268
1, JANUARY 2009
Amplitude Clipping and Iterative Reconstruction of
MIMO-OFDM Signals with Optimum Equalization
Ui-Kun Kwon, Dongsik Kim, and Gi-Hong Im, Senior Member, IEEE
Abstract—Multi-input multi-output orthogonal frequency-
division multiplexing (MIMO-OFDM) has become a promising
candidate for next generation broadband wireless communica-
tions. However, like a single-input single-output (SISO)-OFDM,
one main disadvantage of the MIMO-OFDM is the high peak-
to-average power ratio (PAPR), which can be reduced by using
an amplitude clipping. In this paper, we propose clipped signal
reconstruction methods for the MIMO-OFDMs with spatial
diversity, such as space-time and space-frequency block codes
(STBC/SFBC). The proposed methods are based on the technique
called iterative amplitude reconstruction (IAR) for SISO-OFDM.
It is shown that the IAR can be easily employed for the STBC-
OFDM, but it cannot be directly applied to the SFBC-OFDM,
because the transmitted sequences over different antennas are
dependent due to the use of space-frequency code. We pro-
pose a new SFBC transmitter for clipped OFDM, which has
approximately half the computational complexity of conventional
SFBC-OFDM. The proposed clipping preserves the orthogonality
of transmitted signals, and the clipped signals are iteratively
recovered at the receiver. Further, we theoretically analyze the
performance of IAR with optimum equalization, and also provide
highly accurate channel estimation of the OFDM with amplitude
clipping. Simulation results show that the proposed receivers
effectively recover contaminated OFDM signals with a moderate
computational complexity.
Index Terms—Orthogonal frequency-division multiplexing
(OFDM), amplitude clipping, space-time and space-frequency
block code (STBC/SFBC), iterative amplitude reconstruction
(IAR).
I. I NTRODUCTION
I N VARIOUS fora and organizations for wireless research,
such as the Wireless World Research Forum (WWRF)
and International Telecommunication Union (ITU), there have
been active discussions about 4G or beyond 3G systems to
be deployed around 2010 [1], [2]. Multi-input multi-output
orthogonal frequency-division multiplexing (MIMO-OFDM)
has drawn significant interests as a candidate for the 4G or
beyond 3G modulation techniques, due to the high bandwidth
efficiency and the robustness against multipath fading channels
[4]-[6]. However, MIMO-OFDM, like a single-input single-
output (SISO)-OFDM, has an inherent drawback of high peak-
to-average power ratio (PAPR). High PAPR requires large
dynamic range of the transmit power amplifier and reduces the
power efficiency, and thus the cost of transmitter is increased
and the battery life time is decreased.
Manuscript received September 19, 2007; revised January 24, 2008, and
March 28, 2008; accepted May 14, 2008. The associate editor coordinating
the review of this paper and approving it for publication was H. Ochiai. This
work was supported by Samsung Advanced Institute of Technology.
The authors are with the Department of Electronics and Electrical Engi-
neering, Pohang University of Science and Technology (POSTECH), Pohang
790-784, Korea (e-mail: {kuk2580, mgkds, igh}@postech.ac.kr).
Digital Object Identifier 10.1109/T-WC.2009.071034
1536-1276/09$25.00 
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:54 from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO.
A number of approaches have been proposed to cope with
the SISO-OFDM PAPR problem. First, techniques based on
the channel coding transmit only the codewords with low
PAPR [8], [9]. Such coding techniques offer good PAPR
reduction and coding gain. The critical problem of coding
approach is that for an OFDM system with large number
of subcarriers, either it encounters design difficulties or the
coding rate becomes prohibitively low. Phase rotation is an-
other approach to reduce PAPR, including selective mapping
(SLM) [10], and partial transmit sequence (PTS) [11]. It
generates a set of sufficiently different candidate data blocks,
all representing the same information as the original data
block, and selects the most favorable block for transmission.
Although the phase rotation works with an arbitrary number
of subcarriers and any modulation schemes, it requires high
computational complexity and side information. Deliberate
amplitude clipping [12]-[17] may be one of the most effective
solutions when the number of subcarriers is large. It clearly re-
moves the amplitude peak, and does not introduce redundancy
and power increase. Clipping, however, causes distortion that
degrades the system performance.
To mitigate the harmful effects of the clipping, a few tech-
niques have been proposed. In [19], a scheme called decision-
aided reconstruction (DAR) is proposed, and oversampled
signals are used to compensate for the SNR degradation due
to the clipping [20]. However, the techniques in [19] and [20]
work well at high clipping ratio (CR) values, and the technique
in [20] needs bandwidth expansions. In [21], [22], maximum-
likelihood detection methods are proposed for clipped OFDM
signals. In this paper, based on the DAR, we have proposed a
technique called iterative amplitude reconstruction (IAR) for
coded OFDM, which recovers clipped signals by comparing
the estimates of clipped and nonclipped OFDM samples [18].
Since the IAR is an iterative technique, the reliability of
initial estimates and the propagation error to the next iteration
considerably affect the overall system performance. In this
paper, we theoretically show that the optimum equalization
of IAR increases the reliability of initial estimates, and the
power of propagation error to the next iteration is approx-
imately half of the DAR. Further, we first propose clipped
signal reconstruction methods for MIMO-OFDMs with spa-
tial diversity, such as space-time and space-frequency block
codes (STBC/SFBC). It is shown that the IAR can be easily
employed for the STBC-OFDM. However, unlike the STBC-
OFDM, where independent OFDM sequences are transmitted
over different antennas at certain time, the IAR cannot be
directly applied to the SFBC-OFDM, because the sequences
over different antennas in SFBC-OFDM are dependent due to
c 2009 IEEE
KWON et al.: AMPLITUDE CLIPPING AND ITERATIVE RECONSTRUCTION OF MIMO-OFDM SIGNALS WITH OPTIMUM EQUALIZATION 269

