Qos Based Cac Nsn Paper

7/31/2019 Qos Based Cac Nsn Paper

1/5

QoS-Aware Single Cell Admission Control

for UTRAN LTE Uplink

M. Anas, C. Rosa, F. D. Calabrese, P. H. Michaelsen, K. I. Pedersen, and P. E. Mogensen

Department of Electronic Systems, Aalborg University, DenmarkNokia Siemens Networks, Aalborg, Denmark

[email protected]

Abstract UTRAN Long Term Evolution (LTE) architectureshall support end-to-end quality of service (QoS). To maintain theQoS of in-progress sessions in a cell it is important to admit a newradio bearer only if all the existing sessions and the new bearercan be guaranteed QoS according to their requirements. Henceadmission control (AC) has an important role to play for QoSprovisioning. In this paper we propose an AC algorithm for LTEuplink utilizing the fractional power control formula agreed in3GPP. The proposed AC algorithm is compared with a reference

AC algorithm. It is shown that at 10% unsatisfied user probabilitythe carried traffic gain of the proposed AC algorithm can be upto around 29% over the reference AC algorithm. Furthermore,the proposed AC algorithm inherently tunes itself for varyingload conditions without additional complexity.

I. INTRODUCTION

UTRAN Long Term Evolution (LTE) is based on a de-

centralized architecture with most of the radio resource man-

agement functionalities e.g. admission control (AC), mobility

control etc., embedded in evolved Node-B (eNB) [1]. As

shown in Fig. 1, the AC for LTE uplink is located in the

eNB at Layer 3 will utilize the local cell load information to

make the AC decision. Hence, the focus of this paper is on

the single cell AC.

UTRAN LTE is targeted to efficiently guarantee the quality

of service (QoS) of services such as audio/video streaming,

gaming and voice over IP (VoIP). To provide QoS control, it is

necessary that AC and packet scheduling (PS) are QoS aware

[2]. The QoS aware AC determines whether a new radio bearer

should be granted or denied access based on if required QoS

of the new radio bearer will be fulfilled while guaranteeing

the required QoS of the in-progress sessions [3].

The uplink AC algorithms for WCDMA systems are based

on estimating and maintaining the increase in intra-cell inter-

ference for admitting a new user [4]. As opposed to this, inLTE uplink intra-cell interference is non-existent because of

the use of orthogonal multiple access scheme. Furthermore,

in LTE uplink users are scheduled on the dynamically shared

channel with fast adaptive modulation and coding (AMC)

and hybrid ARQ (HARQ). Therefore, the AC algorithms

for WCDMA system will not be suitable for LTE. An AC

algorithm in downlink for IEEE 802.16e, which has similar

PHY-MAC as LTE, to provide QoS is proposed in [5]. But this

downlink AC algorithm cannot be used in uplink because of an

additional degree of freedom i.e., user transmit power, which

Access Gateway

S1S1

S1

X2

X2 X2

eNB

eNB

eNB

UE

eNB Layer 2Packet Scheduler

eNB Layer 3Admission ControlMobility Control

Fig. 1. UTRAN LTE system architecture

is different for all the users and it varies with transmission

time interval (TTI) due to power control.

In this paper, we propose an AC algorithm for LTE utilizing

the fractional power control (FPC) formula agreed in 3GPP

[6]. Additionally, a reference AC algorithm is developed to

compare the performance of the proposed AC algorithm.

Blocking, outage and unsatisfied user probabilities are used

to evaluate the performance of the proposed and reference AC

algorithms. This paper considers guaranteed bit rate (GBR) as

the only QoS criterion of the bearer, and each user is assumed

to have a single-bearer.

The rest of the paper is organized as follows. In Section

II, reference AC algorithm and a QoS-aware AC algorithm

is proposed for LTE uplink. The proposed AC algorithm is

compared with the reference AC algorithm using a detailed

system level simulator described in Section III. In Section IV,

simulation results are presented illustrating the comparison

of the proposed AC algorithms, and Section V contains the

concluding remarks.

