles of irradiance and incidence relat ed to the LOS path are denot ed by i,u and ψi,u , respectively.The LOS channel of Li-Fi is formulated as [16, eq. ( 10)]: where m=−ln2/ln(cosΦ1/2) is the Lambertian emission order, and Φ1/2 is the radiation angle at which the intensity is hal f of the intensity at the main-beam di rection; Apd denotes the physical area of PD; gf is the gain of the optical filter; and gc(ψi,u) is the optical concentrator gain, which is given by: where n denotes the refractive index, and Ψmax is the semi-angle of the fi eld of view (FOV) of the PD.

C. Wi-Fi Channel Model
The path loss model used for indoor RF propagation consists of the free space loss (slope of 2) up to a breakpoint distance, and a slope of 3.5 after the breakpoint distance [17].Let LFS(⋅) , Xσ and dBP denote the free space loss, the shadow fading and the breakpoint distance, respectively.The path loss is written as

Conventional Access Poin Selection Methods
The SSS is a straightforward method that always selects the AP offering the highest spectrum effici ency.In a homogeneous network, the receiver experiences the s ame l evel of noise power when collecting the signals emitted from di fferent APs.Therefore for the user of interest, the SSS method simply selects the AP that delivers the highest received signal power.In a hybrid network, however, di fferent mechanisms are needed to receive light and radio signals, leading to di fferent noise power per bandwidth between the Li-Fi and Wi-Fi systems.Also, those two systems could use different bandwidths.Thus we adopt signal-tonoise ratio (SNR) instead of signal strength to perform the SSS method in a hybrid network.With AP i sending out the desired signal, the received SNR at user u is denoted by γi,u :

B. Load Balancing
The LB methods, which consider resource availability as well as channel quality, can be classi fied into two categories: channel borrowing and traffic trans fer.Since Li-Fi and Wi-Fi operate at different spectrum, channel borrowing is infeasible in a hybrid Li-Fi/Wi-Fi network.Here we consider a straightforward traffic-trans fer method, while the optimisation-based LB is deem ed as an optimisation method.
Using this LB method, the user is connected to the AP offering the highest SNR if that AP can meet the user's data rate requirement.Otherwise, the user selects the AP that provides the highest user's satisfaction.Note that this AP could still be the one offering the highest SNR.If s everal APs achieve the highest user's satisfaction, the one having the highest SNR is chosen.In other words, the LB method first maximises the user's satis faction, and then maximises the channel quality.The corresponding OF is expressed as:

Optimisation Method
The most commonly used optimisation method is max-sum-log-rat e wise where αi,u is a binary value that indicates the connection status: αi,u=1 means user u is connected to AP i , and αi,u=0 means otherwise.The elements of αi,u for all pairs of AP and us er constitute the

Proposed Access Point Selection Method
In this section, bas ed on the di fferent charact eristics between Li-Fi and Wi-Fi in terms of coverage and capacity, we propose a tailor-made APS method for the hybrid network.The main contribution of this section is three-fold: i) analyse the key issues when conducting the conventional APS methods in a hybrid network; ii) formulate the APS process as a two-stage probl em, which firstly determines the users that need service from Wi-Fi and then performs APS for the remaining users as i f in a homogeneous Li-Fi network; and iii) apply fuzzy logic to the first stage to rank the user's priority of accessing Wi-Fi.
Regarding the second stage, a conventional APS method, such as the SSS and LB, is applicable.

A. Discussion About the APS in a Hybrid Network
With respect to APS, a hybrid network differs from a homogeneous network in two aspects: i) the coverage areas of di fferent systems overlay one another; and ii) the coverage range varies with the AP types.The first point widens the scale of possible options for APS, leading to an exponential increas e in the computational complexity requi red by the optimisation method.See the complexity analysis in Section V-B.Regarding the second point, a Wi-Fi AP has a larger coverage area but less capacity than a Li-Fi AP.In Fig. 3, the Wi-Fi SNR is stronger than the Li-Fi SNR in the green area, which covers 32% of the room, while otherwise in the red area.Considering uniformly distributed users, this means the Wi-Fi AP has to serve 32% users i f the SSS is adopted.Meanwhile, in average, each Li-Fi AP serves less than 6% users.Therefore in this situation the Wi-Fi system is prone to be overloaded, i.e., it cannot meet the data rate demands of all served users.Also, it is worth noting the users nearby a Wi-Fi AP are attract ed to Wi-Fi, even if they are right beneath a Li-Fi AP (e.g., user 1).As a result, the Li-Fi APs close to a Wi-Fi AP are underused.The LB method can relieve the congestion of Wi-Fi by diverting new users to Li-Fi.However, becaus e o f not affecting the AP assignment of existing users, the LB method does not necessarily improve the usage of those underused Li-Fi APs.The lack of effi ciency in Li-Fi raises an open question: assigning what kind of us ers to Wi-Fi (or Li-Fi) is beneficial to the entire hybrid network?Fig. 3 demonstrates some representative users.Due to the presence of ICI in the Li-Fi network, cellcentre users (e.g., user 1) obtain a much higher SINR and thus a much higher spectrum efficiency than cell-edge users (e.g., user 2).Note that both user 1 and 2 would be connected to Wi-Fi if the SSS is applied.To reach the same data rate, user 2 requires more resource than user 1 if they are both switched to Li-Fi.Hence assigning user 2 to Wi-Fi is better than assigning user 1, though user 1 receives a stronger Wi-Fi signal than user 2 does.User 3 is in a situation similar to user 2, but locates in the field where the Wi-Fi SNR is lower than the Li-Fi SNR.
In other words, user 3 is connected to Li-Fi when using the SSS method.Because of receiving a lower Wi-Fi SNR, user 3 has a lower priority than user 2 to use the Wi-Fi resource.

