SINR-based Network Selection for Optimization in Heterogeneous Wireless Networks (HWNs)

Transcription

1 48 J. ICT Res. Appl., Vol. 9, No., 5, 48-6 SINR-based Network Seletion for Optimization in Heterogeneous Wireless Networks (HWNs) Abubakar M. Miyim, Mahamod Ismail & Rosdiadee Nordin Department of Eletrial, Eletronis and Systems Engineering, Faulty of Engineering and Built Environment, Universiti Kebangsaan Malaysia 46 Bangi, Selangor DE Abstrat. To guarantee the phenomenon of Always Best Connetion in heterogeneous wireless networks, a vertial handover optimization is neessary to realize seamless mobility. Reeived signal strength (RSS) from the user equipment (UE) ontains interferene from surrounding base stations, whih happens to be a funtion of the network load of the nearby ells. An expression is derived for the reeived SINR (signal to interferene and noise ratio) as a funtion of traffi load in interfering ells of data networks. A better estimate of the UE SINR is ahieved by taking into aount the ontribution of inter-ell interferene. The proposed sheme affords UE to reeive high throughput with less data rate, and hene benefits users who are loated far from the base station. The proposed sheme demonstrates an improved throughput between the serving base station and the ell boundary. Keywords: interell; interferene; predition; SINR; throughput; vertial handover. BIntrodution As a result of high demand and ompetition, wireless ommuniation servies have been witnessing an exponential growth in reent years. This development has led to the onvergene of multiple wireless (LTE/LTE-A, ellular, WLAN and satellite) networks to form one pervasive wireless aess point in providing diverse servies []. The major onern, however, is how to harmonize the divergene of the networks available in managing user mobility as eah network differs in data rate, transmission range, traffi lass, et. []. Vertial handover, defined as the handover of resoures between the radio base stations (RBSs) of diverse wireless aess tehnologies, is assumed to be the solution for optimizing the handover deision algorithm []. Many of the researh works related to horizontal handover are about the evaluation of RSS for the mobile user equipment (UE) but the fators used in RSS omparison and evaluation do not suffie for optimization of the performane of wireless networks. Metris suh as system performane, servie aggregation, network onditions, servie ost, mobile host and user preferenes need to be aorded attention [4]. Sine RSS-based VHO is not a QoS aware sheme, it annot Reeived April 7 th, 5, Revised August rd, 5, Aepted for publiation August 4 th, 5. Copyright 5 Published by ITB Journal Publisher, ISSN: , DOI:.564/itbj.it.res.appl.5.9..

2 SINR-based Network Seletion for Optimization in HWNs 49 provide better QoS to the user to support multimedia servies [,]. Furthermore, the ahievable data rate for the UE is a funtion of the reeived signal to interferene and noise ratio (SINR). Therefore, a SINR based VHO is not only expeted to ahieve maximum throughputs and minimum dropping probabilities but also to provide a unified radio resoure for the heterogeneous wireless networks. With the proposed vertial handover algorithm, the future RSS an be predited and it alulates the servie ost to determine the right hoie of the network to be handed over to. Being triggered at the end of the handover proedure, the algorithm is employed to take an optimum deision in evaluating the input parameters of the andidate target networks. The rest of this paper is organized as follows: Setion reviews related works, while Setion disusses SINR predition using M (, ) and a neural network. Setion 4 evaluates the performane of the proposed Preditive SINR Handover Algorithm (PSHA). Simulation set-up and analysis of the results obtained therein are disussed in Setion 5. The onlusions of the paper are drawn in Setion VI. Related Works In order to make possible seamless aess in multi-aess wireless networks, designing an effiient vertial handover (VHO) algorithm is always onsidered ritial. Nowadays, the attention of researhers has shifted to the design and deployment of seamless heterogeneous wireless network tehnologies with various works that have appeared overing vertial handover (VHO) algorithms. However, many of the earlier studies on vertial handover onsidered the reeived signal strength (RSS) as the main deision indiator. Two handover lasses of handover proedures exist with RSS as the major riterion: ) handover deided by omparing RSS with a predefined threshold value [5,6]; and ) RSS as an entity used to initiate and trigger handover [7,8]. In [5], a vertial handover deision-making algorithm is formulated, adopting a Markov deision proess (MDP) to maximize the expeted total rewards per onnetion. The proposed approah in [6-8] determines the optimal target network through polynomial regressive RSS predition and MDP analysis. The TCP sender is apable of aurately prediting the bandwidth available in order to enhane the network throughput by using a ross-layer approah. RSS and servie type were ombined to optimize the VHO deision mehanism as formulated in [9]. A fuzzy logi system is onsidered when seleting ells from WLAN/WiMAX/ LTE networks, while the RSS threshold for the VHO deision remains unperturbed thereby aggravating the ping-pong effet. The aforementioned vertial handover shemes using RSS as an indiator have their advantages. The attainable data rates of the UE an be expressed as a