I I . UPLINK ADMISSION CONTROL

A. Reference AC Algorithm

This baseline 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 the predefined average uplink cell throughput

(Rmax) as expressed in (1).

Ki=1

GBRi + GBRnew Rmax (1)

978-1-4244-1645-5/08/$25.00 2008 IEEE 2487


2/5

Drawback of the reference AC algorithm is that it treats all

the users equally and does not differentiate them based on their

location in the cell. Furthermore, fixed value of Rmax does

not represent the actual average uplink cell throughput. For

Rmax this algorithm will admit all the users requestingadmission and is equivalent to the case of no AC.

B. Proposed AC AlgorithmTo fulfill the GBR requirement a user can either be allocated

larger bandwidth and transmit at lower power spectral density

(PSD) i.e., lower modulation and coding scheme (MCS), and

hence lower signal-to-interference-plus-noise ratio (SINR).

Otherwise, a user can be allocated smaller bandwidth and is

required to transmit at higher PSD i.e., higher MCS and hence

higher SINR, to achieve the required GBR. This is intuitive

from the Shannon capacity curve as shown in Fig. 2.

The proposed AC algorithm checks if the current resource

allocation can be modified so as to admit the new user and

satisfy the GBR requirements of all the active users and

the new user. Hence, the admission criterion for the new

user is that the sum of the required number of physicalresource blocks1 (PRBs) per TTI (Ni) by a new user requesting

admission and existing users is less than or equal to the total

number of PRBs in the system bandwidth (Ntot) e.g. 50 PRBs

in 10 MHz [7]. This can be expressed as,

Ki=1

Ni + Nnew Ntot, (2)

where K is the number of existing users in the cell. Hence,

the AC problem is to calculate the required number of PRBs

per TTI of a user while satisfying its GBR requirement and

transmit power constraint.

C. Calculation of Ni and Nnew in (2)

The Ni of the existing users could be measured at the eNB

by using the average number of PRBs allocated to these users.

In this section we calculate the Ni and Nnew assuming that

the required GBR and pathloss2 (P L) of the existing users

and new user requesting admission is known to the AC unit at

the eNB. The modified Shannon formula [8] in (3) is used to

calculate the required average number of PRBs per TTI (Ni)

for a known GBRi and P Li for user i.

S[bits/s/Hz] = BWeff.. log21 +SN R

SN Reff (3)

where, BWeff is the system bandwidth efficiency, SN Reffadjusts for the signal-to-noise ratio (SNR) implementation ef-

ficiency, and is the correction factor which nominally should

be equal to one. As shown in Fig. 3, [BWeff, , S N Reff] =[0.72, 0.68, 0.2 dB] for 1x2 antenna deployment, gives a good

1Physical resource block is the basic time-frequency resource available fordata transmission in LTE. It is equal to 180 kHz per TTI.

2Pathloss between serving eNB and user is measured and averaged over fastfading from the pilots at the user in downlink, and is transmitted in uplinkusing layer 3 signaling.

S2

S1

SINR1 SINR2

[bps/Hz]

MCS1

MCS2

MCS3

MCS4

SINR

Spectral efficiency

Fig. 2. Spectral efficiency vs. SINR curve for different MCS. GBR can bemaintained using lesser number of PRBs at higher MCS i.e., higher spectralefficiency, and hence higher SINR using higher PSD limited by the maximumuser transmit power.

curve fit to the LTE uplink link-level results in [9]. The

required spectral efficiency (Si) for GBRi and Ni PRBs is,

Si =GBRi

Ni.BWPRB

(4)

where, BWPRB = 180 kHz. The SINR of user i with transmit

power Pi is,

SINRi =Pi

Ni.BWPRB.P Li.IoT.N0.N F(5)

where IoT is the total received interference power plus ther-

mal noise from all the users in the neighboring cells divided

by the thermal noise power, N0 is the thermal noise power

density at the antenna, and N F is the noise figure at the eNB.

