Sunteți pe pagina 1din 5

A Multi-user MIMO Resource Scheduling Scheme for Carrier Aggregation Scenario

Na Lei, Caili Guo, Chunyan Feng, Yu Chen


School of Information and Communication Engineering Beijing University of Posts and Telecommunications Beijing , China E-mail: lnjessie@gmail.com, guocaili@gmail.com, cyfeng@bupt.edu.cn, 815012912@qq.com algorithm[35] , without taking care of the PCC, one or more UEs scheduled on a RB may not be transmitted, resulting an invalid multi-user MIMO user group. Moreover, in non adjacent inter band aggregation scenario, where the aggregated carriers belong to different frequency bands, the fading characteristics are different between carriers, such as the path loss and Doppler shift[6] . This can result in spectrum heterogeneity that can be used as frequency selective diversity, which the resource scheduling scheme in one component carrier system can not exploit to optimize the system performance. In this paper, a multi-user MIMO Resource Scheduling Scheme for Carrier Aggregation scenario (RSM-CA) is proposed. Compared to PF, RSM-CA guarantees the resource on the PCC of a UE is primarily allocated to the UE, making sure all user groups are valid. In order to use frequency selective diversity, different users have different grouping users on different carriers aiming to maximize the system throughput. The rest of this paper is organized as follows. Section 2 gives the system model and background of the proposed scheduling scheme. Section 3 elaborates the multi-user MIMO resource scheduling scheme for carrier aggregation. Section 4 presents the performance of the proposed scheme, including system level simulation assumptions and results. Section 5 concludes this paper with a summary of results. II. SYSTEM MODEL AND BACKGROUND A typical cell structure in LTE-A system with an eNB (evolved Node B) and several UEs is shown in Fig.1. For carrier aggregation scenario, the aggregated carriers are divided into several categories, these categories includes eNB CC, which is a cells total number of component carriers that can be allocated to its belonging UEs; UE CC, which is a subset of the eNB CC that UE selects according to its own needs and aggregation ability[2] ; Active CC, which is the UE CC UE actually used in one scheduling process; Deactive CC, corresponding to active CC, the UE CC that is not used by UE in one scheduling process is called deactive CC; PCC, which is the only one carrier of UE CC that is used to transmit the control signaling, and its predened by the system, so during one scheduling process this carrier can not be deactived; SCC, which is the rest of the UE CC except the PCC. In the case of Fig.1, there are ve eNB CC: C1,C2,C3,C4,C5; The UE CC of UE1 is: C1,C2,C3, the UE CC of UE2 is: C1,C3,C4,C5;

AbstractThis paper focuses on the multi-user MIMO resource scheduling of LTE-Advanced system with carrier aggregation. With the special function of a UEs primary component carrier (PCC), the original proportional fair (PF) scheduling scheme can result in invalid user groups for multi-user MIMO transmission, making UEs of the user group can not be transmitted. A multi-user MIMO resource scheduling scheme for carrier aggregation scenario (RSM-CA) is proposed. Compared to PF, RSM-CA guarantees the resource on the PCC of a UE is primarily allocated to the UE, making sure all user groups are valid. Furthermore, frequency selective diversity, which is specic for carrier aggregation scenario, is exploit in RSM-CA to maximize the system throughput. According to the system level simulations of downlink LTE-A, RSM-CA can achieve better system throughput than the original PF scheduling scheme. Keywords-LTE-Advanced, MIMO, proportional fair carrier aggregation, multi-user

I. I NTRODUCTION In order to support wider transmission bandwidths e.g. up to 100MHz, the LTE-Advanced system introduces the carrier aggregation technology, where two or more component carriers belonging to a single frequency band or different frequency bands can be aggregated[1] . With the carrier aggregation technology, it will be possible to schedule a user (UE) on multiple component carriers simultaneously, but a UE can not use all the component carriers in its belonging cell, a certain UE uses only a certain set of the component carriers in one cell according to their own aggregation abilities[2] . Therefore, different UEs have different sets of aggregated carriers, and the control signaling is transmitted on one of the aggregated carriers, this special carrier is called Primary Component Carrier (PCC), which can not be deactivated, that is to say, if a UE doesnt get a Resource Block (RB) from its PCC during the resource scheduling procedure, it can not be transmitted. Thus, some new problems should be considered in resource scheduling procedure. The resource scheduling problem in carrier aggregation scenario is especially severe for multi-user MIMO transmission, which can allow more than one UE to use a RB. While using the classic scheduling scheme, such as proportional fair (PF)
This work is supported by Chinese National Nature Science Foundation (60902047) and the Fundamental Research Funds for the Central Universities (2011RC0113).

