Low Complexity Channel Estimation for 3GPP LTE

Downlink MIMO OFDM Systems

Ömer Çetin∗ , Bahattin Karakaya∗ and Hakan Ali Çırpan‡
∗ Department of Electrical and Electronics Engineering,
İstanbul University, Avcılar Campus, 34320, İstanbul, Turkey
Email: omercetin@ieee.org, bahattin@istanbul.edu.tr
‡ Department of Electronics and Communication Engineering,

İstanbul Technical University, İTU̇ Maslak Campus, 34469, İstanbul, Turkey

Email: cirpanh@itu.edu.tr

AT Transpose of A.
Abstract—Multiple Input Multiple Output - Orthogonal Fre-
AH Conjugate transpose of A.
quency Division Multiplexing (MIMO-OFDM) is a promising diag(a) Diagonal matrix of a.
technique for reaching high data rates targeted in the 3rd IN N × N identity matrix.
Generation Partnership Project - Long Term Evolution (3GPP- 0N N × 1 column vector with all zeros.
LTE). MIMO-OFDM channel estimation schemes play a very E[. ] The expectation operator.
important role on achieving this aim. However, the complexity RN N dimensional space with real numbers.
of the estimators increases exponentially due to the structure of CN N dimensional space with complex numbers.
the MIMO systems. This causes an increase in the computational ⊗ Kronecker product.
burden of the transceivers. As a result, the complexity of the
channel estimators is becoming an important issue in real world
MIMO-OFDM applications. In this paper, the performance of the in Figure 1. M-QAM modulated serial data stream is grouped
low complexity Minimum Mean Square Error (MMSE) channel into Nd sized blocks.
estimator scheme based on Karhunen-Loéve (KL) series expan-
sion coefficients for the 3GPP-LTE Downlink MIMO-OFDM
systems is examined. System level simulations are accomplished A(1)
,n
X(1)
,n
x(1) s(1)
,n
,n
to compare the performances of the estimators under the spatially
correlated channel coefficient variations. !" "#$
Index Terms—3GPP, LTE, MIMO, OFDM, MMSE, Karhunen-
Loéve.
Modulation

I. I NTRODUCTION
OFDM is a high performance candidate for wireless com-
munication systems owing to its many advantages, especially
its performance in frequency-selective fading channels. Besides
that, MIMO transceiver architecture has potential to improve
the system capacity. Hence in 3GPP-LTE, combined MIMO- (N )
s N,nT
( )
X ,nNT
( )

OFDM structure is anticipated to meet the demands of rapidly

A ,nT x(N,nT )
increasing number of applications on wireless mobile networks.
In MIMO-OFDM systems channel state information is essential !" "#$
for the detection and equalization. In [1], 3GPP-LTE physical
channels are described in detailed and modulation types are
standardized. Besides that spatial channel models for MIMO
simulations are described elaborately in [2]. In the time domain
pilot-aided channel estimation, unoccupied subcarriers namely Fig. 1. MIMO-OFDM Transmitter scheme for LTE Downlink
”guard bands”, adversely affect the performance of the estimator
[3]. In order to reduce the complexity of the estimators, MMSE
estimation of the KL series expansion coefficients is examined in A number of modulated Nrs reference signals, namely pilots
[4]–[6]. In this paper, we compare the performances of the MMSE and Nd sized data signals are allocated to the resource elements as
estimations of the KL series expansion coefficients of the 3GPP- defined in [1]. Reference and data signals have been paralleled
LTE MIMO channel. In the simulations, the performances of the into consecutive data blocks in order to constitute an OFDM
MMSE estimator of the KL expansion and LS estimator under symbol. At the pth transmit antenna, the ℓth OFDM symbol can
the different channel correlations are compared. The notation be represented as
used in this paper is as follows: (p) (p) (p) (p)
Aℓ = [Aℓ [0], Aℓ [1], . . . , Aℓ [NBW − 1]]T ∈ ΞNBW (1)
II. S IGNAL M ODEL
The block diagram of the simplified 3GPP-LTE, MIMO- where Ξ denotes the modulation alphabet and NBW = Nd + Nrs
OFDM transmitter system with NT transmit antennas is depicted is the number of the occupied subcarriers at each of the transmit
antenna. Before the N -point Inverse Fast Fourier Transform
(IFFT) block, in order to avoid interference, NGB zeros are

