Correspondence letter Traffc balancng over lcensed and unlcensed bands n heterogeneous networks LI Zhen, CUI Qme, CUI Zhyan, ZHENG We Natonal Engneerng Laboratory for Moble Network Securty, Bejng Unversty of Posts and Telecommuncatons, Bejng 100876, Chna Abstract: capacty n cellular networks. It s necessary to dscover an approprate coexstence between cellular and other rado access to devces, a non-convex problem s modeled to maxmze the throughput as well as guarantee the desred user satsfacton. users nstead of tradtonal competton between cellular and W-F n the unlcensed bands. Fnally, smulaton results show the Key words: Ctaton 1 Introducton Unlke the evoluton of tradtonal cellular penetraton based on spectral effcency, small cells or ultra dense network are proposed by means of network nfrastructure densfcaton to deal wth boosted. There are abundant Unlcensed Natonal Informaton Infrastructure (U-NII) bands, whch can allevate the pressure on the lcensed bands. However, two types of the most wdely used small cells,.e., femto cell and W-F hotspot, cannot reach to the maxmum spectrum utlzaton due to the lack of a sutable ntellgent coexstence n the unlcensed bands. Many companes (e.g., Qualcomm, HuaWe) n ndustry have undertaken research and consdered advanced strateges n Refs coexstence. Access) usng LTE (Long Term Evoluton) for small cells based on carrer aggregaton technology. In the lterature, numerous methods have been proposed
116 Journal of Communcatons and Informaton Networks for establshng coexstence, such as Lsten Before Talk, Carrer-Sensng Adaptve Transmsson,. Regardless of the dfferences and smlartes of these methods, a theoretcal gude to acheve hgher throughput and guarantee user satsfacton s vtal. In ths paper, we gnore the specfc coexstence mplementaton and focus on the theoretcal optmal tme allocaton n the unlcensed bands. In partcular, tme allocaton means the tme rato that every devce can occupy n the unlcensed bands, and power allocaton means the power allocated to every band. Sdevce and wdevce are used to represent the end-user termnal n the LAA and W- F heterogeneous networks respectvely. Sdevces can access both the lcensed and unlcensed bands where wdevces can only use the unlcensed bands. Some studes were performed to mplement traffc balancng for user satsfacton over the lcensed and unlcensed bands under dfferent constrants and scenaros. Benns, et al. proposed a cross-system learnng framework to offload the delay-tolerant data traffc on to the unlcensed bands where no wdevce s assumed n the framework. Elsherf et al. desgned a method to maxmze the total throughput wthout guaranteeng the user satsfacton n whch mult W-F, femto and macro users were consdered. user satsfacton, whch s acheved by usng natural logarthm functon, power budget and optmal power allocaton. Fang, et al. extended the orgnal sngle unlcensed band scenaro to the multple unlcensed bands scenaro wth varable power dstrbuton to the unlcensed bands. However, all the aforementoned studes manly focus on the competton between LAA and W-F and gnore the nner competton among dfferent sdevces n the unlcensed bands. In prevous research work, only sngle-sdevce has been dscussed whch can certanly access the unlcensed bands as long as the a certan sdevce n the mult-sdevces stuaton can access the unlcensed bands or not also depends on the condtons of other compettve sdevces, e.g., avalable lcensed bandwdth and channel qualty. The man contrbuton of ths paper s proposng a optmal power and tme allocaton n a novel scenaro where mult-sdevces and mult-wdevces compete wth each other. These mult-sdevces and multwdevces coexst n heterogeneous networks, and the nner compettve relatonshp among sdevces n the unlcensed bands s evaluated. The rest of ths paper s organzed as follows: the system model and problem formulaton are presented scheme s proposed to solve optmal tme and power compettve relaton among sdevces wth dfferent lcensed bandwdth s analyzed. Smulaton results of 2 System model and problem formulaton 2.1 System model Femto cells represented wth N s and one W-F AP(Access Pont) are consdered n a scenaro where each femto cell s connected wth one sdevce and the W-F AP serves wdevces represented wth N w. Each sdevce can use both the lcensed and unlcensed bands whle wdevces can only work n the unlcensed band through the W-F AP, as shown n Fg.1. Assumng that each sdevce can access k dfferent lcensed bands and coexsts wth wdevces n a certan unlcensed band. K() denotes the set of specfed lcensed bands for sdevce and G() denotes the set of avalable lcensed and unlcensed
