IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 62, NO.

1, JANUARY 2014 293

A Blind Fine Synchronization Scheme for

SC-FDE Systems
Ying-Tsung Lin, Student Member, IEEE, and Sau-Gee Chen, Member, IEEE

Abstract—This work presents a blind fine synchronization certain frequency-domain signals. Since accumulated STO and
scheme, which estimates and compensates residual carrier- RCFO effects can significantly degrade system performances,
frequency offset (RCFO) and symbol timing offset (STO) , for in order to resolve this problem, various data-aided methods
single-carrier frequency-domain equalization (SC-FDE) systems.
Existing fine synchronization schemes for SC-FDE systems rely [6]-[13] have been proposed to perform fine synchronization
on time-domain unique words (UW) sequences as reference in frequency domain, for either OFDM or SC-FDE systems.
signals to assure the estimation accuracy, at the cost of decreased Techniques of linear least-square (LLS) [11] and simple
system throughput. The proposed technique, named simplified weighted least square (SWLS) [8] were proposed to estimate
weighted least-square method for single-carrier systems (SWLS- STO and RCFO based on known reference signals, such as
SC), combines the decision feedback structure and SWLS es-
timator for OFDM systems. Together with specifically derived pilot subcarriers in OFDM systems. The SWLS method has
weighting factors, it has much better estimation accuracy than been proven to be more robust to multipath channels than LLS
the well-known linear least-square (LLS) method for SC-FDE method, because it weights the received signal with estimated
systems, and its BER performance can approach that of the power of channel frequency response such that the estimation
ideal synchronization condition. The proposed technique is more error due to deep-fading frequency response can be minimized.
effective than existing techniques, in terms of both performance
and throughput. Theoretical estimation bounds are also derived Further, [12] proposes an ML estimation method on RCFO and
to verify the effectiveness of the proposed method. STO for OFDM systems. However, these frequency-domain
estimation methods are developed specifically only for OFDM
Index Terms—Synchronization operations, symbol timing off-
set, residual carrier-frequency offset, simple weighted least systems by exploiting pre-defined pilot subcarriers but cannot
square, SC-FDE systems. be applied to SC-FDE systems. In order to apply SWLS
concept to SC-FDE systems, the method in [9] performs joint
I. I NTRODUCTION RCFO and STO estimations by taking advantage of a special
SC-FDE block format defined in [2]. Nevertheless, it only

S INCE recent years, SC-FDE techniques have been pop-

ularly applied to many modern wireless communica-
tion systems, such as IEEE 802.11.ad [1] and IEEE 802.16
works for such a special block format, which requires multiple
unique word (UW) sequences arranged sequentially in a row,
and is not a general solution for the fine-synchronization
Wimax [2]. The data format of a SC-FDE system is gen- problem of SC-FDE systems. More importantly, the insertion
erally composed of data blocks, each preceded with a of multiple UWs causes degradation in system throughput.
cyclic prefix (CP) such that low-complexity and accurate
frequency-domain channel equalization can be effectively ap- Based on our preliminary work [18], this work generalizes
plied. Similar to SC-FDE systems, distributed and localized the concept to SC-FDE systems and presents a blind tech-
single-carrier frequency-division multiple access (SC-FDMA) nique, named SWLS-SC, for joint RCFO and STO estimations
schemes adopted in LTE standard [17] also facilitate low- without needing additional UW sequences. In addition, a
complexity frequency-domain equalization in combating inter- receiver structure based on the proposed SWLS-SC estimator
symbol interferences [3]-[5]. Major advantages of SC-FDE is also illustrated in this work. The proposed technique mainly
over OFDM are its lower peak-to-average power ratio (PAPR) combines techniques of decision feedback [7] and SWLS
and less sensitivity to carrier-frequency offset (CFO). estimator to perform fine synchronization. In addition, to
However, compared to OFDM systems, problems and tech- further deal with random frequency responses of SC signals
niques of fine synchronization for SC-FDE systems are much and multipath effects, a new method for refining weighting
less discussed in the literature. The issue of fine synchroniza- coefficients are also derived. Numerical performance analysis
tion is mainly to estimate and compensate STO and RCFO, shows that SWLS-SC method is robust to severe multipath
which can be estimated jointly by extracting the phases of channels. Moreover, BER simulations demonstrate that the
proposed blind synchronization scheme can produce results
Manuscript received March 18, 2013; revised August 29, 2013. The editor very close to those under ideal synchronization condition
coordinating the review of this paper and approving it for publication was A.
Tonello.
for light-of-sight (LOS) channels [14]. Theoretical estimation
The authors are with the Department of Electronics Engineering and the error of the proposed scheme is also derived and verified with
Institute of Electronics, National Chiao Tung University, Hsinchu city, Taiwan simulations in order to demonstrate its effectiveness.
(e-mail: ytlin.ee97g@nctu.edu.tw, sgchen@cc.nctu.edu.tw).
This work was presented in part at the IEEE International Symposium on This paper is organized as follows. Section II describes the
Circuits and Systems (ISCAS), Seoul, Korea, May 20-23, 2012. signal model of SC-FDE systems under effects of STO and
This work is supported in part by National Science Council, Taiwan under
the grants of NSC 101-2220-E-009-025 and NSC 101-2219-E-009-020. RCFO. In Section III, an existing work [9] on applying SWLS
Digital Object Identifier 10.1109/TCOMM.2013.112413.130209 method to SC-FDE systems is briefly discussed. The proposed
0090-6778/14$31.00 
c 2014 IEEE
294 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 62, NO. 1, JANUARY 2014

