You are on page 1of 6

2014 7th International Symposium on Telecommunications (IST'2014)

Multi-user MIMO in LTE:


An exclusive low complex QRM-MLD detector
with 2×2 antenna configuration

Hamid Rajabi, Student Member, IEEE Naaser Neda, Assistant Professor


Faculty of Electrical and Computer engineering Faculty of Electrical and Computer engineering
Birjand university Birjand university
Birjand, Iran Birjand, Iran
rajabi.hm@gmail.com nneda@birjand.ac.ir

Abstract—Multi-user multiple input and multiple output (MU- receiver sides (MIMO channel) [3], which makes the system
MIMO) is one of the key techniques in 3GPP Long-Term resilient against the pernicious factors in wireless channel like
Evolution (LTE), which enhances the overall system capacity in shadowing, fading, noise and interference [4]. However, one of
the downlink. In order to reduce the effect of inter-user- the key technologies in MIMO is Multi-user MIMO (MU-
interference in this system, it is necessary to utilize an efficient MIMO); a popular technique which is selected for 4G such as
detection method at the user equipment (UE) that besides its LTE-based systems [5]. Applying this technique in the
feasibility at high data rate transmission systems such as LTE, downlink, make it possible to transmit data to multiple users
must have a low computational complexity to be implemented which are co-scheduled at the same time-frequency resources.
into a small-sized UE. This paper proposes a new low complex
So obviously, the MU-MIMO will enhance the overall system
detector for 2×2 MU-MIMO receiver in LTE downlink
capacity. Note that in cause of supporting and scheduling
exclusively, which is based on the QRM-MLD algorithm. We will
see that this presented method achieves an identical BER
multiple users at one Resource Block (RB), the inter-user-
performance to that of the conventional QRM-MLD, however interference will be occurred inevitably. Therefore, in addition
leads to at least 75% reduction in the complexity with respect to to apply an appropriate precoding strategy and user-scheduling
the conventional method, and furthermore, it avoids the technique at the eNB (enhanced Node B; LTE notation for the
unnecessary actions those exist in QRM-MLD while detecting base station), it is also necessary to utilize efficient detection
each co-scheduled UE’s signal in MU-MIMO LTE. Hence, it methods at the UE (LTE notation for user equipment) to extract
would be preferred from hardware implementation point of view. correctly the corresponding data for each user in the downlink.

Keywords— MU-MIMO, LTE, QRM-MLD, complexity. Signal detection at the UE constitutes an essential process
in MU-MIMO transmission system at the downlink. In general,
I. INTRODUCTION maximum likelihood detection (MLD) method has the highest
performance but its computational complexity for MU-MIMO
The evolution process in wireless communications is
is exponentially increased by the number of transmit antennas,
motivated by some agents such as increase in demand for
the signal constellation size and the number of co-scheduled
advanced services, or increasing competence among vendors
UEs. To overcome the high complexity of the MLD, several
and operators, and so on. So, significant improvements really suboptimal detection strategies have been investigated: ML-
would be inevitable for these systems. These improvements are DFE, and Maximum Likelihood Detection with QR
including a very high data rate transmission with high quality Decomposition and M algorithm (QRM-MLD) [6]-[8].
multimedia services. In this regard, the International
Telecommunication Union (ITU) started to define the 4th The QRM-MLD is one of the suboptimal methods that is
generation of wireless communication systems (4G) in 2005, applicable in high data rate of transmissions as expected for
which its most prominent feature is high data rate of LTE and LTE-Advanced [9]. But, in case of MU-MIMO LTE,
transmission (from 100Mbps up to 1Gbps) [1]. The first steps it has a significant complexity, so it is not a suitable method for
to achieve the defined objectives for 4G were taken in 2008 by such a system which needs to implement into a small-sized UE.
3rd Generation Partnership Project (3GPP) through Furthermore, it has unnecessary operations at detecting each
standardization of the Long-Term Evolution (LTE) [2]. This UE’s desired data. To avoid these unnecessary actions in MU-
standard can be considered as an evolutionary path between the MIMO, there is an efficient detector: the Interference-aware
3G and the 4G wireless systems which is based on MIMO- (IA) receiver [10]. However, the IA receiver has still a high
OFDM technology. computational complexity, too. So these challenges instigated
us to present a novel detector; a low complex detector for 2×2
Clearly, one of the most important techniques to approach MU-MIMO system exclusively, which is based on the QRM-
the specified objectives for the 4G (the high bit rate and high MLD algorithm. While this detector shows a similar BER
throughput) is using multiple antennas at both transmitter and performance to that of the conventional QRM-MLD, but it
This research is sponsored by “Iran Telecommunication Research Center
(www.itrc.ac.ir)” with the contract number: ‫ت‬/11312/500.