117 bands for sdevce. t s s the tme rato of unlcensed band occuped by the sdevce whle t w s the tme rato for all wdevces. There are two knds of power allocaton, P (j) u denotes the power allocaton for band j when sdevce accesses the unlcensed bands whle P (j) l denotes the allocaton when sdevce does not. sdevce wdevce femto cell YkHk"OR wdevce femto cell sdevce Farness s an effcent algorthm to balance the users can accept at least a mnmal level of servce. Ths mathematcal problem has been solved and can be presented as: f(x) = ln(x Intutvely, the value of natural logarthm ncreases rapdly when the varable s small and slows down as the varable ncreases, whch s consstent wth U(S) can be formulated by the wdely used natural logarthm functon to acheve a farly balanced estmate of all Based on the system model and problem descrpton, the ntegrated problem can be formulated as follows: Fgure 1 Mult-sdevces and mult-wdevces scenaro For LAA functon s physcally attached to LAA base staton, e.g., the upper lmt of the sum of P (j) l and P (j) u are both equal to total transmt power., 2.2 Problem formulaton The throughput of a certan nth wdevce S,n can be expressed as where R,n s the data rate of wdevce n, and n represents the proporton of t that wdevce n occupes n the unlcensed band. R,n and n are constant when all the wdevces are under the same condtons. Smlarly, the throughput of th sdevce represented by S s, s (1) where (j) s the path loss of sgnal n band j. R s ( ) s the data rate of the sdevce, whch s formulated by Shannon Capacty theorem for smplcty. Another problem s fndng an accurate functon to denote the user satsfacton U(S). Proportonal Constrants and ndcate that the nterference power leaked from the femto cell to ts neghbors can not surpass the maxmum power lmt I n one lcensed band. Constrants, and show that the lmt power allocated by femto cell over the lcensed and unlcensed bands can not exceed the maxmum transmt power. Constrant provdes the upper lmt power of the unlcensed band when t
118 Journal of Communcatons and Informaton Networks s occuped by sdevces. The average traffc load of all the wdevces s denoted as t, and the practcal trac load of wdevces should be less than the average load n order to maxmze the entre throughput, whch s acheved by constrant. Constrants and ntroduces the maxmum usage of the avalable tme denoted by t max t s and t can be regulated to mprove U(S) through the optmal tme allocaton and so do P (j) l and P (j) u through optmal power allocaton. The problem s non-convex because of the jont of power allocaton and tme allocaton, thus n ths paper the problem s dvded nto two sub-problems to acheve the calculaton, whch s known as t s and t where n the second sub-problem optmal tme allocaton s solved wth power allocaton obtaned n prevous sub-problem. A novel two-step traffc balancng scheme s proposed to solve two sub- The scheme wll be dscussed n the next secton. 3 Traffc balancng scheme In ths secton, the problem s dvded nto two subproblems,.e., optmal power allocaton and optmal tme allocaton, detals of each are shown n next two subsectons. Frst optmal power allocatons P (j) l and P (j) u are calculated whle tme allocaton t s and t w s wll be obtaned by water-fllng algorthm. Second, after gettng the optmal power allocatons the tme allocaton concludes an optmal calculaton, and numercal results are obtaned by adjustng parameters. 3.1 Power allocaton unlcensed bands or not, there are two types of power allocaton P (j) l and P (j) u. P (j) u denotes the power allocaton for any band j when sdevce accesses the unlcensed bands whle P (j) l denotes the allocaton for only lcensed band j when sdevce does not, and thus they should be solved ndvdually. P (j) l affects the user satsfacton only through the part. Functon U(S)=ln(S) s strctly ncreasng wth throughput S and t s s fxed n ths step, hence ths part can P (j) l can be descrbed as (6) whch s subject to constrants and. R s ( ) s formulated by Shannon Capacty theorem, that s where B (j) s the bandwdth of lcensed band j. Eq.(7) s convex so P (j) l can be solved usng the Karush- Kuhn-Tucker (KKT) condtons : (7) (8) where (x) + means max(x, 0) and v s chosen to satsfy constrant wth equalty. In the same way optmal P (j) u s calculated where u and are chosen to satsfy the constrant wth equalty and P max s the maxmum power for a unlcensed band. 3.2 Tme allocaton The total data rate of the sdevce s represented as