N N tively. In addition, wi [n] ∼ CN (0, N0 ) denotes the time-

CP Burst Data CP Burst Data domain noise. Furthermore, by taking discrete Fourier trans-
NCP
form (DFT) of (2), the received sample in the k-th frequency
point of the i-th block is
(a)
N N Zi [k] = (ej2π(i(N +NCP )+NCP )φkk /N ej2π(N −1)φkk /N ×
λN (φkk )Hi [k]Xi [k]) + Ii [k] + Wi [k],
UW Burst Data UW UW Burst Data UW
(3)
NCP where Wi [k] ∼ CN (0, N N0 ) and Ii [k] is the inter-carrier
(b) interference (ICI) given by
N/2
Fig. 1. Block structure of the (a) data block without UW, (b) data block Ii [k] = m=−N/2+1,m=k ej2π(i(N +NCP )+NCP )φkm /N
with UW. ej2π(N −1)φkm /N × λN (φkm )Hi [m]Xi [m].
(4)
In (3) and (4), φkm is defined as φkm ≡ (1 + δ)( + m) − k
algorithm along with its theoretical performance analysis is
and λ(φkm ) is:
presented and discussed in Section IV. Simulation results and
conclusions are given in sections V and VI, respectively. sin{π[(1 + δ)( + m) − k]}
λN (φkm ) = . (5)
N sin{ π[(1+δ)(+m)−k]
N }
II. S YSTEM M ODEL F OR SC-FDE S YSTEMS W ITH STO
AND RCFO Note that, in this work, RCFO is assumed to be relatively
small as compared with the frequency sample spacing and,
Fig. 1 shows two different data block formats for a SC- hence, Ii [k] is ignored. Besides, if δ is also assumed to be
FDE system. Fig. 1(a) illustrates a general format of a SC- small, then φkk can be reduced to  + δk. As a result, (3) can
FDE block where cyclic-prefix (CP) is a copy of the last be approximated as
NCP samples in the block of N data samples. With the
1 j2π(i(N +NCP )+NCP )φkk /N j2π(N −1)(+δk)/N
help of CP, frequency-domain equalization can be applied to Zi [k] ≈ Ne e
such systems as in OFDM systems to lessen the multipath Hi [k]Xi [k] + Wi [k],
effect with low computational cost. However, unlike OFDM, (6)
SC-FDE systems do not have pilot subcarriers to help fine In (6), the effect of STO contributes to a slope change in the
synchronizations in frequency domain. Therefore, as shown in phase shift which is proportional to the frequency index k,
Fig. 1(b), the predefined UW sequence is inserted and served while the phase shift caused by RCFO is independent of k.
as a reference sequence to ensure the estimation accuracy
based on existing fine synchronization algorithms. However, III. T HE SWLS J OINT E STIMATIONS OF STO AND RCFO
for higher throughput consideration, this work only considers FOR SC-FDE S YSTEMS
the data block format without UW sequences in Fig. 1(a), and
only its system model is discussed in the remaining part of In this section, an SWLS fine synchronization estimator
this section. [9] for SC-FDE systems is briefly discussed. In Fig. 1(b),
In Fig. 2, a basic block diagram of a SC-FDE system is it is noted that there are two consecutive UW sequences,
shown. One of its major differences with conventional SC where the first one is the last Ncp samples of the first data
systems is the insertion of CP at the transmitter side as well as block and the second one is CP of the second data block.
the CP removal at the receiver side. If denoting z(t) and Ts as Moreover, since the UW sequence is known in advance at
the received signal subject to multipath fading effect h(t) and the receiver, the preceding UW sequence can be served as
the duration of one symbol, respectively, z(t) can be written the CP of its following UW sequence. SWLS estimator [9]
as takes Ncp -point DFT operation of the latter UW sequence for
 (M−1)/Ts
frequency-domain SWLS estimations. Similar to (6), one can
z(t) = h(τ )x(t − τ )dτ + w(t), (1) obtain the following frequency-domain signal Z̃i [k] by taking
0
DFT operation of the i-th received UW sequence, and treating
where x(t), w(t) and M are the transmitted signal, the AWGN it as the CP of each data block:
and the number of channel taps, respectively. By defining
normalized RCFO and STO factors as  and δ, the n-th Z̃i [k] = λNCP (φkk )ej2πi(N +NCP )phikk /NCP
(7)
received sample in the i-th block in the presence of fine + H̃i [k]X̃i [k] + W̃i [k],
synchronization error, RCFO /(N Ts ) and STO δTs , is
where H̃i [k], X̃i [k] and W̃i [k] denote the channel frequency
zi [n] = ej2π[t/(N Ts )] z(t)|t=Ts (1+δ)(i(N +Ncp )+NCP )n response, the UW sequence and combined effects of AWGN
= (ej2π(i(N +NCP )+NCP )(1+δ)n/N and ICI, respectively. Without loss of generality, H̃i [k] is
N/2
× N1 m=−N/2+1 e
j2πnm/N j2πnδm/N
e assumed to be relatively static between successively received
Hi [m]Xi [m]) + wi [n], for 0 ≤ n ≤ N − 1, blocks and X̃i [k] (with its average power Ẽi [k]) is also
(2) known at the receiver. Then, assuming negligible ICI effect,
where Hi [m] and Xi [m] are the channel frequency response the information of and can be extracted by taking the
and transmitted signal at the m-th frequency point, respec- conjugate product of X̃i [k] between two successively received
LIN and CHEN: A BLIND FINE SYNCHRONIZATION SCHEME FOR SC-FDE SYSTEMS 295