the use of space-frequency code. In this paper, we propose PAPR reduction of OFDM signals at the transmitter. The CR
a new SFBC transmitter for the clipped OFDM, which has is defined as
approximately half the computational complexity of conven- A
tional SFBC-OFDM. The proposed clipping preserves the CR = 20 log10 dB (3)
σ
orthogonality of transmitted signals, and the clipped signals
are iteratively recovered at the receiver based on the IAR. where σ 2 = Pin is the input power of xm [k]. Due to
We also provide a channel estimation of the clipped OFDM the central limit theorem with a relatively large number of
by using a sequence, which has constant amplitude in both subcarriers, say more than 100, the amplitude of xm [k] can be
frequency and time domains. Simulation results are presented assumed to have a Rayleigh distribution with the probability
to verify our analysis and to provide further insights into the density function
performance. 2|xm [k]| −|xm [k]|2 /Pin
f (|xm [k]|) = e . (4)
The remainder of this paper is organized as follows. In Sec- Pin
tion II, the system model of clipped OFDM is described, and
Then, the output power of clipped signal is given by
the technique called IAR is presented and analyzed in Section  ∞
III. In Section IV, we extend the IAR into the STBC/SFBC Pout = |x̄m [k]|2 f (|xm [k]|)d|xm [k]|
MIMO-OFDMs. A new SFBC transmitter for clipped OFDM 0
and its iterative signal reconstruction method are proposed in = (1 − e−γ )Pin
2
(5)
Section V. In Section VI, channel estimation and simulation
results of the proposed techniques are presented, and the paper where γ = A/σ. At the receiver, frequency domain channel
is concluded in Section VII. observation can be expressed as
Rm [n] = Hm [n]X̄m [n] + Zm [n], 0≤n≤N −1 (6)
II. S YSTEM M ODEL OF C LIPPED OFDM
where Hm [n] is the channel frequency response (CFR), and
The complex baseband OFDM signal with N subcarriers is Zm [n] is the AWGN with variance N0 .
expressed as
N −1 III. I TERATIVE A MPLITUDE R ECONSTRUCTION (IAR)
1 
x(t) = √ Xm [n]exp {j2πnf0 t} , 0≤t≤T (1) WITH O PTIMUM E QUALIZATION
N n=0
In this section, we present the IAR for the clipped signal
where f0 is the subcarrier spacing, T is the OFDM block reconstruction, and theoretically compare the IAR with the
N −1 DAR.
duration, and {Xm [n]}n=0 is the symbol sequence to be
transmitted for the mth OFDM block. The discrete time
JN −1
OFDM signals, {xm [k]}k=0 , sampled at the time interval A. The Procedure of IAR
N −1
Δt = T /JN , can be obtained by padding {Xm [n]}n=0 with
The procedure of IAR is explained as follows with reference
(J − 1)N zeros and taking JN -point inverse fast Fourier
to the Fig. 1. Here, the receiver is assumed to know the
transform (IFFT). The parameter J is an oversampling factor,
clipping amplitude A, defined in (2).
and J = 1 is for the Nyquist rate sampling1 . Clipping is
performed on the IFFT output sequence as 1) Frequency domain channel observation, Rm [n], is ob-
 tained by performing FFT on the discrete received sam-
−1
xm [k], |xm [k]| ≤ A, ples, {rm [k]}N
k=0 .
x̄m [k] =
A exp{arg(xm [k])}, |xm [k]| > A, 2) Then, the estimate of the clipped sample, x̄ ˆm [k], is
0 ≤ k ≤ JN − 1 (2) obtained and stored in memory by performing IFFT on
ˆ [n]}N −1 , where
{X̄ m n=0
where x̄m [k] is the clipped sample and A is the clipping am- ∗
plitude. If the clipping is performed on the Nyquist sampling ˆ [n] = Hm [n]
X̄ m 2
Rm [n]
rate, it causes clipping noise to fall all in-band and suffers |Hm [n]| + N0 /Pout
considerable peak regrowth after the D/A conversion. When = Wm [n]Rm [n], 0 ≤ n ≤ N − 1. (7)
the clipping is performed on an oversampling rate, however,
Wm [n] is an MMSE equalizer tap coefficient for the
it produces out-of-band radiation, which requires additional
nth subcarrier, and (·)∗ denotes the complex-conjugate
bandpass filter and may also cause peak regrowth. More-
operation.
over, it requires larger size (JN -point) fast Fourier transform (I) −1
(FFT)/IFFT blocks, which increases not only computational 3) Estimate the transmitted symbols {X̂m [n]}N n=0 from
N −1
complexity, but also hardware costs. To overcome these prob- the {Cm [n]Rm [n]}n=0 , where I represents the iteration
lems of clipping, H. Ochiai et al. presented that the Nyquist number and starts with an initial value of I=0. Cm [n]
rate clipping combined with the SLM achieves significant is an optimum equalizer tap coefficient to get the non-
PAPR reduction with moderate increment of complexity [15]. clipped OFDM samples from the received signals, which
Thus, in this paper, we adopt Nyquist rate clipping for the will be detailed at the end of this subsection.
4) IFFT is performed on the decisions in Step 3 to obtain
1 The guard carriers of OFDM systems provide an inherent oversampling. the estimates of the samples before the clipping, thus
(I)
For simplicity, however, it is not considered in this paper. yielding x̂m [k].

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:54 from IEEE Xplore. Restrictions apply.
270 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 1, JANUARY 2009

EQ. ( Wm) IFFT Memory

I=0
S/P FFT EQ. ( C m ) Soft Demapping Soft Interleaving Amplitude
IFFT
& Deinterleaving Decoding & Soft Mapping Reconstruction
I>0

FFT

Fig. 1. The receiver structure with iterative amplitude reconstruction (IAR).

5) Clipped signals are detected by comparing the amplitude complex Gaussian random variables with a zero mean. Then,
(I)
of x̂m [k] to A. Then, the amplitude of clipped signals the frequency domain channel observation in (6) can be
(I) −1
is reconstructed, and a new sequence {ym [k]}N k=0 is
expressed as
generated as
 Rm [n] = Hm [n]X̄m [n] + Zm [n]
(I)
(I) x̄ˆm [k], |x̂m [k]| ≤ A, = Hm [n](αXm [n] + Dm [n]) + Zm [n]
ym [k] = (I)
ˆm [k])}, |x̂(I)
|x̂m [k]| exp{arg(x̄ m [k]| > A, = αHm [n]Xm [n] + Qm [n],
0 ≤ k ≤ N − 1. (8) 0 ≤ n ≤ N −1 (12)
(I) −1
6) The sequence {ym [k]}Nk=0 is converted to the frequency
(I) where Qm [n] is the sum of Hm [n]Dm [n] and Zm [n]. For
domain, yielding Ym [n], and the transmitted signals simplicity, we assume that the input power Pin is equal
(I+1) −1
{X̂m [n]}N n=0 are estimated. to 1 without loss of generality. Since Xm [n], Dm [n], and
7) This completes the Ith iteration, and for more iterations, Zm [n] are uncorrelated and E[|Dm [n]|2 ] = Pout − α2 Pin , the
go back to Step 4 with I=I+1. optimum equalizer tap coefficients, {Cm [n]}N −1
n=0 , are given
Since the clipping affects not the phase but the amplitude of by
OFDM signals, IAR replaces only the amplitude of detected
samples in Step 5. This is why we call this algorithm as itera- (αHm [n])∗
Cm [n] =
tive amplitude reconstruction. From Fig. 1 and the discussion |αHm [n]|2 + E[|Qm [n]|2 ]
∗
above, each iteration of the amplitude reconstruction requires αHm [n]
= 2
a single pair of N -point IFFT/FFT operations. α |Hm [n]| + E[|Hm [n]Dm [n]|2 ] + E[|Zm [n]|2 ]
2

