Two Novel Handover Algorithms with Load Balancing for Heterogeneous Network

Transcription

1 Two Novel Handover Algorithm Load Balancing for Heterogeneou Network Rintaro Yoneya, Abolfazl Mehbodniya and Fumiyuki Adachi Dept. of Communication Engineering, Graduate School of Engineering, Tohoku Univerity, Sendai, Japan , Aza-Aoba, Aramaki, Aoba-ku, Sendai, Miyagi, , Japan (yoneya, Abtract The demand for wirele reource i increaing at high pace. Heterogeneou network (HetNet), i.e., network compoed by bae tation (BS) different coverage area, are ueful olution to cope thi increaing demand. In thi paper, we propoe two handover (HO) algorithm. Firt HO algorithm i baed on uer equipment (UE ) velocity, UE poition, UE received ignal trength (RSS) from BS and traffic load of BS. In the firt phae of the firt HO algorithm, UE employ parameter uch a the ditance of UE from it connected BS, velocity of UE and UE RSS from it connected BS to determine the neceity of HO. If HO i needed, in the econd phae, i.e., HO execution phae, UE elect the new BS baed on the ditance of UE from BS to which UE approache, RSS and average traffic load of BS which i advertied through beacon ignal. UE ue trength of received beacon ignal a RSS. UE tranmit the connection requet ignal to elected BS. Average traffic load mean time average of traffic load. UE get information of it poition and it velocity via GPS. Second HO algorithm i baed on only UE RSS from BS and traffic load of BS. Thi HO algorithm ha imilar tructure a the firt one. However, in the firt phae, only UE intantaneou RSS and average RSS which mean time average of RSS are employed to determine the neceity of HO by UE. In HOEP, UE elect the new BS baed on intantaneou RSS, average RSS and average traffic load of BS. UE tranmit the connection requet ignal to elected BS. A game-theoretic leep mode algorithm i executed parallel to each HO algorithm in BS. The performance of the two propoed algorithm, uch a total number of HO, UE throughput and ytem power conumption i evaluated by mean of computer imulation. Index Term HetNet, handover, bae tation leep mode algorithm, game theory, energy efficiency, mobility I. INTRODUCTION The demand for wirele reource i increaing at high pace. Video treaming and ocial media are mainly reponible for thi increae []. Conequently, traffic load and energy conumption in wirele cellular ytem are increaing accordingly and thee urge the neceity of deigning more energy and pectral efficient ytem. Heterogeneou network (HetNet), coniting of macro cell bae tation (MBS) and mall cell BS (SBS), are proven to be highly effective in increaing the wirele reource [2] [4]. The total conumption energy in HetNet reduce when combined leep mode algorithm which adapt to traffic condition in the network. In [3], a centralized leep mode algorithm i propoed. It i hown that thi algorithm can improve the energy efficiency in HetNet. However, if centralized approache are ued, the number of control ignal increae an increaed information exchange between BS. On the other hand, connecting all BS to data proceing center, e.g. cloud radio acce network (cloud-ran), through high capacity backbone link, i an expenive olution and probably not yet uitable for ome geographic area. Sleep mode The reearch reult preented in thi material have been achieved by Toward Energy-Efficient Hyper-Dene Wirele Network Trillion of Device, the Commiioned Reearch of National Intitute of Information and Communication Technology (NICT), JAPAN and KDDI foundation reearch grant, Energy-Efficient Radio Reource Management for Next Generation Wirele Network. algorithm which rely on elf-ditribution control, do