Aalborg Universitet
Uplink multi-cluster scheduling with MU-MIMO for LTE-advanced with carrier
aggregation
Wang, Hua; Nguyen, Hung Tuan; Rosa, Claudio; Pedersen, Klaus
Published in:
Proceedings of the IEEE Wireless Communications and Networking Conference (WCNC)
DOI (link to publication from Publisher):
10.1109/WCNC.2012.6213960
Publication date:
2012
Document Version
Early version, also known as pre-print
Link to publication from Aalborg University
Citation for published version (APA):
Wang, H., Nguyen, H. T., Rosa, C., & Pedersen, K. (2012). Uplink multi-cluster scheduling with MU-MIMO for
LTE-advanced with carrier aggregation. In Proceedings of the IEEE Wireless Communications and Networking
Conference (WCNC) (pp. 1202 - 1206). IEEE Communications Society. I E E E Wireless Communications and
Networking Conference. Proceedings https://doi.org/10.1109/WCNC.2012.6213960
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
- Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
- You may not further distribute the material or use it for any profit-making activity or commercial gain
- You may freely distribute the URL identifying the publication in the public portal -
Take down policy
If you believe that this document breaches copyright please contact us at vbn@aub.aau.dk providing details, and we will remove access to
the work immediately and investigate your claim.
Uplink Multi-Cluster Scheduling with MU-MIMO
for LTE-Advanced with Carrier Aggregation
Hua Wang
†
, Hung Nguyen
†
, Claudio Rosa
⋄
, and Klaus Pedersen
⋄
†
Radio Access Technology, Aalborg University, Aalborg, Denmark
⋄
Nokia Siemens Networks - Research, Aalborg, Denmark
Email:
†
huw@es.aau.dk
Abstract— LTE-Advanced is the evolutionary path from LTE
Release 8. It is designed to significantly enhance the perfor-
mance of LTE Release 8 in terms of higher peak data rates,
improved system capacity and coverage, and lower latency.
These enhancements allow LTE-Advanced to meet or exceed
the IMT-Advanced requirements and are being considered as
part of LTE Release 10. In this paper, some of the physical
layer enhancement techniques for LTE-Advanced have been
studied including carrier aggregation (CA), uplink multi-cluster
scheduling, and uplink multi-user multiple-input multiple-output
(MU-MIMO) with non-overlapping allocations. A system-level
simulation was conducted to investigate the performance gains
that can be achieved in uplink CA with multi-cluster scheduling
and MU-MIMO. Simulation results show that with proper dif-
ferentiation between power-limited and non-power-limited LTE-
A users, multi-cluster scheduling with CA has similar coverage
performance as in Rel’8, but can achieve substantial gains in
average user throughput compared with Rel’8. MU-MIMO can
further improve the throughput performance, especially when
MU-MIMO is combined with multi-cluster scheduling.
I. INTRODUCTION
Long Term Evolution (LTE) is one of the primary broadband
technologies based on Orthogonal Frequency Division Multi-
plexing (OFDM) for next-generation mobile communication
systems. LTE Release 8 (Rel’8) was finalized in March 2009
providing peak data rates of 300 Mbps in downlink and 75
Mbps in uplink with 20 MHz bandwidth, and allowing flexible
bandwidth operation of up to 20 MHz [1]. To further enhance
the performance, the 3
rd
Generation Partnership Project (3GP-
P) started a new study item in March 2008 on evolving from
LTE towards LTE-Advanced (also known as LTE Release 10),
targeting to meet or exceed the IMT-Advanced requirements
defined by the International Telecommunication Union (ITU),
i.e., peak data rates up to 1 Gbps in downlink and 500 Mbps in
uplink. The study item was closed in March 2010 with a set of
new radio features. These enhancements are being considered
as part of LTE-Advanced in 3GPP Rel’10.
