)
Improvement in Multiuser MIMO channel
I. Introduction
MIMO( multi input multi output) channels
project(3GPP) and long term evolution(LTE) systems. The main aim of multiuser MIMO system
is to suppress the multiuser interference and enhance signal to interference and noise ratio
(SINR). The well known scheme, block diagonalization
interference but at the same time it also reduces the received power.
Eigen valued Based Beamforming
do nothing in the interference rejection. So this project is
Value Based method of precoding
Fig: 1 System model for Multiuser
(EBB) Beamforming precoding design for downlink Capacity
Improvement in Multiuser MIMO channel.
Krishna Ram Budhathoki
MIMO( multi input multi output) channels have been extensively used in 3
rd
generation partner
project(3GPP) and long term evolution(LTE) systems. The main aim of multiuser MIMO system
is to suppress the multiuser interference and enhance signal to interference and noise ratio
known scheme, block diagonalization (BD) could eliminate the multiuser
interference but at the same time it also reduces the received power. On the other side, there is
eamforming (EBB) that would maximizes the received power and would
the interference rejection. So this project is mainly aimed to focus
coding to have the desired received power and signal to noise ratio.
System model for Multiuser MIMO downlink channel
downlink Capacity
generation partner
project(3GPP) and long term evolution(LTE) systems. The main aim of multiuser MIMO system
is to suppress the multiuser interference and enhance signal to interference and noise ratio
(BD) could eliminate the multiuser
On the other side, there is
(EBB) that would maximizes the received power and would
mainly aimed to focus on the Eigen
to have the desired received power and signal to noise ratio.
II. System Description
Throughout this paper, the following notations will be used.
quantities, bold lowercase letters indicate vectors, and boldface uppercase letters designa
matrices; ( (( ( ) )) ) ( (( ( ) )) )
T †
. , . and . denote determinant, transpose, and
Fig.1 shows the system model for the MIMO downlink channel. The incoming stream of data is
denoted by ‘s’ which would be pre
for the transmission.
MIMO (Orthogonal Frequency Division Multiplexing)
T
n antennas and K users with n
{ }
,1 ,
,...
R R K T
n n n × matrix. So for each user there would be separate channel, i.e for user1 the
channel would be
,1
×
R T
n n , for user 2
on.
Determination /Estimation of channel matrix
In this paper we have assumed Channel State Information (CSI) at the transmitter side.
refers to known channel properties of
signal propagates from the transmitter to the receiver and represents the combined effect of, for
example, scattering, fading, and power decay with distance. The CSI makes it possible to adapt
transmissions to current channel conditions, which is crucial for achieving
communication with high data rates
CSI needs to be estimated at the receiver and usually
Since the channel conditions vary, instantaneous CSI needs to be
A popular approach is socalled training sequence (or pilot sequence), where a known signal is
transmitted and the channel matrix
transmitted and received signal.
Let the training sequence be denoted
channel as
By combining the received training signals
becomes
Throughout this paper, the following notations will be used. Normal letters represent scalar
quantities, bold lowercase letters indicate vectors, and boldface uppercase letters designa
denote determinant, transpose, and Hermitian operators, respectively.
Fig.1 shows the system model for the MIMO downlink channel. The incoming stream of data is
’ which would be precoded by precoder matrix ‘M’ and is sent to Base Station
(Orthogonal Frequency Division Multiplexing) OFDM system composed of a BS with
, R k
n antennas for each user. So such a channel is represented by
So for each user there would be separate channel, i.e for user1 the
, for user 2
nd
there would be separate channel matrix
of channel matrix H.
In this paper we have assumed Channel State Information (CSI) at the transmitter side.
refers to known channel properties of a communication link. This information describes how a
from the transmitter to the receiver and represents the combined effect of, for
, and power decay with distance. The CSI makes it possible to adapt
sions to current channel conditions, which is crucial for achieving
data rates in multantenna systems.
CSI needs to be estimated at the receiver and usually quantized and fed back to the transmitter
conditions vary, instantaneous CSI needs to be estimated on a short
called training sequence (or pilot sequence), where a known signal is
transmitted and the channel matrix is estimated using the combined knowledge of
Let the training sequence be denoted , where the vector is transmitted over the
By combining the received training signals for , the total training
Normal letters represent scalar
quantities, bold lowercase letters indicate vectors, and boldface uppercase letters designate
Hermitian operators, respectively.
Fig.1 shows the system model for the MIMO downlink channel. The incoming stream of data is
’ and is sent to Base Station
OFDM system composed of a BS with
antennas for each user. So such a channel is represented by
So for each user there would be separate channel, i.e for user1 the
there would be separate channel matrix
,2
×
R T
n n and so
In this paper we have assumed Channel State Information (CSI) at the transmitter side. (CSI)
This information describes how a
from the transmitter to the receiver and represents the combined effect of, for
, and power decay with distance. The CSI makes it possible to adapt
sions to current channel conditions, which is crucial for achieving reliable
to the transmitter.
on a shortterm basis.
called training sequence (or pilot sequence), where a known signal is
is estimated using the combined knowledge of the
is transmitted over the
, the total training signaling
with the training matrix
channel estimation means that
Now in our precoder design, let
denotes precoding matrix with
symbol ‘s’ with
†
v
E ss I ( =
¸ ¸
.
Then the output could be represented
z y HMs = + = + = + = +
where
1 2
...
T T T
K
y y y y ( =
¸ ¸
K is the number of users.
and
,
1 ×
∈
R k
n
y c is the received vector of user k and
noise(AWGN) with
† 2
σ ( =
¸ ¸
n
E zz I
so we can write ,
1 2
.. ( =
¸ ¸
T
T T T
K
H H H H
1 2
.. ( =
¸ ¸
T T T
K
M M M M
where ( ) 1≤ ≤
k
H k K and (1
k
M k K
user k.
The capacity of user k as obtained by [3]
So the sum capacity of the whole system is given by
( )
1
1
=
= ≤ ≤
∑
k
K
k
C C k K
(
† †
2
2 † †
1,
log
δ
= ≠
= +
 
