A Configurable Dual-Mode Algorithm on Delay

advertisement
2012 IEEE Wireless Communications and Networking Conference: MAC and Cross-Layer Design
A Configurable Dual-Mode Algorithm on Delay-Aware
Low-Computation Scheduling and Resource
Allocation in LTE Downlink
Siyue Sun, Qiyue Yu, Weixiao Meng
Cheng Li
Communication Research Center
Harbin Institute of Technology
Harbin, China, 150001
Email:{sunsiyue, yuqiyue, wxmeng}@hit.edu.cn
Electrical and Computer Engineering, Faculty of
Engineering and Applied Science, Memorial University
St. John's, NF, Canada AIB 3X
Email: licheng@mun.ca
Abstract—Long Term Evolution (LTE) has been proposed as a
promising radio access technology to bring higher peak data
rates and better spectral efficiency. However, scheduling and
resource allocation in LTE still face huge design challenges due to
their complexity. This paper divides the complex problem into
three sub-problems: scheduling pattern, scheduling priority and
quantity of scheduled data. Based on analysis of three subproblems, a configurable dual-mode (CD) algorithm is proposed.
CD algorithm is able to guarantee queuing delay with low loss of
resource utilization and fairness by employing dual-mode
scheduling mechanism. And it can be configured by three
parameters catering to different performance requirements. By
utilizing QoS Class Identifier (QCI) and Channel Quality
Indicator (CQI) defined by LTE, low computation is realized in
CD scheduler. Finally, performance evaluation of the proposed
scheduler is presented. The results and correlative analysis testify
effectiveness of CD algorithm.
Keywords-LTE, scheduling, resource allocation, QCI, CQI
I.
INTRODUCTION
As a new member of the 3GPP family, Long Term
Evolution (LTE), which is an all-IP network, bears higher
peak data rates, better spectral efficiency, increased cell edge
throughput, enhanced support for end-to-end QoS and flexible
spectrum deployment with scalable bandwidth ranging from
1.4 to 20 MHz. Such higher system performance targets and
the evolved system architecture present new challenges and
opportunities for radio resource allocation and scheduling
which undertake the responsibility to assign radio resource to
users according to their quality of service (QoS) attributes,
specific algorithm and quality of radio links.
Scheduling and resource allocation algorithm for LTE
systems has become an increasingly interesting research topic.
Researchers have proposed a variety of different algorithms
considering diversity factors, although the most important
objective of LTE scheduling is to achieve an optimal tradeoff
between utilization and fairness while satisfying QoS
requirements of all users [1].
However, most of the proposed algorithms are far away
from system implementation due to their computation
complexity. In fact, classic algorithms Round Robin (RR)[1]
and Proportional Fair (PF) [2] are most widely applied owing
to their simpleness. Furthermore, diversified design
considerations of these algorithms cause the fact: a specific
algorithm performs better performance only under specific
case. He isn’t fit for ever-changing network status. It’s a huge
challenge for most of current scheduling and resource
allocation algorithms.
In addition, the specification of LTE has provided
directions for scheduling and resource allocation. In [3], QoS
Class Identifier (QCI) is defined to characterize classified QoS
requirements of different traffic. Every QCI is associated with
a priority level which should be an important reference for
scheduling. However, to the best of our knowledge, few
schedulers well utilize QCI as a design consideration in the
published literatures.
In this paper, a configurable dual-mode (CD) scheduling
and resource allocation algorithm for LTE downlink is
proposed. Targeting high universality, CD algorithm can be
configured by three parameters catering to different system
requirements. Dual-mode mechanism is employed by CD
algorithm to achieve a favorable tradeoff among resource
utilization, fairness and guarantee of QoS. Additionally, the
proposed algorithm is able to guarantee queuing delay
according to QCI of the service. And by means of utilizing QCI
and Channel Quality Indicator (CQI) defined by LTE
specification, low computation is realized in CD scheduler.
II.