978-1-4577-1010-0/11/$26.00 2011 IEEE

eNB CC
C1 C2 C3 C4 C5

UE
1 Transmission request Feedback eNB CC 3 Feedback UE CC & CQI
Feedback the result of resource scheduling

eNB

eNB UE1 UE CC C1,C2,C3 Active CC C1,C3 Deactive CC C2 PCC C3 SCC C1,C2 UE2 UE CC C1,C3,C4,C5 Active CC C1,C3,C5 Deactive CC C4 PCC C5 SCC C1,C3,C4

2 Resource scheduling scheme 4

UE1

UE2

Transmit/receive data

C1 C2 C3

C1 C3 C4 C5

Fig. 2.

The Process of Resource Scheduling Procedure

A user group on C1:UE1(SU)+UE2(GU)


Fig. 1. A Typical Cell Structure in LTE-A system

The active CC of UE1 is: C1,C3, the active CC of UE2 is: C1,C3,C5; The deactive CC of UE1 is: C2, the deactive CC of UE2 is: C4; The PCC of UE1 is: C3, the PCC of UE2 is: C5; The SCC of UE1 is: C1,C2, the SCC of UE2 is: C1,C3,C4. As for the multi-user MIMO transmission mode, more than one UE use the same RB to form a user group. In the rest of this paper, user group of two UEs is considered, but the proposed scheme is not limited by the number of the user group. In one user group, the rst scheduled UE is called scheduling user (SU), the UE which is selected to form the user group with SU is called grouping user (GU). In the case of Fig.1, UE1 and UE2 form a user group on one of the RBs of C1, and the SU is UE1, GU is UE2. In LTE/LTE-A system, before every downlink time slot, eNB conducts resource scheduling procedure to allocate resource to its serving UEs[1] , the user groups of multi-user MIMO are also decided during this procedure. Fig.2 shows the process of resource scheduling procedure, which includes: Step 1: UE has transmission needs and informs eNB. Step 2: eNB feedbacks eNB CC. Step 3: UE selects UE CC according to its own needs and aggregation ability[2] and feedbacks its UE CC and CQI (channel quality indicator) to eNB. Step 4: eNB conducts resource scheduling scheme to allocate resource to UEs. Step 5: UE uses the allocated resource to transmit/receive data. In this paper, we focused on the forth step, and proposed a new resource scheduling scheme according to the special needs of carrier aggregation scenario. As the PCC of a transmitted UE can not be deactived, the original PF scheduling scheme can not be used anymore. Taking the system of Fig.1 for example, with the PF scheduling scheme, all the RBs take turns to be allocated to UE which has the highest priority and then nd a matched UE which can give the largest packet size when they are transmitted together on one RB. If UE1 and UE2 are scheduled as a user group on C1, UE1 gets a RB on its PCC C3, but UE2 doesnt get a RB on C5, which is

its PCC. In this case, UE2 can not transmit/receive, so the user group of UE1 and UE2 is invalid, making both UE1 and UE2 can not be transmitted. In order to make all the user groups valid and exploit frequency diversity of carrier aggregation, we proposed a new resource scheduling scheme which is described in details in the next section. III. THE MULTI-USER MIMO RESOURCE SCHEDULING SCHEME FOR CARRIER AGGREGATION SCENARIO In this section, the proposed RSM-CA is presented. RSMCA is based on the original PF scheduling scheme but can be used in the carrier aggregation scenario. While choosing a UE for scheduling, not all of the UEs priorities are calculated, a selection factor is introduced to guarantee the resource on the PCC of a UE is primarily allocated to the UE itself. As for the grouping users which are scheduled on the same RB, a user marking resource allocation (UMRA) procedure is introduced to guarantee the grouping users can be transmitted and the user group is valid. Moreover, different users have different grouping users on different carriers in order to use frequency selective diversity aiming to maximize the system throughput. There are ve steps in RSM-CA, including scheduling RB, getting SU, getting GU, deciding the transmission mode, and updating priority. The ve steps are described as follows: A. Scheduling RB All the RBs of eNB CC take turns to be allocated in the rst step. B. Getting SU Instead of calculating all the UEs priorities, we selectively calculate the priorities to get the scheduled user (SU) in the second step. At each scheduling slot t, the original PF scheduling scheme allocates the RB k to user i who maximizes the ratio of achievable instantaneous data rate over average received data rate. This scheduling policy can be describe as i = arg max
i

Ri (t, k) Ti (t)

(1)

where Ri (t, k)is the instantaneous transmission date on RB k for user i, and is the average data rate for user i. For single

Scheduling RB

User Group
yes Is there marked user? no

CASE I

CASE II

CASE III

Allocate a RB to the marked user on its PCC When scheduling new eNB CC or SU hasn t been grouped

Getting SU

Does SU need grouping?

no

Use the previous group

Valid

Invalid

The SU and GU give the largest packet size on the RB

yes Get a GU

Fig. 3.

