文档库

最新最全的文档下载
当前位置:文档库 > Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

Mohmmad Anas?,Claudio Rosa?,Francesco D.Calabrese?,Klaus I.Pedersen?,and Preben E.Mogensen???Department of Electronic Systems,Aalborg University,Denmark

?Nokia Siemens Networks,Aalborg,Denmark

ma@es.aau.dk

Abstract—Long Term Evolution(LTE)architecture shall sup-port end-to-end quality of service(QoS).For the QoS support and service differentiation it is important that the admission control and packet scheduling functionalities are QoS-aware. In this paper a combined admission control and a decoupled time-frequency domain scheduling framework for LTE uplink is presented.The proposed framework is shown to effectively differentiate QoS user classes in a mixed traf?c scenario.

I.I NTRODUCTION

Long Term Evolution(LTE)is a3GPP standard based on a decentralized architecture with most of the radio resource man-agement functionalities embedded in evolved Node-B(eNB) e.g.admission control(AC),packet scheduling(PS)etc[1]. LTE uplink utilizes the dynamically shared channel with fast link adaptation based on adaptive modulation and coding (AMC),fractional power control(FPC),hybrid automatic repeat request(HARQ),and transmission time interval(TTI) of1ms.LTE is targeted to effectively guarantee the quality of service(QoS)of services such as audio/video streaming, gaming and voice over IP(V oIP).To provide QoS control,it is necessary that AC and PS are QoS-aware[2][3].

The QoS aware AC grants or denies access to a new radio bearer depending on whether the required QoS of the new radio bearer will be ful?lled while guaranteeing the required QoS of the in-progress sessions[4].Furthermore,the QoS aware PS allocates the dynamically shared data channel to the active radio bearer so as to ful?l their required QoS.

The bearer level QoS parameters in LTE are QoS class identi?er(QCI),allocation and retention priority(ARP), guaranteed bit rate(GBR)and aggregate maximum bit rate (AMBR)[5].The QCI is a scalar identi?er which does a mapping to a service type based on bearer priority,packet delay budget,and packet loss rate.The ARP decides whether a GBR bearer can be accepted or has to be rejected in case of resource limitation.Hence,the ARP is primarily used for AC decision.The GBR denotes the bit rate that can be expected to be provided by a GBR bearer.The AMBR is shared by all the non-GBR bearers of a user equipment(UE).These QoS parameters are signaled from the access gateway(aGW)to the eNB for the bearers over the S1interface as shown in Fig.1. This paper studies the performance of combined AC and PS for QoS support and service differentiation.In this paper a QoS aware PS is proposed and it is combined with the AC algorithm designed in[6].This paper considers GBR as the only QoS

aGW eNB UE

Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

Fig.1.QoS parameter settings in LTE

parameter for a bearer,and each user is assumed to have a single bearer.A mixed best effort(BE),and a mixed constant bit rate(CBR)streaming and BE scenarios with different GBR settings are simulated to show that the proposed combined AC and PS is suitable for the QoS differentiation.In practice BE users are non-GBR bearers,but in this paper GBR is used as the target bit rate for BE users to test the proposed algorithms. The rest of the paper is organized as follows.In Section II, QoS aware AC algorithms are described.The QoS aware PS metrics used are discussed in Section III.In Section IV the system level simulator is described.In Section V,simulation results are presented,and Section VI contains the concluding remarks.

II.Q O S A WARE A DMISSION C ONTROL

A.Reference AC Algorithm

The reference AC algorithm decides to admit a new user if the sum of the GBR of the new and the existing users is less than or equal to a prede?ned R max as expressed in(1).

K

i=1

GBR i+GBR new≤R max,(1)

where K is the number of existing users in the cell.The users in a cell require different amount of resources to ful?l their required GBR as it depends on their radio channel quality.A drawback of the reference AC algorithm is that it treats all the users equally and does not differentiate them based on their channel quality.Furthermore,R max is a tunable parameter and does not represent the actual average uplink cell throughput,

which is time-variant as it depends on the resources allocated to the users and their experienced channel quality[2].