Orthogonal Frequency Division Multiplexing (OFDM)
radio technology has been selected as LTE downlink radio
access scheme owing to its high bandwidth scalability, simple
equalization, high robustness against multi-path fading and
high spectral efficiency. Concerning resource allocation,
OFDM-based LTE downlink can be seen as a time-frequency
two-dimensional resource sharing system, as described in Fig.1
(a). Such two-dimensional resource is divided into multiple
resource blocks (RBs). An RB, which last 0.5 ms in the time
domain and 12 consecutive subcarriers in the frequency
domain, is considered as the minimum scheduling unit. Each
LTE frame lasts 10 ms and it is divided into ten equally size
This work was supported by National Science and Technology Major
Projects of China (Grant No. 2011ZX03004-003 and 2011ZX03004-006),
Program for New Century Excellent Talents in University (Grant No. NCET08-0157) and the Fundamental Research Funds for the Central Universities
(Grant No.HIT.NSRIF.201148).
978-1-4673-0437-5/12/$31.00 ©2012 IEEE
SCHEDULING AND RESOURCE ALLOCATION IN LTE
DOWNLINK
1444
sub-frame, called Transmission Time Interval (TTI). Evolved
NodeBs (eNodeBs), the base stations in LTE, executes the RBto-user assignment at its medium access control (MAC) layer
according to the selected scheduling algorithm every TTI, as
shown in Fig. 1(b).
QoS, and scheduling pattern will turn to RB-by-RB pattern
when high resource utilization is more important.
Determination of scheduling priority among users and
among RBs is the core issue of scheduling algorithm. As
already stated, favorable tradeoff among QoS guarantee,
resource utilization and fairness is the general target, but high
computation should be avoided considering the realizability.
The last sub-problem quantity of scheduled data should also
be placed emphasis on, especially under user-by-user
scheduling pattern. The quantity of data sent to each scheduled
user will affect performance of the scheduler apparently.
Assaad et al. [7] proposed a complicated algorithm to calculate
the number of RBs allocated to scheduled user. The algorithm
increases fairness of the scheduler, but also increases
complexity of computation. In fact, guarantee of QoS is the
target of user-by-user pattern. Therefore quantity of scheduled
data should be determined according to QoS requirements of
the scheduled user.
Figure 1.
III.
B. Proposed CD Algorithm
Several assumptions are made before presenting the new
algorithm. Firstly two consecutive RBs in time domain are
assigned to the same user in a TTI. In addition, only resource
allocation in Physical Downlink Shared Channel (PDSCH) is
considered and the number of available RBs is fixed every TTI,
due to time limitations as well as in order to reduce
complexity of the system simulation.
Scheduling and Resource Allocation in LTE Downlink
PROPOSED SCHEDULING AND RESOURCE ALLOCATION
ALGORITHM
A. Strategy Analysis
In order to make the analysis much clear and reasonable,
this paper divides the problem of scheduling and resource
allocation in LTE downlink into three sub-problems:
determination of scheduling pattern, scheduling priority and
quantity of scheduled data.
QoS requirements of traffic to be scheduled are depicted
by QCI defined in [3]. 3GPP has defined nine traffic classes
mainly according to their resource type, priority, packet delay
budget (PDB), and packet error loss rate. Tab. I(Tab. 6.1.7 in
[3]) shows the standardized QCI characteristics with main
QoS requirements and example services.
TABLE I. STANDARDIZED QCI CHARACTERISTICS[3]
In LTE system, radio resource can be allocated using two
patterns: user-by-user and RB-by-RB, which is determined by
scheduling pattern. Numerous literatures ignored this problem
and chose one of the scheduling patterns directly. In [4-6]
priority among users is calculated according to certain
algorithm firstly, and then radio resource is assigned user-byuser in accordance with the priority order. In user-by-user
pattern, scheduling priority is determined according to overall
channel quality of each user, which overlooks its differences
in the frequency and will absolutely affect system performance.
Oppositely, in [7-8] priority among users on each RB is
calculated RB-by-RB according to a certain algorithm, and
then RBs are allocated to the user with the highest priority. In
RB-by-RB pattern, specific allocation order among RBs will
also bring performance loss. Zhu et al. [9] even proposed a
Maximum Deviation Channel First (MDCF) concept to
optimize the allocation order of RBs in order to reduce such
impact on performance. But high computation complexity is
also induced unfortunately.