not need uch information exchange through backhaul communication. In uch algorithm, each BS decide independently to turn wake mode or leep mode depending on it traffic load and conumption power. In [5], an HO algorithm for HetNet i propoed. The HO method in [5] ue eparate ignaling mechanim to identify the BS for connection and to trigger the handover. Our HO algorithm i implemented in parallel a BS leep mode algorithm to control the tranmiion power. UE decide independently the BS they want to connect to. In thi paper, we ue the imilar BS leep mode algorithm a in [], in which a non-cooperative, mixed trategy, game i ued. Here, leep mode refer to idle condition in which BS conume power only for detecting uer equipment (UE). In trategic form game [6], each player, i.e., BS, elect it trategy (action) only to maximize it utility, i.e., a function which evaluate each player outcome. In non-cooperative game, player decide their trategie independently out negotiating other player. Later, we propoed two HO handover (HO) procedure in combination the aforementioned leep mode algorithm. Firt HO algorithm i baed on UE velocity, UE poition, UE received ignal trength (RSS) from BS and traffic load of BS. The HO algorithm comprie of two different phae, i.e., HO neceity etimation phae (HONEP) and HO execution phae (HOEP). In HONEP, parameter uch a the ditance of UE connected BS, velocity of UE and it RSS are employed to determine the neceity of HO. Thi phae help reducing the unneceary HO and achieve a higher energy efficiency. After HONEP, if HO i needed, in HOEP, UE elect the new BS baed on the average ditance of UE and BS to which they approach, RSS and average traffic load of BS which i advertied periodically through beacon ignal. UE ue trength of received beacon ignal a RSS. UE get information of poition and velocity via GPS. Second HO algorithm i baed on only UE RSS and traffic load of BS. Obtaining the information for UE velocity and UE poition via GPS may be difficult. Therefore, we alo propoe thi HO algorithm, which i only baed on RSS and traffic load. Thi HO algorithm comprie of two phae, i.e., HONEP and HOEP, imilar to the firt HO algorithm. In HONEP, only UE intantaneou RSS and average RSS are employed to determine the neceity of HO by UE. In HOEP, UE elect the new BS baed on intantaneou RSS, average RSS and average traffic load. The ret of thi paper i organized a follow. In Section II, ytem model i decribed along power conumption, load, and utility function. Section III dicue our propoed algorithm. Section IV provide the imulation reult and the evaluation of our algorithm. Section V conclude the paper /5/$ IEEE

2 SBS3 UE MBS SBS2 SBS Algorithm : Sleep mode algorithm at BS []. : Initialization: S =,..., S }; 2: while do 3: t t, 4: BS trategy election: a (t) =f (p,i (t )) 5: Calculation of average traffic load ˆν (t) and broadcat to all UE 6: Calculation of traffic load ν (t), power conumption P All (t) and utility u (t) 7: Update of average utility û,i (t), regret ˆr,i (t) and probability ditribution p,i (t) 8: end while Fig.. HetNet topology. B. Traffic load Signal to Interference plu Noie Ratio (SINR) of UE at point z i given by: II. SYSTEM MODEL In thi paper, we focu on the downlink tranmiion in Het- Net, coniting of a MBS and everal SBS, S =,..., S}, which are ditributed uniformly in the macro cell. Fig. how an example realization of uch HetNet cenario. Each BS chooe it trategy (tranmiion power level) from Table I. Tranmiion power of th BS i given according to: P (t) =ξ (t) P TX MAX, () where ξ (t) i the tranmiion power level and PMAX TX i the maximum tranmiion power of th BS. In the