3 5 Abubakar M. Miyim, et al. funtion of the reeived signal to interferene plus noise ratio (SINR) []. In addition, the use of SINR in integrated wireless networks an provide a solution for handover optimization. Unlike RSS-based vertial handovers, SINR an provide a higher average throughput for the user and ahieve the best possible performane of the system. Furthermore, an SINR-based vertial handover is more desirable to support better multimedia QoS. As expeted, the ombined SINR-based vertial handover (CSVH) disussed in [] does aquire a higher throughput ompared with an RSS-based vertial handover, but the algorithm gives no onsideration to other QoS parameters. Thus, we propose multidimensional preditive SINR-based vertial handover (MPSVH) algorithm to provide a seamless vertial handover that supports multi-attribute QoS. In line with the four lasses of servie defined by the Third eneration Partnership Projet (PP), SINR, user preferenes, ost and available bandwidth of eah WLAN aess point (AP) and LTE enodeb radio base station (RBS) were given onsideration. The proposed sheme omprises: ) the ombination of SINR predition by M (, ) and bak propagation (BP) neural network to fix the appropriate time for triggering handover. Additionally, a stability period to redue unneessary handovers is also onsidered. The obtained results demonstrate an improvement not only in seleting the optimal network while onsidering user preferenes and network onditions, but the sheme also performs well with regards to system throughput. SINR Predition Model {M(, ) and ANN (Bak Propagation)} Most of the time, suessful handover exeution is largely dependent on the auray of the predited SINR. On the other hand, wrong or faulty handover deisions our when the SINR estimation is inaurate. This phenomenon is prevalent with networks that adopt only one single predition method for SINR estimation. Reduing randomness by using more than one method helps to signifiantly improve the auray of predition beause the information aquired is a ombination of the results from the various predition methods [6]. The grey model M(n, m), whih desribes the dynami ondut of the grey system and is widely established as M(, ), is quite suitable for predition of highly noisy data suh as the SINR, as disussed in [7] and [9]. However, the results obtained by using M(, ) predition annot meet the requirements in atual situations as SINR hange is highly nonlinear beause of the omplexity of the wireless environment. For a network to approximate omplex non-linear funtions, it requires a learning proess. Therefore, in this work, a multi-layer neural network (MLNN) that employs a bak propagation algorithm and is apable of parallel distribution and self-training was hosen to determine the relationship between the neurons [], even though residual errors are most

4 SINR-based Network Seletion for Optimization in HWNs 5 likely to our. Correting suh anomalies (errors) asks for the oupling of two or more variant models that omplement one another. The M(, ) model was found to be the perfet hoie to enhane the reliability of the predition results. Exploring the qualities of the two model networks M(, ) and the artifiial neural network led to the birth of a nonlinear model that is just perfet to predit the SINR and enhane the auray of the predition, as depited in Figure. Figure Arhiteture of MLNN with bak propagation model. 4 Proposed Predition Tehniques The inputs in Figure represent the predition values of multiple layers of M(m, n). The M(, ) together with a multiple-layer artifiial BP neural network is used to predit SINR. The following fators are onsidered during the modelling and predition proesses: ) time series predition of SINR aomplished by M(, ); ) establishment of ANN-BP, initialization of weight, and offset value; ) updating the weights and offset values of the ANN through repeated training until the error between output value and target value drops below a ertain fixed value; 4) employment of the predition model by applying the predited omponents of M(, ) as input data for the BP neural network in order to obtain the output data. It is indeed lear that the traffi on the downlink requires a higher bit rate (bandwidth) than the uplink traffi and this is in agreement with the requirement of multimedia servies. Therefore, the assumption is that all andidate enodeb stations and aess points (APs) of a heterogeneous wireless network of m enodebs (BS) and n APs may be indexed by to m n and arranged in a set: Z= [BS, BS,,BS m, AP, AP,, AP n ] () The moment the ombined predition A produes an SINR that is lower than the defined threshold, triggering of the handover is set. Determining the