In IAR and DAR, clipped signals are recovered by using the ∗

αHm [n]
estimates of clipped and nonclipped OFDM samples, which = −γ 2 ,
(1−e )|Hm [n]|2 +N0
ˆm [k] and x̂(I)
are x̄ m [k], respectively. While the DAR uses 0 ≤ n ≤ N − 1 (13)
only one equalizer for recovering these two estimates, the
proposed IAR uses two different equalizers, i.e., Wm [n] and based on the MMSE criterion.
Cm [n], with an increase of complexity due to the additional
equalization. From the [15], the clipped OFDM sample x̄m [k]
in (2) can be expressed as the aggregate of the attenuated B. Performance Analysis of Clipped OFDM before the first
OFDM signal and the clipping noise dm [k] as iteration
Since decision error propagates to the next iteration, the
x̄m [k] = αxm [k] + dm [k], 0≤k ≤N −1 (9)
reliability of initial estimates considerably affects the overall
where the attenuation factor α is a function of γ and is given system performance. To get reliable initial estimates of the
by nonclipped OFDM samples, we have derived an optimum
√ MMSE equalizer, Cm [n], in Section III-A. In this subsection,
E{xm [k]x̄m [k]} −γ 2 πγ
α= =1−e + erfc(γ). (10) we evaluate the SER performance of clipped OFDM before
E{xm [k]xm [k]} 2 the first iteration by calculating the means and variances of the
Since xm [k] is assumed to be a complex i.i.d. Gaussian equalizer outputs, and show that the proposed equalizer Cm [n]
random variable with a zero mean, the memoryless property of increases the reliability of the initial estimates compared to the
the system may guarantee dm [k] to be a complex i.i.d. random equalizer Wm [n]. For simplicity, 16-ary quadrature amplitude
variable with a zero mean, but not Gaussian. The clipping modulation (16-QAM) without channel coding is considered,
noise Dm [n] that falls on the nth subcarrier can be written as and the perfect synchronization and carrier recovery at the
N −1
receiver are assumed. The output of equalizer Cm [n] in the
1  Step 3 can be written as
Dm [n] = √ dm [k]WNnk , 0≤n≤N −1 (11)
N k=0
Cm [n]Rm [n]
∗
where WNk  e−j(2πk/N ) . As the number of subcarriers αHm [n](αHm [n]Xm [n]+Hm [n]Dm [n]+Zm [n])
−1 = . (14)
increases and the CR value decreases, {Dm [n]}N
n=0 approach (1−e−γ 2 )|Hm [n]|2 +N0

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:54 from IEEE Xplore. Restrictions apply.
KWON et al.: AMPLITUDE CLIPPING AND ITERATIVE RECONSTRUCTION OF MIMO-OFDM SIGNALS WITH OPTIMUM EQUALIZATION 271

where
0
10

CR=1dB 
  ∞
1 x 1 2
Q(x) = √
erfc = √ e−t /2 dt,
−1
10 2 2 2π x
α2 |Hm [n]|2
μc [n] = . (17)
−2
CR=3dB (1−e 2 )|Hm [n]|2 +N0
−γ
Symbol Error Rate

10

From (16), before the first iteration, the SER performance of

clipped OFDM with the equalizer Cm [n] is given by
−3
10

1  
N −1
SER16-QAM = 1 − (1 − SER4-P AM [n])2 . (18)
N n=0
−4
10
CR=4dB
simulation with Wm
simulation with Cm
Without the Cm [n], the SER performance of clipped OFDM
analysis with Wm
analysis with Cm
with only the equalizer Wm [n] can be similarly obtained.
10 15 20 25 30 35 40 45 Fig. 2 shows the simulated and analyzed SER performances
Average SNR [dB]
of the clipped OFDM with equalizers Cm [n] and Wm [n]
Fig. 2. SER performance of clipped OFDM before the first iteration
before the first iteration. 1024 subcarriers (N=1024), 5MHz
(N=1024, BW=5MHz, 16-QAM, static BU channel, no channel
coding). bandwidth, and a six-tap bad urban (BU) channel are con-
sidered. Perfect channel state information (CSI) is assumed.
As seen in the figure, analyzed and simulated results are in a
good agreement, and the optimum equalizer Cm [n] increases
Since the means of Dm [n] and Zm [n] are zeros, and Xm [n],
the reliability of initial estimates.
Dm [n], and Zm [n] are uncorrelated, the means and variances
of the equalizer outputs are given by
C. Propagation Error of DAR and IAR to the next iteration
μc |Xm [n]=Si Besides the optimum equalizer Cm [n], another major dif-
 E [Cm [n]Rm [n]|Xm [n]=Si ] ference between the DAR and IAR is in the Step 5 of the IAR
α2 |Hm [n]|2 procedure in Section III-A. In this subsection, we discuss the
= Si , effectiveness of the Step 5. In the case of DAR, any detection
(1−e−γ 2 )|Hm [n]|2 +N0
of the clipped sample is replaced with the estimated version,
σc2 |Xm [n]=Si (I) (I)
  x̂m [k]. Let the estimate of the time domain sample x̂m [k]
 E |Cm [n]Rm [n]|2 |Xm [n]=Si − μ2c |Xm [n]=Si be the sum of the transmitted sample xm [k] and the residual
    (I)
α2 |Hm [n]|4 E |Dm [n]|2 + α2 |Hm [n]|2 E |Zm [n]|2 error pm [k] as
=  2
(1−e−γ 2 )|Hm [n]|2 +N0

  x̂(I) (I)
m [k] = xm [k] + pm [k], 0 ≤ k ≤ N − 1. (19)
2
α2 |Hm [n]|2 (1−e−γ ) − α2 |Hm [n]|2 + N0 (I)
=  2 Here, all the power of pm [k] propagates to the next iteration
(1−e−γ 2 )|Hm [n]|2 +N0 in the case of DAR. If we assume perfect channel gains and
= σc2 [n] (15) high SNR, the reconstructed samples of the IAR in Step 5 can
be rewritten as
where {Si }15i=0 are the 16-QAM symbols, and the subscript (I)
[k] = |x̂(I) ˆ
ym m [k]| exp{arg(x̄m [k])}
“c ” represents the equalizer Cm [n]. From (15), it is observed
that the variance σc2 |Xm [n]=Si is the same for all Si . ≈ |x̂(I)
m [k]| exp{arg(x̄m [k])}

In order to obtain the SER performance of initial estimates, = |x̂(I)

m [k]| exp{arg(xm [k])}
(I) (I)
we derive the SER for the horizontal and vertical 4-ary pulse = xm [k] + vm [k], if |x̂i,k | > A (20)
amplitude modulation (4-PAM) first, and then extend the result
(I)
into the 16-QAM.
 Since
 we have assumed the unit input where vm [k] is the error component propagating to the
(I) (I)
power, i.e., E |Xm [n]|2 = 1, amplitude levels and decision next iteration. Since |pm [k]|  |xm [k]| in general, vm [k]
thresholds of the horizontal and vertical 4-PAMs for 16-QAM (I)
can be considered as the projection of pm [k] onto xm [k]
are given by {− √310 , − √110 , √110 , √310 } and {− √210 , 0, √210 }, (see Fig. 3). For the OFDM systems, frequency selective
respectively. Assuming equally probable symbol alphabets, channel can be modelled as N parallel independent flat chan-
and from (15), the SER for 4-PAM can be obtained as follows nels. Therefore, the decision error in the frequency domain,
(I) (I)
  3  Pm [n] = X̂m [n]−Xm [n], can be assumed to be a zero-mean
√ μc [n] − √2 independent random variable such that the time domain error
1 10 10
SER4-P AM [n] = Q √ (I)
pm [k] approaches a circular symmetric zero-mean Gaussian
2 σc [n]/ 2
 2   1  process for large N . As a result, the IAR is more robust against
√
10
− √110 μc [n] √ μc [n]
10 the decision error than the DAR because the power of vm [k]
(I)
+Q √ +Q √ (16) (I)
σc [n]/ 2 σc [n]/ 2 is almost half of the pm [k].