978-1-4799-5359-2/14/$31.00 ©2014 IEEE

1216
represents a very low complexity and removes the unnecessary Gaussian noises of variance at each receive antenna. The
operations. Therefore, it will be beneficial from hardware parameter “ ” denotes second entry of precoders in (1) i.e.
implementation point of view just as it is specified for LTE and
LTE-Advanced detectors. 1, (3)
Regarding notations, we will use . , and . ,I for real and
imaginary parts of a complex number respectively. |. | and . Selection of the precoder for each UE would ensure
indicate norm of scalar and vector while . , . and . maximization of its desired signal strength, that is |
indicate transpose, conjugate and conjugate transpose | for the first receive antenna of each UE, while selection
respectively. of the UE pairs according to the above scheduling, would
ensure minimization of the interference strength seen by each
The rest of this paper is organized as follows. We have a UE, that is | | for the first receive antenna of each
glimpse on the MU-MIMO LTE system in section II. In section UE. Though this precoding and scheduling strategy would
III, the conventional QRM-MLD technique for 2×2 MU- ensure minimization of the interference, the residual
MIMO LTE is described. We present the novel signal detector interference would still be significant. Therefore UEs need to
with low complexity for 2×2 MU-MIMO LTE in section IV. In employ an efficient detection method in their receivers.
section V, we compare the computational complexity and show
its reduction. Besides, computer simulations are conducted to III. CONVENTIONAL QRM-MLD DETECTOR FOR 2×2 MU-
show the BER similarity between the conventional and the MIMO LTE
proposed detector. Finally in section VI, conclusion is In general, for a 2×2 MU-MIMO system with 2 co-
remarked. scheduled UEs at same REs, the received signal is expressed
II. 2×2 MU-MIMO IN LTE: A QUICK REVIEW ON SYSTEM as:
MODEL
(4)
MU-MIMO system has the capability of data transmission
simultaneously and in the same frequency resources for
multiple users. In other words, in this system a linear Where denotes the received signal vector by both antennas of
combination of multiple streams for multiple UEs transmitted one of the 2 co-scheduled UEs: ; and are
at each Resource Element (RE). Clearly, using this strategy, the received signals by first and second antennas respectively.
overall system capacity will be increased. However, in order to is an 2×2 effective channel matrix that is the product of
minimize the inter-user-interference in MU-MIMO system, it is the 2×2 channel matrix and the 2×2 precoding matrix, is a
necessary to utilize the precoding and scheduling at the eNB. In 2×1 transmitted complex modulated symbols vector:
LTE standard, this operation is done in a closed-loop scheme , and denotes the circularly symmetric white
with channel state information (CSI) feedback system to the Gaussian noise vector: that each entry of this
eNB and according to the codebooks. For example, if there are vector has variance . So, (4) can be represented as:
2 transmit antennas at the eNB, one of four precoders in (1) is
employed [11].
(5)
1 1 1 1
, , , ; √ 1 (1)
√ 1 √ 1 √ √