This paper argues that the two patterns should be
employed in a scheduling algorithm simultaneously, and the
choice between them is determined by current performance
target. User-by-user pattern is used aiming at guarantee of
Q Reso
C urce
I Type
Priority
2
1
4
2
GBR
3
3
4
5
5
1
6
7
8
9
NonGBR
6
7
8
9
PDB
100
ms
150
ms
50
ms
300
ms
100
ms
300
ms
100
ms
300
ms
Packet
Error Loss
Rate
Example Services
10-2
Conversational Voice
10-3
Conversational Video
10-3
Real Time Gaming
10-6
Non-Conversational
Video
10-6
IMS Signaling
10-6
Video, TCP-based
10-3
Voice, Video,
Interactive Gaming
10-6
Video, TCP-based
This paper also assumes that radio channel quality of each
user on each RB is quantified by so-called CQI defined in [10],
and the value of CQI is credible, because the adaption of CQI
in base station side is out of range of this paper. The CQI
indices and their interpretations are given in Tab.II (Tab.
1445
7.2.3-1 in [10]).
TABLE II.
CQI index
modulation
0
delay between a Policy and Charging Enforcement Function
(PCEF) and a radio base station should be subtracted from a
given PDB to derive the PDB that applies to the radio
interface. Therefore DMAX of each traffic can be obtained
according to its QCI. Scheduling and resource allocation
policies in the two modes are listed in Tab.III.
4-BIT CQI TABLE [10]
code rate × 1024
efficiency
out of range
1
QPSK
78
0.1523
2
QPSK
120
0.2344
3
QPSK
193
0.3770
4
QPSK
308
0.6016
5
QPSK
449
0.8770
6
QPSK
602
1.1758
7
16QAM
378
1.4766
8
16QAM
490
1.9141
9
16QAM
616
2.4063
10
64QAM
466
2.7305
11
64QAM
567
3.3223
12
64QAM
666
3.9023
13
64QAM
772
4.5234
14
64QAM
873
5.1152
15
64QAM
948
5.5547
TABLE III.
The value of CQI is calculated by UE according to SNR
estimation. Achievable data rate Ri , k on RB i for user k can be
calculated by (1).
RB
Ri , k = effciencyi , k × N scRB × N symbol
E-mode
N-mode
Design Target
QoS and utilization
Utilization and fairness
scheduling priority
QCI and CQI
Simplified configurable
Proportional Fair
scheduling pattern
User-by-user
RB-by-RB
quantity of
scheduled data
Emergency packets
According to CQI
Guarantee of QoS is the main target of E-mode, because
there is traffic on the verge of out of QoS in current TTI. Set
of users to be scheduled covers the users which possess
EMERGENCY TRAFFIC, and radio resource is allocated userby-user. According to [3], if the PDB can no longer be met for
one or more traffic across all UEs that have sufficient radio
channel quality, then PriorityQCI defined by Tab.I shall be
used as the priority among EMERGENCY TRAFFIC which
can be stated by (3).
(1)
Where effciencyi , k is obtained according to corresponding
RB
denotes consecutive subcarriers included
CQI and Tab.II. N SC
RB
denotes number
in a RB in the frequency domain and N symbol
of consecutive OFDM symbols included in a RB in time
domain.
In order to guarantee QoS for each user with low loss of
resource utilization and fairness, a dual-mode scheduling
algorithm, including EMERGENCY MODE (E-mode) and
NORMAL MODE (N-mode), is proposed. Each TTI the
scheduler may turn into E-mode, N-mode or E-mode followed
N-mode, which is determined by status of buffer queue for
traffic. If there exists any EMERGENCY TRAFFIC in current
TTI, E-mode will be activated, otherwise N-mode will be
triggered. EMERGENCY TRAFFIC means at least the head
packet in its buffer queue is EMERGENCY PACKET whose
queuing delay satisfies (2).