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:54 from IEEE Xplore. Restrictions apply.
272 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 1, JANUARY 2009

 
where Λ2i [n] = Λ2i+1 [n] = 2p=1 2q=1 |Hpq,2i [n]|2 , Z2i [n]

Imag
2  ∗ ∗

= H1q,2i [n]Zq,2i [n] − H2q,2i [n]Zq,2i+1 [n] , and

q=1 
2  ∗ ∗

Z2i+1 [n] = q=1 H2q,2i [n]Zq,2i [n] + H1q,2i [n]Zq,2i+1 [n] .
Here, the STBC combined signal X p,2i [n] can be seen as a
frequency domain channel observation of the SISO-OFDM
with equivalent CFR of Λ2i [n], since the clipped signals of
each transmit antenna, X̄1,2i [n] and X̄2,2i [n], are completely
separated after the STBC combining. Therefore, the clipped
STBC-OFDM signals can be easily recovered by using the
A Real
IAR for SISO-OFDM.
(I) (I)
Fig. 3. The comparison of propagation errors; pi,k (DAR), and vi,k B. IAR for Clipped SFBC-OFDM
(IAR).
The SFBC encoder generates the coded symbol Xp,m [n]
for the neighboring subcarriers, n = 2v, 2v + 1, v =
IV. IAR FOR STBC/SFBC MIMO-OFDM S 0, 1, ..., N/2 − 1, as follows [6]
   
In this section, we extend the IAR for SISO-OFDM into X1,m [2v] X1,m [2v + 1] S [2v] Sm [2v + 1]
 ∗m ∗ .
the two-by-two MIMO-OFDMs with spatial diversity, such as X2,m [2v] X2,m [2v + 1] Sm [2v + 1] −Sm [2v]
STBC and SFBC. (24)
−1
Similar to the STBC-OFDM, {xp,m [k]}N k=0 are obtained by
N −1
taking N -point IFFT on the symbols {Xp,m [n]}n=0 , and the
A. IAR for Clipped STBC-OFDM clipping is performed as in (2). Assuming that the CFRs
Let {Xp,m [n]}N −1 between adjacent subcarriers are approximately constant, i.e.,
n=0 , p = 1, 2, m = 2i, 2i + 1, denote the
transmitted symbol sequence of the pth transmit antenna at the Hpq,m [2v] ≈ Hpq,m [2v + 1], the SFBC combined signals at
mth OFDM block duration. The STBC encoder generates the the receiver are obtained as
 
coded symbol Xp,m [n] of the nth subcarrier as follows [3] Sm [2v]
   
X1,2i [n] X1,2i+1 [n] S2i [n] S2i+1∗
[n] Sm [2v + 1]
 ∗ ,   
X2,2i [n] X2,2i+1 [n] S2i+1 [n] −S2i [n] 2 ∗
H1q,m ∗
[2v]Rq,m [2v] − H2q,m [2v]Rq,m [2v + 1]
(21) = 2  q=1
∗ ∗
 ,
where Sm = [Sm [0], Sm [1], ..., Sm [N − 1]] is a channel coded q=1 H2q,m [2v]Rq,m [2v] + H1q,m [2v]Rq,m [2v + 1]
symbol vector for the mth OFDM block duration. The discrete 0 ≤ v ≤ N/2−1. (25)
−1
time OFDM samples, {xp,m [k]}N k=0 , are obtained by taking Substituting (22), (24) and the frequency domain representa-
−1
N -point IFFT on the STBC encoded symbols {Xp,m [n]}N n=0 . tion of (9) into (25), we have
Then, clipping is performed on the IFFT output sequences
as in (2), yielding {x̄p,m [k]}N −1 Sm [2v]
k=0 . At the receiver, frequency
domain channel observation at the qth receive antenna can be 2 
 2

expressed as =α |Hpq,m [2v]|2 Sm [2v]

p=1 q=1
2 
2
 
Rq,m [n] = Hpq,m [n]X̄p,m [n] + Zq,m [n], + |H1q,m [2v]|2 D1,m [2v]−|H2q,m [2v]|2 D2,m
∗
[2v+1]
p=1
q=1
0≤n≤N −1 (22) ∗
+ H1q,m ∗
[2v]H2q,m [2v](D2,m [2v]−D1,m [2v+1])

where Hpq,m [n] is the CFR between pth transmit and qth ∗ ∗
+ H1q,m [2v]Zq,m [2v]−H2q,m [2v]Zq,m [2v+1] ,
receive antennas, and Zq,m [n] is the AWGN with variance
N0 . If we assume that the CFRs between two consecutive
OFDM symbols are approximately constant, i.e., Hpq,2i [n] ≈ Sm [2v+1]
Hpq,2i+1 [n], the STBC combined signals are obtained as [3] 2 2
  =α |Hpq,m [2v]|2 Sm [2v+1]
S2i [n] p=1 q=1
S2i+1 [n] 2 

  + |H1q,m [2v]|2 D1,m [2v+1]+|H2q,m[2v]|2 D2,m
∗
[2v]
1,2i [n]
X
= 2,2i [n]
q=1
X
   ∗   ∗
+ H1q,m ∗
[2v]H2q,m [2v](D1,m [2v]+D2,m [2v+1])
2 ∗
H [n]R [n] − H 2q,2i [n]R [n] 
= q=1
2  1q,2i
∗
q,2i q,2i+1
∗
 ∗ ∗
q=1 H2q,2i [n]Rq,2i [n] + H1q,2i [n]Rq,2i+1 [n] + H2q,m [2v]Zq,m [2v]+H1q,m [2v]Zq,m [2v+1] . (26)
  
Λ2i [n]X̄1,2i [n] + Z2i [n] In order to recover the clipped signals of each transmit
≈  , 0 ≤ n ≤ N −1 (23) 1,m [n] and X
2,m [n] are required at the receiver like
Λ2i+1 [n]X̄2,2i [n] + Z2i+1 [n] antenna, X

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:54 from IEEE Xplore. Restrictions apply.
KWON et al.: AMPLITUDE CLIPPING AND ITERATIVE RECONSTRUCTION OF MIMO-OFDM SIGNALS WITH OPTIMUM EQUALIZATION 273

N SFBC
Source S/P IFFT X 2 Clipping
2 Encoding

(a)
~ ˆ
Sm Sm N sˆme , sˆmo
EQ. ( Wm ) IFFT X 2 Memory
S/P FFT 2

(I) (I)

SFBC MRRC
I=0
Soft Demapping Soft Interleaving Ŝ (I)
m N sˆ em , sˆ om Amplitude
EQ. ( C m ) IFFT X 2
Combining & Deinterleaving Decoding & Soft Mapping 2 Reconstruction
I>0
(I) (I)
S/P FFT Ym(I) N y em , y om
FFT X 2
2

(b)
Fig. 4. The transceiver structure for the proposed SFBC-OFDM with amplitude clipping. (a) Transmitter. (b) Receiver.