It has been concluded within 3GPP that the power control

for the physical uplink shared channel (PUSCH) will be FPC

consisting of open loop power control along with aperiodic

closed-loop adjustments as in (6) [6][10].

Pi = min {Pmax, P0 + 10log10 Ni + Li + MCS + f(i)}(6)

where Pi is total user power in dBm, P0 is a user specific

parameter, Ni is the number of assigned PRBs to the user,

-10 -8 -6 -4 -2 0 2 4 6 8 10 12 140

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

2.2

2.4

SNR [dB]

Spec

tra

le

fficiency

[b

ps

/Hz

]

1x2 SIMO; 10 MHz

Modified Shannon [0.72, 0.68, 0.2 dB]

Fig. 3. Modified Shannon fit curve using the link-level results in [9].

2488


3/5

TABLE I

SIMULATION PARAMETERS

Parameter Assumptions

Cellular layout Hexagonal grid, 19 sites, 3 cells per siteeNB receiver 2 receive ant ennas per cellInter site distance 500 m (Macro case 1) [7]Pathloss 128.1 + 37.6log10(R in km) dBLog-normal shadowing Standard deviation = 8 dB

Shadowing correlation 1.0 for intra-site, 0.5 for inter-sitePenetration loss 20 dBFast fading Typical Urban (TU3)System bandwidth 10 MHz (50 PRBs, 180 kHz per PRB)TTI 1 msUser maximum power 24 dBm (250 mW)User noise figure 9 dBHARQ Synchronous, AdaptiveBLER Target 20%Power control Fractional power controlPo, -59 dBm, 0.6Link adaptation Fast AMCAvail able MCSs BPSK [1/5, 1/3]

QPSK [1/4, 1/3, 1/2, 2/3, 3/4]16QAM [2/3, 3/4, 5/6]

User arrival Poisson processUser arrival rate [4, 5, 6, 7, 8, 10] users/cell/s

Traffic model Finite buffer model, 1 Mbit buffer sizeGBR requirement 256 kbpsNumber of admitted-calls simulated 10000

is the pathloss compensation factor, Li is the pathloss in dB,

MCS is signaled by the radio resource control and i is auser specific correction value depending on f().

Assuming user is operating within the transmit power dy-

namic range, the PSD (i) in mWatts per PRB of user i using

FPC formula in (6) is,

10log10 i = P0 + Li + MCS + f(i) (7)

In this paper MCS and f(i) terms are set to zero.Limiting the transmit power of by i in (7) modifies the SINR

definition in (5) as,

SINRi

=i

BWPRB.P Li.IoT.N0.N F(8)

which gives the closed form solution for Ni using (3), (4),

and (8) as,

Ni =GBRi

BWPRB.BWeff.. log2

1 +SINRi

SNReff (9)

Ni and Nnew in (2) are calculated using (9) to make the

AC decision. Additionally, if new.Nnew > Pmax the new

user is denied admission since it is power limited. Since, in

Macro case 1 scenario few users will be power limited, and

hence the blocked users due to the transmit power constraint

will be negligible. Admission denials due to the transmit

power constraint will be higher in the Macro case 3 scenario.

The proposed AC algorithm in (2) is referred to as the

FPC based AC.

III. SIMULATION METHODOLOGY

The performance evaluation is done using a detailed multi-

cell system level simulator which follows the guidelines in [7].

The users are created in the system according to a Poisson

call arrival process. If the AC decision criterion proposed in

Section II is fulfilled the user is admitted otherwise the user

is blocked. A finite buffer best-effort traffic model is used,

where each user uploads a 1 Mbit packet call. All the users in

the network are assumed to have the same GBR requirements.

The session is terminated as soon as the upload is completed.

The channel model includes distance dependent pathloss,

log-normal shadowing and frequency selective fast fading. 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 profile for user speed of 3 kmph.