In the QRM-MLD detector, the 2×2 effective channel


Here, we consider 2 UEs with 2 receive antennas in the
downlink that have been scheduled at the same REs. In order to matrix is decomposed into an 2×2 unitary matrix and
minimize the inter-user-interference, these 2 co-scheduled UEs an 2×2 upper-triangular matrix , in the form of (6),
have to be selected as their requested precoders must have
minimum cross-correlation. So here, according to a sample of (6)
LTE codebook in (1), it’s only adequate that the second entries 0
of two co-scheduled UE’s precoders must be mirror of each
other. Therefore, with a 2×2 MU-MIMO system for each UE in Multiplying both sides of (4) with , (7) is obtained, i.e.
the downlink, the received signal at each RE is in the form (2),
i.e. (7)

1 1 Where , . Then the Maximum Likelihood


(2)
(ML) metric can be expressed as:

Where and are the received symbols by each UE at its


first and second antennas respectively. , , and 0
are channel coefficients between transmit antennas and the | |
UE’s receive antennas, and are transmitted complex
| | (8)
modulated symbols for UE-1 and 2 respectively and and
are the zero mean circularly symmetric complex white

1217
Applying the M algorithm to (8), we select a candidate set conventional QRM-MLD detector, with decomposition of
for the transmitted signal vector . After the candidate and multiplying both sides of (4) with , we have:
set selection, signal vector that minimizes the ML metric is
chosen as the best estimate of the transmitted signal vector. , , ̃
Hence, here we have two steps: 0 (11)
, ̃
1) First step: First, we describe the candidate set selection
method for . Using (9), the ML metric values are calculated Also we have two steps here:
for all constellation points; thus, |C| times metric calculations 1) First step: Regarding (11) and using (12) and (13), we
are performed, where |C| is the constellation size. draw and respectively, then according to the location of
each one in the constellation, choose M candidates by
| | (9) selecting one of the constellation regions in Fig. 1 that
contains and separately.
Among |C| candidates for , we select M of them that
correspond to M smallest values of (9). Let , denotes (12)
,
these M selected values and then , is one of them.
2) Second step: In this step, among M × |C| vectors for ,
(13)
,
, , we select M of them that correspond to M
smallest values of (10), i.e. In Fig. 1, we see a 16-QAM constellation but all other
constellations can be handled in a similar way. The
(10) constellation points are denoted as 1, 2, … , 16. When M = 4,
, ,
nine regions are constructed from the points. For example, if
or are located in region “a”, the candidate set is {1, 2, 5, 6}.
Let , , denotes these M selected vectors. Finally
through these M candidate vectors, just one of them that By selecting a candidate set in the proposed method, we
minimizes the metric in (8) will be selected as the final can avoid the computation in (9) and sorting the metric values,
estimated transmitted vector for both UE-1&2 which received thereby reducing the complexity. Using the proposed candidate
by one of the UEs. set method, the candidate set identical to the conventional
method is selected. This is because that the simple division
The performance of QRM-MLD depends on the parameter shows an identical performance to that of the MLD signal
M. As M increases, its performance approaches MLD detector in flat fading SISO (single input and single output)
performance at the sacrifice of the complexity. Regardless of channels.
SNR and channel condition number, the complexity of QRM-
MLD is fixed for the given value of M, which makes its 2) Second step: In this step, let’s consider the ML metric
hardware implementation simpler. as:
Although the conventional QRM-MLD is less-complex
than the MLD, it still requires high computations from , , , (14)
perspective of hardware implementation in small-sized UEs
those exist in LTE and other upper standards, furthermore in After extension (14) and using the relation | | | |
this method, each UE while detecting its own signal, also has | | 2 we have:
to detect its co-scheduled UE’s signal; we know this is an
unnecessary action. In the next section, based on the QRM- | | | | , , ,
MLD, we present a proposal of a novel signal detector for 2×2
MU-MIMO LTE system which requires much low complexity 2 , 2 , 2 , (15)
and resolves this problem while it has a BER performance near
to the QRM-MLD. Where,