the STBC-OFDM. However, in the case of SFBC-OFDM, it is Since sem [k] and som [k] are periodic in k with period N/2, we
difficult to derive those signals from the {Sm [2v], Sm [2v+1]}, can replace them with sem [(k)(N/2) ] and som [(k)(N/2) ]. Here,
∗
since D1,m [2v] = −D2,m [2v + 1] and D1,m [2v + 1] = we clip sem [(k)(N/2) ] and som [(k)(N/2) ] instead of x1,m [k].
∗
D2,m [2v], which means that the clipping destroys the orthog- Then, the transmitted signals of the first antenna can be written
onality of SFBC-OFDM transmitted signals. In other words, as
X̄1,m [n] and X̄2,m [n] can not be completely separated at 1  
the receiver, because Sm [2v] and Sm [2v + 1] in (26) are x̄1,m [k] = √ s̄em [(k)(N/2) ] + WN−k s̄om [(k)(N/2) ] ,
2
mixed with the clipping noises from each transmit antenna.
0 ≤ k ≤ N − 1. (29)
Therefore, it is difficult to obtain robust estimates of the
clipped and nonclipped OFDM samples for recovering the Because conjugate operation does not change the PAPR prop-
amplitude of clipped signals. erties, transmitted signals of the second antenna can be derived
from the first antenna signals using the SFBC in (24) and
V. P ROPOSED SFBC-OFDM WITH A MPLITUDE C LIPPING discrete Fourier transform (DFT) symmetry property [23], i.e.,
s∗ [−k]N ⇔ S ∗ [n], k, n = 0, 1, ..., N − 1, as follows
In Section IV-B, we have shown that the IAR for SISO-
OFDM cannot be directly applied to the SFBC-OFDM, be- 1  ∗ ∗

x̄2,m [k] = √ s̄om [(−k)(N/2) ] − WN−k s̄em [(−k)(N/2) ] ,
cause the clipping destroys the orthogonality of SFBC-OFDM 2
transmitted signals. In this section, we propose a new SFBC 0 ≤ k ≤ N − 1. (30)
transmitter for clipped OFDM and its signal reconstruction
method at the receiver. This implies that only N/2 multiplications and N additions
are required to derive x̄2,m [k] from x̄1,m [k]. The block
diagram of proposed SFBC transmitter for clipped OFDM
A. Proposed Clipping for SFBC-OFDM is shown in Fig. 4 (a), which has approximately half the
By separating {X1,m [n]}N −1
n=0 of the SFBC-OFDM into
computational complexity of the conventional SFBC-OFDM
even and odd elements, time domain signals of the first transmitter, especially when the number of subcarriers is large.
antenna can be written as Note that the proposed clipping preserves the orthogonality of
N −1
transmitted signals, and the clipped signals, s̄em [k] and s̄om [k],
1  can be completely separated after the SFBC combining at
x1,m [k] = √ X1,m [n]WN−nk
N n=0 the receiver, while the addition of separately clipped signals
increases the PAPR of transmitted signals. Fig. 5 shows
1   
(N/2)−1
=√ Sm [2v] + WN−k Sm [2v + 1] W(N/2)
−vk the PAPR complementary cumulative distribution functions
N v=0 (CCDFs) of clipped OFDM and proposed clipped SFBC-
1   OFDM with N =128 at CR=0 dB. The ideal bandlimited
= √ sem [k] + WN−k som [k] , 0 ≤ k ≤ N − 1 analog OFDM signals are approximated by oversampling the
2
(27) discrete signals by a factor of sixteen. It is observed that the
PAPR of proposed SFBC-OFDM is approximately 1 dB higher
where sem [k] and som [k] are represented as than the clipped OFDM at CCDF = 10−3 .
 (N/2)−1
2 
−vk
sem [k] = Sm [2v]WN/2 , B. IAR for Proposed SFBC-OFDM
N v=0
 (N/2)−1 In this subsection, we describe the IAR for the proposed
2 
som [k] = Sm [2v + −vk
1]WN/2 . (28) SFBC-OFDM. The procedure of the proposed receiver is
N v=0 explained as follows with reference to Fig. 4 (b).

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:54 from IEEE Xplore. Restrictions apply.
274 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 1, JANUARY 2009

Cm [n]
0
10
OFDM (w/o clipping)

αΛm [n]σS2
OFDM (CR=0dB)
Proposed SFBC−OFDM (CR=0dB)
= 2 
(αΛm [n])2 σS +E[|Λm [n]Dm [n]|2 ]+E[|Zm [n]|2 ]
Complementary Cumulative Distribution Function

2
αΛm [n]σS
−1
10
=
(αΛm [n])2 σS2 +Λ2m [n]{(1−e−γ 2 −α2 )σS2 }+Λm [n]N0
ασS2
= , 0 ≤ n ≤ N − 1 (34)
−2
10 (1 − e −γ 2
)Λm [n]σS2 + N0
where σS2 ( E[|Sm [n]|2 ]) is the power of Sm [n].
(I) (I)
4) ŝem [k] and ŝom [k] are obtained by performing two N/2-
(I)
point IFFTs on even and odd elements of the Ŝm [n],
−3
10

and the clipped signals are detected by comparing the

(I) (I)
amplitude of ŝem [k] and ŝom [k] to A. Then, the clipped
−4
10
signals are reconstructed, and two new sequences with
e(I) N/2−1 o(I) N/2−1
block length N/2, {ym [k]}k=0 and {ym [k]}k=0 ,
2 3 4 5 6 7 8 9 10 11 12
Peak−to−Average Power Ratio [dB]

Fig. 5. PAPR CCDFs of the clipped OFDM and proposed SFBC- are generated as
OFDM. (I)
{ym
e(o)
[k]}

e(o)(I)
s̄ˆm [k],
e(o)
1) From the received signals, SFBC combined signals are |ŝm [k]| ≤ A,
= e(o)(I) e(o)(I)
obtained as |ŝm ˆ e(o)
[k]| exp{arg(s̄m [k])}, |ŝm [k]| > A,
 

Sm [2v] 0 ≤ k ≤ N/2 − 1. (35)
Sm [2v + 1] 5) ym
e(I)
[k] and ymo (I)
[k] are converted to the frequency
  ∗  (I) (I)
2 ∗
H1q,m [2v]Rq,m [2v] − H2q,m [2v]Rq,m [2v + 1] domain, yielding Ym [2v] and Ym [2v + 1], and the
= 2  q=1  (I+1)
channel coded OFDM symbol vector {Ŝm [n]}N −1
∗ ∗ n=0 is
q=1 H2q,m [2v]Rq,m [2v] + H1q,m [2v]Rq,m [2v + 1]
   estimated.
Λm [2v]S̄m [2v] + Zm [2v] 6) This completes the Ith iteration, and for more iterations,
≈  ,
Λm [2v + 1]S̄m [2v + 1] + Zm [2v + 1] go back to Step 4 with I=I+1.
0 ≤ v ≤ N/2 − 1 (31) From Fig. 4 (b) and the discussion above, each iteration
requires two pairs of N/2-point IFFT/FFT operations, which
2 2Λm [2v]
where =
2
Λm [2v + 1]

=
has even less computational complexity than that of the IAR
|H [2v]| , Z [2v] =
p=12  q=1
∗
pq,m m
for SISO-OFDM. Note that we have derived the {Cm [n]}N −1
H 1q,m [2v]Z [2v] − H 2q,m [2v] n=0