Dqueuing ≥ DMAX − Dth
SCHEDULING STRATEGIES IN THE TWO MODES
(2)
Where Dqueuing denotes queuing delay of packet in buffer
queue, DMAX denotes the maximal queuing delay for current
traffic according to QoS requirements, and Dth is delay
threshold which is a configurable parameter indicating
guarantee degree of QoS. Apparently, the bigger Dth is, the
higher guarantee degree is. DMAX is determined by the actual
demand and in this paper it is calculated according to QCI. As
noted following Tab. 6.1.7 in [3], a delay of 20 ms for the
PE − mode =
1
PriorityQCI
(CQI > CQI th )
(3)
Where CQI th is also a configurable parameter denoting
threshold of sufficient radio channel quality. The value of
CQI th will not only affect resource utilization but also affect
guarantee of QoS and fairness.
It is noteworthy that PriorityQCI only equals one of nine
integers. Therefore, ordering among EMERGENCY TRAFFIC
(whose user possesses sufficient radio channel quality)
according to PE − mode can be substituted by classifying each
EMERGENCY TRAFFIC into nine groups according to
its PriorityQCI (group number equals to PriorityQCI ). And then
choosing scheduled EMERGENCY TRAFFIC from Group One
to Nine (EMERGENCY TRAFFIC with same PriorityQCI is
selected randomly). Therefore computation can be
significantly reduced through well utilization of QCI.
For the scheduled EMERGENCY TRAFFIC, all of its
EMERGENCY PACKET will be sent if there are enough RBs
and the corresponding user possesses sufficient radio channel
quality. RBs with better radio channel quality for the
corresponding user are assigned in order to improve resource
utilization. In this paper, CQI is used to quantify channel
quality. And same to PriorityQCI , ordering among RBs
according to their channel quality can be replaced by
classifying RBs into sixteen groups according to their CQI and
1446
then choosing scheduled RBs from Group Sixteen to One.
Therefore, without computation of priority and ordering,
computation of E-mode is very low.
If there is no EMERGENCY TRAFFIC or all of the
EMERGENCY PACKETs have been assigned RBs, N-mode
will be activated. Favorable tradeoff between high resource
utilization and fairness is the main target of N-mode, because
guarantee of QoS in E-mode may induce impact on utilization
and fairness. Set of users to be scheduled covers the users
which possess available data. Radio resource is allocated RBby-RB in N-mode. RB i which hasn’t been allocated will be
assigned to the user with highest priority determined by (4).
PNi ,−kmode =
CQI i , k α
Ratek
(4)
Where Rate k denotes average obtained rate of user k and it
can be calculated by (5).
Ratek = (1 −
1
1
) × Ratek (t − 1) + × Ratek (t − 1)
Tw
Tw
Nuser
Throughput =
In fact, (4) is simplified and configurable PF algorithm.
It’s well known that PF scheduler [2] provides an attractive
tradeoff between maximizing the average sum throughput and
providing fairness to the users. In this paper, achievable
instantaneous rate on RB i in PF algorithm is replace by its
CQI in order to reduce computation. α is a configurable
weighting factor used to control the tradeoff between overall
system resource utilization and data-rate fairness among the
users. Quantity of scheduled data for each scheduled user in
N-mode is determined by CQI of each allocated RB and the
allocation results.
By well utilizing QCI and CQI parameters, computation of
priority and ordering among such priorities are avoided in Emode, so the computation can be very low. N-mode induces
more computation to compensate for resource utilization and
fairness loss which is sacrificed to guarantee QoS in E-mode.
However, such computation of N-mode is still lower than that
of traditional PF scheduler owing to utilization of CQI.
i
i =1
(6)
TimeSIM
Long-term rate fairness is selected to estimate fairness
among users, and it is calculated by standard deviation of
average data rate of users in this paper. Average data rate of
user k is calculated by (7). Timeactivated indicates TTIs when
user k is activated. N RB is the total number of available RBs,
and Ri , k is achievable data rate on RB i for user k which can
be calculated by (1). δ i , k (t ) ∈ {0,1} indicates allocation results
in t-th TTI. If δ i , k (t ) equals to 1, RB i is assigned to user k in
t-th TTI, otherwise the opposite.
Timeactivated N RB
∑ ∑R
i,k
k
Ravge
=
t =1
⋅ δ i , k (t )
i =1
(7)
Timeactivated
In this paper the data whose queuing delay is longer than
that required by QCI is named bad data. In order to measure
the guarantee of QoS, average amount of bad data every TTI
should be calculated. This metrics presents performance of the
scheduler from a system view. Moreover, average amount of
bad data of traffic belonging to different priority grades which
are classified by QCI, is also estimated. It is more meaningful
for evaluating QoS-guaranteed scheduler.
