8/8/2019 icpwc02

1/5

Performance of Code Allocation Algorithms on UMTS Uplink with Mixed Voice/Data Trafc

Balakrishna J. Prabhu, A. ChockalingamDepartment of ECE

Indian Institute of Science, Bangalore 560012e-mail: achockal@.ece.iisc.ernet.in

D. JayaramCommunications and Embedded Systems Group

L & T Infotech Ltd., Bangalore 560001e-mail: Jayaram.D@lntinfotech.com

ABSTRACT

In this paper, we are concerned with the performance evalua-tion of code allocation algorithms in Universal Mobile Telecom-munication System (UMTS) networks. UMTS networks willoffer multiple services (voice, data, video, etc.) with differentquality-of-service (QoS) requirements to mobile users. In thispaper, we evaluate the performance of different code (rate) allo-cation algorithms on the UMTS uplink in a mixed voice/data traf-c scenario. Two different code allocation algorithms are consid-ered: one based on the overall buffer occupancy at the user termi-nal, and the other based on dividing the available codes equallyamong the requesting users. Further, for a data-only system, weevaluate the performance of two algorithms for rate and powerallocation based on received signal-to-interference ratio (SIR) atthe base station.

I. INTRODUCTION

UMTS networks will offer multiple services (voice, data,video, etc.) with different quality-of-service (QoS) requirementsto mobile users [1]-[3]. UMTS networks support high speed ra-dio access (up to 2 Mbps) based on Wideband Code DivisionMultiple Access (WCDMA). The users, during call establish-ment, request the network the desired QoS for the connectionin terms of data rate, delay, priority, reliability, etc. The network allocates the available resources (transmission rate and transmitpower) to different users based on certain resource allocation pol-icy. Rate allocation to each user can be varied either by allocat-ing multiple spreading codes with constant spreading factor or byvarying the spreading factor. In WCDMA, the physical channelsfollow a layered structure of radio frames and time slots . Eachradio frame is of 10ms duration and consists of 15 slots. Themaximum number of bits sent per radio frame depends on therate allocated to the user. The network can dynamically vary therate allocation to different users on a frame-by-frame basis.

The radio interface protocol layers which are involved in theradio resource allocation are shown in Fig. 1 [4]. The radio re-source control (RRC) in Layer 3 (L3) is responsible for the sig-naling and control information exchange between the user and

network to effect allocation and deallocation of radio resources[5]. The data transport services offered by the physical layer(Layer 1 L1) [6] is achieved through the use of transport chan-nels via the media access control (MAC) [7] sublayer in Layer2 (L2). Each user terminal may simultaneously have multipletransport channels multiplexed on to one or more physical chan-

This work was supported in part by L & T Infotech Ltd., Bangalore 560001under project No. SID/PC14032.

Medium Access Control (MAC)

Physical Layer

Radio Link Control (RLC)

RRC

Control Plane

L3

L2

L1

Transport channels

Logical Channels

User Plane

Fig. 1. Radio interface protocol architecture

nels [8]. For example, a 64 Kbps packet switched (PS) data trans-port channel and a 12.2 Kbps circuit switched (CS) voice trans-port channel can be multiplexed on to a 128 Kbps physical chan-nel. Transport channels are dened by how and with what fea-tures (e.g., transport block size, transmission time interval, etc.,which are dened in the following paragraph) data is transferredover the air interface.