∗
q=1  q,m
 2  in Step 3 to get reliable estimates of the nonclipped OFDM
Zq,m [2v + 1] , and Zm [2v + 1] = q=1 H2q,m [2v] samples before the first iteration, and replace only the ampli-
∗ ∗
Zq,m [2v] + H1q,m [2v]Zq,m [2v+1] . tude of clipped samples in Step 4, since the clipping affects
2) Estimates of the clipped samples, s̄ˆem [k] and s̄ˆom [k], not the phase but the amplitude of the signals.
are obtained and stored in memory by performing two
N/2-point IFFTs on even and odd elements of the VI. C HANNEL E STIMATION AND S IMULATION R ESULTS
{S̄ˆm [n]}N −1
n=0 , where A. Block-type Channel Estimation for Clipped OFDM
σS̄2
S̄ˆm [n] = Sm [n] = Wm [n]Sm [n], In this paper, the channel estimation is based on the
Λm [n]σS̄2 + N0 method proposed in [24], but without the recursive process for
0 ≤ n ≤ N − 1. (32) reducing the computational complexity. In the SISO-OFDM,
−1 we have used the Chu sequence as a pilot symbol block,
{Wm [n]}N n=0 are the MMSE equalizer tap coefficients
which has constant amplitude in both frequency and time
for the clipped OFDM samples.
domains [25]. Due to the unit PAPR property, utilizing the
3) Estimate the channel coded OFDM symbol vec-
(I) −1  N −1 Chu sequence as a pilot symbol block precludes the problem
tor {Ŝm [n]}N n=0 from the {Cm [n]Sm [n]}n=0 , where of nonlinear distortion caused by the clipping when the CR
Cm [n] is the optimum MMSE equalizer tap coefficient
value is greater than 0dB. The nth element of a length-N Chu
for the nonclipped OFDM samples, and the number of
sequence is given by
iteration starts with I = 0. Since the SFBC combined  jπrn2 /N
signals in Step 1 can be represented as e , for even N
CN [n] = (36)
Sm [n] ≈ Λm [n]S̄m [n] + Zm [n]
 ejπrn(n+1)/N , for odd N

= Λm [n](αSm [n] + Dm [n]) + Zm [n],


where r is relatively prime to N .
Channel estimation for the MIMO-OFDM can be done by
0≤n≤N −1 (33)
transmitting a pilot symbol block from one antenna at a time

and Sm [n], Dm [n], and Zm [n] are uncorrelated, while the remaining transmit antennas are idle. This method,
−1
{Cm [n]}N
n=0 are obtained as follows however, becomes inefficient when the number of transmit

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:54 from IEEE Xplore. Restrictions apply.
KWON et al.: AMPLITUDE CLIPPING AND ITERATIVE RECONSTRUCTION OF MIMO-OFDM SIGNALS WITH OPTIMUM EQUALIZATION 275

0
10 2

calculated
0 simulated

−1
10
−2

−2
10 −4

Mean Squared Error (dB)

Bit Error Rate

−6
−3
10 Wm
−8

−4
10 −10

−12
SISO−OFDM (No clipping, Perfect CSI)
−5
10 IAR for SISO−OFDM w/o Cm (I=0, Perfect CSI)
IAR for SISO−OFDM with Cm (I=0, Perfect CSI)
IAR for SISO−OFDM with Cm (I=1, Perfect CSI) −14
IAR for SISO−OFDM with Cm (I=2, Perfect CSI) Cm
IAR for SISO−OFDM with Cm (I=2, Estimated CSI)
−6
10 −16
4 6 8 10 12 14 16 18 0 100 200 300 400 500 600 700 800 900 1000
Average SNR [dB] Subchannel Index

Fig. 6. BER performance of the IAR for SISO-OFDM over BU Fig. 7. Mean squared error of clipped OFDM before the first iteration
channel (N=1024, 16-QAM, CR=0dB, fd ts =0.0001). (N=1024, 16-QAM, BU channel, SNR=20dB, CR=0dB).

antennas is large. To cope with this problem, we design 10

orthogonal pilot symbol blocks for each transmit antenna.

Using this design, only one OFDM block is needed to estimate −1
10

the entire MIMO channels. In the case of two transmit and two
receive antennas, a simple way of designing such pilot symbol −2
10
blocks is by sending pilot symbols on even subcarriers of the
first antenna, and zeros on those subcarriers for the second
Bit Error Rate

antenna. For the odd subcarriers, the second antenna sends

−3
10

pilot symbols, while the first antenna transmits zeros. In order

to have constant amplitude pilot sequences in time domain, −4
10

like the SISO-OFDM, the Chu sequence with length-N/2 can

be used as pilot symbols for the MIMO-OFDM as follows −5
SISO−OFDM (No clipping, Perfect CSI)

  10 IAR for SISO−OFDM with Cm (I=0, Perfect CSI)

IAR for SISO−OFDM with Cm (I=2, Perfect CSI)
Xp1,m = CN/2 [0], 0, CN/2 [1], 0, ...., CN/2 [N/2 − 1], 0 , IAR for SISO−OFDM with Cm (I=2, Estimated CSI)
  DAR for SISO−OFDM with Cm (I=2, Perfect CSI)

Xp2,m = 0, CN/2 [0], 0, CN/2 [1], ...., 0, CN/2 [N/2 − 1] (37)

where the subscript “p ” represents the pilot symbol block.
Note that xp1,m (time domain representation of Xp1,m ) ia a
simple repetition of the time domain Chu sequence with
length-N/2, and xp2,m is the phase rotated version of the
xp1,m . Therefore, both xp1,m and xp2,m have the desirable unit
PAPR property. The nulled subcarriers are reconstructed at the
receiver by using the interpolation algorithm described in [24].
B. Simulation Results
The BER performance of IAR for clipped SISO and MIMO-
OFDMs are investigated through computer simulations. We
consider coded OFDMs with 1024 subcarriers (N=1024), 16-
QAM constellation, 5MHz bandwidth, and a 1/2-rate convolu-
tional code with constraint length of 3. Every frame consists of
one pilot symbol block and 10 OFDM data blocks. Decoding
is carried out by using the maximum a posteriori (MAP)
algorithm [26], which minimizes the bit error probability.
Cyclic prefix is appended to each OFDM block to eliminate
the inter-symbol interference caused by multipath channels,
and 6-tap typical urban (TU) and BU channels with 0.0001
normalized Doppler frequency (fd ts ) are used. The total
transmit power from MIMO-OFDM antennas is set to the
same as the power of SISO-OFDM, and MMSE equalizer is
applied to the nonclipped OFDM systems.
DAR for SISO−OFDM with Cm (I=2, Estimated CSI)
−6
10
4 6 8 10 12 14 16 18
Average SNR [dB]
Fig. 8. BER performance comparison of the IAR and DAR over TU
channel (N=1024, 16-QAM, CR=0dB, fd ts =0.0001).
Fig. 6 shows the performance of IAR for SISO-OFDM with
and without Cm [n] over the BU channel (see Step 3 of the
IAR in Section III-A). Only Wm [n] is used for the system
without optimum equalization of Cm [n]. It is observed that the
optimum equalization method increases the reliability of the
initial estimates, and the receiver with the optimum equalizer
dramatically improves the overall system performance. With
only two iterations, the BER of IAR with Cm [n] approaches
the nonclipped system at SNR=18dB. In order to explain the
effect optimum equalization more clearly, the mean squared
errors (MSEs) of the clipped system with and without Cm [n]
are given in Fig. 7. The MSEs are measured at the outputs of
the equalizers. It shows that the Cm [n] reduces the MSEs
about 2dB at all subcarriers except the nulled subcarriers.
The reduction of the MSE leads to a reliable estimation
of the transmitted symbols before the first iteration, which
facilitates the iterative IAR process. In Fig. 8, the performance
of IAR and DAR over TU channel is compared, where the
optimum equalization is used for both of the schemes. The
IAR outperforms the DAR. As analyzed in Section III-C, the