978-1-4577-9538-2/11/$26.00 ©2011 IEEE 1441

padded to the unused subcarriers at the edges of the spectrum where FL ∈ C N ×L can be constituted by taking the first L
as guard bands, where NGB is equal to N − NBW . Only the columns of the N -point DFT matrix F in (5). From (8) and
specified OFDM symbols contain the reference signal for the (9) entire MIMO channel frequency response H ∈ C M , where
channel estimation. Considering only the reference signal inserted M ≡ NT NR N , can be expressed as below
OFDM symbols in the data-aided channel estimation, the OFDM [ T T
]
T T
symbol indicator ℓ could be omitted for the sake of simplicity. H = H(1) , H(2) , ..., H(NR ) . (10)
Reference and data signals with guard band before the IFFT
block can be represented by X(p) ∈ ΞN as below In the above expression, it is assumed that the whole CFR is
spatially uncorrelated. In order to drive an expression of the
(p) T (p) T spatially correlated channel frequency response, HS ∈ C M in
X(p) = [0, Alef t , 0TNF F T −NBW , Aright ]T (2) terms of H; we have to define the spatial correlation matrix
where R ∈ C NR NT ×NR NT as below [7]

(p) T NBW R = RT X ⊗ RRX . (11)

Alef t = [A(p) [ ], . . . , A(p) [NBW − 1]]T
2 NT ×NT NR ×NR
(p) T NBW Above, RT X ∈ C and RRX ∈ C are the
Aright = [A(p) [0], A(p) [1], . . . , A(p) [ − 1]]T . (3) normalized spatial correlation matrices of the transmit and the
2
receive antennas respectively. From (10) and (11) the spatially
N -point IFFT block is drived by the OFDM symbol X(p) to correlated channel frequency response can be denoted as
generate time domain transmitted signal x(p) is found to be √
HS = R ⊗ H . (12)
x(p) = FX(p) (4)
where (1)
r ,n y (1) (1)
  ,n Y ,n
1 1 ... 1

6&7&%&25& #/82*+
!"
 ej2π/N ej2π(N −1)/N 

,91%*51/(2
 1 ... 
#$"
%&'()*+
F= .. .. .. ..  (5)
 

,-.*+/0*1/(2 3 4&1&51/(2
. . . .
1 ej2π(N −1)/N ... ej2π(N −1)(N −1)/N

,;1/'*1/(2
~ (1)

!:*22&+
is the DFT matrix. In order to get rid of the inter-symbol Y ,n
interference, Cyclic Prefix (CP) with a size of NCP ≥ L is added ~ (N R )
to the signal x(p) to obtain the time domain transmitted signal N (N ) Y ,n
(N R )
r ,n y(,n R ) Y ,n R
s(p) ∈ C N +NCP . L represents the channel impulse response

6&7&%&25& #/82*+
(CIR) vector length of the MIMO channel between the pth !"

,91%*51/(2
transmit antenna and[ the qth receive antenna where it] can be #$"
%&'()*+
expressed as h(qp) = h(qp) [0], h(qp) [1], · · · , h(qp) [L − 1] ∈ C L .
In Figure 2, MIMO-OFDM receiver scheme with NR anten-
nas is depicted. At each receive antenna, the signal r(q) ∈
Fig. 2. MIMO-OFDM Receiver scheme for LTE Downlink
C N +NCP +L−1 can be expressed as the sum of the convolved
signals s(p) and h(qp) from the transmit antennas. Before the
FFT block at each receiver, unnecessary terms are removed and Denoting
the signal combined into the subsequent blocks with a length of N T
termed as y(q) . In the light of above time domain expressions, the Y = [Y1T , Y2T , ..., YN
T
R
] (13)
frequency domain representation of the received signal at the qth
as the entire received signal vector Y ∈ C NR N
the whole signal
antenna Y(q) = [Y (q) [0], Y (q) [1], ..., Y (q) [N − 1]]T ∈ C N can be
model can be written as in [5]:
expressed as
Y(q) = X H(q) + Wn (6) Y = (INR ⊗ X )HS + W . (14)
where Wn ∈ C , is the additive white gaussian noise (AWGN)
N
where W ∈ C NR N is the zero mean white gaussian noise with
at the qth receive antenna with zero mean and covariance matrix a covariance matrix of CW = E[WWH ] = σ 2 INR N .
CWn = E[Wn Wn H ] = σ 2 IN . X is the transmitted signal
from all of the antennas in the frequency domain and can be III. C HANNEL E STIMATION
represented as below
In this section, LS channel estimation and MMSE estimation
] ∈ ΞN ×NT N ,
(1) (2) (N )
X = [Xdiag , Xdiag , ..., Xdiag
T
(7) of the Karhunen-Loéve expansion coefficients will be derived.
Assume that only the symbols which contain the reference signal
where Xdiag = diag{X(p) } ∈ ΞN ×N and p ∈ 1, 2, ..., NT . H(q) ∈
(p) are selected. In the frequency domain signal representation, there
C NT N is the combined channel frequency response (CFR) from exist Nrs reference signals. Denoting
all of the transmit antennas to the qth receive antenna, expressed Ỹ = (INR ⊗ X˜ )H̃S + W̃ , (15)
by [ ]
T T T
(qNT ) T
H(q) = H(q1) , H(q2) , ..., H . (8) as the received signal vector at the reference signal positions
where Ỹ ∈ C Np and Np = NT NR Nrs is the total number
H(qp) ∈ C N ’s are the CFR between the pth transmit antenna of reference signal in the MIMO scheme. In (15) X˜ can be
and the qth receive antenna and can be obtained as below expressed as

