Towards Low Latency in 5G HetNets: A Bayesian Cell Selection / User Association Approach

Size: px

Start display at page:

Download "Towards Low Latency in 5G HetNets: A Bayesian Cell Selection / User Association Approach"

Tobias Pope
5 years ago
Views:

1 Towards Low Latecy i 5G HetNets: A Bayesia Cell Selectio / User Associatio Approach Mohamed Elkourdi, Asim Mazi, ad Richard D. Gitli, Life Fellow, IEEE Iovatio i Wireless Iformatio Networkig Lab (iwinlab) Departmet of Electrical Egieerig, Uiversity of South Florida, Tampa, Florida 3362, USA {elkourdi, asimmazi}@mail.usf.edu, richgitli@usf.edu Abstract Expadig the cellular ecosystem to support a immese umber of coected devices ad creatig a platform that accommodates a wide rage of emergig services of differet traffic types ad Quality of Service (QoS) metrics are amog the 5G s headlie features. Oe of the key 5G performace metrics is ultra-low latecy to eable ew delay-sesitive use cases. Some etwork architectural amedmets are proposed to achieve the 5G ultra-low latecy objective. With these paradigm shifts i system architecture, it is of cardial importace to rethi the cell selectio / user associatio process to achieve substatial improvemet i system performace over covetioal maximum sigal-to- iterferece plus oise ratio (Max-SINR) ad cell rage expasio (CRE) algorithms employed i Log Term Evolutio- Advaced (LTE-Advaced). I this paper, a ovel Bayesia cell selectio / user associatio algorithm, icorporatig the access odes capabilities ad the user equipmet () traffic type, is proposed i order to maximize the probability of proper associatio ad cosequetly ehace the system performace i terms of achieved latecy. Simulatio results show that Bayesia game approach attais the 5G low ed-to-ed latecy target with a probability exceedig 8%. I. INTRODUCTION F ifth-geeratio (5G) etworks are expected to support a broad domai of emergig services with various Quality of Service (QoS) requiremets, i.e., from arrow-badwidth, delay-sesitive services to badwidth-hugry, delay-tolerat services. To address these disparate services, several solutios have recetly bee proposed. For istace, Coordiated Multi- Poit (CoMP) [] is evisioed as oe of the promiet 5G solutios by improvig the service for cell edge users sufferig from high levels of iterferece particularly i multitier etworks. CoMP has bee stadardized i LTE-Advaced (Release.) [2]. Moreover, 5G is aticipated to witess a icrease i the heterogeeity ad desity of access odes (ANs) (e.g. ultradese etworks (UDNs)) [3] as a solutio to cope with the tremedous growth i the umber of devices coected to the etwork ad cosequetly boost the system s capacity [4]. However, i UDNs the multi-tier iterferece becomes more severe. Hece, the radio resource allocatio process should be carried out i a cetral uit (CU), while takig ito accout the bigger etwork picture [5]. For istace, [6] proposes exploitig the temporal ad spatial traffic fluctuatios i the etwork to reduce the iterferece levels by turig off some ANs with low or o traffic loads. Recetly, the Cloud-Radio Access Network (C-RAN) architecture has attracted attetio as a key eabler i implemetig iterferce avoicdece ad cacellatio algortihem for flexiable multi-tier 5G etworks [7]. Despite the C-RAN s cetralizatio throughput gai, which is achieved by poolig the computatioal resources, there is a challege regardig the delay-sesitive traffic due to the large etwork latecy accruig from trasferrig the data through the core etwork ad the Iteret backboe. To satisfy the low latecy requiremets of delay-sesitive traffic, some computig processes, applicatio servers ad cachig capabilities ca be migrated from the cloud to the edge of the etwork as a evolved architecture [8]. This architecture is commoly referred to as a Fog-Radio Access Network (F-RAN) [9], []. Hece, some of the traffic, i particular delay-sesitive traffic, ca get served at the F- RAN odes ad do ot eed to travel to the core etwork ad the Iteret. This results i a sigificat reductio i the etwork latecy []. As a atural result of the expected icrease of heterogeeity i 5G etworks, user equipmet () will have a large rage of coectivity optios with differet characteristics such as power cosumptio, latecy budget, ad the achievable data rate. To best exploit the available opportuities, both the s traffic type ad the characteristics of available ANs should be take ito accout at the cell selectio / user associatio stage. I traditioal cellular systems, the process of cell selectio / user associatio is based o the AN that ca provide the highest sigal-to-iterferece-plus-oise ratio (SINR) [2]. However, this approach is geerally ot optimum i multitier etworks with diverse traffic. To alleviate this problem, [3] itroduces the cell rage expasio (CRE) approach for Low Power Nodes (LPNs) via a biasig method such that the High Power Node (HPN) trasmit power is reduced o a group of sub-carriers i order to eable better coverage o the same group of sub-carriers for a overlaid LPN. Aother approach for cell associatio is to employ user-perceived rate cosiderig the SINR ad the etwork load [4]. However, these papers did ot take ito accout the differet types of traffic. Ideed, the selectio ad associatio procedure based o the maximum SINR ad CRE criteria oly might degrade the performace of the system from a latecy perspective (i.e. whe s of delay-tolerat traffic get associated with Fog-Low Power Nodes (F-LPNs) or whe s of delaysesitive traffic are associated with a HPN).