Refer Fig. 2 which illustrates the following denitions. Thebasic unit of data exchange between L1 and MAC for L1 pro-

cessing is called a transport block (TB). The transport block sizeis the number of bits in a TB. A set of TBs exchanged betweenL1 and MAC during the same frame using the same transportchannel is called a transport block set (TBS). The TBS size is thenumber of bits in the TBS. The TB size and the TBS size are cho-sen by the MAC at the user terminal. The periodicity at which aTBS is transferred by L1 on to the air interface is called transmis-sion time interval (TTI). The allowed values of TTI are 10, 20 40,and 80 ms. The MAC delivers one TBS to L1 every TTI. The for-mat (TB size, TBS size, type/rate of coding, size of CRC) offeredby L1 to MAC (and vice versa) for the delivery of TBS during aTTI on a given transport channel is called transport format (TF).The transport format determines the transport channel bit rate be-fore L1 processing. For example, let TB size = 336 bits (320 bitspayload + 16 bits header), TBS size = 2 TBs per TTI, and TTI =10 ms. Then the transport channel bit rate (with header) is givenby 336*2/10 = 67.2 Kbps. The user bit rate (without header) isgiven by 320*2/10 = 64 Kbps. Thus, it can be seen that vari-able bit rate transmission on a single transport channel can beachieved by changing (from one TTI to the other) either the TBSsize only, or both the TB size and the TBS size. A set of transport

8/8/2019 icpwc02

2/5

formats associated with a transport channel is called a transport format set (TFS). A combination of TFs on different transportchannels in a given TTI is called transport format combination(TFC). A set of transport format combinations allowed by thenetwork is called the transport format combination set (TFCS).The network informs the TFCS to the user terminal to be usedon the uplink transmission by the user terminal. The MAC atthe user terminal then chooses between the different TFCs spec-ied in the TFCS. The user terminal MAC can (based on certaincriteria) choose different TFCs for different TTIs.

Transport Block

Transmission Time Interval

TB

TBTB

TB

TBTB TB

TB

(TTI)

TransportChannel 2

TransportChannel 1

Transport Format Set(TFS)

Transport Format Combination (TFC)

TTI

Transport Format(TF)

Transport Format Combination Set (TFCS)

Set (TBS)Transport Block

Fig. 2. Transport channel de nitions

In this paper, we are interested in evaluating the performanceof different code (rate) allocation algorithms on the UMTS up-link in a mixed voice/data trafc scenario. Two different codeallocation algorithms are considered: one based on the overallbuffer occupancy at the user terminal, and the other based on di-viding the available codes equally among the requesting users. Inthe buffer occupancy based algorithm, the rate allocation is doneis such a way that the user with a larger buffer occupancy will beassigned a higher rate. We also evaluate the performance of SIRbased rateand powerallocation algorithms in a data-only system.

II. SYSTEM MODEL

We consider a mixed voice/data system in which there are

mobile user terminals. Each user can generate a CS voice calland/or a PS data call. That is, at any given time, a terminal canhave a) neither a voice nor a data call, b) a voice call and no datacall, c) a data call and no voice call, and d) simultaneously botha voice as well as a data call. Voice call arrivals are assumedto follow a Poisson distribution, i.e., the voice call inter-arrivaltime is exponentially distributed with mean

. The voice callholding time is also assumed to be exponential with mean call

holding time

. Since speech consists of alternating active andsilence periods, we model the voice source as a ON/OFF sourcewhere the ON and OFF periods are assumed to be exponentiallydistributed with means and , respectively. During a PSdatacall, packets arrive in bursts wherethe burst inter-arrival timeis assumed to be exponentially distributed with mean

, andthe number of transport blocks (TB) per burst is assumed to begeometrically distributed with mean .

A. Code Allocation AlgorithmsThe network allocates channel rates to users by assigning dif-

ferent number of spreading codes with different spreading fac-tors. The spreading factor can be in the range 4 to 256 on theUMTS uplink. In other words, with a 3.84 Mcps chip rate, theminimum channel rate can be !# " $& %( '0 )2 14 3 57 6& 8& 9A @B )2 8 Kbps. Thischannel rate will typically be larger than the user information ratebecause of overhead bits due to coding, CRC, etc.