following, we are going to explain the BS energy conumption and load model along the BS utility which i a function of the two latter. A. Conumption power When one BS i in leep mode, the BS conume power only for detecting UE in the macro cell. The conumption power of th BS at time t i given by [7]: P All (t) = Pradio +P bae χ = P Idle P (t) ηχ( χ feed ) + P Back (leep mode) + P Idle (wake mode), (2) χ =( χ DC )( χ main )( χ cool ), (3) where P radio, P bae and P Back are conumption power in radio frequency, baeband unit and backbone network. χ DC, χ main, χ cool and χ feed are loe in DC-DC converion, main upply, cooling unit and the feeder. η i the power amplifier efficiency. TABLE I TRANSMISSION POWER LEVELS. Identification Number of Strategy i Tranmiion Level ξ (t) Power 0 2 /3 3 2/3 4 ς (z,t) = S/ P (t)g (z) P (t)g (z)+n, (4) where g (z) i UE channel gain at point z and connected to th BS. N i the noie variance. Data rate of UE at point z i given by: D (z,t) =w log 2 ( + ς (z,t)), (5) where w i the channel bandwidth. Traffic load denity of UE at point z i given by [8]: ϑ (z,t) = κ (z)v (z) D (z,t), (6) where κ (z) i the packet arrival rate and v (z) i the average packet ize of UE at point. Traffic load, which indicate the utilization rate of BS cell capacity i given by: ν (t) = z L ϑ (z,t), (7) where L i the et of all UE connected to th BS. C. Utility The maller a BS conumption power, P All (t) and traffic load, ν (t) are, the better condition for the BS i. Therefore, utility of th BS conit of it conumption power, P All (t) and traffic load, ν (t) according to: u (t) = (φ P All (t)/pmax TX + ϕ ν (t)), (8) where φ and ϕ (φ > 0,ϕ > 0) are weighting factor of conumption power and traffic load. Thee figure define the influence of conumption power and load. III. PROPOSED ALGORITHMS We ue the imilar BS leep mode algorithm in [], a hown in Algorithm. The propoed algorithm, which i executed at UE i hown in Algorithm 2.

3 Algorithm 2 : Aociation algorithm at UE. : if UE in t currently connected to any BS then 2: UE chooe a new BS, (z,t) (HOEP) 3: ele 4: Decide whether to HO or not (HONEP) 5: if HO i neceary then 6: UE elect a new BS, (z,t) (HOEP) 7: ele 8: UE doen t change it BS 9: end if 0: end if Algorithm 3 : Strategy election at BS. : Input: p,j (t ) (j =, 2, 3, 4) 2: Output: i 3: Select r (0 <r<) randomly 4: if p, (t ) >rthen 5: i = 6: ele 2 7: if p,j (t ) >rthen j= 8: i =2 9: ele 3 0: if p,j (t ) >rthen j= : i =3 2: ele 3: i =4 5: end if 6: end if û,i (t +)=û,i (t)+ι b (t +) (t) (u (t) û,i (t)), ˆr,i (t +)=ˆr,i (t)+τ (t +) (û,i (t) u (t) ˆr,i (t)), p,i (t +)=p,i (t)+ϱ (t +) (G,i (ˆr,i (t)) p,i (t)), () and (t) = (if a (t +)=a (t)) 0 (if a (t +) a (t)), G,i (ˆr,i (t)) = exp (ε ˆr,i (t)) i A exp (ε ˆr,i (t)) (2), (3) where G,i (ˆr,i (t)) i the Boltzmann ditribution and ε i the temperature parameter. ι (t), τ (t) and ϱ (t) are learning rate which follow a form like /t c (c: power parameter) and hould meet the following criterion. ι (m) =+, ϱ (m) =+, τ 2 (m) < +, τ (t) ι (t) =0, τ (m) =+, ι 2 (m) < +, ϱ 2 (m) < +, ϱ (t) τ (t) =0. (4) A. BS trategy election Each BS ha a et of trategie, A = a (,),..., a (,A) } and a (t) i the trategy choen by th BS at time t. It elect it trategy their trategie probability ditribution, p,i (t ) at time t according to: a (t) =f(p,i (t )), (9) where f i the converion function from probability ditribution to trategy and i elaborated in Algorithm 3. A previouly decribed, trategie define tranmiion power level of BS, ξ (t). It hould be noted that MBS elect it trategy only from i =and i =4. B. Average traffic load Each BS calculate itaverage load, i.e., traffic load time