The packet scheduling is done as a two step algorithm,

first time-domain (TD) scheduling is used to select the users

which will then be multiplexed using frequency-domain (FD)

scheduling [11]. In this paper a GBR-aware packet scheduler

is used in TD, which prioritizes the users which are far from

their GBR requirement based on the metric in (10). Ri is the

past average throughput of user i calculated using exponential

average filtering [12].

MTD,i =

GBRiRi

GBRi < Ri

1.0 GBRi Ri(10)

The FD packet scheduler allocates the fixed number of

PRBs to the users selected by the TD scheduler according

to proportional fair metric as [11],

MFD,i[k] =

di[k]

Rsch,i , (11)

where di[k] is the estimated achievable throughput for user ion PRB k, and Rsch,i is an estimate of user throughput if user

i was scheduled every TTI [12]. The allocated bandwidth per

user is assumed to be fixed and the same for all the scheduled

users. In this paper 8 users are multiplexed per TTI, giving 6

PRBs per user per TTI. The total number of PRBs used for

data transmission is 48 PRBs while 2 PRBs are reserved for

control transmission.

The link adaptation selects the most suitable MCS based

on SINR estimations over the allocated bandwidth. Such

estimations are obtained from the sounding reference signals

transmitted by the user and used at the eNB to extract near-instantaneous frequency selective channel state information

(CSI). It is assumed that the CSI is available at the eNB

every TTI over the entire system bandwidth for all the active

users with a given bandwidth resolution. The main simulation

parameters listed in Table I are according to the assumptions

in [7].

IV. RESULTS

The performance of the proposed AC algorithms is eval-

uated using the blocking and outage probabilities. Blocking

2489


4/5

4 5 6 7 8 9 100

0.1

0.2

0.3

0.4(a)

Offered traffic (User arrival rate) [users/cell/s]

Bloc

kingpro

ba

bility Ref AC - 2 Mbps

Ref AC - 5 Mbps

Ref AC - 25 Mbps

FPC AC

4 5 6 7 8 9 100

0.1

0.2

0.3

0.4(b)


Ou

tagepro

ba

bility

Ref AC - 2 Mbps

Ref AC - 5 Mbps

Ref AC - 25 Mbps

FPC AC

Fig. 4. (a) Blocking probability vs. offered traffic, (b) outage probability vs. of-fered traffic for FPC based AC and reference AC algorithms. GBR = 256 kbps.

probability (Pb) is defined as the ratio of the number of

blocked users to the total number of new users requesting

admission. Outage probability (Po) is calculated as the ratio

of the number of users not fulfilling their GBR requirement,

to the total number of admitted users. The unsatisfied user

probability (Pu) is calculated as,

Pu = 1 (1 Pb)(1 Po) (12)

In Fig. 4(a) we notice that the blocking probability increases

with the increasing user arrival rate (offered traffic) for dif-

ferent AC algorithms. Even at very low offered traffic FPCbased AC denies admission to the users whose QoS cannot

be fulfilled. Furthermore, the reference AC makes admission

decision without taking into account the average channel

condition of users requesting admission. The Rmax parameter

for the reference AC is set as [2, 5, 25] Mbps. For the reference

0 2 4 6 8 10 12 14 16 18 200

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Average Call Length [s]

CDF

Ref AC - 2 Mbps

Ref AC - 5 Mbps

Ref AC - 25 Mbps

FPC AC

Fig. 5. CDF of average call length for FPC based AC and reference ACalgorithms. GBR = 256 kbps, arrival rate = 8 users/cell/s.

