Proceedings of Asia-Pacific Conference on Communications 2007
Channel and Queue Aware Scheduling for RealTime Service in Multiuser MIMO OFDM system Guangyi LIU1, Jianhua ZHANG2, Jianchi ZHU2, Weidong WANG2, Research Institute of China Mobile1, Beijing University of Posts&Telecoms2， Email：
[email protected],
[email protected] Abstract- For real time service, not only the data rate, but also the packet delay should be guaranteed. In this paper, by exploiting the queue status information and channel status information, a novel framework of joint spatial-frequency scheduling for real time service in MIMO OFDMA system is proposed, and some joint spatial-frequency scheduling algorithm is proposed based on Modified Largest Weighted Delay First (M-LWDF), Quality Guaranteed (QG) and Packet Loss Ratio (PLR) priority function. Comparing the simulation results of MIMO 1 × 2 case, M-LWDF achieves best performance; QG and M-LWDF scheduler may support 20 video user with 500kbps data rate in 10MHz bandwidth, and the spectrum efficiency may approach 0.9bps/Hz. while for MIMO 2 × 4 results, QG scheduler may achieve best performance comprehensively, and more than 50 video user can be supported in 10MHz, and the spectrum efficiency may approach 2.5bps/Hz. Key words: MIMO. OFDMA, Mu-ZFB, Real Time Service
I. INTRODUCTION To fulfill the requirements of 3G LTE, OFDMA is proposed for LTE downlink [1] for its excellent capability to mitigate the frequency selective fading and provide high spectrum efficiency. Further, OFDMA provides a natural multiple access method by assigning different users with orthogonal subcarriers, and multiuser diversity gain in frequency domain can be exploited [2]. By configuring multiple antennas at both ends of communication link, the MIMO channel capacity may be improved to be approximate to the minimum number of the antennas at the transmitter and receiver [3]. Exploiting the CSI at the transmitter, VBLAST can also approach the capacity of MIMO channel. Besides, spatial multiuser diversity and multiplexing can be exploited to achieve better cell throughput. [4] [5] have proposed the improved round robin and greedy antenna scheduler respectively for multiuser downlink with VBLAST to exploit the spatial multiuser diversity. Naturally, by joint spatial-frequency scheduling [6] [7], the multiuser diversity gain can be achieved in spatial and frequency domain to maximize the system throughput. In past work of this filed, most paper is focused on maximizing the system throughput when the scheduler is channel aware [3] [4]. However, for different packet service, the burst characteristic of the packets varies much and influences the system throughput much. For Unconstrained Delay Data (UDD) service, the delay of the packet is not serious, but for
Real Time service, the packet with delay exceeding the pre-defined value will be dropped, which leads to bad use experience. So for real time service, usually the user data rate and packet delay should both be guaranteed. Obviously, if the packet status in queue is monitored and the packets may be transmitted in time, then QoS can be guaranteed. In [8], the queue aware scheduling is considered in OFDMA, and better QoS can be provided. In this paper, based on multiuser Zero Force Beamforming (ZFB) [9], the joint spatial-frequency scheduling based on channel and queue aware for MIMO OFDMA system is investigated, and novel Quality Guaranteed (QG), modified M-LWDF [10] algorithms are proposed and compared to Packet Loss Ratio (PLR) algorithm [11] for real time service. From the simulation results, the performance of QG approaches that of M-LWDF and better than that of PLR. For MIMO 1×2 case, QG and M-LWDF scheduler may support 20 video user with 500kbps data rate in 10MHz bandwidth, and the spectrum efficiency may approach 0.9bps ; while for MIMO 2×4 case, more than 50 video user can be supported in 10MHz, and the spectrum efficiency may approach 2.5bps/Hz. II. SYSTEM MODEL OF MULTIUSER ZFB For MIMO OFDMA system, MIMO channel on every subcarrier can be regarded as flat fading MIMO, so the MIMO channel of user i on subcarrier n can be expressed as Η i ,n , and the multiuser ZFB on subcarrier n can be described as Figure 1. Assume the general multiuser MIMO ˆ = H channel matrix is H n 1, n ... H i , n ... H K , n , where H i , n is the MIMO channel matrix of user i with
M R × M T dimension, and M T , M R are the transmitter and receiver antenna number respectively. Then one antenna can be selected from every user as following. Antenna selection Algorithm (subopt) Step a: the MIMO channel matrix of M T users selected constructs a generalized MIMO channel matrix, and one receiver antenna is selected, which maximize the MISO channel capacity between it and the transmitter antennas; Then delete the left receiver antennas of the selected user from the generalized MIMO channel matrix. Step b: Select one receiver antenna from the left, and maximize the capacity of the MIMO channel between the selected receiver antenna and the transmitter.