w(t)
xi[n] x(t)
Input Modulation CP Channel
DAC
Bitstream Mapping Insertion h(t)

z(t)
zi[n]
Decoded Data Freq Domain CP Synchronization
IFFT FFT ADC
Bitstream Decoder Equalization Removal Error

Channel
Estimation

Fig. 2. A block diagram of conventional SC-FDE systems

data blocks as A. The SWLS-SC Estimator

∗
M̃i [k] = Z̃i [k]Z̃i−1 [k] Fig. 3 shows a receiver structure based on SWLS-SC
j2π(N +NCP )φkk /NCP
≈e | H̃i [k] |2 Ẽi [k] + noise, estimator. Prior to minimum-mean-square-error frequency-
(8) domain equalizer (MMSE-FDE) [4][5], three additional mod-
where the noise term is the sum of all other interfering effects. ules, including an interpolator, a frequency compensator and
Subsequently, the total phase shift due to STO and RCFO can DFT window adjustment, are required for fine synchroniza-
be obtained as tion, compared with Fig. 2. As shown, the frequency compen-
θ̃i [k] = NCP sator module is to eliminate the RCFO effect in the received
2π(N +NCP ) arg(M̃i [k]) + ẽi [k], (9)
signal by utilizing the estimated RCFO. Besides, DFT window
where ẽi [k] denotes the overall phase error at the k-th fre- adjustment literally means to adjust the DFT sample window
quency point of the i-th received block. To avoid effects and remove CP from the received signal, by using the esti-
of the direct-current offset, the technique in [9] excludes mated STO. The interpolator serves to oversample the received
frequency samples at the DC with indices k ∈ K, where signal in order to achieve better decoding performance for
K = {k : k = ±1, ±2, · · · , ±Q}. Parameter Q is chosen either SC or SC-FDE systems.
empirically as 0.4NCP as suggested in [9] based on the In Fig. 3, X̂i [k] is defined as the frequency-domain signal of
tradeoff between the computational complexity and estimation x̂i [n], and is equivalent to Xi [k] + Pi [k], where Pi [k] denotes
performance. As a result, the RCFO and STO estimates due the overall decoding error. Different from SWLS estimator,
to SWLS estimator are given as the proposed estimator extracts RCFO and STO by taking
 conjugate product Mi [k] of X̂i [k] and Yi [k] as shown below
NCP k∈K C̃i [k]θ̃i [k]
˜i =  , (10)
2π(N + NCP ) k∈K C̃i [k] Mi [k] = X̂i∗ [k]Yi [k]
j2π(i(N +N
CP )+NCP )φkk j2π(i(N −1)(+δk)
1
and  ≈ Ne
N e N (12)
k∈K k C̃i [k]θ̃i [k]
2 ∗ ∗ ∗
NCP |Hi [k]| X̂i [k]Xi [k] 1 Hi [k]X̂i [k]Wi [k]
δ̃i =  . (11) × |Hi [k]|2 +1/ηi + N |Hi [k]|2 +1/ηi .
2π(N + NCP ) 2
k∈K k C̃i [k]
Next, assuming that the RCFO and STO of previously received
Coefficient C̃i [k] is derived to be roughly equal to H̃i [k]. block are well estimated and compensated, then (12) can be
The purpose of this weighting coefficient is to reduce the approximated as
effect for those heavily-faded frequency samples owing to the
frequency-selective channel. 1 j2π(i(N −1)(+δk) |Hi [k]|2 |X̂i∗ [k]|2
Mi [k] ≈ Ne
N
|Hi [k]|2 +1/ηi
However, when applying SWLS estimator to the frame ∗ ∗
1 Hi [k]P̂i [k]Xi [k]
structure as in Fig. 1(a), the assumption of constant signal + N |Hi [k]|2 +1/ηi
Hi∗ [k]Xi∗ [k]Wi [k] (13)
power in (8) fails to hold, because transmitted time-domain + N1 |H 2
i [k]| +1/ηi
symbols are randomly generated and independent between ∗ ∗
1 Hi [k]P̂i [k]Wi [k]
+ N |Hi [k]|2 +1/ηi .
successive data blocks. Hence, SWLS estimator is not a
general solution for all frame formats of SC-FDE systems. In high SNR conditions, is equal or very close to zero.
Therefore, the extracted phase of (13) can be further simplified
IV. T HE P ROPOSED SWLS-SC E STIMATOR FOR SC-FDE as
S YSTEMS θ̂i [k] = arg(Mi [k])
(14)
In this section, the proposed SWLS-SC estimator along with ≈ 2π(N −1)(+δk)
N + êi [k],
its corresponding receiver structure for the data block format
of Fig. 1(a) will be introduced, followed by its theoretical where êi [k] is the overall phase error due to AWGN and
performance analysis and discussion. decoding error. Similar to (10) and (11), estimated RCFO and
296 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 62, NO. 1, JANUARY 2014