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:54 from IEEE Xplore. Restrictions apply.
276 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 1, JANUARY 2009

−1 0
10 10
SISO−OFDM (No clipping, Perfect CSI)
IAR for SISO−OFDM w/o Cm (I=0, Perfect CSI)
IAR for SISO−OFDM with Cm (I=0, Perfect CSI)
IAR for SISO−OFDM with Cm (I=2, Perfect CSI) −1
10
DAR for SISO−OFDM with Cm (I=2, Perfect CSI)
−2
10

−2
10

−3
10 −3

Bit Error Rate

Bit Error Rate

10

−4
−4
10
10

−5
10

−5
10 2X2 SFBC−OFDM (No clipping, Perfect CSI)
−6 IAR for proposed 2X2 SFBC−OFDM w/o Cm (I=0, Perfect CSI)
10 IAR for proposed 2X2 SFBC−OFDM with Cm (I=0, Perfect CSI)
IAR for proposed 2X2 SFBC−OFDM with Cm (I=3, Perfect CSI)
IAR for proposed 2X2 SFBC−OFDM with Cm (I=3, Estimated CSI)
IAR for conventional 2X2 SFBC−OFDM with Cm (I=3, Perfect CSI)
−6 −7
10 10
0 0.5 1 1.5 2 2.5 3 3.5 4 0 2 4 6 8 10 12
Clipping Ratio [dB] Average SNR [dB]

Fig. 9. BER performance comparison of the IAR and DAR over BU Fig. 11. BER performance of the IAR for SFBC MIMO-OFDM over
channel (N=1024, 16-QAM, SNR=16dB, fd ts =0.0001). BU channel (N=1024, 16-QAM, CR=0dB, fd ts =0.0001).

0
10
optimum equalizer Cm [n] increases the reliability of initial
−1
10
estimates, and thus the IAR for proposed SFBC-OFDM sig-
nificantly improves the overall system performance. The result
−2
of IAR for conventional SFBC-OFDM in Section IV-B is also
10
1,m [n]
given in the figure. For the amplitude reconstruction, X
−3

and X2,m [n] are obtained from the SFBC encoding of Sm [2v]
and Sm [2v + 1] in (28). Due to the destruction of orthogo-
Bit Error Rate

10

−4
10
nality, it shows approximately one order of BER degradation
compared to the proposed SFBC-OFDM at SNR=12dB. In
−5
addition, it requires two pairs of N -point IFFT/FFT operations
10
per each iteration, while the proposed receiver requires two
pairs of N/2-point IFFT/FFT operations. In Figs. 10 and 11,
2X1 STBC−OFDM (No clipping, Perfect CSI)
−6
2X2 STBC−OFDM (No clipping, Perfect CSI)
10 IAR for 2X2 STBC−OFDM w/o Cm (I=0, Perfect CSI)
IAR for 2X2 STBC−OFDM with Cm (I=0, Perfect CSI) the performance of MIMO-OFDMs with channel estimation
IAR for 2X2 STBC−OFDM with Cm (I=3, Perfect CSI)

−7
IAR for 2X2 STBC−OFDM with Cm (I=3, Estimated CSI) is also given. Note that the clipped systems with channel
10
0 2 4 6
Average SNR [dB]
8 10 12 estimation have slight performance degradation compared to
Fig. 10. BER performance of the IAR for STBC MIMO-OFDM over the systems with perfect CSI, and highly accurate channel
BU channel (N=1024, 16-QAM, CR=0dB, fd ts =0.0001). estimation for the MIMO-OFDM employing a clipping can be
done by using the Chu sequence design described in Section
IV-A, which has the desirable unit PAPR property.
difference of performance between the IAR and DAR is due
to the residual error propagating to the next iteration (see Step VII. C ONCLUSIONS
5 of the IAR in Section III-A). In Fig. 9, the results of Figs. We first proposed clipped signal reconstruction methods
6, 7 and 8 are summarized by the BER curves for different for MIMO-OFDMs based on the IAR. Since the IAR is an
CRs. The advantage of Cm [n] is maintained through all CR iterative technique, the performance of IAR largely depends
values, and highlighted especially at low CR values. Due to on the reliability of initial estimates and the propagation
the propagation error to the next iteration, the DAR works error to the next iteration. Theoretical analysis showed that
well only at high CR values, while the IAR approaches the the optimum equalization of IAR increases the reliability of
nonclipped SISO-OFDM through all CR values. In Figs. 6 initial estimates and its amplitude reconstruction halves the
and 8, the performance of SISO-OFDMs with the channel power of propagation error to the next iteration compared
estimation is also given. It is observed that there is almost no to the DAR. Further, we showed that the IAR can not be
performance degradation compared to the systems with perfect directly employed for the SFBC-OFDM due to the dependency
CSI. of sequences over each transmit antenna, while it can be
Fig. 10 shows the BER performance of IAR for STBC- easily applied to the STBC-OFDM. We proposed a new SFBC
OFDM. The performance of clipped 2×2 STBC-OFDM with- transmitter for clipped OFDM, which has approximately half
out the signal reconstruction is even worse than that of the 2×1 the computational complexity of conventional SFBC-OFDM
STBC-OFDM. However, with three iterations, the proposed and can also be applied to nonclipped OFDM. The proposed
receiver approaches the nonclipped 2×2 STBC-OFDM with clipping preserves the orthogonality of transmitted signals, and
only 0.3dB SNR gap at BER=10−6 . The performance of IAR the clipped signals were iteratively recovered at the receiver.
for SFBC-OFDM is shown in Fig. 11. It is observed that the Finally, we presented that accurate channel estimation of the