H(qp) = FL h(qp) , (9) X˜ = [X̃(1) , X̃(2) , ..., X̃(NT ) ] ∈ ΦNT Kp ×NT Kp

1442
 TX1 TX2
Γ can be denoted as
Resource Block Resource Block
( )−1
k=11 D D D D D D k=11 D D D D D D R Γ = Λg Λg + σ 2 INp
D D D D D D D D D D D D Reference λg1 λgNp
D
R1 D D D0 D D D0 D D R0
D D D Symbol = diag{ ,..., }, (23)
D D D D D D D D D D D D
λg1 + σ 2 λgNp + σ 2
D D D D D D D D D D D D
D0 D D D D D R0
D D D D0 D D
0
where λg1 , λg2 , · · · , λgNp ’s as the singular values of the Λg .
R1 Null
D D D D D D D D D D D D
Assuming the rank of the matrix Λg as , the MMSE estimator
Symbol
D D D D D D D D D D D D of the optimum truncated KL expansion can be defined as
R1
D D D D0 D D D
0 D D R0
D D D
D λg1 λgr
D D D D D D D D D D D D Γtr = diag{ ,..., , 0, . . . , 0} . (24)
D D D D D D D D D D D D
Data λg1 + σ 2 λgr + σ 2
Symbol
k=0 D0 D D R1
D D D k=0 D
R0 D D D0 D D
As a result, the MMSE estimator of the KL expansion coefficient
requires only the division operations of the coefficients instead
=0 =5 =0 =5
Frequency

Frequency

of huge matrix multiplications.

IV. S IMULATION
Time Time Full channel impulse response is extracted from the one-
dimensional linear interpolation of the reference signal at the
Fig. 3. 3GPP-LTE MIMO scheme downlink signal pattern pilot points. In [2], LTE MIMO Channel parameters are defined
for the simulations. In Table I. LTE downlink air interface
specifications are defined for different bandwidths. A Bandwidth
of 3 MHz is chosen in the simulations. This choice means that
A. Least Squares Estimator DL
working with a number of NRB = 15 resource blocks. From I,
The LS estimator of the reference signals can be derived as DFT size and the number of occupied subcarriers are defined as
N = 256, NBW = 180 respectively. It is obvious that the total
−1
H̃S = (ΘH Θ) ΘH Ỹ , (16) number of the guard bands is parameterized as NGB = 76.
Modulation type chosen as BPSK. In order to generate the
where Θ ∈ C Np ×Np can be obtained as Θ = INR ⊗ X˜. In reference signal, L-31 Gold Sequence Pseudo Random Number
LTE MIMO structure, the reference signals are orthogonal to array is used as depicted in [1]. 2x2 MIMO channel type is
each other. In Figure 3. a simple reference signal orthogonality chosen. Equalization is performed by LS estimated and interpo-
is depicted. Thus (ΘH Θ)−1 in (16) can be easily derived as in lated reference signals. A number of r coefficient is computed
[6] below by the optimum truncated KL expansion coefficient estimator.
H̃S = ΘH Ỹ . (17) Reduced rank truncated KL estimator finds r−1 coefficients. The
transmit and receive antenna’s correlation coefficients are defined
B. Karhunen-Loéve Expansion as α = 0.3 and β = 0.3 respectively. The transmit antenna
correlation matrix RT X and the receive antenna correlation
The frequency response of the MIMO channel at reference matrix RRX are defined as
signal locations H̃S , is a zero-mean random variable with covari- [ ]
ance matrix CH . KL transformation makes the channel vector 1 α
RT X =
orthogonal such that it can be represented by KL coefficients α 1
basis vectors as below [ ]
1 β
RRX =
∑
Np
β 1
H̃S = gl ψl = Ψg , (18)
l=1
In Figure 4; the performances of the LS estimator, optimal rank
and reduced rank truncated KL expansion MMSE estimators
[where ψi ’s are ]orthonormal basis vectors constitute Ψ = are compared in Bit Error Rate (BER) sense. There exists an
ψ1 , ψ2 , · · · , ψNp and gl ’s are the KL expansion coefficients error floor for the reduced rank KL MMSE estimator greater
[ ]T
defined in vector g = g1 , g2 , · · · , gNp . Consequently defining than the values of a 20 dB Eb/N0. In Figure 5; BER results are
E[ggH ] = Λg ; the channel covariance matrix at reference signal examined according to different correlation coefficients for the
positions can be easily expressed as same LTE spatial channel model [7].