H i [k ]
yˆi [n] Zi [ k ]
Data MMSE DFT Window Frequency zi [n]
Decoding IDFT
FDE
DFT Adjustment Compensator
Interpolator

Yi [k ] δˆi Loop δˆi'

STO/RCFO Filter
Symbol
mapping
DFT Estimations εˆi Loop εˆi'
xˆ [n]
i
Xˆ i [ k ] Filter

Fig. 3. The receiver structure based on the SWLS-SC estimator for SC-FDE systems.

STO due to SWLS-SC estimator can be derived as where

N/2−1  
1 1
N k=−N/2+1,k=0 Ci [k]θ̂i [k] W̃i [k] = Xi∗ [k]Wi [k]. (21)
i =
ˆ N/2−1 , (15) N 1 + 1/ηi [k]
2π(N − 1) Ci [k]
k=−N/2+1,k=0
In order to analyze the mean-square-error (MSE) of RCFO
and estimation, one can firstly multiply (20) with a normalization
N/2−1 factor α, which is defined as
N k=−N/2+1,k=0 kCi [k]θ̂i [k]
δ̂i = N/2−1 , (16) (1 + ηi )/ηi
2π(N − 1) k 2 Ci [k] α ≡ N e−j(2π(N −1))/N . (22)
k=−N/2+1,k=0 | Xi [k] |2
where 2 Then, (20) becomes
Ci [k] ≈| Hi [k] |2 ||Ŵ
X̂i [k]|
2
i [k]| (17)
=| Hi [k] |2 η̂i [k], αMi [k] ≈ ej(2π(N −1)δk)/N + αW̃i [k], (23)