2 The mai cotributios of this paper are threefold. First, a system model that supports diverse traffic types is preseted. Followig that a ovel cell selectio / user associatio algorithm based o Bayesia game from the domai of Game Theory is proposed, while icorporatig the the traffic type ad ANs capabilities. Fially the superiority of the proposed Bayesia cell selectio / user associatio algorithm is prove iterms of the achieved latecy ad the probability of proper assocaitio for delay-sesitive ad delay-tolerat traffic. The remider of this paper is orgaized as follows. Sectio II describes the system model. Sectio III defies the problem ad the proposed Bayesia Game approach. I this part, the utility fuctios of the ad the etwork are defied. Sectio IV ivestigates the performace of the proposed Bayesia game algorithm through simulatios. Fially, the cotributios of the paper are summarized i sectio V. II. SYSTEM MODEL We cosider a simplified two-tier heterogeeous etwork (HetNet) cosistig of oe HPN overlaid by several F-LPNs with cachig ad computatio capabilities as show i Fig.. The set of all radio access odes (ANs) i the HetNet is defied as N = {,,..., N }, where represets the HPN ad the subset L = {,..., N } deotes the F-LPNs. The F-LPNs are radomly distributed i the service area with spatial desity of λ (F-LPNs/km 2 ). The set of all s uder the coverage area of the two-tier HetNet is deoted by U = {u,..., u K }. Each k U requests a service class defied as the tuple φ k = (η k, τ k ), where η k, τ k are the k required data rate ad latecy respectively. Hece, the s traffic i the proposed system model, ca be maily classified as delay-sesitive (DS) or delay-tolerat (DT) accordig to their latecy requiremet τ. The sigal to iterferece plus oise ratio (SINR) at k, associated with access ode N whose trasmittig i dowli o the resoruce block r, is expressed as γ r h P r = r /() h P r + W N, () Where P r is defied as the trasmitted power from access ode to k o resource block r R, R is the total umber of resource blocks (RBs), W is the badwidth of each RB, N is the thermal oise spectral power, r /() are the access odes which are usig the resource block r ad causig iterferece o the k associated with the access ode, ad h is the chael betwee access ode ad the k. The chael model icorporates the effects of smallscale fadig ad large-scale fadig (the latter icludes path loss ad shadowig). The achievable dowli data rate at the k from AN o sigle resource block r is give by R r = W log 2 ( + γ r ). (2) Sigalig Data F-LPN CoMP amog F-LPN ad HPN Radio Access Cotroller (RAC) HPN CoMP amog F-LPNs Fig.. System Model: Two-tier 5G HetNet cosistig of RAC, HPN ad F-LPNs. Hece, the total achieved data rate from all the access odes associated with k ca be writte as R k = ψr r, r (3) where {, } ad ψr {, } are the user associatio ad the resource allocatio cotrol variables respectively. To meet the data rate requiremet, we assume that the k ca be associated with M access odes, which is referred to as CoMP trasmissio M. (4) Thus, the AN serves the k if the miimum data rate requiremet i the service class tuple is guarateed. Hece we ca write R k = ψr r r η k. (5) The access odes assig the eeded resource blocks to k to satisfy (5). The umber of resource blocks is give by ηk η k Γ = =, (6) R k α k ψr Rr where is the ceilig fuctio. Similarly, to meet the latecy requiremet of k τ k, (7) where is the iverse of the statistical roud trip delay-time (RTT) from access ode to k. As metioed previously, the F-LPNs are fitted with computatio ad cachig capabilities that are brought close to the etwork edge. Therefore, the statistical RTT of the F-LPN is assumed to be much smaller tha the statistical RTT of the HPN F-LPN > HPN [5]. The time required by a processor to serve a request is ot icluded. It is assumed that with the advacemet i computig powers ad by aggregatig multiple CPUs i a cetral uit, the processig latecy ca be eglected compared to the other latecy compoets.