-------------------------------------------------------------------------------------------------------This work is funded by the 863 project of China under grant No.2006AA01Z258.
1-4244-1374-5/07/$25.00 ©2007 IEEE
509
Step c: Repeat step b until every user obtains one receiver antenna. Assume that the antennas of K users are selected to receive, and then the virtual MIMO channel matrix between transmitter and the UEs selected can be constructed as: H n = H1, n ... H i, n ... H K , n
(1)
i
Where H is MIMO channel response of UE i on the selected antenna. Then weight matrix for ZFB is [9]: B n = H †n ( H n H †n ) Dn −1
(
(2)
)
Where Dn = diag d1,n ,...d k , n ,..., d M R , n is the diagonal matrix which keeps the transmit power unchanged after beamforming, and † means the hermit transpose. M R is the antenna number selected at the virtual UE, it is also the independent data stream number. dk,n =
1
( H H † ) −1 n n k , k
(3)
III. JOINT SPATIAL-FREQUENCY CHANNEL AND QUEUE AWARE SCHEDULING
In this paper, the distributed control is considered, and no multi-cell coordination is necessary. The power of Node B is distributed uniformly on every spatial sub-channel of every subcarrier, and thus the problem is simplified as multiuser subcarrier allocation and multiuser antenna selection. Then the joint spatial-frequency scheduling algorithm based on channel and queue aware can be summed as following: Step 1: On every subcarrier, the user priority is calculated by M-LWDF or QG or PLR function below. Step 2: On every subcarrier, M T users with highest priority value are chosen to construct the user set U n and ˆ = H general MIMO matrix H n 1, n ... H i , n ... H K , n . Step 3: By antenna selection algorithm proposed in section II, and the spatial sub-channel gain d k , n is calculated. Step 4: Every user’s averaged data rate ri is updated as following: (1 − α )ri + α Ri , if user k is served . ri = else (1 − α )ri , Where 0 < α < 1 is the forgetting factor.
(4)
N
Ri = ∑ ri , n is the served data rate in current scheduling n =1
period, ri, n is the data rate on subcarrier n of user i . If subcarrier n is not allocated to user i , then ri,n = 0 . The priority functions proposed are described below: A. M-LWDF For real time service, M-LWDF [10] is regarded as the optimal scheduling for real time service in CDMA system. 510
For multiuser OFDMA system, it is modified further to adopt the multi-channel allocation simultaneously. The scheduling of M-LWDF takes into account the maximum service delay Wi ( t ) and channel status information of user i .The user priority function for subcarrier n is defined as:
µi ,n ( t ) = bW i i (t )
rˆi , n ( t ) ri
(5)
Where Wi ( t ) is the delay of the first packet in the queue
i of user i . bi = − ( log σ i ) / WMax , σ i is the maximum i i , Pr {Wi ( t ) > WMax probability of Wi ( t ) exceeds WMax } < σi .
rˆi ,n (t ) is the transmission capability of user i on subcarrier n at time t and can be calculated as Shannon capacity ; ri is the average data rate of user i . rˆi ,n (t ) and ri vary with time and decided by the modulation and channel condition. The user with highest priority value µi , n ( t ) obtains the chance to transmit on subcarrier n . If all service is the i same type, and have the same QoS, then WMax and σ i are the same for all user, and the priority can be simplified as:
µi,n ( t ) = Wi ( t )
rˆi,n (t ) ri
(6)
B. PLR [11] has proposed a Packet Loss Ratio (PLR) algorithm for real time service. It calculates a priority for every user on every subcarrier, and allocates the subcarrier according to the user’s priority, and the subcarrier is allocated to the user with highest priority value. The priority calculation takes into account the user packet delay Wi ( t ) , channel
status rˆi , n ( t ) , the user average throughput ri , and the user
packet loss ratio PLRi ( t ) . The priority function on subcarrier n is defined as following:
i rˆi , n ( t ) PLRmax i , if PLRi ( t ) > PLRmax Wi ( t ) r PLR t i i ( ) ˆ r (t ) k i µi, n ( t ) = Wi ( t ) i, n , if PLRi ( t ) < k ≤ PLRmax i r PLR i max rˆi ,n ( t ) PLRi ( t ) Wi ( t ) otherwise , i ri PLRmax (7) i Where PLRmax is the maximum packet loss ratio defined the user,， k is a none-zero constant which is far smaller i than PLRmax .