average, according to: ˆν (t) =ˆν (t ) + n(t) (ν (t ) ˆν (t )), (0) where n(t) i the learning rate and indicate the impact of intantaneou value of load on average load. n(t) i choen in a way to make computation of average traffic load adequately lower than the UE aociation peed. If average traffic load change rapidly, UE change their connected BS frequently. In thi cae, it may reult in detabilizing the algorithm. C. Computation of probability ditribution For ith trategy of th BS, average utility, û,i (t+),regret, ˆr,i (t+), and probability ditribution, p,i (t+), are updated according to [].: D. HO Algorithm at UE Each UE receive the average traffic load, ˆν (t), through beacon ignal and RSS, P (t), from all BS in the macro cell. UE ue trength of received beacon ignal a RSS. The two propoed HO algorithm ue thi information for the HO deciion proce. In each algorithm, UE decide neceity of HO at firt. Nextly, UE implement HO if HO i needed. ) Firt propoed HO algorithm which ue the UE poition and velocity data: Fig. 2 how the geo-relational model of UE and it connected BS. In thi figure, v(t)( 0) i the velocity of UE and v b (t) ( <v b (t) < + ) i it velocity component in the direction of the connected BS. d(t) i the UE ditance to it connected BS. UE get data of it poition and velocity via GPS and get data of BS poition through beacon ignal. In thi algorithm, each UE decide about HO baed on ditance from it connected BS, velocity of UE and it RSS. Algorithm 4 how the HONEP phae at UE. In HONEP phae, UE alway earche for a new BS if UE i connected to MBS. Thi i becaue we want to prevent UE to keep connecting to MBS for long time. If UE i connected to SBS, UE implement HONEP a follow. If d(t) >r SBS (r SBS : mall cell radiu) condition i met, UE earche for a new BS. The reaon for thi i that we want to prevent UE to keep connecting to BS far from UE. If UE i connected to SBS and d(t) r SBS condition i met, UE implement HONEP a follow. If v b (t) < 0, d(t) d TH (d TH : ditance threhold) and P (t) P TH (P TH : RSS threhold ) condition are met, UE earche for a new BS.

4 d(t) Connected BS b 90 v b ( t) = v( t)coθ v(t) θ ( 0 θ < 2π ) UE Fig. 2. Geo-relational model of UE and it connected BS Algorithm 4 : Firt propoed algorithm; HONEP uing data of UE poition and velocity, at UE. : if UE i connected to MBS then 2: Alway earch for a new BS uing (5) 3: ele 4: (UE i connected to SBS) 5: if d(t) >r SBS (r SBS : mall cell radiu) then 6: earch for a new BS uing (5) 7: ele 8: if v b (t) < 0, d(t) d TH (d TH : ditance threhold) and P (t) P TH (P TH : RSS threhold ) then 9: earch for a new BS uing (5) 0: ele : Do not change current connected BS 2: end if 3: end if For new UE or whenever HO i needed, each UE at point z elect the BS, (z,t), baed on the following criterion: (z,t) = arg max (ˆν (t)+ς ) ϖ P (t) (d (t)) λ }, S (5) where ς i offet of th BS. ϖ (ϖ >0) and λ (λ >0) are coefficient which define the influence of average traffic load, ˆν (t), and the ditance between UE and th BS, d (t). UE tranmit the connection requet ignal to elected BS. 2) Second propoed HO algorithm which doe not ue the UE poition and velocity data: Obtaining the information for UE velocity and UE poition via GPS may be difficult. Therefore, we alo propoe econd HO algorithm, which i only baed on RSS and traffic load. In thi algorithm, at firt, UE compute each BS average RSS, which indicate time average of RSS from each BS, according to: (t) = (t ) (P (t)/ (t )) γ, (6) where γ i learning rate and indicate the impact of intantaneou RSS, P (t). UE implement HONEP a hown in algorithm 5. In HONEP phae, UE alway earche for a new BS if UE i connected to