B. Simulation Results
The most relevant simulation parameters are summarized
in Tab. IV. Different configuration of traffic model, CQI and
QCI will bring different simulation results. Owing to space
limitations of the paper, only two typical configurations of
simulation scenarios are presented. During the simulation each
user initiates only one service traffic.
TABLE IV.
PERFORMANCE EVALUATION
A. Performance Metrics
To evaluate the effectiveness of the proposed downlink
scheduling and resource allocation scheme, several metrics are
computed.
System throughput characterizes the average amount of
data that is transmitted by the radio network in a certain
∑M
Where M i denotes total amount of transmitted data of user
i during TimeSIM . N user is total number of activated users.
(5)
Where Tw is a smoothing average factor which is set to
1000 ms generally. Ratek (t − 1) is the average obtained rate of
user k in last TTI. Rate(t − 1) is instantaneous rate in last TTI
and it can calculated by Ri , k in (1) and allocation results of last
TTI.
IV.
amount of time. It is estimated to evaluate system resource
utilization in this paper, and it is calculated by (6).
SIMULATION PARAMETERS
Number of
available RBs
Number of subcarriers per RB
Number of OFDM symbols
per RB
40
12
7
Traffic rate
QCI
CQI
simulation
scenario A
constant rate
and packet
size; the rate
is high
1~9: each value of
QCI corresponds
with equal number
of users
higher QCI
corresponds
with better
CQI
simulation
scenario B
constant rate
and packet
size; the rate
is low
idem
CQI value is
distributed
uniformly
Tab. V shows average amount of bad data belonging to
different QCIs in simulation scenario A. As a comparison, PF
1447
algorithm was also implemented and evaluated in the same
simulation model. As can be seen from the results, CD
algorithm can guarantee QoS according to QCI priority grade,
while under PF algorithm each service has no difference in the
aspect of QoS guarantee.
Number of users
QCI
1
CD
PF
5
45
90
135
180
0
0
0
0
5
0
48.4
110715296.6
149029207
9
0
152.7
110257872.8
149449989.9
1
0
171716.3
83652108.8
132625875.6
5
0
0
62935113.0
102488913.6
9
0
0
62866933.8
102380968.3
4
Fairness
6
0
Figure 2.
100
200
CDD
PF
300
3
200
4
5
85
7
6
Throughput (kbps)
System throughput under different scheduling algorithm
4100
180
4000
160
3900
Throughput
Fairness
1
1.5
2
2.5
3
3.5
4
4.5
140
120
5
α
1500
10000
1000
5000
500
Throughput
Fairness
Figure 5.
System throughput under different scheduling algorithm
System throughput under different scheduling algorithm
15000
0
2
In Fig. 4 and 5, the influences of the configurable
3
200
3800
0.5
CDD
PF
300
The results illustrated in Fig. 3-5 prove the configurability
CD algorithm. As shown in Fig. 3, when the delay threshold
Dth is set bigger, the average amount of bad data becomes
smaller resulting in higher guarantee degree of QoS; and at
the same time, the standard deviation of average data rate of
users becomes higher indicating worse long term data rate
fairness. Therefore, the delay threshold Dth is a configurable
parameter controlling the tradeoff between guarantee degree
of QoS and fairness.
2
4200
Figure 4.
100
1
Figure 3.
4
4
0
Dth (ms)
5
8
86
1
Throughput (kbps)
12
Bad data
2
Fig. 2 shows the variation of system throughput with the
increase in number of users. As can be seen from the plots, the
values of throughput under two scenarios and two scheduling
algorithms all increase with increase in number of users and
get their saturation values at a certain point. It's worth noting
that throughput under CD algorithm is higher in scenario A,
while throughput under PF algorithm is higher in scenario B.
The results prove that a fixed scheduling algorithm is unable
to always achieve good performance under the ever changing
network scenario. Configurability and adaptability are
essential for scheduling algorithm. CD scheme can be
configured by three parameters and it can change its mode
under different system status. It will bring better performance
than other scheme with single fixed strategy.