+

\ ¹
∑
k k k k
k
K
n k k k k
i i k
H M M H
C I
I H M M H
and the noise matrix . With this notation,
should be recovered from the knowledge of
et
R T
n n
H C
×
= denotes such channel fading matrix and
coding matrix with estimation,
†
T
n
E MM I ( =
¸ ¸
where υ is the dimension of data
Then the output could be represented from the basestation as
(1)
(2)
is the received vector of user k and
1 ×
∈
R
n
z c is additive white
† 2
σ
n
E zz I , and
2
σ
n
is the variance of noise.
(3)
(4)
) 1≤ ≤ M k K are the channel fading matrix and pre
ity of user k as obtained by [3]
(5)
So the sum capacity of the whole system is given by
(6)
)
† †
2 † †
 

\ ¹
k k k k
n k k k k
H M M H
I H M M H
With this notation,
and .
fading matrix and
t
n v
M C
×
∈
is the dimension of data
white Gaussian
coding matrix of
A. Eigenvalue based Beamforming
This is the common and well known method of precoding technique.In the method, the
precoding matrix for any user k is obtained from the singular value decomposition(SVD) of
k
H
†
=
k k k k
H U S V
(7)
Where’ U’ and ‘V’ are the unitary matrices and ‘S’ be the diagonal matrix. In this method
k
V is
taken as precoding matrix
,
EBB
k
M . So the capacity of user k is calculated as
( )
† †
2
2 † †
1,
log
δ
= ≠
= +
 