3 III. PROBLEM FORMULATION AND PROPOSED BAYESIAN A. Probelem Formulatio GAME APPROACH I this subsectio, the problem of proper cell selectio / user associatio is formulated. The utility fuctios are defied to represet the etwork s resource utilizatio ad the s degree of satisfactio with the Quality of Service (QoS) icludig latecy. The aim is to maximize the s utility fuctio for improved QoS satisfactio ad the etwork s resource utilizatio with respect to it s prefereces. Therefore, the ad the etwork utility fuctios, are respectively writte as U k (α, ψ) = α k U Net = ω (θ k ) ψr r r θ k, (8) α k N, (9) where N is the total umber of available resources i AN ad ω (θ k ) is represetig the AN preferce for k traffic type θ k. The objective of the is to maximize (8) as max α,ψ U k (α, ψ), subject to (4), (5), (6), (7). () Similarly, the objective of the etwork is to maximize (9) as max : U Net, ω () subject to, ψ, r η k < C, where C is the cap o the data rate allowed for k by access ode. We modeled the cell selectio / user associatio problem as a Bayesia Game 2 [6]. The motivatio for selectig a Bayesia game to model this problem is as follows. First, performig a joit optimizatio for both the ad the etwork usig regular optimizatio methods ca be complex ad computatioally itesive. Also, as previously metioed, some of the proposed solutios such as the desificatio i deployig LPNs ad brigig the cotet closer to the user, are desiged to accommodate the expected icrease i data rates ad to lower the ed-to-ed latecy for certai use cases. However, achievig the optimum system performace may ot be feasible by just applyig those solutios due to the lack of AN s a priori iformatio about the s exact traffic type (at the access etwork, rather at the core etworklevel). Furthermore, assumig that the radio access odes kow the exact type of the s traffic i real-time via the core etwork level is urealistic sice this takes cosiderable time ad makes achievig low latecy quite ulikely. Hece, selectig a Bayesia game to model this problem, where perfect kowledge about s traffic types is ot available at the access ode, is well justified. 2 The proposed cell selectio / user associatio Bayesia Game algorithm is implemeted i a cetral uit hypervisor referred to as the Radio Access Cotroller (RAC) which maages multiple access odes. B. Proposed Bayesia Game Approach Defiitio: The cell selectio / user associatio Bayesia game is defied i the strategic form as G = (P, Θ, A, P, U), where: Players (P): Set of two players P = {u k, }: the k ad the AN, respectively. Types (Θ): Set of possible types for k accordig to its traffic Θ = {θ k, = DS, θ k,2 = DT }, where θ k,j is the type j for the k. Actios (A): The space of all possible combiatioal actios A = A A 2. Where, the s actio space A = {H = selects HPN, L = selects LPN} ad the AN s actio space A 2 = {S = serve, C = CoMP}. Prior Probabilities (P): The probability P(θ j ) over the types of users. Utility fuctios (U ): The ad the etwork utility fuctios as defied i (8) ad (9), respectively. The player i s strategy is mappig s i : Θ i A i, which represets the player i s actio for each possible type. I our Bayesia model, we started with a ubiased prior probability over the types. However, a give AN oly kows its type ad strategy, ad does ot kow the strategies selected by the s or their actual traffic type. The expected utility of a player i uder strategy profile s is E(U i ) = U i (s(θ j ), θ j )P(θ j ). (2) θ j Θ Hece, the expected utility (payoff) of is E(Uk ) = P(θ j ) ψ r R r θ +( P(θ j )) ψr r r θ. Similarly, the expected utility (payoff) of etwork is ( E(U Net ) = P(θ j ) ω (θ) ( +( P(θ j )) ω (θ) α k N α k N ) ). (3) (4) The k is of type θ k,, whe Uk (L, S) > Uk (H, S) ad Uk (L, C) > Uk (H, C). Cosequetly strategy L strictly domiates strategy H. O the other had, the is of type θ k,2, whe Uk (H, S) > Uk (L, S) ad Uk (H, C) > Uk (L, C). The, strategy H strictly domiates strategy L. Startig with a ubiased belief (a priori probability) about the types of the s, the the expected utility (payoff) of the domiat strategies 3 for both traffic types is assumed to be eve (at the begiig the etwork is ot biased i its choice for ay type of users). Hece, usig TABLE I, the posterior probability P ca be calculated as i (5).