Carrier aggregation (CA) is one of the key features for LTE-
Advanced to support a much wider transmission bandwidth
up to 100 MHz compared with legacy LTE Rel’8, enabling
peak date rates requirement to be satisfied. This is achieved by
aggregating two or more individual component carriers (CCs)
of the same or different bandwidth belonging to contiguous or
non-contiguous frequency bands, subject to spectrum availabil-
ity and the user equipment (UE)’s capability [2]. In addition
to bandwidth extension, multi-cluster transmission has been
proposed to improve the spectral efficiency in uplink. With
multi-cluster transmission, a UE can be allocated to maximum
of 2 clusters not adjacent to each other, so it has higher
scheduling flexibility compared with Single Carrier Frequency
Division Multiple Access (SC-FDMA), which is the multiple
access scheme in LTE uplink. With CA or multi-cluster
transmission, the single carrier property in the uplink is no
longer preserved. As a result, the Peak-to-Average Power Ratio
(PAPR) increases, which will cause an effective reduction of
the maximum UE transmission power [1]. This effect together
with channel estimation loss experienced with multi-CC multi-
cluster transmission might reduce the throughput for cell-
edge users, since they usually experience unfavorable channel
conditions and are limited by the transmission power. There-
fore, the assignment of CCs and clusters to users has to be
carefully designed. LTE-Advanced supports enhanced multi-
user multiple-input multiple-output (MU-MIMO) with non-
overlapping allocations. Up to four layers of quasi-orthogonal
UE-specific reference signals is available for MU-MIMO
enabling co-scheduling of up to four UEs in the same time-
frequency resource, which can provide substantial gains in
sector throughput. Each of these features has been studied
individually in [3]-[5], but to the authors’ knowledge, there
has been no study to evaluate the overall performance by
taking all of them into considerations. The objective of this
paper is to perform a system-level study to investigate the
performance gains that can be achieved in uplink CA with
multi-cluster scheduling and MU-MIMO, both the overall gain
and individual gains under different traffic models.
The rest of the paper is organized as follows. Section II
gives a general overview of CA in LTE-Advanced, as well
as multi-cluster scheduling, power back-off model, coverage
improvement, and MU-MIMO. Section III outlines the simu-
lation methodology and main assumptions. Simulation results
and performance analysis are presented in Section IV. Finally,
some conclusions are drawn in Section V.
II. RADIO RESOURCE MANAGEMENT
The radio resource management (RRM) framework for
multi-CC LTE-Advanced system is illustrated in Fig. 1. Sep-
arate RRM blocks operate independently on each CC. It has
been agreed within 3GPP working group to adopt independent
Link Adaption (LA) and Hybrid ARQ (HARQ) per CC in
coherence with LTE Rel’8 assumptions [2]. Such strategy
$GPLVVLRQFRQWURO
/&&6HOHFWLRQ
/LQNDGDSWDWLRQ
+$54
3+<#&&
/LQNDGDSWDWLRQ
+$54
3+<#&&
/LQNDGDSWDWLRQ
+$54
3+<#&&1
«
3DFNHW VFKHGXOLQJ 3DFNHW VFKHGXOLQJ 3DFNHW VFKHGXOLQJ
3RZHUFRQWURO
3RZHUFRQWURO 3RZHUFRQWURO
Fig. 1. RRM framework of a multi-component carrier LTE-Advanced system
maintains the backward compatibility so that an LTE Rel’8
terminal can work in an LTE-Advanced system. The admission
control module in base station decides whether an incoming
connection should be accepted or not. Then the CC selection
module allocates one or multiple CCs to the incoming user
based on the Quality-of-Service (QoS) requirements, terminal
capability, etc. For the layer 2 packet scheduler, since a
user may be allocated on multiple CCs, the per-CC time
and frequency domain packet scheduler could support joint
scheduling across multiple assigned CCs [4] to achieve better
performance in terms of fairness and coverage. Since the
UEs are limited by the transmission power, power control is
also an important issue in uplink. It is worth mentioning that
admission control, CC-selection, and packet scheduling are not
part of the standard, but are vendor specific.
A. Multi-Cluster Scheduling in Uplink
In downlink, 3GPP uses Orthogonal Frequency Division
Multiple Access (OFDMA) due to its high spectral efficiency
and good performance in link adaptation and frequency do-
main scheduling. But one of the main challenges in OFDMA
is the high PAPR of the transmitted signal [6]. Therefore in
uplink, 3GPP uses SC-FDMA for multiple access. SC-FDMA
uses many of the OFDMA principles to achieve high spectral
efficiency. In SC-FDMA, UEs can only be scheduled on one
set of continuous sub-carriers. In the time domain, only a
single modulation symbol is sent at a time. This allows SC-
FDMA to reach a very low signal PAPR, facilitating efficient
power amplifiers in mobile devices. Since UEs can only be
scheduled on a continuous part of the spectrum, SC-FDMA
has less frequency diversity gain compared to OFDMA.