5 5 Abubakar M. Miyim, et al. network optimization depends on a ost funtion for every network user based on the vertial handover algorithm regarding attributes like: user preferenes, SINR, bandwidth availability and monetary ost of traffi for users. 4. Signal to Interferene plus Noise Ratio (SINR) It is possible to obtain the maximum attainable data rate for the given arrier bandwidth and SINR when the Shannon apaity formula is applied. The highest data rate attainable (Δ AP and Δ BS ) from WLAN and LTE-A networks respetively for a onneted mobile user are easily repliated by the reeived SINR from the two networks ζ AP and ζ BS respetively: ζ AP AP = κ AP log Φ AP ζ BS BS = κ BS log Φ BS Ф AP and Ф BS are the loss fators of the hannel odes for both WLAN and LTE- A networks respetively. When downlink traffi has the same data rate, i.e. Δ AP = Δ BS and is offered users by the two (WLAN and LTE-A) networks, then an expeted relationship is established between ζ AP and ζ BS, whih is expressed thus: ζ BS κ AP κ ζ BS AP =Φ BS Φ AP At a ertain moment in time, the values of the reeived SINR from LTE-A enodeb j by a user i an be expressed as: ζ BS ji, The SINR ( ζ BS P ji, BS ji, = m P P P P AP j, i k=, k j ( BS ) ( ) ki, BSk BS χ ji, BS j BS ji, ) reeived from the WLAN by a user i is expressed as: () () (4) (5) ζ APji, AP P ji, APj = n P P ( ) ki, k BS AP AP k=, k j (6)

6 SINR-based Network Seletion for Optimization in HWNs 5 In order to ahieve the same data rate via enodeb (BS), the reeived SINR from APs (S Ap, i ) is onsidered as the input and so beomes the SINR: S κ AP S BS AP, i κ ' AP, i =Φ BS Φ AP To determine the propagation ondition, a maro-ell propagation model with path loss is adopted here for urban and suburban areas []. ( ) 58.8 log( ) 7.6log( ) ( ) PL db = f d S db (8) Where f is the arrier frequeny, d is the distane between the user and BS or AP, and S orresponds to the log-normal shadowing with s = db standard deviation. 4. SINR Estimation Referring to Figure, onsider a terminal loated in setor and assume that the terminal is being served by base station. Assume that in addition to BS, the terminal has BSs and and are in its ative set. The ative set of a terminal is the set of BSs whose pilot signal an be orretly deoded by the terminal. The terminal reeives inter-ell interferene from setors and. Let i be the forward link SINR of setor i as measured by the terminal using the urrent sheme [8], [4] (i.e. by using Eq. (4) and assuming % load in the rest of the setors). (7) Figure Topology of a typial ell [9].

7 54 Abubakar M. Miyim, et al. (9) In our proposed sheme, the terminal uses the above values of i that are known from pilot measurements. The above sets of equations are then solved to obtain the hannel gains i as follows: () Let the atual SINR at the terminal for the forward link of setor i be β i. This SINR takes into aount the traffi load in the interfering setors. Using Eq. (4), we obtain: () Sine i is already known from (6), the atual SINR, β i, for eah of the setors an be obtained by using the above expressions one ρ i are known. Thus, we do not propose altering the urrent implementation of the pilot measurement proedure, whih makes our SINR estimation sheme easy to implement within the urrent framework. However, it is required of eah enodeb to keep trak of its traffi load (ρ i ) sine the derivation of Eq. () requires the probability of the time slot being busy. Even with multi-slot pakets, the load measurement sheme only needs to know whether a time slot is empty or busy. A long-term ( ) ( ) ( ) = N AT T A N AT T A N AT T A = ( ) ( ) ( ) = N AT A T N AT A T N AT A T β β β

8 SINR-based Network Seletion for Optimization in HWNs 55 averaging exponential filter an then be used by eah BS i to determine its average load as explained in the following setion. 4. Cell Seletion The fat that the terminals see a time-varying wireless hannel over the link should be taken into aount when sheduling data. In all the aforementioned works, as well as several other related works, no speial onsideration is given to the fat that the inter-ell interferene has an important role to play in determining the SINR of a user. In the previous setion, SINR estimation used a sheme that is disadvantaged as it does not provide an aurate estimate of the inter-ell interferene in the SINR. This is beause under the sheme, users loated near the ell boundary may report a lower data rate than the atual supportable rate as a result of inaurate rate estimation. When a user is about to hand over to a neighbouring setor using ell-seletion, it measures the SINR of all the BSs in its ative set to determine the best serving setor. It is possible that due to proximity, the SINR of the UE for BS tends to be higher than that of BS ( > ). However, the real SINR of the UE for BS may be higher than that of BS (β > β ). Thus, the user ontinues to remain in setor instead of handing over to setor. Illustrating the simulation results using i instead of β i may result in sub-optimum ell seletion, where the resultant SINR of the UE in the hosen ell may be lower when i are used for ell seletion riterion instead of β i. 5 Simulation Setup While it is easy to extend the simulation settings to aount for more than two interfering setors, we note that the signals of two neighbouring setors are substantially stronger than the interfering signal reeived from other distant setors; hene, only two interfering setors were simulated. Corresponding to eah user and eah base station (serving and interfering base stations), a timevarying hannel gain i was generated in eah time slot. The user SINR was a funtion of random variables i, and i expressed as: i = d n i ςi Wi i () In Eq. (7), the first term in the produt is deterministi (for a fixed user loation) orresponding to path loss, while the seond term is a random variable orresponding to lognormal shadowing loss. Here, ξ i is a aussian random variable with mean and variane σ. ITU path loss models for vehiular ( kmph) and pedestrian ( kph) users [7] were adopted. Shadowing was orrelated