MBS. Thi i becaue we want to prevent UE from keep connecting to MBS for long time. If UE i connected to SBS, UE implement HONEP a following. If (t) <P TH condition i met, UE earche for a new BS. The reaon for thi i that we want to prevent UE from keep connecting to BS far from UE. If UE i connected to SBS and (t) P TH condition i met, UE implement HONEP a Algorithm 5 : Second propoed algorithm; HONEP out uing data of UE poition and velocity, at UE. : if UE i connected to MBS then 2: Search for a new BS 3: ele 4: (UE i connected to SBS) 5: if (t) <P TH then 6: Search for a new BS uing (7) 7: ele 8: if (t) <P2 TH, (t) < P (t) <P2 TH (P2 TH (P2 TH >P TH (t ) and ): RSS threhold 2)) then 9: Search for a new BS uing (7) 0: ele : Do not change current connected BS 2: end if 3: end if follow; If (t) < P2 TH, (t) < (t ) and P (t) < P2 TH (P2 TH (P2 TH > P TH ): RSS threhold 2)) condition are met, UE earche for a new BS. UE being not connected to any BS or UE needing HO at point z, elect BS (z,t) to connect to baed on the following criteria: (z,t) = arg max (ˆν (t)+ς ) ϖ (t) Ṕ(t), (7) S Ṕ (t) = Pa if th BS i SBS and (t) > Otherwie, (t ) (8) where P a >, ς i an offet and P a i the inflation contant. UE tranmit the connection requet ignal to elected BS. IV. COMPUTER SIMULATION We ued MATLAB oftware for thi computer imulation. In thi computer imulation, all UE move around the macro cell during the whole imulation time. Simulation parameter are ummarized in Table 2. UE velocitie have a Gauian ditribution. Total imulation time i Time interval of leep mode algorithm and HO algorithm are. Thi indicate that the algorithm run totally 0000 time. Time interval for calculating average RSS i 0.. In the imulation reult, propoed HO algorithm indicate the HO algorithm baed on UE velocity, UE poition, UE RSS from BS and traffic load of BS. Propoed HO algorithm 2 indicate HO algorithm baed on only UE RSS from BS and traffic load of BS. Fig. 3 how the total number of HO per v different number of UE. Total number of HO in propoed HO algorithm 2 i bigger than the one in our propoed HO algorithm. Thi i mainly due to occurring of Ping-Pong effect in algorithm which rely more on RSS. Pleae note that Ping-Pong indirectly contribute to higher energy lo. Fig. 4 how the total conumption power in HetNet v different number of UE. We compare the reult the cae that all BS are in wake mode and communicate in maximum tranmiion power. The total conumption power both in propoed HO algorithm and 2 i maller than the cae where all BS are in wake mode. Therefore, we deduct that leep mode algorithm work well in both cae of uing propoed HO algorithm and 2. For intance, auming 40 UE, total conumption power in propoed HO algorithm 2 i maller than propoed HO algorithm. Thi i due to the effect of RSS in propoed HO algorithm 2 which i bigger than

5 Total time of handover per Propoed HO algorithm Propoed HO algorithm Number of UE Fig. 3. Total number of HO per v different number of UE at an average velocity of 4 km/h. 400 Total conumption power in HetNet[W] Propoed HO algorithm Propoed HO algorithm 2 Alway ON Number of UE Fig. 4. The total conumption power in HetNet v different number of UE at an average velocity of 4 km/h. 2 Average throughput per UE Propoed HO algorithm Propoed HO algorithm Number of UE Fig. 5. The average throughput per UE v different number of UE at an average velocity of 4 km/h. that the propoed HO algorithm. A a reult, UE elect BS higher RSS more frequently. However, in the preence of 50 SBS, total conumption power in propoed HO algorithm 2 i higher than propoed HO algorithm. Thi i becaue the effect of RSS in propoed HO algorithm 