and η̂i [k] denotes the estimated SNR at the k-th frequency where αW̃i [k] ∼ CN (0, N0 /Es ). Note that roughly only STO
point. As given in the Appendix, (17) shows the quasi-optimal factor and scaled noise remain in (23), which constitute the
weighting coefficient for SWLS-SC estimator. In addition error in estimating the RCFO. Next, if δ is assumed to be small
to reducing estimation error in frequency-selective channels, enough, the RCFO estimation error e,i for the i-th received
(17) also contains instantaneous SNR information, which can block can be expressed as
enhance estimation accuracy particularly at the locations of e,i ≡ ˆi −  N/2−1
deep frequency notches resulted from the transmitted SC Ci [k] (αMi [k]) (24)
N N/2−1
signal itself. Besides, estimations of (15) and (16) degenerates = 2π(N −1)
k=−N/2+1,k=0
.
Ci [k]
k=−N/2+1,k=0
to an LLS estimation only when is equal to 1 for all k.
In order to further increase the estimation accuracy, the If further assuming that weighting coefficient Ci [k] can be
technique of closed-loop tracking is incorporated and the averagely approximated as Es /N0 , then (24) can be rewritten
resultant estimated RCFO and STO are as
N/2−1 Es 
  (αMi [k])
i = ˆ
ˆ i + α ˆ
i , (18) N
e,i ≈ 2π(N −1) k=−N/2+1,k
N
N/2−1 =0 0 Es
k=−N/2+1,k=0 N0
and N/2−1 (25)
  ≈ 2π(NN−1)2 k=−N/2+1,k=0 (αMi [k])
δ̂i = δ̂i + αδ δ̂i , (19)  N/2−1
≈ 2π(NN−1)2 k=−N/2+1,k=0 (αW̃i [k])
Finally, the estimated RCFO and STO are then passed to mod-
ules of frequency compensator and DFT window adjustment where (•) denotes the imaginary part of a complex variable.
for fine synchronization processing. The derivation reduction to the last line of (25) is by using the
approximation sin(θ) ≈ θ due to small STO. Then, knowing
that (αW̃i [m]) ∼ N (0, N0 /2Es ), one can show that e,i is
B. Theoretical Performance Analysis unbiased. As a result, the MSE for one-shot RCFO estimation
If√ denoting the average amplitude of xi [n] in Fig. (2) error is
as Es for all n, then the average power of Xi [n] in (2) MSE(ˆ i ) ≡ E[e2,i ]
N2 (26)
can be shown to be N Es . In (13), the STO and RCFO of = 8π2 (N −1) 3 (E /N )
s 0
previously received block are assumed to be well compensated
Similarly, one can also follow procedures from (23)-(26) to
and negligible in the performance analysis. Under AWGN
obtain the MSE for one-shot STO estimation as
channel (Hi [k]=1) and error-free reference feedback signal
Mi [k] becomes MSE(δ̂i ) ≡ E[e2δ,i ]
3N (27)
  = 2π2 (N −1)3 (N −2)(Es /N0 )
ej2π(N −1)(+δk)/N ηi
Mi [k] ≈ | Xi [k] |2 +W̃i [k], Following (26) and (27), theoretical MSEs of the incorporated
N 1 + ηi
(20) closed-loop tracking scheme can also be derived by referring
LIN and CHEN: A BLIND FINE SYNCHRONIZATION SCHEME FOR SC-FDE SYSTEMS 297

10
0 TABLE I
S IMULATION P ROFILE

-1 Channel Models CM1.1(LOS) and CM2.1(NLOS) [12]

10
Symbol Rate 1.7GHz for SC mode
Equalizer MMSE-FDE
10
-2
LLS DFT Length 512
RMSE

Cyclic Prefix 128

-3 α and αδ 0.1
10
SWLS-SC
16QAM-NLOS
16QAM-LOS
10
-4
Subsequently, in order to evaluate estimation performance,
BPSK-NLOS
BPSK-LOS root-mean-square error (RMSE) is adopted as the comparison
measure. However, since SWLS estimator for SC systems
-5
10
-0.15 -0.1 -0.05 0 0.05 0.1 0.15
cannot be performed without inserting known sequences, this
RCFO (ε) work only compares LLS estimator as well as the perfect
(a) synchronization scenario for evaluating the effectiveness of the
proposed scheme. Next, LOS and non-LOS (NLOS) channel
10
-4
models defined in [14] are adopted to simulate the robustness
of SWLS-SC under severe channel conditions.
Fig. 4 depicts RMSEs of joint RCFO and STO estimations
under different levels of RCFOs where STO and SNR are
fixed at 20ppm and 24dB, respectively. In Fig. 4(a), simulation
results due to LLS estimator almost overlap with each other
LLS regardless of channel conditions and modulation schemes.
RMSE

10
-5
The reason is that, the frequency response of the reference
signal is random and deeply faded at certain frequency points
where their SNRs are not high enough for accurate estimation
and the performance is thus severely degraded. Note that,
16QAM-NLOS by employing SWLS-SC estimator, RMSE performance is
SWLS-SC 16QAM-LOS
BPSK-NLOS
improved by nearly an order of magnitude, compared with
BPSK-LOS LLS method when RCFO is small, because weighting co-
10
-6 efficients effectively reduce the estimation error caused by
-0.15 -0.1 -0.05 0 0.05 0.1 0.15
RCFO (ε)
either the multipath fading effect or the reference signal itself.
However, for the case of 16QAM with SWLS-SC, the RMSE
(b)
of the estimated STO increases with the CFO value, because
Fig. 4. RMSEs of (a) RCFO and (b) STO estimations versus RCFO, in the constellation of 16QAM is more sensitive to ICI effect,
stationary multipath fading channels, SNR is 24dB and δ is 20ppm. and hence, the quasi-optimality of the weighting coefficients
in (17) is decreased. Fig. 4(b) compares estimated STOs in
different channel conditions and modulation schemes. The
to [16]. Therefore, in the case of small tracking error, steady- results also clearly show that the proposed scheme is much
 