IV. LOW COMPLEX QRM-MLD FOR 2×2 MU-MIMO LTE , , , , , , ,


Here, we propose a new detector for downlink 2×2 MU-
MIMO LTE which based on QRM-MLD and however it is , , , , , , , (16)
less-complex with respect to the QRM-MLD with a similar
error performance to it. Besides, in this method, each UE | | and | | can be ignored as they are independent of the
without detecting the signal of its co-scheduled UE, directly minimization operation. If we consider the UE-1 as the desired
extract only its own signal. UE, then obtaining , and , is an unnecessary action that
A. Instruction UE-1doesn’t need to do it. To resolve this problem, we operate
We use an 2×2 antenna configuration for the eNB and each as follows.
UE and suppose 2 UEs scheduling at same REs. Similar to the

1218
selection in the second step shows the same performance. So,
the error performance for can be expressed as:

| ,
| , 1 (18)

Where is the probability that the transmitted , is


included in the candidate set, is the error probability for
, | , is the probability that , is included in the
candidate set but an error occurs, and | , is the
probability that , is not included in the candidate set and
an error occur. The probabilities | , and
| , depend only on the given channel, not on the
signal detection methods. Therefore the two detection methods
Fig. 1. Selecting the candidate set in the first step of the proposed method for at the same channel condition, have an identical error
16-QAM modulation.
performance if the probability is the same. This can be
expressed as:
The values of , and , which minimize (15) need to be in
the opposite directions of and respectively, thereby
| ,
avoiding search on the alphabets of and reducing one
complex dimension in the detection, then the reference | , 1 (19)
optimization problem in this method is expressed as:
In deriving (19), we know | , 1. After studying
(19), we can see that the probability that , is included in the
, , , , , ,
candidate set is the sum of the probability that , is
, , , , , , included in the candidate set and the , is included in the
2 , , , 2 , , , 2| | , candidate set; and the sum of the probability that , is not
2| | , (17) included in the candidate set and , is included in the
candidate set. Since the two methods choose the same
Depending on drawn from (12), be in which one of the candidate set for , , is the same for the two methods.
regions in Fig. 1, we have different possible sets for The possible performance difference can only be caused by the
, , , ; if is located in one of the regions “a”, “c”, “g” | , values in the two methods. However, we
and “i”, we have 4 possible sets {[1,1],[1,3],[3,1],[3,3]} for expect that the probability that , is included in the
, , , . similarly, if is located in region “b” or “h”, candidate set is very low when , is not included in the
we have {[1,1],[1,3]}, in region “d” or “f”, we have candidate set for both methods. Consequently, the QRM-MLD
{[1,1],[3,1]}, and finally if is located in region “e”, we have method and the proposed low complex method show an
only 1 possible set {[1,1]} for , , , . Finally, identical error performance.
considering the M point for drawn from (13), and one of the
V. COMPLEXITY AND BER PERFORMANCE COMPARISONS
above possible sets for real and imaginary parts of , we
calculate the metric in (17) and choose the value of that In this section, we compare the proposed detector with the
correspond to the smallest metric. Hence, this is the detection conventional QRM-MLD method, from the perspective of
process only for the transmitted symbol of UE-1. So as we see, computational complexity and Bit Error Rate.
without detecting the exact value of , and just using the A. Complexity
absolute real and imaginary part of its location in the
constellation, we have a significant reduction in computational The complexity of the conventional QRM-MLD method for
complexity for detecting symbols of the desired user; UE-1. 2×2 MU-MIMO in LTE with 2 co-scheduled UEs, is described
in TABLE I; at each step, the metric calculations need to be
Note that increasing each one of the parameters: performed. In the first step, the metric calculations involving
constellation size |C|, the number of co-scheduled UEs, the are performed for each constellation point; we assumed that
parameter M and the number of transmit antennas, will increase the constellation points are composed of an integer real part
the computational complexity. While increasing the parameter and an integer imaginary part, thus, the complexity is 2|C| real
M, enhances the accuracy of detection, on the other hand, by multiplications. In the second step, M |C| branches need to be
increasing the complexity, will approach the performance of considered. Here, we need to calculate an additional branch
this detection method to the MLD. metric that involves two real multiplications. Therefore, the
B. Error performance total complexity is 2(M + 1) |C| real multiplications.
The M selected candidates for in both methods are The complexity of the proposed detector is described in
similar (the first step of each method). Therefore, the two TABLE II. In the first step, the complexity is “0” because the
methods show a similar error performance if the candidate branch metrics are not calculated: furthermore, the division by