+

\ ¹
∑
EBB EBB
EBB k k k k
k
K
EBB EBB
n k i i k
i i k
H M M H
C I
I H M M H
(8)
In above matrix decomposition, U (
r r
N N × ) and V (
t t
N N × ) are the unitary matrixes, S has the
same size of H and is the diagonal matrix which Eigen values { }
i
λ
.
The above capacity equation would maximize the useful power of each user but it fails to
eliminate the interference caused by other users.
B. Block Diagonalization.
BD is the method we use to make to eliminate the interference due to multiusers in the network.
BD works often uses the zero forcing (ZR) algorithm[4]. So the zero forcing constraint for user k
is defined as,
1 1 1
.... ...
∧
− +
( =
¸ ¸
T
T T T T
k
k k K
H H H H H (9)
For the cancellation of the interference we should have
0 for i j = ≠
BD
i j
H M . (10)
So , the precoding matrix
BD
K
M should lie in the null space of
∧
k H
†
(1) (0)
∧ ∧ ∧ ∧ ∧
(
=
(
¸ ¸
k k k k k H U S V V ’ (11)
So,
0
∧
k
k
H V is the projection of
k
H in the null space of
∧
k H ,where
(0)
∧
k V is the null space of
∧
k H
so we define SVD again as,
0
†
(1) (0)
∧
( =
¸ ¸
k
k k k k k
H V U S V V , (12)
So we can find out the precoding matrix of user k is
(0) (1)
∧ ∧
=
BD
k k
k
M V V
So for the Block Diagonalization (BD) method, the channel capacity of user k is defined as
( )
† †
2
2
log
δ
= +
BD BD
BD k k k k
k
n
H M M H
C I
I
(13)
From this scheme, we would have lower power leading to capacity loss for low SNR. So we
would purpose a combination of EBB and BD method of precoding.
III. Example for the EBB
Consider the M=3, number of transmission antenna, N= 2 ,no of receiver antenna so the channel
matrix would be of size 2 by 3. With variance
2
.5 σ = w , then random Gaussian normal
distributed Rayleigh channel is modeled by complex 2 by 3 matrix as below.
0.0285  1.0431i 0.4023 + 0.0838i 0.2669 + 1.0207i
H
0.4788  0.1655i 0.1808 + 0.2226i 0.2092  0.2482i
( ( ( (
= == =
( ( ( (
¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸
Using the SVD method of matrix decomposition, we obtain
0.9321 + 0.0000i 0.3622  0.0000i
U
0.0268  0.3612i 0.0690 + 0.9296i
( ( ( (
= == =
( ( ( (
¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸
1.6472 0 0
S
0 0.3189 0
( ( ( (
= == =
( ( ( (
¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸
0.0279  0.6925i 0.6184 + 0.1752i 0.2302  0.2313i
V 0.2794 + 0.0834i 0.2312  0.3836i 0.5839  0.6111i
0.2020 + 0.6275i 0.3750 + 0.4956i 0.4191  0.0593i
( ( ( (
( ( ( (
= == =
( ( ( (
( ( ( (
¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸
So, we take
k
V as
EBB
k
M
.
So,
0.0279  0.6925i 0.6184 + 0.1752i 0.2302  0.2313i
0.2794 + 0.0834i 0.2312  0.3836i 0.5839  0.6111i
0.2020 + 0.6275i 0.3750 + 0.4956i 0.4191  0.0593i
EBB
k
M
( ( ( (
( ( ( (
= == =
( ( ( (
( ( ( (
¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸
is the precoding matrix.
This precoding matrix has many advantages like
1. We can use it to modulate the data to be sent so that the other unintended user doesn’t
know about the precoder matrix so they could not decode it even if they were able to tab
the information. Data without precoder information is like noise.
2. Only the intended receiver which already knows the precoder matrix could be able to
decode it using a decoder matrix.
Since there are three transmitters, dimension of s be 3 by 1 as follow (generated random for
eg)
0.1499  0.5247i
0.1682 + 0.7653i
0.7126  0.0930i
s
( ( ( (
( ( ( (
= == =
( ( ( (
( ( ( (
¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸
Then, the output transmitted matrix y would be
z y HMs = + = + = + = +
where z is the noise vector, so
0.2496 + 0.7172i
0.5361  0.0794i
y
( ( ( (
= == =
( ( ( (
¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸
Even if this y is received by other unintended user, they could not decode it.
At the intended receiver, there are techniques for decoder design ( which is out of the scope
of this project) for retrieving the information out of y.
From eqn(8), the capacity for the antenna configuration of 2 by 3 is calculated to be 4.5
bps/sec/Hz for the SNR of 10dB. Simulation result shows that capacity could be increased with
higher antenna configuration.
EBB is the most commonly used precoding method in order to maximize the received power
and enabling security for the information to be transmitted.
IV. Conclusion
We have shown with example that for 3GPP LTE , multi user MIMO system, EBB method of
precoding is better for maximizing the received power at the receiver. We can show that the
performance would increase with the increase of number of receiver antennas.
References:
1. Block diagonalisation and Eigenvalue based beamforming precoding design for downlink
capacity improvement in multiuser MIMO channel,Hongxuan Liu, Xiantao Cheng,
Zhizum Zhou and GangWu
2. W.Fezali and J.G. Proakis, “ Adaptive SVD algorithm for covariance matrix
eigenstructure computation,”
3. P. Viswanath and D. Tse, ”Sum capacity of the vector Gaussian broadcast channel and
uplinkdownlink duality,” IEEE Trans. Inform. Theory., vol. 49, no. 8, pp. 19121921,
August 2003
4. .Q. H. Spencer, A. L. Swindlehurst, and M. Haardt, ” ZeroForcing Methods for
Downlink Spatial Multiplexing in Multiuser MIMO Channels,” IEEE Trans. Signal,
Process., vol. 52, no. 2, pp461471, February 2004.