C. QG For real time service, the service has strict constraint on delay, but permits some packet drop ratio. To maximize the user number serving by the system, system may permit every user has a small packet drop ratio less than the pre-defined
one. So we proposed a modified scheduling algorithm (QG) for real time service, the user priority is defined as:
ri
× f ( PLRi ,Wi ( t ) )
(8)
Where,
10Wi ( t ) / WMax ×10 PLRi / PLRMax , if PLRMax ≥ PLRi f ( PLRi , Wi ( t )) = W (t ) / W ( 2 PLRMax − PLRi ) / PLRMax , 10 i Max × 10 if PLR < PLR Max i (9) When PLRi is small, the priority value of QG is less than that
of
Proportional
Fairness;
when
PLRi
approaches PLRMax , the priority increases fast; when PLRi exceeds PLRMax , the priority is decreased to avoid wasting the limited radio resource. IV.
SIMULATION PARAMETER
In the simulation, MCS adopted is given in Table 1, the convolution coding are combined with QPSK, 16QAM to create 5 MCS. No H-ARQ is considered in link level simulation, but the chase combining is adopted in system level simulation. The suitable MCS is selected from the Table 1 to transmit data symbol on every subcarrier. The SNR threshold for MCS as Table 1 is obtained. Table 1 The MCS and the SNR threshold MCS
Mod
Code Ratio
data bits
SNR threshold
1
QPSK
1/3
2/3
0.5dB
2
QPSK
1/2
1
3.7dB
3
QPSK
3/4
3/2
6.3dB
4
16QAM
1/2
2
10dB
5
16QAM
3/4
3
15.2dB
If the user has 5% packets are dropped during the communication, then the user is regarded as unsatisfied. The other system simulation parameters are as Table 3. Table 3. System parameters Parameter Assumption Carrier Frequency 2GHz Band width 10MHz Sample Frequency 15.36 MHz Sub-carrier spacing 15 kHz CP length(µs/samples) 7.29/14 FFT Size 1024 Occupied Subcarriers number 601 Subcarrier Group number 75 Inter-site distance 2Km Cell number 27 (9 clusters) Distance-dependent path loss L=128.1 + 37.6log10(R) Shadowing standard deviation 8 dB Correlation distance of Shadowing 50 m Shadowing Between cells 0.5 correlation Between sectors 1.0 Penetration Loss 20dB Channel model Typical Urban (TU) PDP,SCM Total BS TX power (Ptotal) 43dBm Minimum distance between UE >= 35 meters and cell User data rate 2Mbps H-ARQ Chase combining
V. SIMULATION RESULTS In this section, the simulation results of the joint spatialfrequency scheduling based on modified M-LWDF, PLR and QG algorithms for real time service are presented as Figure 1, Figure 2 and Figure 3. UE Drop Rate VS UE number RT 0.14
QG 2x4 QG 1x2 M-LWDF 1x2 M-LWDF 2x4 PLR 1x2 PLR 2x4
0.12
0.1
The same frame structure as LCR TDD for LTE TDD [12] and the soft frequency reuse scheme is adopted and optimized as [6] is adopted. The 500kbit/s video traffic model is defined as Table 2.
UE Drop Ratio
µ i ,n ( t ) =
rˆi , n ( t )
0.08
0.06
0.04
0.02
Table 2 traffic model parameters Video Stream model Value Inter-arrival time between the 100ms beginning of each video-frame Number of video-packet in a frame 8 Video packet size 400byte Truncated Pareto Inter-arrival between video-packets in K=2.5 alpha = 1.2 a video frame M= 12.5ms Video packet max delay time 200ms Video average data rate 500kbit/s Video minimum data rate 128kbit/s Video length 120s
0
5
10
15
20
25
30
35
40
UE number per cell
Figure 1 user number vs. drop ratio
From Figure 1, for MIMO 1×2 case, M-LWDF has best performance, and QG is better than PLR, and when system load increases, QG converges to M-LWDF. For MIMO 2× 4 case, QG is little bit better than M-LWDF, and PLR has worst performance. From Figure 2, or MIMO 1×2 case, MLWDF has highest average user data rate, and QG is higher 511
than PLR. For MIMO 2×4 case, QG is almost the same as M-LWDF, and PLR is the lowest one. UE Data Rate VS UE number RT
0.9bps ; while for MIMO 2×4 case, more than 50 video user can be supported in 10MHz, and the spectrum efficiency may approach 2.5bps/Hz.