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:54 from IEEE Xplore. Restrictions apply.
KWON et al.: AMPLITUDE CLIPPING AND ITERATIVE RECONSTRUCTION OF MIMO-OFDM SIGNALS WITH OPTIMUM EQUALIZATION
clipped OFDM systems can be done with a sequence, which
has constant amplitude in both frequency and time domains.
Although we focused on two transmit antennas, the proposed
techniques can be easily extended to quasi-orthogonal signal
designs with four and eight transmit antennas [27]. Extensive
simulation results show that the proposed receivers effectively
recover the amplitude of clipped OFDMs with a moderate
computational complexity.
R EFERENCES
[1] Y. S. Kim, B. J. Jeong, J. H. Chung, C. S. Hwang, J. S. Ryu, K. H.
Kim, and Y. K. Kim, “Beyond 3G: vision, requirement, and enabling
technologies,” IEEE Commun. Mag., vol. 41, pp. 120-124, Mar. 2003.
[2] J. Chuang and N. Sollenberger, “Beyond 3G: wideband wireless data
access based on OFDM and dynamic packet assignment,” IEEE Commun.
Mag., vol. 38, pp. 78-87, July 2000.
[3] S. M. Alamouti, “A simple transmit diversity technique for wireless
communications,” IEEE J. Select. Areas Commun., vol. 16, pp. 1451-
1458, Oct. 1998.
[4] D. Agrawal, V. Tarokh, A. Naguib, and N. Seshadri, “Space-time coded
OFDM for high data-rate wireless communication over wideband chan-
nels,” in Proc. IEEE Vehic. Tech. Conf., vol. 13, pp. 2232-2236, May
1998.
[5] H. C. Won and G. H. Im, “Iterative cyclic prefix reconstruction and
channel estimation for a STBC OFDM system,” IEEE Commun. Lett.,
vol. 9, pp. 307-309, Apr. 2005.
[6] K. Lee and D. Williams, “A space-frequency transmitter diversity tech-
nique for OFDM systems,” in Proc. IEEE Global. Telecom. Conf., vol.
3, pp. 1473-1477, Nov. 2000.
[7] J. H. Jang, H. C. Won, and G. H. Im, “Cyclic prefixed single carrier
transmission with SFBC over mobile wireless channels,” IEEE Signal
Processing Lett., vol. 13, pp. 261-264, May 2006.
[8] A. E. Jones, T. A. Wilkinson, and S. K. Barton, “Block coding scheme
for reduction of peak to mean envelope power ratio of multicarrier
transmission schemes,” Elect. Lett., vol. 30, pp. 2098-2099, Dec. 1994.
[9] G. Yue and X. Wang, “A hybrid PAPR reduction scheme for coded
OFDM,” IEEE Wireless Commun., vol. 5, pp. 2712-2722, Oct. 2006.
[10] R. Bauml, R. Fischer, and J. Huber, “Reducting the peak-to-average
power ratio of multicarrier modulation by selected mapping,” Electron.
Lett., vol. 32, pp. 2056-2057, Oct. 1996.
[11] S. Muller and J. Huber, “OFDM with reduced peak-to-average power
ratio by optimum combination of patial transmit sequences,” Electron.
Lett., vol. 33, pp. 368-369, Feb. 1997.
[12] X. Li and L. J. Cimini Jr., “Effects of clipping and filtering on the
performance of OFDM,” in Proc. IEEE Vehic. Tech. Conf., vol. 3, pp.
1634-1638, May 1997.
[13] E. Cost, M. Midrio, and S. Pupolin, “Impact of amplifier nonlinearities
on OFDM transmission system performance,” IEEE Commun. Lett., vol.
3, pp. 37-39, Feb. 1999.
[14] M. Friese, “On the degradation of OFDM-signals due to peak-clipping in
optimally predistorted power amplifiers,” in Proc. IEEE Global. Telecom.
Conf., vol. 2, pp. 8-12, Nov. 1998.
[15] H. Ochiai and H. Imai, “Performance of the deliberate clipping with
adaptive symbol selection for strictly band-limited OFDM systems,” IEEE
J. Select. Areas Commun., vol. 18, pp. 2270-2277, Nov. 2000.
[16] L. Wang and C. Tellambura, “A simplified clipping and filtering tech-
nique for PAR reduction in OFDM systems,” IEEE Signal Processing
Lett., vol. 12, pp. 453-456, June 2005.
[17] U. K. Kwon, G. H. Im, and E. S. Kim, “An iteration technique for
recovering insufficient CP and clipped OFDM signals,” IEEE Singal
Processing Lett., vol. 14, pp. 317-320, May 2007.
[18] U. K. Kwon and G. H. Im, “Iterative amplitude reconstruction of clipped
OFDM signals with optimum equalization,” Electron. Lett., vol. 42, pp.
1189-1190, Sept. 2006.
[19] D. Kim and G. L. Stüber, “Clipping noise mitigation for OFDM by
decision-aided reconstruction,” IEEE Commun. Lett., vol. 3, pp. 4-6, Jan.
1999.
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:54 from IEEE Xplore. Restrictions apply.
277
[20] H. Saedi, M. Sharif, and F. Marvasti, “Clipping noise cancellation in
OFDM systems using oversampled signal reconstruction,” IEEE Com-
mun. Lett., vol. 6, pp. 73-75, Feb. 2002.
[21] J. Tellado, L. Hoo, and J. Cioffi, “Maximum-likelihood detection of
nonlinearly distorted multicarrier symbols by iterative decoding,” IEEE
Trans. Commun., vol. 51, pp. 218-228, Feb. 2003.
[22] P. Zillmann, W. Rave, and G. Fettweis, “Soft Detection and Decoding
of Clipped and Filtered COFDM Signals,” in Proc. IEEE Vehic. Tech.
Conf., pp. 1598-1602, Apr. 2007.
[23] J. G. Proakis, Digital Communications. McGraw-Hill, 2001.
[24] J. Siew, R. J. Piechocki, A. Nix, and S. Armour, “A channel estimation
method for MIMO-OFDM systems,” in Proc. London Communications
Symp. 2002-UCL, pp. 101-104, Sept. 2002.
[25] D. Chu, “Polyphase codes with good periodic correlation properties,”
IEEE Trans. Inform. Theory, vol. 18, pp. 531-532, July 1972.
[26] J. P. Woodard and L. Hanzo, “Comparative study of turbo decoding
techniques: an overview,” IEEE Trans. Veh. Technol., vol. 49, pp. 2208-
2233, Nov. 2000.
[27] H. Jafarkhani, “A quasi-orthogonal space-time code,” IEEE Trans.
Commun, vol. 49, pp. 1-4, Jan. 2001.
Ui-Kun Kwon received the B.S. degree in electronic
and electrical engineering from Kyungpook National
University, Daegu, Korea, in 2003. He is currently
working toward the Ph.D. degree at Communica-
tions Research Laboratory, Pohang University of
Science and Technology (POSTECH), Pohang, Ko-
rea. His current research interests are adaptive sig-
nal processing, cooperative communications, and
MIMO systems.
Dongsik Kim received the B.S. degree in electronic
and electrical engineering from Pohang University
of Science and Technology (POSTECH), Kyungbuk,
Korea, in 2005. He is currently working toward the
Ph.D. degree at Communications Research Labora-
tory, POSTECH. His current research interests are
cooperative communications and MIMO systems.
Gi-Hong Im (M’87-SM’94) was with AT&T Bell
Laboratories, Holmdel, NJ, where he was respon-
sible of design and implementation of high-speed
digital transmission systems for loop plant, local
area network and broadband access applications
(1990-1996). He has authored or co-authored more
than twenty standards contributions to standards
organizations such as ANSI T1E1.4, ETSI, IEEE
802.9, ANSI X3T9.5, and the ATM Forum. These
contributions have led to the adoption of three AT&T
proposals for new standards for high-speed LANs
and broadband access. In 1995, he was appointed as Distinguished Member
of Technical Staff at AT&T Bell Laboratories. Since 1996, he has been with
Postech as a professor. From 1996 to 2000, he was a Bell Laboratories
Technical Consultant. From 2002 to 2003, he was a visiting vice president
of Samsung Electronics, where he worked on 4G wireless communication
systems. His current research interests include signal processing and digital
communications with applications to high-speed digital transmission systems.
Dr. Im received the 1996 Leonard G. Abraham Prize Paper Award from the
IEEE Communications Society, and the 2000 LG Award from LG Electronics,
and the 2005 National Scientist Award from Korea government. He serves
as an editor for the IEEE T RANSACTIONS ON C OMMUNICATIONS and an
associate editor for the IEEE C OMMUNICATIONS L ETTERS . He holds ten
U.S. patents with seven more patents pending.