state MSEs for ˆi and δ̂i are less sensitive to ICI effect than LLS estimator regardless of
 α2 channel conditions.
MSE(ˆ
i ) = MSE(ˆ
i ) (28) In addition, similar simulations were also conducted to
2 − α
compare STO and RCFO RMSE performances versus STO in
and Fig. 5, in which a constant RCFO,  = 0.02, is injected and the
 α2δ
MSE(δ̂i ) = MSE(δ̂i ) (29) SNR is fixed at 24dB. Results show that the proposed SWLS-
2 − αδ SC estimator outperforms LLS estimator for both estimations
in low ICI conditions. Particularly, one can observe that STO
V. S IMULATION R ESULTS AND A NALYSIS and RCFO estimates are less sensitive to STO even when STO
is 40ppm. The main reason is that the ICI effect due to STO
The simulation profile is listed in Table I based on the is only about several tens ppm of the sub-frequency spacing,
system parameters of the IEEE 802.11.ad SC mode. In the comparing with the amount caused by the given RCFO effect,
simulation, normalized minimum square error (NMSE) as  = 0.02 which is also equal to 20000ppm of the sub-frequency
defined by (30) of the reference feedback signal is firstly spacing.
analyzed. Next, joint estimation performances versus SNR as well
N/2−1 2 as their theoretical bounds are provided in Fig. 6. Due
k=−N/2+1,k=0 | X̂i [k] − Xi [k] |
NMSE = N/2−1 . (30) to multipath channels, both LLS and SWLS-SC estimators
2
k=−N/2+1,k=0 | Xi [k] | deviate from the theoretical bounds. In Fig, 6(a), SWLS-SC
298 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 62, NO. 1, JANUARY 2014

-1 -2
10 10
16QAM-NLOS
16QAM-LOS
BPSK-NLOS
BPSK-LOS

LLS
-2 -3
10 10

RMSE
RMSE

-3 -4
10 10
16QAM-NLOS
16QAM-LOS
BPSK-NLOS
SWLS-SC BPSK-LOS
Theory
-4 -5
10 10
-40 -30 -20 -10 0 10 20 30 40 0 5 10 15 20 25 30
STO(ppm) SNR

(a) (a)

-4
-4 10
10
16QAM-NLOS
16QAM-LOS
BPSK-NLOS
BPSK-LOS

-5
10

LLS
RMSE
RMSE

-5
10

-6
10 16QAM-NLOS
16QAM-LOS
BPSK-NLOS
BPSK-LOS
SWLS-SC Theory

-6 -7
10 10
-40 -30 -20 -10 0 10 20 30 40 0 5 10 15 20 25 30
STO(ppm) SNR

(b) (b)

Fig. 5. RMSEs of (a) RCFO and (b) STO estimations versus STO, in Fig. 6. RMSEs of (a) RCFO and (b) STO estimations versus SNR, in
stationary multipath fading channels, SNR is 24dB and  is 0.02. stationary multipath fading channels,  = 0.02 and δ = 20ppm.

estimator is shown to perform much closer to the theoretical Euclidean distance between adjacent constellation points are
bounds especially in high SNR regions. However, for the closer than PSK schemes, the overall NMSE is much less than
16QAM case, its RCFO estimation performance degrades PSK schemes. In Fig. 7(a), one can observe that NMSEs of
to LLS one in the low SNR region, because the reference the reference signal X̂i [k] are respectively as low as -34dB, -
signal is much less accurate due to high BER. Hence, once 37dB, -42 dB and -44dB for BPSK, QPSK, 8PSK and 16QAM
BER is small enough, estimation performances for 16QAM modulation schemes when SER is at 10−4 . Furthermore,
and BPSK quickly converges the derived bounds. On the NMSEs for various DFT lengths are also analyzed in Fig.
other hand, Fig. 6(b) shows simulation results of RMSEs 7(b) for BPSK example. Simulation results show that NMSEs
versus SNR for STO estimations. It can be easily observed of different DFT points are almost the same. Therefore, for
that SWLS-SC estimator also outperforms LLS estimator. applications with different DFT lengths, the reference signals
Fig. 7(a) and Fig. 7(b) show NMSE comparison results for remain accurate enough for SWLS-SC estimator. It is because
different modulation schemes and different DFT points of that the time-domain decoding error after DFT operation is
the reference signal, respectively. Note that, each incorrectly distributed throughout the entire bandwidth and its resultant
decoded BPSK, QPSK and 8PSK symbols contribute minimal effect to each individual frequency point is much less severe
phase errors of π, π/2 and π/8 to their respective time- than the fine synchronization algorithm for OFDM systems in
domain symbols. Therefore, under the condition of the same [7] which adopts decision-feedback structure.
symbol error rate (SER), the reference signal utilizing lower Fig. 8 shows performance comparisons of fine synchro-
modulation orders would have higher NMSEs than those with nization schemes based on decision-feedback structure for
higher modulation orders. As for QAM modulations, since the SC-FDE and OFDM systems. Simulation parameters for the
LIN and CHEN: A BLIND FINE SYNCHRONIZATION SCHEME FOR SC-FDE SYSTEMS 299