In our model, we consider that the network allocates channelrates to users in integer multiples of CA D , where CE D is the

minimum channel rate that can be assigned to a user. Here, wetake C D F @G 9& 1 Kbps which corresponds to a spreading fac-tor of 64. This 60 Kbps channel rate can carry a 12.2 Kbps CSvoice call or 16 Kbps PS data trafc, both with their associatedoverhead bits due to coding, CRC, etc [9]. The values of theavailable channel rates then are 60, 120, 180, and 240 Kbps. Weassume that

I H

spreading codes each corresponding to rate CP D

are available with the network for allocation. The network canchange the allocation of spreading codes to different users ev-ery TTI. We assume that the network gives priority to voice callswhile allocating spreading codes, i.e., available codes are rst al-lotted to voice calls and the remaining codes are allotted to datacalls. The algorithms presented in the following are for assigningcodes for the data calls. A code allotted to a voice call is heldfor the entire call duration. In user terminals with simultaneousvoice and data calls, silence periods in voice can be used to senddata packets. For terminals with only data calls, codes can beallocated/deallocated every TTI.

A.1 Algorithm I

In this algorithm, the network allots codes to users based onthe data buffer occupancy at the individual user terminal. Thenetwork allocates codes in such a way that the user with a largerbuffer occupancy will be assigned a higher rate.

Let

be the number of users in the system. Let Q H

be the total

number of codes of rate corresponding toCR D

available in thesystem. Let , )T SU V S W H

, be the maximum number of codesof rate corresponding to CW D that can be allotted to a given user.It is assumed that the network has the knowledge of buffer occu-pancy in all the participating user terminals. The network rstassigns codes to newly arrived voice calls. The network then ar-ranges the users in descending order of data buffer occupancy.Let

be the number of transport blocks in the buffer of the X7 Y

user in the ordered list. Let ` be the number of codes allotted tothe Xa Y

user in this list. Note that ` can be either 1 or 0 depend-ing on whether the X Y

user has a voice call or not. The numberof the remaining codes available for allocation to data calls (in-cluding data calls in user terminals with simultaneous voice and

data), bd c , is then given by

b cT @

I Hf eh g

i

pd q

`

" (1)

The algorithm performs code allocation to the largest bufferoccupancy user rst, the second largest buffer occupancy usernext, and so on. This process is continued until the least bufferoccupancy user gets the code allocation or the available codes are

8/8/2019 icpwc02

3/5

8/8/2019 icpwc02

4/5

0

50

100

150

200

250

300

0 0.5 1 1.5 2 2.5 3 3.5 4

A

v e r a g e

d a

t a b u r s

t d e

l a y

( f r a m e s

)

Data burst arrival rate (bursts/sec/node)

T_i=120s. Algo IT_i=120s. Algo IIT_i=180s. Algo I

T_i=180s. Algo IIT_i=240s. Algo I

T_i=240s. Algo IIT_i=360s. Algo I

T_i=360s. Algo IIT_i=600s. Algo I

T_i=600s. Algo IIT_i=1500s. Algo I

T_i=1500s. Algo II

Fig. 3. Average data burst delay vs. data burst arrival rate.G

.G

.

occupancy user the chances of picking the combination TFC10gets increased compared to the equal share allotment AlgorithmII, which results in more number of bits sent per code allotted.This effect will be more as the burst arrival rate gets increased.Hence, the mean delay performance improvement of Algorithm

I over Algorithm II is more visible at higher arrival rates. Also,as the voice call arrival rate increases (e.g.,

@ )2 6& 1 secs), thedelay performance of data bursts degrades as the voice calls getpriority in code allocation.

IV. SIR BASED ALGORITHMS

In the previous section, the code allocation algorithms did notconsider the received SIR at base station. Also, power alloca-tion to the user terminals was not considered while allocating thetransmission rate. In this section, we evaluate the performance of two rate and power allocation algorithms based on received SIRat the base station for a data-only system. In the rst algorithm,the user terminal is allocated a transmission rate as well as trans-mit power by the base station. In the second algorithm, the basestation allocates only the transmission rate and the user terminaltransmits at the maximum allowable transmit power, b D . Inboth the algorithms allocations are made in such a way that a cer-tain BER performance criterion is met at the base station for eachuser. The received SIR at the base station for mobile X is given by

C

@

b

E

g

s

p

c!