B.Fractional Power Control(FPC)based AC Algorithm The FPC based AC algorithm checks if the sum of the required number of physical resource blocks1(PRBs)per TTI (N i)by existing users and a new user requesting admission is less than or equal to the total number of PRBs in the system bandwidth(N tot).This can be expressed as

K

i=1

N i+N new≤N tot.(2)

Hence,the AC problem is to calculate the required number of PRBs per TTI of a user while satisfying its GBR require-ment and transmit power constraint.

The N i of the existing users can be estimated at the eNB by using the average scheduled throughput per PRB information, while the N new needs to be estimated using the pathloss(P L) and required GBR information.The pathloss of a user can be estimated at the eNB by using the downlink reference signal received power(RSRP)measurement signaled by the user over the radio resource control(RRC)[7].

In this study the N i and N new in(2)are calculated using the closed form solution derived in[6].

III.Q O S A WARE P ACKET S CHEDULING

The packet scheduling is done as a two step algorithm,?rst time-domain(TD)scheduling is used to select the users which will then be multiplexed using frequency-domain(FD) scheduling as shown in Fig.2[8].

A.Time Domain Packet Scheduling(TDPS)

In this paper a GBR-aware packet scheduler is used in TD,which prioritize the users according to the metric in(3) giving highest priority to the user which is farthest below its GBR requirement.R i is the past average throughput of user i calculated using exponential average?ltering[9].

M TD,i=GBR i

R i

(3)

B.Frequency Domain Packet Scheduling(FDPS)

The users selected by the TD scheduler are allocated PRBs based on the FD metric.In this study?exible number of PRBs are allocated per user by using an adaptive transmission bandwidth based scheduling to maximize the sum of the FD metric[10].The following FD scheduling metrics are proposed,

1Physical resource block is the basic time-frequency resource available for data transmission in LTE.It is equal to180kHz per TTI.

Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

Fig.2.QoS-aware packet scheduler design

1)Proportional Fair Scheduled FDPS Metric:The FD packet scheduler allocates the PRBs to the users selected by the TD scheduler according to the proportional fair scheduled metric as in[8].This metric is expressed as

M FD,i[k]=

?d

i[k]

R sch,i

,(4)

where?d i[k]is the estimated achievable throughput for user i on PRB k,and R sch,i is an average of scheduled user throughput,i.e.the scheduled throughput is averaged over the TTIs where user i is multiplexed in frequency domain[9]. The reason for using the proportional fair scheduled metric instead of the typical proportional fair(PF)metric is that the PF metric tends to bias the higher GBR users by giving them lower priority in FD.However,in the special case of single GBR bearers in the system the PF and PF-scheduled metric will behave similarly.

2)GBR-aware FDPS Metric:In the situation when all the bearers are multiplexed in frequency domain by the TD sched-uler,the FD metric should be able to differentiate between the GBR bearers.Hence,a FD metric in(5)is proposed which modi?es the metric in(4)so as to prioritize the bearers based on both PF-scheduled and GBR-aware metric.This metric tends to give higher priority to a user which is far below its GBR requirement.

M FD,i[k]=

?d

i[k]

R sch,i

·

GBR i

R i

(5)

IV.S IMULATION M ETHODOLOGY

The performance evaluation is done using a detailed multi-cell system level simulator which follows the guidelines in[11].The main simulation parameters are listed in Table I. The users are created in the system according to a Poisson call arrival process.If the AC decision criterion proposed in Section II is ful?lled,the user is admitted,otherwise the user is blocked.A?nite buffer traf?c model is used as BE with a GBR requirement,where each user uploads a1Mbit packet call.The session is terminated as soon as the upload is completed.This represents a?le transfer protocol(FTP)type traf?c model with a GBR requirement.A CBR streaming traf?c model is used for a GBR bearer.For each CBR bearer a?xed amount of data packets are generated at the user with a constant packet size and constant inter-arrival time.The traf?c model settings are given in Table II.