B. Proposed APS Method
The APS for a hybrid Li -Fi and Wi-Fi network is formulated as a two-stage problem: i) determine the users that need to be served by Wi-Fi ; and ii) conduct APS for the remaining users as i f they are in a stand-alone Li-Fi network.We apply fuzzy logic (FL) to ful fil the task of the first stage, while the SSS or LB can be used in the second st age.Correspondingly, the formed methods are referred to as the FL-SSS and FL-LB.In the following context, a FL system is developed to measure how well a user should be assigned to Wi-Fi.

Simulation Results
In this section, Monte Carlo simulations are conducted to validate the performance of the proposed method in comparison with the conventional methods.Consider an indoor s cenario with 4 rooms as shown in Fig. the closest two.The users are randomly distributed with a uniform probability distribution.In addition, the number of available Wi-Fi channels is assumed equal to the number of Wi-Fi APs, except when analysing its effects on the network performance.

Performance Comparison
Fig. 3 presents the users' satisfaction and fairness of various methods when the average required data rat e is 10 Mbps.As shown in Fig. 9(a), the propos ed method can signifi cantly increas e the users' satisfaction over the SSS and LB, especially for a large number of users.When 30 users are present, using the SSS can meet the data requirements for only 74.6% of the users.This value is increased to 87.4% by employing the LB instead of the SSS.When using the FL-SSS and FL-LB, the proportion of satis fied users is 96.1% and 91.9%, respectively.Note that there is a cross point between the curves o f the FL-SSS and FL-LB.This is because using the LB in the proposed method can improve the perform ance of deeply-uns ed users, by decreasing the number of s atis fied users.In Fig. 9(b), the fairness among users is shown for di fferent numbers of users.Two outcomes are observed: i) the fairness of all methods equals 1 given a small number of us ers, e.g., Nu=10 ; ii) as the number of users increases, the fairness decreases for all methods, but the fairness of the FL-LB decreases much slower than that of the other methods.At Nu=100 , the fai rness of the FL-LB achieves 0.95, while the remaining methods have a fairness below 0.9.

Conclusion
In this paper, a two-stage APS method was propos ed for hybrid Li-Fi and Wi-Fi networks, by exploiting the distinguishing characteristics between those two networks.The proposed m ethod at fi rst determines the users that need servi ce from Wi-Fi, and then assigns the remaining users as if in a homogeneous Li-Fi network.The concept of fuzzy logic is applied in the first st age to rank the us er's priority o f accessing Wi-Fi.In the second stage the SSS or LB can be employed, and the propos ed method is named the FL-SSS or FL-LB correspondingly.Based on experimental results and complexity analysis, it is shown that compared to the optimisation method, the proposed method achieves a near-optimal throughput at significantly reduced complexity.In addition, the FL-LB marginally outperforms the FL-SSS with a slight increase in complexity.Compared with the SSS and LB, results show that FL-LB can improve the network throughput by 24% and 11%, respectively.Future research will involve cellular network in the context of a hybrid network.

Fig.1Schematic diagram
Fig.1Schematic diagram of an indoor hybrid Li-Fi and Wi-Fi network.

3 Fig. 2 .
Fig. 2. Geometry of indoor VLC downlinkpropagation For the NLOS path, only first-order refl ections are taken into account for the purpose of simplicity.A first-order reflection consists of two s egments: i) from the AP to a small area w on the wall; and ii) from w to the user.The Euclidean distances of those segments are denoted by di,w and dw,u , respectively.

Fig. 3 .
Fig. 3. Representative users for APS in a hybrid Li-Fi and Wi-Fi network.

p-ISSN: 2348-6848 e-ISSN: 2348-795X Volume 07 Issue 03 March 2020 Available
1, and each room is square with a side length of 10 m.On the ceiling of each room, 16 Li-Fi APs are placed in a layout of a square m atrix, with a s eparation of 2.5 m between online: http://edupediapublications.org/journals/index.php/IJR/ P a g e | 5