119 (16) R l and R u are constants after the optmal power allocaton P *(j) l and P *(j) u can be represented as (17) and KKT condtons. The optmal tme allocaton of sdevces and all wdevces can be denoted by pluggng (10) Constrant 7 should be acheved wth equalty to maxmze the user satsfacton (18) (11) Eq.(10) can be smplfed further by pluggng Eq.(11) nto Eq.(10) as shown below: Then the dervatve of Eq.(11) s taken wth respect to t s, and set the dervatve to zero, thus obtan There are N s dervatves ndcatng the optmal tme allocaton of N s sdevces, and are added together to obtan an equaton about where s chosen to satsfy Eq.(16). In Algorthm 1, a method s proposed to calculate the optmal tme allocaton faster, as mproves quckly to make approaches, and then adjusts to acheve the precse calculaton. * t s 1: Input data rate R u and R l,. 6: Return t s * = t s. The sum of t s * can be denoted by consderng the and constrants as follows: Gven that the specfed and t w * s fxed, The varaton of optmal tme allocaton s observed by adjustng the number of wdevces and the bandwdths of lcensed bands whle smplfed smulaton parameters (fxed sgnal-to-nose(snr), etc) are adopted, and then the phenomenon s analyzed by prevous formulas. Two sdevces are assumed to coexsts wth wdevces n one unlcensed band, and each sdevce has two lcensed bands wth the same bandwdth. In addton, the value of t max s
s 120 Journal of Communcatons and Informaton Networks tot N w t w = 0:8. B l s the total bandwdth of lcensed bands for a sdevce whle Bu s n ths case. The bandwdth of a certan lcensed band vares from 1~10 MHz. t 0.30 0.25 0.20 0.15 0.10 0.05 t s wth 1 wdevce t s wth 2 wdevce t s wth 3 wdevce t s wth 4 wdevce t s wth 5 wdevce 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 tot B l /B u wdevces As grows, the optmal value of t s * decreases and then tends to be a constant. The constant s equal to. Eq.(18) can be used to explan the phenomenon n mathematcs only f The bandwdth and data rate of a certan sdevce have a lnear relaton n ths step, e.g., R() B(), as shown n Eq.(7). Thus to obtan the optmal value of t s *. can ntutvely replace allocaton. t w s assumed to be less than t max n most The formula holds when s small, thus t s * tends. tme allocaton when two sdevces compete n unlcensed bands wth dfferent lcensed bandwdths. t w = 0.8 and the values of the other parameters are the same as those n the prevous scenaro. refers to the total means the total bandwdth of the second sdevce. There are two knds of varatons n ths fgure, the lcensed bandwdths represented as and respectvely. The lower mesh pattern represents optmal tme allocaton for the frst sdevce whle the upper one represents allocaton for the second sdevce. tme allocaton 0.8 0.6 0.4 0.2 0 1.0 0.5 tot tot B L2 /B L1 0 0 0.5 1.5 1.0 tot B L1 /B u vces wth dfferent lcensed bandwdths 2.0 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 ncreases wth the mprovement of, makng t When tends to be two and approaches easer to obtan the constant value for optmal tme
121 to access the unlcensed band whle the t s * of the second sdevce s equal to t max t. The optmal tme allocatons for two sdevces show an enormous sdevce can easly access the unlcensed band unless as well as dfference from the vew of when the band-wdth 4 Smulatons and analyss of the second sdevce s much smaller when concde wth the ncrease of. Ths phenomenon can be explaned by the soluton of optmal traffc balancng algorthm n Eq.(18). After the sum of the optmal tme allocaton s calculated n, the allocaton for a specfc sdevce depends only on the value of. The water n the water-fllng algorthm wll frst fll the lower place, and may not fll the hgher lace f the water runs out. It has been proved that can be replaced by, so the of frst sdevce ncreases far more than that of the other one when s tends to be two(the sdevce wll not have any tme allocaton due to the Thus t c bandwdths of sdevces and ther compettors s vtal to the optmal tme allocaton. In generally, the optmal tme allocaton for a certan sdevce mproves wth the ncrease n ts bandwdth only f t reserves a small part of the whole bandwdth for all sdevces, * otherwse the optmal t s reduces as ts andwdth ncreases. The sdevce