TABLE I

S YSTEM S IMULATION P ARAMETERS AND A SSUMPTIONS

Parameter Assumptions

Cellular layout Hexagonal grid,19sites,3cells per site

eNB receiver2receive antennas per cell

Inter site distance500m(Macro case1)[11]

Pathloss128.1+37.6log10(R in km)dB

Log-normal shadowing Standard deviation=8dB

Shadowing correlation 1.0for intra-site,0.5for inter-site

Penetration loss20dB

Fast fading Typical Urban(TU3)

System bandwidth10MHz(50PRBs,180kHz per PRB)

TTI1ms

User maximum power24dBm(250mW)

User noise?gure9dB

HARQ Synchronous,Adaptive

BLER Target20%

Power control Fractional power control

P o,α-59dBm,0.6

Link adaptation Fast AMC

Available MCSs QPSK[1/10,1/6,1/4,1/3,1/2,2/3,3/4]

16QAM[2/3,3/4,5/6]

User arrival Poisson process

User arrival rates[4,5,6,7,8,10]users/cell/s

Number of successful-

calls simulated10000

TABLE II

T RAFFIC M ODEL S IMULATION P ARAMETERS

Parameter Settings

Best effort(BE)traf?c

Traf?c type Finite buffer

Buffer size1Mbit

Constant bit rate(CBR)streaming

GBR512kbps

Packet size4096bits

Inter packet arrival time8ms

Buffer size2Mbits

The link-to-system level mapping is based on the actual value interface(A VI)method.It is assumed that the distance dependent pathloss and shadowing are maintained constant for each user.Moreover,the fast fading is updated every TTI based on the Typical Urban(TU)power delay pro?le for user speed of3kmph.

The system model includes synchronous and adaptive HARQ with chase combining.The power control is im-plemented according to the FPC formula standardized in 3GPP[12].The optional closed-loop adjustments are not considered,thus the power is set as

P=min{P max,P0+10log10M+αL},(6) where P max is the maximum user transmit power,P0is a cell speci?c parameter,M is the number of PRBs allocated to the user,αis the pathloss compensation factor,L is the pathloss measured by the user in dB.

The allocated bandwidth per user is assumed to be adaptive between1–48PRBs for all the scheduled users.In this paper 8users are multiplexed per TTI in frequency domain.The

Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

total

0200400600800100012001400160018002000

1

Average user throughput[kbps]

C

D

F

Fig.3.CDF of individual user throughput for a mixed GBR of[64,256]kbps as in Case I.FPC based AC,user arrival rate=8

Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

users/cell/s.

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Average user throughput[kbps]

C

D

F

Fig.4.CDF of individual user throughput for a mixed GBR of[64,1000]kbps as in Case II.FPC based AC,user arrival rate=8users/cell/s.

number of PRBs used for data transmission is48PRBs while 2PRBs are reserved for control transmission.

V.P ERFORMANCE R ESULTS

The performance of two proposed variants of FDPS are compared for the reference AC(R max=5Mbps)and FPC based AC.The performance is evaluated using the average user throughput,average cell throughput,blocking probability, and outage probability for the three cases listed in Table III. Blocking probability is de?ned as the ratio of the number of blocked users to the total number of users requesting admission.Outage probability is calculated as the ratio of the

TABLE III

U SER C LASS P ROBABILITY D ISTRIBUTION

User class Case I Case II Case III

BE user with GBR=64kbps50%50%50%

BE user with GBR=256kbps50%0%0%

BE user with GBR=1000kbps0%50%0%

CBR user with GBR=512kbps0%0%50%

5

10

1520253035

40

Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

1Number of users per cell

C D F

Fig.5.CDF of number of users per cell in Case II.FPC based AC,user arrival rate =8users/cell/s.

-130-120-110

-100-90-80-70-60-50

Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

Pathgain [dB]

B l o c k i n g p r o b a b i l i t y

Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

Pathgain [dB]

O u t a g e p r o b a b i l i t y

Fig. 6.(a)Blocking probability vs.pathgain,(b)Outage probability vs.pathgain for individual user classes in Case II.FPC based AC,GBR-aware FDPS,user arrival rate =8users/cell/s.

number of users not ful?lling their GBR requirement to the total number of admitted users.