450
system throughput VS UE number 18
445 16
system throughput (Mbps)
UE Data Rate (kbps)
440 435 430 425 QG 2x4 QG 1x2 M-LWDF 1x2 M-LWDF 2x4 PLR 1x2 PLR 2x4
420 415 410
5
10
15
14 12 10 8 QG 2x4 QG 1x2 M-LWDF 1x2 M-LWDF 2x4 PLR 1x2 PLR 2x4
6 4
20
25
30
35
40
2
UE number per cell
5
10
15
20
25
30
35
40
UE number per cell
Figure 2 averaged user data rate vs. user number
From Figure 3, QG and M-LWDF almost has the same system throughput, and PLR is the lowest one. The system throughput increases with the user number proportionally. Since the video service has constant data rate, little multiuser diversity gain from scheduling can be observed. For MIMO 1×2 case, QG and M-LWDF scheduler may support 20 video user with 500kbps data rate in 10MHz bandwidth, and the spectrum efficiency may approach 0.9bps ; while for MIMO 2×4 case, more than 50 video user can be supported in 10MHz as predicted from Figure 4, and the spectrum efficiency may approach 2.5bps/Hz. VI. CONCLUSION
Figure 3 system throughput vs. user number
REFERENCES [1] [2]
[3]
[4]
[5]
In this paper, a framework of a joint spatial-frequency scheduling based on channel and queue aware for real time service in MIMO OFDMA system is proposed, and a joint spatial-frequency scheduling algorithm is proposed based on M-LWDF, QG and PLR priority function. From the simulation results of multi-cell scenario, QG based scheduling has best performance comprehensively than the other two algorithms in MIMO 2 × 4 case. QG based scheduling has smaller user drop ratio than M-LWDF and PLR based scheduler, and the almost same average user rate and system throughput performance as M-LWDF scheduler. For MIMO 1×2 case, M-LWDF based scheduler has best performance than the other two algorithm. QG based scheduling has almost the same system throughput as MLWDF scheduler, and little worse user drop ratio and average user data rate performance than M-LWDF scheduler. The system throughput increases with user number proportionally, and little multiuser diversity gain can be observed since video service is delay sensitive. For MIMO 1 × 2 case, QG and M-LWDF scheduler may support 20 video user with 500kbps data rate in 10MHz bandwidth, and the spectrum efficiency may approach 512
[6] [7]
[8]
[9]
[10]
[11]
[12]
3GPP TSG RAN WG1 #42, R1-050789, Text Proposal on “TDD UL/DL based on OFDMA” for TR 25.814. Cheong Yui Wong, Cheng, R. S, et al., “Multiuser OFDM with Adaptive Subcarrier, Bit, and Power Allocation,” IEEE Journal on Selected Areas in communication, vol. 17, pp. 1747 -1758, Oct. 1999. G. Foschini, M. Gans, “On limits of wireless communications in a fading environment when using multiple antennas”, IEEE Wireless Personal Communications, March 1998, 6(3): 311-335. Oh-soon Shin, Kwang Bok(Ed) Lee, “Antenna-Assisted Round Robin Scheduling for MIMO Cellular Systems”, IEEE Communications Letters, vol. 7, no. 3, pp. 109-111, Mar. 200. Manish airy, Sanjay Shakkottai and Robert W. Health, “Spatial Greedy scheduling in multiuser MIMO wireless system”, Proc. of IEEE Asilomar Conf. on Signals, Systems, and Computers, vol. 1, pp. 982 - 986, Pacific Grove, CA, USA, Nov. 9-12, 2003. Guangyi Liu , et al, “Enhanced perforamnce of TD-SCDMA LTE system With Multiuser MIMO in downlink”, IEEE APCC 2006. Guangyi Liu, Jianhua Zhang, et al, “Joint Space-Frequency opportunistic Multiuser Diversity for VBLAST OFDMA with Partial CSI”, IEEE ICCCAS 2006. Vincent K. N. La, et al, “Channel Adaptive Technologies and Cross Layer Designs for Wireless Systems with Multiple Antennas Theory and Applications”, Feb.2006, Wiley. Jinsu Kim1, Sungwoo Park1, et al, “A Scheduling Algorithm Combined with Zero-forcing Beamforming for a Multiuser MIMO Wireless System”, IEEE VTC 2005Fall. Eun Ho Choi; Wan Choi; Andrews, J.,“Throughput of the 1x EV-DO system with various scheduling algorithms”, Spread Spectrum Techniques and Applications, 2004 IEEE Eighth International Symposium on 30 Aug.-2 Sept. 2004 Page(s):359 – 363. Seokjoo shin, Byung-Han Ryu, “Packet Loss fair scheduling scheme for Real-Time Traffic in OFDMA systems”, ETRI Journal, vol 26, No.5, Oct 2004. 3GPP TSG RAN WG1#42, R1-050800, “Numerology and Frame Structure of EUTRA TDD based on OFDMA and text proposal for TR 25.814 “, CATT, RITT, ZTE, Huawei.