1219
and was not counted because the region boundaries in TABLE I. COMPLEXITY OF THE CONVENTIONAL QRM-MLD FOR 2×2
MU-MIMO LTE WITH 2 CO-SCHEDULED UES.
Fig. 1 only need to be multiplied by and and these two
values are real numbers. In the second step, note that using step operation complexity
simulation
(17), the maximum number of branches for is |C| / 4 and the time
exact number of them for is M. So, there are M |C| / 4 1 (9) 2| |
2 (10) 2 | | 561.43 (s)
branches (maximum) instead of M |C| branches. Therefore, the Total 2 1 | |
total complexity of the proposed method is up to 2×M |C| / 4
real multiplications. TABLE II. COMPLEXITY OF THE PROPOSED LOW COMPLEX QRM-MLD
FOR 2×2 MU-MIMO LTE WITH 2 CO-SCHEDULED UES.
In TABLE III, numerical results of complexity comparison
between the conventional and the proposed method for various step operation complexity
simulation
values of M, are listed. This comparison has been done under time
the highest order of modulation (64-QAM). For instance, Selecting
1 and from 0
when M = 9 the complexity of the conventional QRM-MLD (12) , (13)
and the proposed method are 1280 and 288 (maximum) real 2 | |⁄4 103.89 (s)
multiplications, respectively. So in this case, our proposed 2 (17)
(maximum)
detector results in 77.5% reduction in the complexity. For M = 2 | |⁄4
Total
64 (maximum possible value of M), note that because of M = (maximum)
|C|, the first steps of both methods will be absurd. In this case,
the conventional method acts as MLD however the complexity TABLE III. NUMERICAL COMPLEXITY COMPARISON BETWEEN TWO
METHODS, FOR 64-QAM MODULATION AND VARIOUS VALUES OF M.
of the proposed method is still low and the reduction ratio is
fixed on about 75%, too. So in general, we see at least 75% (col. 1) (col. 2) (col. 3) (col. 4)
complexity reduction even for high values of M. proposed
conventional method complexity
B. Bit Error Rate QRM-MLD (maximum reduction ratio
possible .
The error performance of the detection methods is shown in complexity)
( %)
.
Fig. 2 and simulation parameters are listed in TABLE IV. As it
is predicted, the MLD shows the best performance and M=4 640 128 80%
increasing the value of M for both QRM-MLD and proposed M=9 1280 288 77.5%
method, will force them to approach MLD. In summary, based
on the comparison of the complexity and error performances in M = 16 2176 512 76.4%
Fig. 2, we can conclude that the proposed detector can be the
preferred choice over the conventional QRM-MLD for 2×2 M = 25 3328 800 75.9%
MU-MIMO because it achieves a similar error performance to M = 36 4736 1152 75.6%
that of the conventional method while requiring much less
computational complexity. M = 49 6400 1568 75.5%
8192 (acts as
VI. CONCLUSION M = 64 MLD)
2048 75.3%
MU-MIMO system is very likely to be employed in the
fourth generation (4G) of mobile telecommunication systems
and its related standards such as 3GPP LTE, to increase the TABLE IV. SIMULATION PARAMETERS
overall system capacity. In this paper, we proposed a novel Parameter value
detector for 2×2 MU-MIMO LTE system which is based on the
QRM-MLD. Our observations show that the proposed method Transmission BW 11 MHz
for MU-MIMO, achieves a similar BER performance to that of
the conventional QRM-MLD, however this method eliminates Subcarrier spacing 15 kHz
the unnecessary operation in detecting the desired UE’s signal
so that each UE directly extract its own data. Furthermore, the FFT size 1024
computational complexity in this detector is significantly modulation 16-QAM
reduced with respect to the conventional method; from 75% up
to 80% reduction for various values of M. we see that even in Number of Tx antennas 2
high values of M, its complexity reduction ratio stays at about
Number of Rx antennas 2
75%. Thus this MU-MIMO detector is really appropriate to
implement into a small-sized mobile; such a mobile units that Number of co-scheduled UEs 2
is expected to utilize in LTE and LTE-Advanced.
Channel model flat fading
ACKNOWLEDGMENT
Channel estimation Perfect
H. Rajabi and N. Neda thank “Iran Telecommunication
Research Center (www.itrc.ac.ir)” for sponsoring this research
with contract number: ‫ت‬/11312/500.