Fig.3shows the CDF of user throughput of different user classes in Case I for combined FPC based AC and proposed FDPS metrics.We notice that GBR-aware FDPS (GBRwt)has fairer throughput distribution compared to PF-scheduled FDPS (PFsch),but the gain in outage performance is negligible.This is because the two user classes in Case I have relatively low GBR requirements,and therefore more than the maximum number of users scheduled per TTI (8users)in frequency domain are active.Hence GBR-aware TDPS is suf?cient to differentiate between the users and GBR-aware FDPS becomes redundant.

Fig.4shows the CDF of user throughput of different user classes in Case II for combined FPC based AC and

Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

0Carried traffic (Average cell throughput)[Mbps]

B l o c k i n g p r o b a b i l i t y

Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

0.050.10.150.20.250.30.350.4

0.45Carried traffic (Average cell throughput)[Mbps]

O u t a g e p r o b a b i l i t y

Fig.7.(a)Blocking probability vs.carried traf?c,(b)Outage probability vs.carried traf?c for individual user classes in Case II.

proposed FDPS metrics.We observe that GBR-aware FDPS improves the outage performance of 1Mbps GBR user class signi?cantly.In this case on average there are less number of users per cell compared to the maximum number of users scheduled per TTI in FD as shown in Fig.5.Therefore,if the relative difference between the GBR requirements is high in a mixed traf?c scenario proposed GBR-aware FDPS metric is necessary to differentiate between users.

Fig.6shows the blocking and outage probabilities versus pathgain (including distance dependent pathgain,shadowing,and antenna gain)for different user classes in Case II.For FPC based AC the blocking probability increases rapidly below certain pathgain value.Fig.6(b)shows that FPC based AC effectively blocks the user if its QoS cannot be satis?ed.We as well notice that the blocking and outage probabilities are dependent on the GBR requirement of user class since FPC based AC takes pathgain and GBR requirements to make AC decision.

Fig.7shows the blocking and outage probabilities versus carried traf?c (average cell throughput)for different user classes in Case II using FPC based AC.The blocking proba-bility for 1Mbps user class is higher than 64kbps user class,this is due to the fact that more resources are required by 1Mbps user class,as well as the blocking depends on the load conditions in the cell.We observe that the outage probability is best for the GBR-aware FDPS for both 64kbps and 1Mbps

Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

00.050.10.150.20.250.3

0.35Carried traffic (Average cell throughput)[Mbps]

B l o c k i n g p r o b a b i l i t y

(a)

Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

0.050.10.150.20.250.3

0.35Carried traffic (Average cell throughput)[Mbps]

O u t a g e p r o b a b i l i t y

(b)

Fig.8.(a)Blocking probability vs.carried traf?c,(b)Outage probability vs.carried traf?c for Case II.

user class.

Fig.8shows the blocking and outage probabilities versus carried traf?c for Case II.The combination of FPC based AC and GBR-aware FDPS performs best among the studied AC and PS combinations.This result shows that both AC and PS need to be QoS aware for the ef?cient QoS control.

Fig.9shows the CDF of user throughput of individual user class for combined CBR and BE traf?c as in Case III,with GBR-aware FDPS.In this result both CBR and BE users are admitted only if the FPC based AC criterion is ful?lled.The CBR users in outage is shown to decrease using FPC based AC,with no effect on the outage performance of BE users.This shows that the combined FPC based AC and GBR-aware FDPS is useful in a realistic CBR and BE traf?c mix scenario.