For LTE-A uplink, additional multiple access schemes have
been proposed which include N × SC-FDMA, also known as
multi-cluster transmission. In multi-cluster transmission, the
minimum resource allocation unit is a sub-band, which consist-
s of integer number of physical resource blocks (PRBs)
1
. Sev-
eral contiguous sub-bands can be seen as a cluster, and a UE
can be allocated to multiple clusters not adjacent to each other.
Multi-cluster transmission has low signal PAPR compared to
1
In LTE, a PRB consists of 12 sub-carriers, each of which is 15kHz, and
is thus equal to 180kHz
UE power
limited?
end
Generate scheduling
metric table
Yes
No
Yes
No
Select UE and sub-band
with highest metric
Cluster number
exceeded?
Allocate current sub-
band
Expand the BW to
adjacent sub-band
Remove UE from
scheduling matrix
Another UE has
higher metric?
UE power
limited?
Any UE or sub-
band left?
Start
Select an adjacent
sub-band
Yes
Yes
No
No
Yes
No
UE Tx bandwidth
expansion
Fig. 2. Proposed multi-cluster scheduling algorithm in uplink LTE-Advanced
OFDMA, and higher scheduling flexibility compared to SC-
FDMA. It can be considered as a compromise of OFDMA and
SC-FDMA. It is worth mentioning that 3GPP only supports
up to 2 clusters per CC for multi-cluster transmission.
B. Resource Allocation and Scheduling
In [7], the authors proposed an adaptive transmission band-
width (ATB) based packet scheduling algorithm for SC-FDMA
in uplink, which tightly couples the bandwidth allocation
and the packet scheduling together to exploit the bandwidth
flexibility. The basic idea behind ATB is to produce an
allocation table which closely follows the envelope of the
UEs’ scheduling metrics. In this study, we extended the ATB
algorithm so that it is catered to multi-cluster transmission. At
each scheduling instance, the algorithm generates a scheduling
matrix of each UE on each sub-band. The algorithm first se-
lects a UE with the highest scheduling metric and then checks
the power constraint of that UE. If it is out of power with
new resource allocation, remove that UE from the candidate
list and restart from the beginning. After the power constraint
check, the algorithm checks the number of allocated clusters.
If the new allocation causes the cluster number to exceed the
maximum number limitation, the current scheduling metric
grid is disabled and the algorithm restarts from the beginning.
Otherwise, the scheduler allocates the current sub-band to
that UE and expands its transmission bandwidth until either
another UE has a higher scheduling metric on the adjacent
sub-band or the maximum transmission power of that UE is
exceeded. Then the algorithm restarts from the beginning and
continues the loop until either all UEs have been scheduled
or there are no resources left. A detailed description of the
proposed multi-cluster scheduling is illustrated in Fig. 2.
C. Maximum Power Reduction for Non-contiguous Alloca-
tions
An important issue in non-contiguous resource allocation
(e.g., multi-CC or multi-cluster transmission) in uplink is the
increased PAPR and other RF related issues. Studies have
shown that the PAPR increases when a UE is transmitting over
non-contiguous allocations simultaneously, which requires a
larger power back-off in the power amplifier together with
other RF transmission requirements, thereby reducing the
maximum transmission power at the UE [8]. Maximum Power
Reduction (MPR) proposed in [9] is used to model the
power reduction at UE due to increased PAPR and other RF
imperfections. Defining a MPR scheme for non-contiguous
resource allocation is challenging because there are many
dimensions in the signal that affect the required back off,
such as the number of clusters, size of clusters, frequency
separation between clusters, etc. In this study, we adopt the
MPR scheme proposed in [9] that can be used for single
and dual CC cases no matter how many clusters are being
allocated. The MPR value (in dB) is calculated based on the
ratio between the allocated PRBs and the aggregated system
transmission bandwidth, specified as follows:
P
MPR
=
6.2 0 < A ≤ 0.05
7 − 16A 0.05 < A ≤ 0.25
3.83 − 3.33A 0.25 < A ≤ 0.4
2.83 − 0.83A 0.4 < A ≤ 1
(1)
where A = N
PRB alloc
/N
PRB agg
is the ratio between the
allocated PRBs and total available PRBs.
Therefore, if a UE is scheduled for transmission only on
one cluster, there is no additional power back-off. Otherwise,
the UE’s maximum power is reduced by P
MPR
in dB.
D. CC Selection and Coverage Improvement
The main difference of LTE-A RRM framework compared
to Rel’8 is the CC-selection functionality which is responsible
for configuring a CC set for each UE based on their QoS
requirements, UE capability, etc. The legacy Rel’8 users can
only be assigned on one CC, while LTE-A users can be
assigned on multiple CCs. The CC-selection functionality is
important to perform load balancing among CCs, as well as
to optimize the system performance.