7
87
3
0
16
x 10
Long-term data rate fairness
(kbps)
Algorit
hm
AVERAGE AMOUNT OF BAD DATA (KBPS)
3
4
5
6
7
CQI th
8
9
0
10
Long-term data rate fairness
(kbps)
TABLE V.
weighting factor α and threshold of sufficient radio channel
quality CQI th on system throughput and fairness are depicted.
System throughput will increase with the increase of α or
CQI th , while the standard deviation of average data rate of
users also increases indicating decrease of fairness. Therefore,
both α and CQI th are able to control the tradeoff between
system throughput and fairness.
System throughput under different scheduling algorithm
From the simulation results above, it comes to the
conclusion that with three configuration parameters and the
dual-mode scheme, CD algorithm can not only perform a good
tradeoff among guarantee of QoS, resource utilization and
fairness, but also adjust such tradeoff catering to different
system and performance requirements.
V.
CONCLUSION
System resource utilization, fairness and guarantee of QoS
are all design considerations of scheduling, but they are
apparently irreconcilable. What’s worse, low computation also
1448
brings tight constraints on complexity of scheduling algorithm.
In order to make things much clear and reasonable, this paper
divides such complex problem into three sub-problems:
scheduling pattern, scheduling priority and quantity of
scheduled data. Functions, importance and different resolutions
of three sub-problems are well analyzed. Based on the analysis,
CD algorithm that tries to guarantee QoS with low loss of
resource utilization and fairness and low computation is
proposed. Dual-mode scheme is employed to balance guarantee
of QoS, resource utilization and fairness, and three parameters
are configured to adapt different performance targets. Two
parameters QCI and CQI defined by LTE are well utilized by
CD algorithm, which reduces cost of computation significantly.
Performance of the proposed scheduler is compared with that
of PF scheduler. The evaluation results and the correlative
analysis testify effectiveness of CD algorithm. Further
simulation will be carried out in our future work in order to
explore configuration characteristics of CD algorithm.
REFERENCES
[1]
[2]
E. Dahlman, S. Parkvall, J. Skold, and P. Beming, “3G Evolution HSPA
and LTE for Mobile Broadband,”Academic Press, 2008.
S.-B. Lee et al.,“Proportional fair frequency-domain packet scheduling
for 3GPP LTE uplink,” INFOCOM, 2009, pp. 2611-2615.
[3]
3GPP TS 23.203 V11.1.0, “Policy and charging control architecture,”
Release 11, Mar. 2011.
[4] Proebster M., Mueller C.M. and Bakker H., “Adaptive Fairness Control
for a Proportional Fair LTE Scheduler,” 2010 IEEE 21st International
Symposium on Personal Indoor and Mobile Radio Communications,
Sept. 2010,pp.1504-1509.
[5] Xiaowei Li et al.,“Adaptive PF Scheduling Algorithm in LTE Cellular
System,” 2010 International Conference on Information and
Communication Technology Convergence (ICTC 2010), Nov. 2010,
pp.501-504.
[6] Adibah Mohd Ramli H. et al., “Resource Allocation Technique for
Video Streaming Applications in the LTE System,”19th Annual
Wireless and Optical Communications Conference (WOCC 2010), May
2010, pp. 1-5.
[7] Assaad M., Mourad A., “New Frequency-Time Scheduling Algorithms
for 3GPP/LTE-like OFDMA Air Interface in the Downlink,” IEEE
Vehicular Technology Conference, May 2008, pp.1964-1969.
[8] Delgado O., Jaumard B., “Scheduling and Resource Allocation for
Multiclass Services in LTE Uplink systems,” IEEE 6th International
Conference on Wireless and Mobile Computing, Networking and
Communications (WiMob 2010), Oct. 2010, pp. 355-360.
[9] Xinning Zhu et al., “QoS-Guaranteed Scheduling and Resource
Allocation Algorithm for IEEE 802.16 OFDMA System,” ICC 2008,
May 2008, pp. 3463-3468.
[10] 3GPP TS 36.213 V10.0.1, “Physical Layer Procedures (Release 10) ,”
Jan. 2011.
1449
Download