4 5 6 7 8 9 1 0

0

0.1

0.2

0.3

0.4(a)


Unsa

tis

fie

duser

pro

ba

bility

Ref AC - 2 Mbps

Ref AC - 5 Mbps

Ref AC - 25 Mbps

FPC AC

4 5 6 7 8 9 1 03

4

5

6

7

8(b)

Offered traffic (User arrival rate) [users/cell/s]Averagece

llth

roug

hpu

t[M

bps

]

Ref AC - 2 Mbps

Ref AC - 5 Mbps

Ref AC - 25 Mbps

FPC AC

29%

Fig. 6. (a) Blocking plus outage probability vs. offered traffic, (b) average cellthroughput vs. offered traffic for FPC based AC and reference AC algorithms.GBR = 256 kbps.

4 5 6 7 8 9 100

10

20

30

Offered traffic (User arrival rate) [users/cell/s]Averagenum

ber

of

users

per

ce

ll

Ref AC - 2 Mbps

Ref AC - 5 Mbps

Ref AC - 25 Mbps

FPC AC

Fig. 7. Average number of users per cell vs. offered traffic for FPC based ACand reference AC algorithms. GBR = 256 kbps.

AC algorithm, the blocking probability decreases while the

outage probability increases for the increasing value of Rmaxas shown in Fig. 4(b). The FPC based AC is shown to be

better in terms of the blocking probability while maintaining

the outage probability within 1.5%.

For the increasing Rmax, reference AC block lesser users at

low and moderate offered traffic but their blocking probability

tends to converges for high offered traffic as seen in Fig. 4(a).

This is because for increasing Rmax the users tend to stay in

the system for longer time as shown in Fig. 5. Moreover, at

higher offered traffic the blocking probability is lower for the

FPC based AC, because the users are admitted such that their

average call length is limited by certain maximum, to fulfillthe GBR requirement.

Fig. 6 shows the unsatisfied user probability and average

cell throughput (carried traffic) for different AC algorithms.

We notice that the proposed FPC based AC is the best among

the studied AC algorithms in terms of low unsatisfied user

probability and high carried traffic. At 10% of the unsatisfied

user probability, the FPC based AC can support around 29%

more carried traffic over the reference AC - 2 Mbps as shown

in Fig. 6. Similarly, at 10% of the unsatisfied user probability

the carried traffic gain for using the FPC based AC is around

2490


5/5

-140 -120 -100 -80 -60 -400

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Pathgain [dB]

CDF

Ref AC - 2 Mbps

Ref AC - 5 Mbps

Ref AC - 25 Mbps

FPC AC

Fig. 8. CDF of pathgain of admitted users for FPC based AC and referenceAC algorithms. GBR = 256 kbps, arrival rate = 8 users/cell/s.

4 5 6 7 8 9 100

200

400

600

800

1000

1200

Offered traffic (User arrival) [users/cell/s]

95%

coverageuser

throughpu

t[k

bps

]Ref AC - 2 Mbps

Ref AC - 5 Mbps

Ref AC - 25 Mbps

FPC AC

GBR

Fig. 9. 95% coverage user throughput vs. offered traffic for FPC based ACand reference AC algorithms. GBR = 256 kbps.

5.5% and 13% over the reference AC - 5 Mbps and referenceAC - 25 Mbps respectively. It is important to note that the

Rmax for the reference AC is a tunable parameter and is not

the actual average uplink cell throughput.

For reference AC - 25 Mbps the number of users grow

rapidly for arrival rates 7 users/cell/s and higher as shown in

Fig. 7. This is because for very high Rmax the reference AC

tend to behave like system with no AC.

It is shown in Fig. 8 that for the FPC based AC the pathgain

distribution of users in the cell is modified for the cell edge

(low pathgain) users. This is because the FPC based AC

denies admission to a user with a certain GBR requirement

at the cell edge with higher probability compared to the userlocated at the cell center. Moreover, we notice that the pathgain

distribution of the users is same for the reference AC algorithm

because this algorithm treat all the users requesting admission

equally regardless of their channel conditions.

Fig. 9 compares the 95% coverage user throughput for

different AC algorithms. We notice that the FPC based AC

and reference AC - 2 Mbps are the only evaluated algorithms

for which the 95% coverage user throughput is always higher

than the GBR. But at 10% of the unsatisfied user probability

the carried traffic gain of using the FPC based AC over the

reference AC - 2 Mbps is around 29%.