0 -1
10
16QAM-NLOS
BPSK SWLS-SC 16QAM-LOS
-10 QPSK SWLS-SC
BPSK-NLOS
8PSK SWLS-SC
16QAM SWLS-SC
BPSK-LOS
16QAM-NLOS-OFDM
-20 -2
10 16QAM-LOS-OFDM
BPSK-NLOS-OFDM
NMSE(dB)

RMSE
-30 BPSK-LOS-OFDM

-40 -3
10

-50

-60 -6 -4
-4 -2 0 10
10 10 10 10 0 5 10 15 20 25 30
BER SNR
(a) (a)

0
-4
10
-10

-20 64
128 -5
256 10
NMSE(dB)

512
-30
1024
RMSE

16QAM-NLOS
-40 16QAM-LOS
-6 BPSK-NLOS
10
BPSK-LOS
-50 16QAM-NLOS-OFDM
16QAM-LOS-OFDM
BPSK-NLOS-OFDM
-60 -6 -4 -2 0
BPSK-LOS-OFDM
10 10 10 10 -7
Symbol Error Rate 10
0 5 10 15 20 25 30
(b) SNR
(b)
Fig. 7. NMSE comparisons for (a) different modulation schemes with 512-
point DFT, and (b) different DFT points for BPSK case. Fig. 8. Comparative simulation results for (a) RCFO and (b) STO estimations
for =0.02 and δ=20ppm.

OFDM case are the same as the SC-FDE one for fair compar-
ison consideration. The SWLS-SC estimator provides better
estimation accuracies than OFDM systems under various notice that the performance improvement of BPSK case is
modulation schemes and channel conditions. The performance greater than that of 16QAM case. The main reason is that
gap between these two systems is especially larger for higher every single error bit of the BPSK modulation induces a 180
modulation schemes because of more severe error-propagation degree phase error to its time-domain symbol while one error
effect in OFDM system. Note that differences of signal powers bit for 16QAM case only induces much less phase error due
between subcarriers in OFDM systems vary much less than to the Gray-coding scheme.
that of SC-FDE systems. Therefore, fine-synchronization per- Performance comparisons of SWLS-SC and LLS estimators
formance of OFDM systems outperforms the LLS estimator subject to channel estimation error are illustrated in Fig. 10,
of SC-FDE system as shown in Fig. 6. where ”Ch Err” denotes the power ratio of estimation error
Comparisons of BER performance are illustrated in Fig. over original channel response. Results show that channel
9. Fig. 9 (a) gives the result due to 16QAM scheme. It estimation error is a crucial factor for fine synchronizations,
shows that, the fine synchronization with SWLS-SC estimator especially for residual carrier-frequency-offset (RCFO) esti-
only degrades from the ideal synchronization by 0.2dB and mations. In summary, the proposed estimator maintain better
0.5dB for LOS and NLOS cases, respectively. In addition, robustness than the conventional LLS one. Nevertheless, in
Fig. 9(b) presents the BER performance for BPSK case. IEEE 802.11.ad system, channel estimation error [19][20]
Performance degradations of SWLS-SC estimations from the is very small due to well-designed preamble sequences and
ideal estimations are much less than the LLS one. One can generally slow-fading environments.
300 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 62, NO. 1, JANUARY 2014

-1 -2
10 10

-2
10

-3
10
BER

RMSE
10
-3 LLS, Ideal
SWLS-SC, Ideal
-4 LLS, Ch Err 3.2%
10
16QAM_NLOS_Ideal SWLS-SC, Ch Err 3.2%
16QAM_LOS_Ideal LLS, Ch Err 10%
16QAM_NLOS SWLS-SC SWLS-SC, Ch Err 10%
-5 16QAM_LOS SWLS-SC
10 LLS, Ch Err 32%
16QAM_NLOS LLS SWLS,Ch Err 32%
16QAM_LOS LLS