4 Coects to H L P = ω (θ k,j ) TABLE I PAYOFF MATRIX Delay Sesitive (DS) Delay Tolerat (DT) P -P Network Network S C S C (Uk (H, ) (Uk (H, ) (Uk (H, ) (Uk (H, ) (Uk (L, ) (Uk (L, ) (Uk (L, ) (Uk (L, ) Γ N \ α N \ k N Γ ω α k N (θ k,j ) Γ ω (θ k,j ) α k N α k N. (5) \ α + k N α k N.9 Bayesia Game approach Max-SINR approach respectively. The path loss is based o TS model [7], for a urba eviromet as Probability of Latecy > Latecy (sec) Fig. 2. CCDFs of achieved latecy for delay-sesitive traffic IV. SIMULATION RESULTS AND ANALYSIS I this sectio, we preset the simulatio setup ad discuss the performace results of the proposed Bayesia cell selectio / user associatio algorithm. The performace of the proposed Bayesia cell selectio approach is evaluated i terms of the probability of proper associatio ad the achieved latecy with respect to covetioal CRE ad Max-SINR based cell selectio / user associatio algorithms used i LTE-Advaced. I the simulatio, proper associatio is defied as the average umber of delay-sesitive ad delay-tolerat traffic associated with LPNS ad HPN, respectively. The simulatio parameters are preseted i TABLE II. I the simulatio setup, we assume that there is oe HPN ad F-LPNs that are radomly deployed over the service regio. The s are radomly distributed followig a homogeeous PPP ad each could be oe of the two types DS or DT. The DT ad DS applicatios traffic demads are modeled as a uiform radom variable o [5,] Mbps ad [.,4] Mbps, 3 I game theory, player i s strategy is called a domiat strategy if it has a higher payoff tha the payoff of all other strategies, that is U i > U j i, regardless of the actios of the other players. P L = 4( 4 3 hb) log (d) 8 log (hb) +2 log (f) + 8 db, (6) where d is the separatio betwee the ad the AN, hb the height of the AN atea ad the f is the carrier frequecy. The shadow fadig is logormal distributed ad the fast fadig is based o the Wier II model [8]. I simulatio, the statistical RTT for the F-LPN (/ F-LPN ) is modeled as a uiform radom variable o [.5,.5] ms ad for HPN (/ HPN ) as the sum of three uiform radom variables, amely: radio access [.5,.5] ms, the core etwork [,2] ms ad [5,] ms for the Iteret. After performig simulatios, we used the CCDF of achieved latecies of the delay-sesitive traffic to evaluate the Bayesia game i compariso with the Max-SINR approach. As show i Fig. 2, the miimum achieved latecy usig the SINR approach is 6.8 ms due to the improper associatio that happes whe delay-sesitive traffic is associated with HPN usig the max-sinr criterio while the Bayesia approach achieves latecies < ms. I additio to latecy, we study the proper associatio of differet schemes. Fig. 3 illustrates the cumulative distributio fuctios (CDFs) of proper associatios for the proposed Bayesia game compared with the CRE ad Max-SINR approaches. It is oted from Fig. 3 that the miimum percetage of proper associatios for the delay-sesitive traffic of the Bayesia approach is greater tha or equal to 82%, which outperforms the proper associatio of the covetioal Max- SINR ad the CRE approaches. Similarly, usig the CDF of the delay-tolerat traffic as illustrated i Fig. 4, we observed that the CRE ad Max-SINR approaches fall behid the Bayesia game. The Bayesia game approach attais a proper associatio greater tha or equal to 8% for delay-tolerat traffic. V. CONCLUSION The three solutios proposed for 5G, amely: Dese LPNs, F-RANs ad CoMP are ot sufficiet i themselves to meet the