wth small lcensed bands always has a better chance to access the unlcensed bands for the demand of proportonal farness. The concluson of the mult-sdevces scenaro s qute dfferent from that of sngle-sdevce scenaro, n whch the sngle In ths secton the traffc balancng tme allocaton scheme s evaluated through Algorthm 1, and wll n practcal scenaros. As our propose s to sacrfce the throughput to acheve hgher user satsfacton, the performances of these schemes are compared based on user satsfacton and throughput, whch are shown A stuaton n whch there are two sdevces coexstng wth two wdevces n the unlcensed band and each sdevce can access two lcensed bands s consdered. The nterference n every lcensed band comes from a neghborng sdevce whch s connected to another femto cell, and the neghborng sdevce s the femto cell. Meanwhle, there s a neghborng AP Tab.1 summarzes the parameters used n the subsequent smulatons. parameters t max I transmt power nose power path loss: sdevce to femto Table 1 Smulaton paraters value 100 dbm cell or wdevce to W-F AP R) + 0.7R path loss: sdevce to nearbysdevce or wdevce to nearbyw-f AP R) +L shadow (L shadow
122 Journal of Communcatons and Informaton Networks Smulaton results are used to obtan the average through-put of the sdevces when they have dfferent * traffc loads. t s can be adjusted by controllng the access to the nterval and transmsson duraton n Ref. transmts n ts own turn and send ABS(Almost Blank Sub-frame) when leaves the unlcensed band. stuaton, where the contrastve schemes have fxed equal tme allocatons and optmal power allocatons. It can be seen that the traffc balancng scheme provdes a bg mprovement to the user satsfacton. The yellow lne keeps at a low poston because the two sdevces occupy the unlcensed bands for a very long duraton and the throughput of the wdevces can not be guaranteed. The cross of green lne and cyanblue lne can be solved by the same explanaton, as well as the leap of the blue lne. The phenomenon presents that the balancng of avalable bandwdth for every devce s vtal to the user satsfacton. For ths reason the dstrbuton of lnes bascally scheme wth less bandwdth to sdevces can stll approach the green lne because the channel condton average throughput 10 10 7 9 8 7 6 5 4 optmal scheme 3 t s = 0.1 for sdevce 2 t s = 0.2 for sdevce t s = 0.3 for sdevce 1 t s = 0.35 for sdevce 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 tot B l /B u 9.0 5 Concluson and dscusson user satsfacton 8.9 8.8 8.7 8.6 optmal scheme t s = 0.1 for each sdevce t s = 0.2 for each sdevce 8.5 t s = 0.3 for each sdevce t s = 0.35 for each sdevce 8.4 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 tot B l /B u satsfacton s sacrfcng the throughput. Small cell tme unlcensed band s occuped by the sdevces, the more average throughput can the scheme acheves. balancng scheme by proportonal farness and calculated the soluton of power allocaton and tme allocaton, whch can gude the desgn of specfc coexstence schemes. The compettve relatons among sdevces n the unlcensed bands are analyzed usng the water-fllng algorthm, and smulatons the whole user satsfacton. We wll conduct further research on the coexstence scenaro n whch multsdevces connect to the same femto cell and thus share only one power budget, and the power allocaton and tme allocaton wll be dscussed together. References
123 About the authors LI Zhen receved hs B.E. from Shandong s an M.S. student n Bejng Unversty of Posts and Telecommuncaton, Bejng, Chna. Hs research nterests nclude the co-ordnaton between LTE and W-F n wreless communcatons. (Emal: CUI Zhyan receved her B.E. from Wuhan Currently she s an M.S. student n Bejng Unversty of Posts and Telecommuncaton, Bejng, Chna. Her research nterests nclude the energy-
124 Journal of Communcatons and Informaton Networks ZHENG We receved hs B.E. from Bejng Unversty of Posts and Telecommuncaton, Bejng, Bejng Unversty of Posts and Telecommuncaton, Bejng, Chna.Hs research nterests nclude analyss and modelng of heterogeneous network. CUI Qme her B.E. and M.S. wth Honors n Electronc Engneerng from Hunan Unversty, Changsha, Communcatons and Informaton Engneerng at Bejng Unversty of Posts and Telecommuncatons She s a full professor of School of Informaton and Communcaton Engneerng (SICE) n BUPT, Chna. She has nterests focus on transmsson theory and networkng technology communcatons.she has publshed more than 100 research artcles