9 56 Abubakar M. Miyim, et al. over eah time slot depending on the speed of the user as per udmundson s model [7]. The shadow orrelation distane was the same for both the vehiular and the pedestrian model, and depended only on the environment (urban or suburban). Time orrelation of Rayleigh fading was determined by the speed of the terminal and simulated using the filtered aussian noise tehnique from [5]. The simulation parameters are listed in Table. Carrier frequeny, f Parameters Log-normal shadowing, σ Shadow orrelation distane Noise spetral density, N Table Simulation Parameters. 6 MHz db 5. m -74 dbm/hz Values Radius of the setor, R R of ell overage Penetration loss (pedestrian only) Pedestrian path loss in db Vehiular path loss in db Km :9R = :9 Km db log(f ) 49 4 log(d) log(f ) 58:8 7:6 log(d) User loations were seleted equidistantly from the two interfering base stations. At eah UE terminal, the SINR predition sheme was simulated. In this sheme, the terminal has an averaging filter bank for eah paket type. The filter bank uses the SINR measurements from the past to predit the future SINR for eah paket type. In a real-time network, in addition to the SINR predition algorithm, there is a ontrol loop that adjusts the predited SINR values based on measured paket error rate. The UE makes the SINR preditions more onservative when the measured paket error rate is high. 5. Simulation Results In all our simulations, although the time variations of the user hannel as a funtion of user speed were simulated (time-varying shadowing and fading), it was assumed that the loation of the mobile user did not hange. This is beause the objetive of our study was to demonstrate that the users loated near the boundary of the ell benefit from our proposed sheme. Suh an approah is often used by network designers to study how throughput degrades as a funtion of the distane of the user from the serving base station. Figure shows the reeived signal strength with and without noise generation for a single user. Spreading the signals transmitted in the hannel led to reonstrution of the RSS from noise generated with maximal length sequene.

10 SINR-based Network Seletion for Optimization in HWNs 57 Figure RSS of hannel bandwidth with and without noise as a funtion of time. The bit error rate (BER) urve is steaper and omes to zero quikly. For keeping the desired BER at an aeptable level of., a orresponding -.5dB of data interferene and -db of interferene level at rate R were needed. Figure 4 is a graph of the BER plotted as a funtion of the SINR where a.5 db level with less network load indiates a signifiant improvement in the SINR. BER vs SINR.6 SINR BER.5 BER(%) SINR (db) Figure 4 Comparison of retransmission pakets. Figure 5 shows a sharp linear plot of data rate against SINR. With the SINR at -.5, a orresponding value of Kbps was reorded. The plot linearly

11 58 Abubakar M. Miyim, et al. inreases until the data rate reahes Kbps at -db SINR, indiating minimum paket error rate. Using the sheme along with hybrid ARQ, a good throughput improvement was obtained. However making use of two shemes for almost all the settings, the paket error rates were kept below % and never went above %, as an be seen in Figure 6. Thus the proposed sheme enhaned the funtioning of hyrbrid ARQ by providing more aurate initial SINR estimates. Figure 5 Data rate as a funtion of SINR. Figure 6 Inreased throughput. Using a different sheme for SINR estimation (i.e. using i as SINR estimates) instead of the known method (i.e. using β i as SINR estimates), the set of simulations yielded sub-optimum ell seletion. The network loads in the three