User Marking Resource Allocation (UMRA) Procedure

Getting GU Find or not? no SU use the RB exclusively

component carrier scenario, all the UEs priority are calculated and compared, but for carrier aggregation scenario, different UEs have different UE CC, so not all the UEs priorities can be calculated for a RB, and considering the PCCs function we should grantee UE is primarily scheduled on its PCCs RB. Therefore, we modify the scheduling policy as follows: Ri (t, k) i = arg max wi i Ti (t)

Delect the GU from all the UEs

yes UMRA

no The user group valid or not? yes Deciding the transmission mode

(2)

where wi is a selection factor, wi [0, 1], it indicates whether to calculate UE s priority, in order to get its value, three situations should be considered: 1) If UE hasnt have RB on its PCC, and the scheduling RB in the current slot belongs to the UEs PCC, wi = 1; 2) If UE hasnt have RB on its PCC, and the scheduling RB in the current slot doesnt belong to the UEs PCC,wi = 0; 3) If UE has have RB on its PCC, and the scheduling RB in the current slot belongs to the UEs UE CC, wi = 1 . Using the selection factor, UEs priority is rst calculated on its PCC to grantee its transmission. C. Getting GU While nding the group user (GU) of the SU in the third step, we use a user marking resource allocation (UMRA) procedure to guarantee the GU can be transmitted and the user group is valid. In order to make the user group valid, we should make sure the users in one group can be transmitted, that is, they have all have RB on their PCC. With the second step, we grantee the SU can be transmitted, so whether the user group is valid depends on the GU. As shown in Fig.3, UMRA includes three cases: CASE I: The scheduling RB in the current slot belongs to the GUs PCC or the GU has already have RB on its PCC. CASE II: GU hasnt have RB on its PCC, and there is RB left on its PCC which can be allocated to the GU, the GU is marked in order to give it its PCCs RB in the next scheduling slot. CASE III: GU hasnt have RB on its PCC, and there isnt RB left on its PCC. The marked GU grantees it can be rst scheduled in the next scheduling slot to make its current user group valid.

Update priority

Fig. 4.

The Overall MU-MIMO-RSM-CA Process

D. Deciding transmission mode The transmission mode is decided in the forth step, if the packet size of the SU-MIMO mode is larger than the MUMIMO mode, the SU-MIMO mode is selected, otherwise, the MU-MIMO mode is selected. E. Updating priority In the last step, after RB k is allocated, the average data rate Ti (t) is updated as following: Ti (t + 1) = (1 1 1 )Ti (t) + Ri (t, k)b(i) Tc Tc 1 if i = i , 0 if i = i . (3)

{ b(i) =

(4)

where Tc is the observation window length of the average transmission rate in terms of TTI (Transmission Time Interval). The overall RSM-CA process is described as Fig.4. In Fig. 4, the ellipse area corresponding to the third stepgetting GU-in Fig. 3, and in order to use frequency selective diversity, SU nds different GU on different carriers aiming to maximize the system throughput.

IV. SIMULATION RESULTS AND ANALYSIS This section evaluates the performance of the RSM-CA. The simulation is for downlink of LTE-Advanced system. The bandwidth of each CC is 10MHz, and each CC has 50 RBs to allocate to UEs. Table I summarize the simulation parameters.
TABLE I S YSTEM
SIMULATION PARAMETERS [7]

Simulation Parameters Site layout Inter-site distance Minimum distance between UE and cell User location UE speeds of interest Channel Model Thermal Noise Spectral Density Penetration Loss Total TX power Antenna pattern Trafc model

Settings 7 cells wrap-around 500 m 35 m Uniformly dropped in all cells 3 km/h Spatial Channel Model -174 dbm/Hz 20 db 40 dBm (40W) 1*1 Full buffer

CC), some UEs can not get RBs to transmit, the system can only satisfy a certain number of UEs, which is not larger than the total resource number, so the throughput stops to increase. Compared to PF, RSM-CA can have more total throughput (about 20 Mbps) because there are more valid groups can be transmitted and frequency selective diversity is exploited to maximize the system throughput. Some uctuations can be seen from Fig.6, that is because each time the user total number changes, we re-drop the UEs randomly and with their positions become different their channel state are different, which leads to different receiving SINRs, when most UEsSINR are very low, the throughput of more UEs may be smaller than less UEs .
150 MUMIMO without CA CAMUMIMO with 2 CCs CAMUMIMO with 3 CCs RSMCA with 2 CCs RSMCA with 3 CCs 100