1220
[2] http://www.3gpp.org.
[3] I. E. Telatar, “Capacity of multi-antenna Gaussian channels,” European
10
0
Trans. on Telecommu., vol. 10, no. 6, pp. 585–595, 1995.
[4] G. Foschini and M. J. Gans, “On the limits of wireless communications
in a fading environment when using multiple atnennas,”Wireless
10
-1 Personal Communications, vol. 6, no. 3, pp.311–355, 1998.
[5] J. Duplicy, B. Badic, R. Balraj, R. Ghaffar, “MU-MIMO in LTE
Systems,” EURASIP Journal on Wireless Commu. and Net., vol. 2011,
Article ID 496763, March 2011.
Bit Error Rate (BER)

-2
10
[6] W. Choi, R. Negi, and J.M. Cioffi, “Combined ML and DFE decoding
for the V-BLAST system,” in Proc. IEEE ICC, pp. 1243-1248, June
2000.
-3
10 [7] J. Yue, K. J. Kim, J. D. Gibson, and R. A. Iltis, “Channel Estimation and
Data Detection for MIMO-OFDM Systems,” in Proc. GLOBECOM
MLD 2003, pp. 581-585, 2003.
10
-4 QRM-MLD , M = 9 [8] S. Nagayama and T. Hattori, “A proposal of QRM-MLD for Reduced
Proposed detector , M = 9 Complexity of MLD to detect MIMO signals in Fading Environment,”
QRM-MLD , M = 4
Vehicular Technology Conference, VTC-2006 Fall, June 2006.
Proposed detector , M = 4
10
-5 [9] H. Kawai, K. Higuichi, N. Maeda, M. Sawahashi, T. Ito, Y. Kakura, A.
0 2 4 6 8 10 12 14 16 18 20 Ushirokawa, and H. Seki, “Likelihood Function for QRM-MLD suitable
Fig. 2. BER performance comparison of the MLD, conventional QRM-MLD for soft-decision turbo decoding and its performance for OFCDM
and the proposed low complex QRM-MLD for M = 4, 9 and 16-QAM MIMO multiplexing in multipath fading channel,” IEICE Trans.
modulation. Commun., vol. E88-B, no. 1, pp. 47-57, Jan. 2005.
[10] R. Ghaffar, R. Knopp, “Interference-Aware Receiver Structure for
Multi-User MIMO and LTE,” EURASIP Journal on Wireless Commu.
REFERENCES and Net., vol. 2011, July 2011.
[1] D. Sacristian, J. F. Monserrat, J. Penuelas, D. Calabuig, S. Garrigas, and [11] J. Lee, Jin-Kyu Han, and J. (Charlie) Zhang, “MIMO Technologies in
N. Cardona, “On the Way towards Fourth-Generation Mobile: 3GPP 3GPP LTE and LTE-Advanced,“ EURASIP Journal on Wireless
LTE and LTE-Advanced,” EURASIP Journal on Wireless Commu. and Commu. and Net., vol. 2009, Article ID 302092, May 2009.
Net., vol. 2009, Article ID 354089, June 2009.

1221

You might also like