2 i maller. Fig. 5 how the average throughput per UE v different number of UE. We oberve that average throughput per UE in propoed HO algorithm and 2 are almot the ame, which implie that both HO algorithm do not depend on throughput. The overall conumption power and throughput in both propoed HO algorithm are almot identical. However, the propoed HO algorithm ha lightly better performance than propoed algorithm 2 in term of number of HO. Therefore, when the poition and velocity of UE are available via GPS, TABLE II SIMULATION PARAMETERS. Parameter Value Network Noie Variance N 68 dbm/hz Arrival Rate κ (z) 80 kbp MBS Maximum Tranmiion Power PMAX TX 46 dbm Minimum MBS-SBS Ditance 75 m Cell radiu r MBS 250 m SBS Number of SBS 7 Maximum Tranmiion Power PMAX TX 30 dbm Minimum SBS-SBS Ditance 40 m Cell radiu r SBS 40 m Path lo (d: ditance of BS and UE (m)) (unit: db) MBS - UE log 0 (d) [] SBS - UE log 0 (d) [] Algorithm Parameter Weighting Coefficient for Power Conumption 0, 5 and Traffic Load, φ, ϕ Learning Rate of Average Load n(t) /t 0.9 Learning Rate Exponent c for ι, τ, ϱ 0.6,0.7,0.8 Boltzmann Temperature ε 0 Power Threhold P TH, P2 TH 60 dbm, 50 dbm Ditance Threhold d TH 20m Learning Rate of Average Power γ 0.93 Weighting Exponent of Traffic Load for BS Selection ϖ Offet ς 0.5 Inflating Value P a 7dB propoed HO algorithm yield better performance. V. CONCLUSION In thi paper, two different type of handover (HO) algorithm were propoed. Firt HO algorithm i baed on uer equipment (UE ) velocity, UE poition, UE received ignal trength (RSS) from BS and traffic load of BS. Second HO algorithm i baed on only UE RSS from BS and traffic load of BS. Each HO algorithm ha two tage. In the firt tage, each UE etimate the neceity of HO. If HO i needed, in the econd tage, UE elect the BS to be connected to. Simulation reult indicate that firt HO algorithm outperform econd HO algorithm from number of HO point of view. However the overall conumption power and throughput in both propoed HO algorithm are almot identical. A a reult, the firt HO algorithm perform better if UE can acce it poition and velocity information via GPS. REFERENCES [] S. Samarakoon, M. Benni, W. Saad, and M. Latva-aho, Opportunitic leep mode trategie in wirele mall cell network, in Proc. of IEEE International Conf. on Comm. 204(ICC 4), Jun. 204, pp [2] K. M. S. Huq, S. Mumtaz, M. Alam, J. Rodriguez, and R. L. Aguiar, Frequency allocation for hetnet comp: Energy efficiency analyi, in Proc. of Wirele Comm. Sytem (ISWCS 203), Aug. 203, pp. 5. [3] S.Zhou, A.J.Goldmith, and Z.Niu, On optimal relay placement and leep control to improve energy efficiency in cellular network, in Proc. of IEEE International Conf. on Comm. (ICC ), Jun. 20, pp. 6. [4] H. Zhang, C. Jiang, N. C. Beaulieu, X. Chu, X. Wen, and M. Tao, Reource allocation in pectrum-haring ofdma femtocell heterogeneou ervice, IEEE Tran. on Comm., vol. 62, no. 7, pp , Jun [5] Z. Guohua, P. Legg, and G. Hui, A network controlled handover mechanim and it optimization in lte heterogeneou network, in Wirele Communication and Networking Conference (WCNC), 203 IEEE, April 203, pp [6] A. Okada, Game Theory. Yuhikaku, 996. [7] G. Koudouridi and H. Li, Ditributed power on-off optimiation for heterogeneou network - a comparion of autonomou and cooperative optimiation, in in Proc. of IEEE 7th International Workhop on Computer Aided Modeling and Deign of Communication Link and Network (CAMAD), Sep. 202, pp [8] H. Kim, G. de Veciana, X. Yang, and M. Venkatachalam, Ditributed α-optimal uer aociation and cell load balancing in wirele network, Networking, IEEE/ACM Tranaction on, vol. 20, no., pp , Feb. 202.