# "p

b

$ % ' &) (

(5)

where b is the transmit power of mobile X , ( is the channel gainfor mobile X , and ( is the thermal noise power. The channel gain,

E

, is given byE

@1 0

3 2

5 4

) 1

7 6

t9 8

q

c

(6)

where 0 is the distance of mobile X from the base station, ( isthe path loss exponent, and @ is the normally distributed shadowfading random variable for mobile X . The received AC B 5

formobile X is then given by,

AB

@ C

$ D

C

(7)

where C is the rate of the code assigned to user X . In order tomeet a given BER, the received A B 5

should be greater than agiven threshold value, E

Y

.

A. SIR based algorithm with power control

The user terminals with backlogged data packets are orderedaccording to their received SIR at the base station. The code allo-cation is then done in descending order of the received SIR, i.e.,the codes are allotted to the user with the highest SIR rst, and soon. The SIR estimates of user terminals which transmitted pack-ets in the previous frame can be measured on the trafc channeland made available at the base station. The SIR estimates of theuser terminals with new message arrivals can be measured fromthe resource request transmission made by them on the randomaccess channel. The resource request transmission for new mes-sage arrivals is done using the maximum permissible transmitpower, b D .

Once the rate allocation is done, the base station calculatesthe minimum overall transmit power and the individual transmitpower for each mobile in order to meet the given AF B 5

thresh-old value, E

Y

. The objective of the base station is to nd thepower allocations to each mobile so that

R X

i

9 GI H

b

(8)

subject tob

E

Y

4

C

D

4

)

E

4

i

P GI H # "p

b

&) ( R Q

XT S

(9)

b

S bD

" (10)

where

is the set of user terminals which have been allottedrates in this iteration. In order to minimize

s

b

, we use theequality constraint in Eqn. (9) to get

b

@U E

Y

4

C

D

4

)

E

4

i

P GI H # "p

b

$ % ' &) (i V Q

XT S

(11)

In matrix notation, WC X` Y@

(7 aX (12)where G is given by

bc

c

c

d

E

wf e

g

t

w

q

hp i

e

E

g

4$ 4! 4

e

E r q

e

w

g

e

g

t

g

q

hi

4$ 4! 4

e

q

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .e

w

e

g

4$ 4! 4

r q

e

g

t

q

q

hi

su t

t

t

v

and P = w b w b g 4$ 4! 4 b x q y . is the number of user termi-nals which have been a code in this iteration.

If the power allocation to any user terminal exceeds the max-imum permissible transmit power level, b D , its rate of trans-mission is reduced and the power vector is again calculated. Thisprocess is continued till a feasible power vector is found.

B. SIR based algorithm without power control

In this case, the user terminals are allowed to transmit with themaximum power level, b D , and no power optimization is done

8/8/2019 icpwc02

5/5

at the base station. The requesting user terminals are arrangedin descending order of received SIR. Maximum possible rate isallotted to the user terminal with the best received SIR. Rate isthen assigned to the user terminal with the next best SIR, andso on till either the codes are exhausted or all the user terminalshave been assigned codes. The base station then calculates thereceived A B 5

for each user terminal using this rate allocationand transmit power b D

P

. If the received A B 5

for any userterminal is less than E

Y

, the rate allotted to that user terminal isreduced till the received A B 5

exceeds EY

.

Cell radius 1 km 4 dB

B 20 dBmG

12G

@ 16

T 9 dBTABLE V

S IMULATION PARAMETERS FOR SIR BASED ALGORITHMS

We evaluate the performance of the above SIR based rate andpower algorithms through simulations. The transport format setavailable to the mobile is given in Table II. The shadow fadingexperienced by each node is assumed to be Gaussian with mean

1 and variance and is correlated with correlation of 1 " $ . The

simulation parameters are given in Table V. The path loss modelis taken to be0

0 2 @ )2 6& $ " )

&

! " 9

4

!

q

c

0

C 2

(13)

where R is the distance of the user terminal from the base stationin km [10].