In downlink, allocating more CCs to an LTE-A user general-
ly results in a higher throughput thanks to the larger transmis-
sion bandwidth and higher transmission power. However, this
is not always the case in uplink. The main difference between
uplink and downlink is the transmission power constraint of
a UE. For power limited cell edge users, even if they are
assigned on multiple CCs, they do not have sufficient power to
exploit the increased transmission bandwidth. Furthermore, the
impact of increased PAPR will introduce additional reduction
of maximum UE transmission power as mentioned previously.
For power limited cell edge users transmitting at (or close to)
maximum transmission power, such cost might counterbalance
the gain brought by multi-CC multi-cluster transmission, and
even results in a coverage loss compared to the case where
the SC-FDMA properties of the transmitted signals are main-
tained (single-CC single-cluster assignment). Therefore, multi-
CC multi-cluster transmission shall generally be restricted to
users with good channel conditions. An effective pathloss-
threshold based CC-selection algorithm was proposed in [5]
to distinguish between power-limited and non-power-limited
LTE-A users. The derived path loss threshold is:
L
threshold
= L
95%
−
10 log
10
(K) + P
MPR
α
(2)
where L
95%
is the estimated 95-percentile user path loss, K
is the total number of CCs, α is the path loss compensation
factor, and P
MPR
is the average MPR.
LTE-A users whose path loss is lower than the threshold are
considered to be power-limited and are assigned on single-CC
and single-cluster, otherwise they are considered to be non-
power-limited and can be assigned on multi-CC and multi-
cluster. By doing so, cell-edge users will not experience any
performance loss from being scheduled over multiple CCs
and clusters, while non-power-limited LTE-A users can benefit
from the advantages of CA and multi-cluster scheduling. For
Rel’8 users, they are assigned on one CC with least load for
load balancing and single-cluster.
E. Multi-user MIMO
MU UEs selection and scheduling are based on the SU
UE scheduled list described in Section II-C and II-B. Only
scheduled UEs which have the estimated SINR larger than
a pairing threshold are selected into a primary MU-MIMO
UE list. This condition is applied to avoid the scheduling of
UEs at the cell-edge in MU-MIMO mode. For each primary
MU-MIMO UE, a pairing MU-MIMO UE is found from the
candidate list Ω
C
. The primary UEs and the UEs scheduled
for retransmission are excluded from the candidate list. The
pairing UE which has the highest MU-MIMO metric will
be scheduled for MU-MIMO transmission together with the
primary UE.
UE
pair
i
= arg max
k∈Ω
C
β
k
· (1 − γ
k,i
) (3)
where β
k
is the scheduling metric of pairing UE k, and γ
k,i
is
the cross correlation between the precoder of the primary UE
i and the pairing UE k. It should be noted here that, both the
user priority and the UE mutual information are considered in
this MU-MIMO scheduling metric.
Since the MU-MIMO transmissions are carried out from
two different UEs, no power splitting or power sharing as
in DL MU-MIMO transmission is required. The selection and
scheduling of MU-MIMO UEs are processed until all resource
is used or all UEs are scheduled.