VI.C ONCLUSIONS

In this paper a combined AC and PS framework for QoS provisioning in LTE uplink is proposed.The combined FPC based AC and GBR-aware PS is shown to be able to guarantee the respective QoS requirements of different user classes in a mixed traf?c scenario.Furthermore,GBR-aware FDPS is shown to improve the outage performance compared to the PF-scheduled FDPS in the case of a mixed BE traf?c scenario with relatively high difference in the GBR requirements as in Case II.The proposed framework is also shown to effectively differentiate between user classes for a realistic mix of CBR

200

Combined Admission Control and Scheduling for QoS Differentiation in LTE Uplink

400

6008001000120014001600180000.1

0.20.30.40.50.6

0.70.80.91Average user throughput [kbps]

C D F

Fig.9.CDF of individual user throughput for mixed CBR (GBR =512kbps)and BE (GBR =64kbps)as in Case http://www.wendangku.net/doc/b80f5121ba1aa8114531d90d.htmler arrival rate =8users/cell/s.

and BE traf?c.The FPC based AC blocks the bearer with a very low pathgain,and ful?l the required GBR of admitted bearers with a very low outage probability.Unlike FPC based AC,the reference AC admits a bearer irrespective of its channel condition and very low pathgain bearers are eventually served with a signi?cantly higher outage probability.Future study will focus on the PS algorithms to differentiate and prioritize the GBR bearers over non-GBR bearers.

R EFERENCES

[1] A.Toskala,H.Holma,K.Pajukoski,and E.Tiirola,“UTRAN Long

Term Evolution in 3GPP”,IEEE 17th International Symposium on Personal,Indoor and Mobile Radio Communications ,2006.

[2]K.I.Pedersen,P.E.Mogensen,and T.E.Kolding,“Overview of QoS

Options for HSDPA”,IEEE Communications Magazine ,July 2006.[3]P.Hosein,“A Class-Based Admission Control Algorithm for Shared

Wireless Channels Supporting QoS Services”,5th IFIP TC6Interna-tional Conference on Mobile and Wireless Communications Networks ,October 2003.

[4]3GPP TS 36.300V8.3.0,“Evolved Universal Terrestrial Radio Ac-cess (E-UTRA)and Evolved Universal Terrestrial Radio Access (E-UTRAN);Overall Description;Stage 2”,January 2008.

[5]3GPP TS 23.401V8.0.0,“General Packet Radio Service (GPRS)

Enhancements for Evolved Universal Terrestrial Radio Access Network (E-UTRAN)Access”,December 2007.

[6]M.Anas,C.Rosa,F.D.Calabrese,P.H.Michaelsen,K.I.Pedersen,

and P.E.Mogensen “QoS-Aware Single Cell Admission Control for UTRAN LTE Uplink”,IEEE 67th Vehicular Technology Conference ,May 2008.

[7]M.Anas,F.D.Calabrese,P.E.¨Ostling,K.I.Pedersen,and P.E.Mo-gensen “Performance Analysis of Handover Measurements and Layer 3

Filtering for UTRAN LTE”,IEEE 18th International Symposium on Personal,Indoor and Mobile Radio Communications ,September 2007.[8]G.Monghal,K.I.Pedersen,I.Z.Kovacs,and P.E.Mogensen,“QoS

Oriented Time and Frequency Domain Packet Schedulers for the UTRAN Long Term Evolution”,IEEE 67th Vehicular Technology Conference ,May 2008.

[9]T.E.Kolding,“QoS-Aware Proportional Fair Packet Scheduling with

Required Activity Detection”,IEEE 64th Vehicular Technology Con-ference ,September 2006.

[10] F.D.Calabrese,C.Rosa,M.Anas,P.H.Michaelsen,K.I.Pedersen,

and P.E.Mogensen,“Adaptive Transmission Bandwidth Based Packet Scheduling for LTE Uplink”,IEEE 68th Vehicular Technology Con-ference ,September 2008.

[11]3GPP TR 25.814V7.1.0,“Physical Layer Aspect for Evolved Universal

Terrestrial Radio Access (UTRA)”,September 2006.

[12]3GPP TS 36.213V8.1.0,“Evolved Universal Terrestrial Radio Access

(E-UTRA);Physical Layer Procedures (Release 8)”,November 2007.