Professional Documents
Culture Documents
Abstract
This paper reports a study of the carrier frequency offset (CFO) estimation and
compensation in the uplink of the orthogonal frequency division multiple access
(OFDMA) system with interleaved subcarrier assignment. The presence of CFOs
between the transmitters and the uplink receiver will destroy the orthogonality among
the different subcarriers, resulting in inter-carrier interference and multiuser interference.
This paper proposes a pseudo oinverse CFO compensation method based on the signal
structure. The proposed method can compensate the CFOs of all users simultaneously
and isolate the signals from all users at the same time. this method provides a
considerable signal-to-noise ratio gain on the bit error rate performance and has a
relatively low implementation complexity.
(3)
and each subchannel has P N/Q subcarriers. For 2 ( m ) ( m )
( q
2
H p( m ) S p
j )n (m) j pn
subchannel is composed of subcarriers with indexes where n 0,1,..., N 1 , and where ( m ) (0.5,0.5) is
{q, Q q, .,(P-1 )Q q}, q 0,1, ,Q-1 . the normalized CFO of the m-th user. Let
Let superscript (.) ( m ) denotes the m-th user. Fig.1 {f ( m ) }Mm0 represent the CFOs between users and the
depicts the transmitter structure for the m-th user, with uplink receiver and f denote the subcarrier spacing.
s (m) [ S 0(m) S1(m) S P(m)1 ] denoting the information block Then ( m ) is defined as
that the m-th user will transmit during one OFDMA ( m ) f ( m ) f (4)
block. This block is first mapped into a set of N Considering a system with M users, the multiple user
modulation symbols: {X i(m) }iN01 , where, signal structure is given by
M M P1 2
S (m) , r (n) r ( m ) (n) H p( m ) S p
(m) j ( pQ q ( m ) ( m ) ) n
X (m)
g,p
i pQ q ( m )
(1) m1 m1 p 0
e N (5)
g,i
0 , otherwise Since r ( m ) (n) has a special periodic-like structure
with every P samples according to
And p 0,1, P-1 , and q ( m ) denotes the j 2v ( q ( m ) ( m ) ) / Q
subchannel index of user m. In the OFDMA system, r ( m ) (n vP) e r ( m ) ( n) (6)
the length of the cyclic prefix (CP) is equivalent to where v is an integer. Then {r (n)} M
m0
can be
N CP samples, and is assumed to be longer than the arranged into a Q P matrix:
maximum channel delay spread. After an N-point r(P 1)
r(0)
inverse fast Fourier transform (IFFT) and CP insertion r(P)
at the transmitter, the n-th element of the time-domain r(2P 1)
A
block vector x ( m ) of the m-th user is given by (7)
N 1
x n( m ) X i( m ) e
j 2nk
N , N CP n N 1 (2) r(N P) r(N 1) QP
i 0
VS V{UΘ{UFP )}
where represents the Schur product or element-by
element product, S U( BFP ) with
u (1) b (1)
(2) (2)
u b
U ,B
(M) (M)
u MP b MP
From Eq.(12), it is known that Ψ is a Hermitian compensates the CFOs of all users simultaneously and
j 2pˆ( m ) / P
requires 4QM 2 CMs for SVD processing[10] and
[Uˆ ]m , p e (m 1,2,...., M , p 0,1,....P 1) ,
MPQ CMs for the multiplication of Vˆ † Y . In Steps 2
and (.) denotes the conjugate operation. and 3, the operations of Uˆ ΘSˆ and the P-point FFT
Step 3 After P-point FFT processing, the “frequency of each user require MP and ( P/ 2) Log 2 P CM,
domain” signal, i.e., the estimation of B is given by
respectively. Thus, the total number of CMs required
post-multiplying B̂' by FPH , which leads to by the Pinv compensator is approximately
Bˆ B ˆ ' F H B [U
P
ˆ ΘV
ˆ † W]F H
P
(18) 4QM 2 MPQ MP M ( P/ 2) Log 2 P .Table 1 shows
Finally, the symbols of each user can be recovered by the comparison between the complexity of the
using zero-forcing equalizer, according to proposed Pinv method for certain values of
N , Q ,and M . It is easy to see that the Pinv method
sˆ (m) bˆ (m) D † P (h (m) ) (19)
has a much lower complexity For small number of
Where bˆ (m) sˆ (m) D P (h (m) ) is the m-th row of users
B̂ , m 1,2,...., M
Table 1 Number of complex multiplications
and where sˆ (m) Sˆ0(m) Sˆ1(m) Sˆ P(m)1 .
The proposed pseudoinverse method compensates the
CFOs of all users simultaneously in the first step and Method 4 users 8 users 12 users
isolates at the same time the signals of each user, ( M 4) ( M 8) ( M 12)
thereby reducing interference between adjacent users.
In the second step, the CFOs of all users can be N 2048 Pinv method 11 520 25 088 40 704
further compensated. Therefore, the newly proposed
method is not sensitive to large CFOs and has good Q 16
BER performance in the presence of multiple users.
N 2048 Pinv method 11 264 26 624 46 080
Q 32
5. Simulation Results
-3
10
mobile broadband wireless access systems— Amendment for
physical and medium access control layers for combined fixed and
-4
10 mobile operation in licensed bands. New York: IEEE, 2005.
-4
10
0 2 4 6 8 10 12
Number of users
1) Coarse Estimation
First, we introduce SVD (Eigen decomposition)
to ̂ .Then the number of users can be estimated by
means of the AIC , or the MDL [10], criterion using
Fig.6 The relation between the effective CFOs and
the Eigen values of ̂ .After knowing the user the subchannel indexes
number, the coarse estimation of { ( m ) }M , {ˆ ( m ) }M
m1 m1
inc 0.001 / Q ;
Step 3: calculate fˆ ( m ) using Eq.(22) ;
fˆ ( m ) ˆ ( m ) f (Qˆ ( m ) qˆ ( m ) ) f
s (22)
Step 4: return to Step 1 until all the CFOs Of M̂
users are estimated.