Fig. 5 shows the number of invalid user groups with the increase of user total number in ve different scenarios: 1. MU-MIMO without CA, i.e. there is only one CC in the system; 2. CA-MU-MIMO with 2 CCs, i.e. each UE can aggregate 2 CCs, using original PF scheduling scheme to do resource schedule; 3. CA-MU-MIMO with 3 CCs, i.e. each UE can aggregate 3 CCs, using original PF scheduling scheme to do resource schedule; 4. RSM-CA with 2 CCs, i.e. each UE can aggregate 2 CCs, using the proposed RSM-CA to do resource schedule; 5. RSM-CA with 3 CCs, i.e. each UE can aggregate 3 CCs, using the proposed RSM-CA to do resource schedule. We can see that, the number of invalid user group decreases with the increment of the number of aggregated CC, because there are more resource can be allocated to UEs. So for scenario 1, when user total number is 160, there are about 30 user groups are invalid, and the number linearly increases with the increment of user total number, when the system can aggregate one more CC, there are 50 more user groups (100 more UEs) can be transmitted. On the other side, while using the original PF scheduling scheme, when user total number exceeds RB number, some UEs can not get RB on their PCC, resulting in invalid user group. As shown in Fig. 5, when user total number exceeds 200 with 2 CCs or user total number exceeds 300 with 3 CCs, some user groups can not be transmitted, and the number of invalid user groups linearly increases with the increment of user total number. Compared to scenario 2 and 3, for scenario 4 and 5, the proposed RSMCA makes sure there is no invalid user group and all the scheduled UEs can be transmitted. For the last four scenarios of Fig. 5, Fig. 6 shows the total throughput of the system. The system throughput increases when there are more CC aggregated, because more UEs can get resource to be transmitted. When the number of UEs exceeds the number of RBs (i.e. 200 for 2 CC and 300 for 3

Invalid user group number

50

0 150

200

250 300 User total number

350

400

Fig. 5. The Number of Invalid User Group with The Increase of User Total Number in Five Different Scenarios

160

140 Total throughput/Mbps

CAMUMIMO with 2 CCs CAMUMIMO with 3 CCs RSMCA with 2 CCs RSMCA with 3 CCs

120

100

80

60

40 100

150

200

250 300 User total number

350

400

Fig. 6.

The Total Throughput of The System for Four Different Scenarios

V. C ONCLUSION A multi-user MIMO resource scheduling scheme is proposed in this pager for LTE-Advanced system with carrier

aggregation technology. Simulation results demonstrate that the RSM-CA can achieve better system throughput than the original PF scheduling scheme, especially when the number of UEs exceeds the number of RBs. The proposed scheme grantees all scheduled UEs can be transmitted according to 3GPP standard and makes use of frequency selective diversity aiming to maximize the system throughput. R EFERENCES
[1] 3GPP, 3rd Generation Partnership Project; Technical Specication Group Radio Access Network; Further Advancements for E-UTRA; Physical Layer Aspects(Release 9),,in TR 36.814 V0.4.1, ed, 2009. [2] 3GPP, 3rd Generation Partnership Project;Technical Specication Group Radio Access Network;Evolved Universal Terrestrial Radio Access (E-UTRA);Carrier Aggregation;Base Station (BS) radio transmission and reception(Release 10), , in TR 36.808, ed, 2010. [3] W. C. Chung, et al., A low-complexity beamforming-based scheduling to downlink OFDMA/SDMA systems with multimedia trafc, Wireless Networks, vol. 17, pp. 611-620, 2011. [4] S. Jagabathula and D. Shah, Fair Scheduling in Networks Through Packet Election Information Theory, IEEE Transactions on, vol. 57, pp. 1368-1381, 2011. [5] J. W. Jung, et al., Group Based Proportional Fairness Scheduling with Imperfect Channel Quality Indicator in OFDMA Systems, IEICE Transactions on Communications, vol. 94, pp. 599-602, 2011. [6] 3GPP, Doppler Impact of Higher Carrier Frequencies on LTE-A Uplink , R1-090283 2009. [7] 3GPP, 3rd Generation Partnership Project;Technical Specication Group Radio Access Network;Physical layer aspect for evolved Universal Terrestrial Radio Access(UTRA)(Release 7), ,in TR 25.814, ed, 2006.

S-ar putea să vă placă și