III. SIMULATION ASSUMPTIONS
The performance evaluation is based on a detailed multi-
cell system level simulator which follows the guidelines in
[10]. The simulation scenario is 3GPP Macro-cell case #1
with 7 sites and 3 sectors per site using the wrap-around
technique. Spatial Channel Model (SCM) and 3D antenna
pattern with default tilt of 15 degrees are used. Two contiguous
CCs, each with 20 MHz bandwidth, are configured to form
a wide band of 40 MHz. On each CC, up to two clusters
are supported for uplink transmission. Imperfect Sounding
Reference Signal (SRS) measurements and channel estimation
Parameters Settings
Propagation scenario 3GPP Macro case #1
Layout 7 sites - 3 sectors/site - wrap around
Channel profile SCM channel model with 3D antenna
Component carriers 2 × 20 MHz contiguous @ 2GHz band
96 available PRBs per CC
Number of clusters 1 or 2 clusters
PRBs per sub-band 2 PRBs
Sounding resolution 2 PRBs
Sounding period 10 tti
Sounding method Imperfect SRS with channel estimation error
eNode-B receiver 4-Rx MMSE
UE Tx bandwidth ATB for multi-cluster
Packet scheduling Throughput based Proportional Fair
Bursty traffic model Finite buffer with Poisson arrival
Fixed file size of 2 Mbits per UE
Offered load: [5 : 5 : 45] Mbps
Full-buf. traffic model Full buffer with fixed number of UEs
Available MCSs BPSK (R=1/5,1/3)
QPSK (R=1/4,1/3,1/2,2/3,3/4)
16QAM (R=1/2,2/3,3/4,5/6)
Max UE power 200 mW [23 dBm]
HARQ Synchronous and adaptive
BLER target 20%
Link adaptation Fast AMC
α 0.6
Power spectral density Load Adaptive Power Control
Average MPR P
MPR
6 dB
TABLE I
SUMMARY OF MAIN SIMULATI ON PARAMETERS
error are assumed. The link to system level mapping is
based on the actual value interface (AVI) method [11]. It is
assumed that distance-dependent path loss and shadowing are
maintained constant for each UE, but fast fading is updated
every TTI independently on each CC based on the ITU Typical
Urban power delay profile and UEs’ speed. Both full buffer
and bursty traffic models are considered. In full-buffer traffic
model, we assume each sector has a fixed number of UEs
with full buffer. In bursty traffic model, each call in a sector
follows a Poisson arrival process with a finite buffer of 2
Mbits payload. The offered load per cell can be obtained
by multiplying the user arrival rate with the payload size.
Proportional fair scheduling in frequency domain is used
together with multi-cluster bandwidth allocation. Same open
loop power control settings are applied on each CC, while
independent Load Adaptive Power Control (LAPC) [12] is
enabled on each CC to dynamically update the UE power
spectral density P
0
based on the variable load conditions.
The average UE power back-off P
MPR
is set to be 6 dB
(quite ”aggressive” setting when considering the power back-
off model in eqn.1). Table I summarizes the main parameter
settings used in the system-level simulations.
IV. SIMULATION RESULTS
We start our analysis by first looking at the scenario with
fixed number of UEs per sector and full-buffer traffic model.
Only one CC is configured with 10 MHz bandwidth in this
scenario. Fig. 3 shows the average throughput gain versus
different number of UEs per sector. The reference scenario
is single cluster scheduling without MU-MIMO. The gain of
1 2 3 4 5 6 7 8 9 10
0
10
20
30
40
50
60
Number of UE
Average throughput gain [%]
dual−cluster gain over single−cluster
MU−MIMO gain over single−cluster
overall gain over single−cluster
Fig. 3. Average user throughput gain versus different number of UEs per
sector, 1 × 10 MHz with full-buffer traffic model
dual-cluster scheduling over single cluster scheduling increas-
es as the number of UEs increases until reaching the maximum
value at certain point, i.e., 6 UEs per sector in our case. Then
the dual-cluster scheduling gain gradually decreases as the
number of UEs increases. That is because when the number of
UEs per sector is low, with dual-cluster scheduling UEs have
more chance to exploit frequency diversity than single cluster.
But when the number of UEs per sector is high, the gain
brought by dual-cluster scheduling is decreasing due to multi-
user diversity gain. MU-MIMO gain increases monotonically
as the number of UEs increases due to the reason that higher
number of UEs will bring higher multi-user diversity. When
MU-MIMO is combined with dual-cluster scheduling, the
average throughput gain can be up to 56% with 10 UEs per
sector compared with single cluster scheduling without MU-
MIMO. It is much higher than the sum of gains brought by
dual-cluster scheduling alone and MU-MIMO alone because
dual-cluster scheduling allows to fully exploit the gain of
MU-MIMO. Therefore it is recommended that dual-cluster
scheduling is used in combination with MU-MIMO.
Next we evaluate the performance of multi-cluster schedul-
ing with MU-MIMO and CA in bursty traffic model. Two
CCs, each with 20 MHz bandwidth, are configured. Fig. 4
shows the cell edge user throughput versus the offered load
in different scenarios. It is shown that the coverage of LTE-
A UEs is almost the same as that of Rel’8 UEs. In other
words, there is no gain in coverage by applying multi-cluster
scheduling or CA in uplink. That is because at the cell edge,
UEs usually experience high path loss and are limited by the
maximum transmission power. In fact, with LAPC deployed,
cell edge UEs are configured to transmit with maximum power.
Even if those cell edge LTE-A UEs are assigned to multiple
clusters and CCs, they do not have sufficient power to exploit
the increased transmission bandwidth, and may even result in
a performance loss due to the effect of MPR. Therefore, cell
edge LTE-A UEs are assigned on only one cluster and one