12 SINR-based Network Seletion for Optimization in HWNs 59 setors were ρ =., ρ =., ρ =.. These load values were hosen to illustrate the UE s nearest to BS ( >, ) as well as the effet of network load that results in SINR β i suh that β > β, β. This is demonstrated in Figure 7 with plot,, β and β indiating that instantaneous SINR estimates are time averaged values. Figure 7 Predited SINR for different ells. Using β i to make ell seletion deisions, the UE would then hand over to setor instead of staying in setor, thereby having a higher SINR (β ). Thus the proposed sheme an prove benefiial to users during ell seletion. 6 Conlusions A sheme to improve SINR estimation at the user terminal, where the measurements take into aount the atual (and not the worst-ase) inter-ell interferene, has been desribed. The proposed sheme requires the base stations to use traffi measurement (network load) and send this information to the mobile terminals. The terminals then use this information along with an improved SINR estimation sheme to obtain an aurate estimate of their urrent hannel state. In simulations, the proposed sheme outperformed other SINR estimation shemes by providing better throughput to users, espeially the ones loated farther from the serving base station. Also observed is that the gains of the proposed sheme are more pronouned for vehiular users than for

13 6 Abubakar M. Miyim, et al. pedestrian users. The proposal is more benefiial to users when making ell seletion hoies during handover. Aknowledgments The authors would like to aknowledge the Universiti Kebangsaan Malaysia (UKM) for their finanial support of this projet under the ode DPP--6. Also aknowledged is the ontribution of the reviewers in giving useful advie to improve the quality of this paper. Referenes [] Han, D., Andersen, D., Kaminsky, M., Papagiannaki, K. & Seshan, S., Aess Point Loalization Using Loal Signal Strength radient, Conferene Proeedings on Passive and Ative Network Measurement, 5448, Berlin Heidelberg, pp. 99-8, 9. [] Kassar, M., Kervella, B. & Pujolle,., An Overview of Vertial Handover Deision Strategies in Heterogeneous Wireless Networks, Computer Communiations,, pp. 67-6, 8. [] Rouil, R., olmie, N. & Montavont, N., Media Independent Handover Transport Using Cross-Layer Optimized SCTP, Computer Communiations, (9), pp ,. [4] Salih, Y.K., See, O.H. & Yussof, S., A Fuzzy Preditive Handover Mehanism Based on MIH Links Triggering in Heterogeneous Wireless Networks, presented at the International Conferene on Software and Computer Appliations, Singapore, pp. 5-9,. [5] Zahran, A.H., Liang, B. & Saleh, A., Signal Thresholds Adaptation for Vertial Handoff in Heterogeneous Wireless Networks, ACM/Spring Mobile Networks and Appliations, (4), pp , 6. [6] Miyim, A.M., Ismail, M., Nordin, R. & Ismail, M.T., Regressive Predition Approah to Vertial Handover in Fourth eneration Wireless Networks, Journal of ICT Researh Appliations, 8(), pp. - 48, 4. [7] Chang, B.J. & Chen, J.F., Cross-Layer-Based Adaptive Vertial Handoff with Preditive RSS in Heterogeneous Wireless Networks, IEEE Transations on Vehiular Tehnology, 57(6), pp , 8. [8] Kunarak, S. & Suleesathira, R., Preditive RSS with fuzzy logi based Vertial Handoff Algorithm in Heterogeneous Wireless Networks, International Symposium on Communiations and Information Tehnologies, Austin Tx, USA, pp. 5-4,. [9] Kayaan, E., Ulutas, B. & Kaynak, O., ray System Theory-based Models in Time Series Predition, Journal of Expert System with Appliation, 7(), pp ,.

14 SINR-based Network Seletion for Optimization in HWNs 6 [] Shgluof, I., Ismail, M. & Nordin, R., Effiient Femtoell Deployment Under Maroell Coverage in LTE-Advaned System, IEEE International Conferene on Computing, Management and Teleommuniations (ComManTel), Ho Chi Minh City, Vietnam, pp. 6-65,. DOI:.9/ComManTel [] Yang, K., ondal, I. & Qiu, B., Combined SINR based Vertial Handoff Algorithm for Next eneration Heterogeneous Wireless Networks, Proeedings of IEEE lobal Teleommuniations (LOBECOM), Washington, DC, USA, pp , 7. [] Hagan, T., Demuth, B. & Beale, H., Neural Network Design, Thomson Learning, Asia Pte Ltd, Singapore,. [] El-Fadeel,.A., El-Sawy, A.E. & Adib, M.J., C4. Vertial Handoff in Heterogeneous Wireless Networks with Preditive SINR Using M(, ), in Radio Siene Conferene (NRSC), 9th National, pp ,. [4] Yang, K., ondal, I. & Qiu, B., Multi-Dimensional Adaptive SINR based Vertial Handoff for Heterogeneous Wireless Networks, IEEE Communiations Letters, (6), pp , 8. [5] Hristov, V., Improving Fairness in CDMA-HDR Networks, Sientifi Researh of South-West University Blagoevgrad, Bulgaria, 5, pp , 7.