V. CONCLUSION

In this paper, an AC algorithm for UTRAN LTE uplink,

utilizing FPC formula agreed in 3GPP, is proposed along with

a reference AC algorithm. The FPC based AC determines if

a user requesting admission can be accepted based on the

pathgain so as to fulfill the QoS of the new and existing users.The results show that at 10% unsatisfied user probability the

carried traffic gain of the FPC based AC can be up to 29% over

the reference AC. The FPC based AC is based on a closed form

solution and hence the complexity of the proposed AC and the

reference AC algorithms is of the same order. The FPC based

AC algorithm is robust and tunes itself inherently to the load

conditions, unlike Rmax tuning in reference AC. Hence, the

FPC based AC is a good QoS-aware AC algorithm for LTE

uplink. Similar AC design approach could also be used for

other NextG standards which uses orthogonal multiple access

techniques and FPC in uplink.

Further work is needed to study the QoS differentiation

for mixed GBR and best effort users case. Additionally,the combined offered and carried traffic gain in using FPC

based AC for realistic traffic model e.g., streaming, need to

be investigated. Moreover, in real situation the closed loop

adjustments of FPC should be taken into account.

ACKNOWLEDGMENT

The authors would like to thank Basuki Endah Priyanto for

providing the link-level results used in Fig. 3.

REFERENCES

[1] A. Toskala, H. Holma, K. Pajukoski, and E. Tiirola, UTRAN LongTerm Evolution in 3GPP, IEEE 17th International Symposium onPersonal, Indoor and Mobile Radio Communications, 2006.

[2] K.I. Pedersen, P.E. Mogensen, and T.E. Kolding, Overview of QoSOptions for HSDPA, IEEE Communications Magazine, July 2006.[3] 3GPP TS 36.300 V8.3.0, Evolved Universal Terrestrial Radio Ac-

cess (E-UTRA) and Evolved Universal Terrestrial Radio Access (E-UTRAN); Overall description; Stage 2, January 2008.

[4] H. Holma and J. Laakso, Uplink Admission Control and Soft Capacitywith MUD in CDMA, IEEE 50th Vehicular Technology Conference,September 1999.

[5] S.S. Jeong, J.A. Han, and W.S. Jeon, Adaptive Connection AdmissionControl Scheme for High Data Rate Mobile Networks, IEEE 62nd

Vehicular Technology Conference, September 2005.[6] 3GPP R1-073224, Way Forward on Power Control of PUSCH, TSG-

RAN WG1 49-bis, June 2007.[7] 3GPP TR 25.814 V7.1.0, Physical Layer Aspect for Evolved Universal

Terrestrial Radio Access (UTRA), October 2006.[8] P.E. Mogensen et.al., LTE Capacity Compared to the Shannon

Bound, IEEE 65th Vehicular Technology Conference, April 2007.

[9] B.E. Priyanto, H. Codina, S. Rene, T.B. Sorensen, and P.E. Mogensen,Initial Performance Evaluation of DFT-Spread OFDM based SC-FDMA for UTRA LTE Uplink, IEEE 65th Vehicular TechnologyConference, April 2007.

[10] C.U. Castellanos et.al., Performance of Uplink Fractional Power Con-trol in UTRAN LTE, IEEE 67th Vehicular Technology Conference,May 2008.

[11] G. Monghal, K.I. Pedersen, I.Z. Kovacs, and P.E. Mogensen, QoSOriented Time and Frequency Domain Packet Schedulers for theUTRAN Long Term Evolution, IEEE 67th Vehicular TechnologyConference, May 2008.

[12] T.E. Kolding, QoS-Aware Proportional Fair Packet Scheduling withRequired Activity Detection, IEEE 64th Vehicular Technology Con-

ference, September 2006.

2491

Documents

Qos Based Cac Nsn Paper