CH = E[H̃S (H̃S )H ] (19) V. C ONCLUSION

= ΨΛg Ψ H
. (20) The KL expansion MMSE estimator exhibits a better perfor-
mance than the LS estimator in the view of BER criteria. Reduced
In the above equation, if the Λg matrix is diagonal, ΨΛg ΨH rank truncated KL expansion coefficient estimator performance
expression becomes the Singular Value Decomposition of the CH is also better than the LS at low Eb/N0 values. However at
matrix. Thus we can express (15) in a different form high Eb/N0 values LS estimator exhibits a higher performance
than the truncated KL expansion estimator. In addition to
Ỹ = ΘΨg + W̃ . (21) increased performance, KL expansion MMSE estimator is also a
computationally efficient structure for the LTE downlink MIMO-
Finally, the MMSE estimator of the KL expansion coefficients OFDM systems.
can be expressed as in [5]
( )−1 H H VI. ACKNOWLEGMENT
ĝ = Λg Λg + σ 2 INp Ψ Θ Ỹ
This work is supported by The Scientific and Technological Re-
= ΓΨH ΘH Ỹ . (22) search Council of Turkey (TUBITAK) under grant no. 108E054.

1443
 Channel Bandwidth 1.4 MHz 3 MHz 5 MHz 10 MHz 15 MHz 20 MHz
Number of Resource Blocks NRB 6 15 25 50 75 100
Number of used subcarriers 72 180 300 600 900 1200
DFT Size 128 256 512 1024 1536 2048

TABLE I
3GPP-LTE RESOURCE CONFIGURATION [1]

0
10
LS estimated Channel
MMSE estimation of KL coefficients Communications Applications Conference (SIU), 2010 IEEE 18th,
MMSE estimation of truncated KL coefficients
Perfect Channel knowledge pp. 129-132, 2010.
10
−1
[7] “Correlation-Based Spatial Channel Modeling,” [Online]. Available:
http://spcprev.spirentcom.com/documents/5204.pdf

−2
10
BER

−3
10

−4
10

5 10 15 20 25 30 35 40
Eb / No [dB]

Fig. 4. BER results for estimated and Perfect channel state informations

0
10
α = β = 0.9
α = β = 0.7
α = β = 0.5
α = β = 0.4
α = β = 0.3

−1
10
BER

−2
10

−3
10

−4
10
5 10 15 20 25 30 35 40
Eb / No [dB]

Fig. 5. BER curves for the LTE 2x2 MIMO-OFDM system with different
correlation coefficients

R EFERENCES
[1] “3GPP TS 36.211 V9.0.0 Evolved Universal Terrestrial Radio Access
(E-UTRA); Physical Channels and Modulation (Release 9), Technical
report, 3GPP, December 2009 ,” 2009.
[2] “3GPP TR 25.996 V9.0.0 Spatial Channel Model for Multiple Input
Multiple Output (MIMO) simulations (Release 9), Technical report,
3GPP, December 2009,” 2009.
[3] Seongwook Song and A. Singer, “Pilot-Aided OFDM Channel Esti-
mation in the Presence of the Guard Band,” Communications, IEEE
Transactions on, vol. 55, 2007, pp. 1459-1465.
[4] M. Stege, P. Zillmann, and G. Fettweis, “MIMO channel estimation
with dimension reduction,” Fifth Symposium on Wireless Personal
Multimedia Communication (WPMC), 2002.
[5] B. Karakaya, H. Cirpan, and E. Panayirci, “Channel estimation for
MIMO-OFDM systems in fixed broadband wireless applications,”
Signal Processing and Its Applications, 2007. ISSPA 2007. 9th
International Symposium on, 2007, pp. 1-4.
[6] O. Cetin, B. Karakaya, and H. A. Cirpan, “Channel Estimation
for 3GPP LTE MIMO-OFDMA systems,” in Signal Processing and

1444