10

15

20

25

30

35

40

0 0.5 1 1.5 2 2.5 3 3.5

A v e r a g e

d a t a

b u r s

t d e

l a y

( f r a m e s

)

Data burst arrival rate (messages/sec/node)

With power controlWithout power control

Fig. 4. Average data burst delay vs. data burst arrival rate.G

.G

.

C. Results and Discussion

Fig. 4 shows the average data burst delay versus the data burstarrival rate for the SIR based algorithms. It is observed that theSIR based algorithm with power control performs better in termsof average data burst delay for a given arrival rate. The interfer-ence from the transmitting nodes is lesser when power controlis used than when there is no power control. This allows morenodes to transmit at the same time resulting in better resource(code) utilization and hence lower delays. Fig. 5 shows the aver-agedata burstdelay for individual nodes as a function of their dis-tance from the base station for a particular simulation scenario.

It is observed that the algorithm with power control ensures thatnodes which are far away see the same delay as the nodes whichare closer to the base station. In the no power control case, all thenodes transmit at b D

P

resulting in a high interference from thenearby nodes. This causes the far off nodes to wait till the endof the transmission from closer nodes before being able to sendtheir data bursts. Hence, the far off nodes experience a higherdelay than the nearby nodes.

10

15

20

25

30

35

40

45

50

100 200 300 400 500 600 700 800 900 1000

A v e r a g e

d a

t a b u r s

t d e

l a y

( f r a m e s

)

Distance (m)

With power controlWithout power control

Fig. 5. Average data burst delay vs. distance.G

.G

. Data burstarrival rate = 1 message/sec/node.

V. CONCLUSION

In this paper, we evaluated the performance on two code al-location algorithms for mixed voice/data trafc on the UMTSuplink. Two different code allocation algorithms were consid-ered: one based on buffer occupancy at the user terminal, and theother based on dividing the available codes equally among therequesting users. We showed that the allocation based on bufferoccupancy performs marginally better than the equal share algo-rithm. For a data-only system, we evaluated the performance of SIR based algorithms with and without power control. It wasshown that the system with power control performs better thanthe system without power control.

REFERENCES

[1] H. Holma and A. Toskala, WCDMA for UMTS: Radio Access for "$ #@ Gen-

eration Mobile Communications , John Wiley & Sons, 2000.[2] H. Kaaranen et al , UMTS Networks, Architecture, Mobility and Services

John Wiley & Sons, Chichester, 2002.[3] J. Laiho, A. Wacker, and T. Novasad, Radio Network Planning and Opti-

misation for UMTS , John Wiley & Sons, Chichester, 2002.[4] J. P. Castro, The UMTS Network and Radio Access Technology, Air Inter-

face Techniques for Future Mobile Systems , John Wiley & Sons, Chich-ester, 2001.

[5] 3GPP TS 25.331, Radio Resource Control (RRC) Protocol Speci ca-tion, version 4.3.0, Release 4, 2001.

[6] 3GPP TS 25.201, Physical Layer - General Description, version 4.1.0,Release 4, 2001.

[7] 3GPP TS 25.321, Medium Access Control (MAC) Protocol Speci ca-tion, version 4.3.0, Release 4, 2001.

[8] 3GPP TS 25.211, Physical Channels and Mapping of Transport Channelsonto Physical Channels (FDD), version 4.3.0, Release 4, 2001.

[9] 3GPP TS 34.108, Common Test Environments for User Equipment (UE)Conformance Testing, version 4.1.0, Release 1999.

[10] 3GPP TR 101.112 (UMTS 30.03): Selection Procedures for the Choiceof Radio Transmission Technologies of the UMTS, version 3.2.0, Release1998.