5 Probability of Proper Associatio < (x) % Probability of Proper Associatio < (x) % CRE approach Bayesia Game approach Max-SINR approach Proper Associatio (x) % Fig. 3. CDFs of proper associatio for delay-sesitive traffic CRE approach Max-SINR approach Bayesia Game approach Proper Associatio (x) % Fig. 4. CDFs of proper associatio for delay-tolerat traffic diverse requiremets for a wide rage of emergig applicatios ad i particular the low-latecy target. For this reaso, it is essetial to rethi the traditioal cell selectio/ user associatio procedures, by cosiderig both the traffic type ad the access ode capabilities. I this paper, a ovel method for cell selectio/ user associatio for 5G heterogeeous etworks is proposed usig Bayesia game. The utility fuctios of the user equipmet ad the etwork are defied based o the achievable data rate, the statistical RTT ad the access ode s traffic prefereces. Simulatio results demostrate that the proposed Bayesia game algorithm provides a sigificat improvemets i terms of the probability of proper associatio (as a fuctio of traffic type) ad the achieved latecy compared with Max-SINR criterio ad CRE approach. Such a methodology ca be quite importat i achievig the 5G low latecy objective. TABLE II SIMULATION PARAMETERS Parameter Value Service area 4 m 4 m Number of ANs N = L = +9 F-LPNs desity (λ) 2 F-LPN/km 2 Number of s per AN 4 Total trasmit power of ANs {46.2,4} dbm Trasmit atea height of ANs (hb) 2 m Carrier frequecy (f ) 2.4 GHz Badwidth 2 MHz ACKNOWLEDGMENT The authors tha Dr. Ere Balevi for the fruitful discussio ad valuable suggestios. REFERENCES [] A. Ghosh, R. Ratasuk, B. Modal, N. Magalvedhe, ad T. Thomas, LTE-advaced: ext-geeratio wireless broadbad techology [ivited paper], IEEE Wireless Commuicatios, vol. 7, o. 3, pp. 22, Jue 2. [2] 3GPP, Coordiated Multi-Poit Operatio for LTE Physical Layer Aspects (Release ), TR 36.89, December 2. [3] J. G. Adrews et al., What will 5G be? IEEE Joural o Selected Areas i Commuicatios, vol. 32, o. 6, pp , Jue 24. [4] 3GPP, Service requiremets for ext geeratio ew services ad markets; stage (release 5), TS 22.26, August 26. [5] A. Checko et al., Cloud RAN for mobile etworks;a techology overview, IEEE Commuicatios Surveys Tutorials, vol. 7, o., pp , Firstquarter 25. [6] C. L. I, C. Rowell, S. Ha, Z. Xu, G. Li, ad Z. Pa, Toward gree ad soft: a 5G perspective, IEEE Commuicatios Magazie, vol. 52, o. 2, pp , February 24. [7] M. Peg, Y. Li, J. Jiag, J. Li, ad C. Wag, Heterogeeous cloud radio access etworks: a ew perspective for ehacig spectral ad eergy efficiecies, IEEE Wireless Commuicatios, vol. 2, o. 6, pp , December 24. [8] A. Segupta, R. Tado, ad O. Simeoe, Cache aided wireless etworks: Tradeoffs betwee storage ad latecy, i 26 Aual Coferece o Iformatio Sciece ad Systems (CISS), March 26, pp [9] M. Peg, S. Ya, K. Zhag, ad C. Wag, Fog-computig-based radio access etworks: issues ad challeges, IEEE Network, vol. 3, o. 4, pp , July 26. [] Y. Y. Shih, W. H. Chug, A. C. Pag, T. C. Chiu, ad H. Y. Wei, Eablig low-latecy applicatios i fog-radio access etworks, IEEE Network, vol. 3, o., pp , Jauary 27. [] T. C. Chiu, W. H. Chug, A. C. Pag, Y. J. Yu, ad P. H. Ye, Ultralow latecy service provisio i 5G fog-radio access etworks, i 26 IEEE 27th Aual Iteratioal Symposium o Persoal, Idoor, ad Mobile Radio Commuicatios (PIMRC), Sept 26, pp. 6. [2] S. Ahmadi, LTE-Advaced: A Practical Systems Approach to Uderstadig 3GPP LTE Releases ad Radio Access Techologies, 23. [3] 3GPP, Rage Expasio Techiques for HetNets, R 2453, Qualcomm Icorporated, October 22. [4] Q. Ye, B. Rog, Y. Che, M. Al-Shalash, C. Caramais, ad J. G. Adrews, User associatio for load balacig i heterogeeous cellular etworks, IEEE Trasactios o Wireless Commuicatios, vol. 2, o. 6, pp , Jue 23. [5] GSMA ad Itelligece, Uderstadig 5G: Perspectives o future techological advacemets i mobile, GSMA Itelligece Uderstadig 5G, o. December, pp. 3 5, 24. [6] Y. Shoham ad K. Leyto-Brow, Multiaget Systems: Algorithmic, Game-Theoretic, ad Logical Foudatios. Cambridge Uiv. Press, 28. [7] 3GPP, E-UTRA; LTE RF system scearios, TS , 28. [8] P. Kysti, J. Meiil et al., IST WINNER II, EBITG,TUI, UOULU, CU/CRC, NOKIA, Tech. Rep., September 27.

A New Space-Repetition Code Based on One Bit Feedback Compared to Alamouti Space-Time Code

A New Space-Repetition Code Based on One Bit Feedback Compared to Alamouti Space-Time Code Proceedigs of the 4th WSEAS It. Coferece o Electromagetics, Wireless ad Optical Commuicatios, Veice, Italy, November 0-, 006 107 A New Space-Repetitio Code Based o Oe Bit Feedback Compared to Alamouti