-4
14 16 18 20 22 24 26 28 10
SNR 0 5 10 15 20 25 30
SNR
(a)
(a)
0
10 10
-4

BPSK_NLOS_Ideal
BPSK_LOS_Ideal
-1
10 BPSK_NLOS SWLS-SC
BPSK_LOS SWLS-SC
BPSK_NLOS LLS
-2
10 BPSK_LOS LLS 10
-5
BER

RMSE

-3
10
LLS, Ideal
SWLS-SC, Ideal
-4
10 10
-6 LLS, Ch Err 3.2%
SWLS-SC, Ch Err 3.2%
LLS, Ch Err 10%
-5 SWLS-SC, Ch Err 10%
10
LLS, Ch Err 32%
SWLS,Ch Err 32%
-6
10 10
-7
0 5 10 15 20 0 5 10 15 20 25 30
SNR SNR
(b) (b)
Fig. 9. Error rate performance for (a) 16QAM and (b) BPSK modulation Fig. 10. Performance comparisons subject to estimation error on (a) RCFO
schemes, =0.02 and δ=20ppm. and (b) STO estimations for the case of 16QAM scheme and LOS channel,
given that =0.02 and δ=20ppm.

VI. C ONCLUSION
A PPENDIX
In this work, an effective SWLS-SC joint estimation method
The quasi-optimal weighting coefficients in (17) are derived
for STO and RCFO is proposed. With the proposed estimator,
based on (13) and (14) as
one can perform fine-synchronizations without using UW
for SC-FDE systems. It is developed based on the SWLS θ̂i [k] = arg(M
estimator and assisted with a decision-feedback structure  i [k]) 
|Hi [k]|2 +1/ηi
= arg N Ẑi [k] |H 2
i [k]| |Xi [k]|
2
to obtain the reference signal. Moreover, the derivation of   
jπ(N −1)(+δk)/N Hi [k]X̂i [k]Ŵi∗ [k]
quasi-optimal weights specifically for SC-FDE system is also = arg e 1+ |Hi [k]|2 |Xi [k]|2
given along with its theoretical bound for estimation error. (31)
Simulation results show that the BER performance of the Assuming high SNR, the approximation, sinθ ≈ θ, can be
proposed estimation is close to those with ideal synchro- applied to simplify the derivation. As a result, with error-free
nizations. Additionally, the feedback reference signal is also reference signal, êi [k] in (14), for small RCFO and STO, is
demonstrated to be accurate enough for various modulation approximately equivalent to
schemes and different lengths of DFT operations without the   
H [k]X̂ [k]Ŵi∗ [k]
need of inserting known sequences. Therefore, by using the êi [k] ≈ arg ejπ(N −1)(+δk)/N |Hi i [k]|i2 |Xi [k]| 2
proposed estimation method, both the system throughput and H [k]X̂ [k]Ŵi∗ [k]
(32)
performance can be simultaneously increased. ≈ |Hi [k]|i2 |X̂ [k]| 2
.
i i
LIN and CHEN: A BLIND FINE SYNCHRONIZATION SCHEME FOR SC-FDE SYSTEMS 301

In [15], it shows that the weighted least-square (WLS) es- [14] V. Erceg et al., “TGn channel models,” IEEE 902.11.document 11-
timation has minimum variance in estimation error if the 03/0940r4.
[15] L. L. Scharf, Statistical Signal Processing, Detection, Estimation, and
coefficient, Di [k], in (16) and (17) is adopted as Time Series Analysis. Addison-Wesley, 1991.
−1 [16] U. Mengali and A. N. D’Andrea, Synchronization Techniques for Digital
Di [k] = E [êi [k]ê∗i [k]] Receivers. Plenum, 1997.
2
≈| Hi [k] |2 ||Ŵ
X̂i [k]|
2
(33) [17] 3GPP TR 36.913, “Technical specification group radio access network;
i [k]| requirements for further advancements for evolved universal terrestrial
=| Hi [k] |2 η̂i [k]. radio access (EUTRA).”
[18] Y. T. Lin and S. G. Chen, “An efficient blind fine synchronization
R EFERENCES scheme for SCBT systems,” in Proc. 2012 IEEE ISCAS, pp. 1066–1070.
[19] Y. T. Lin, C. H. Kuo, and S. G. Chen, “Efficient channel and SNR
[1] “IEEE 802.11ad PAR Doument.” Available:
estimation schemes and their analysis for the mmWave single carrier
http://www.ieee802.org/11/Reports/tgad update.htm.
system,” Int. J. Electrical Engineering, vol. 18, no. 5, pp. 2486-2494,
[2] IEEE P802.16-2004/Cor1/D5 (Draft Corrigendum to IEEE Std. 802.16-
Oct. 2011.
2004), “Corrigendum to IEEE standard for local and metropolitan are-
[20] M. Lei and Y. Huang, “CFR and SNR estimation based on complemen-
anetworks — Part 16: Air interface for fixed broadband wireless access
tary Golay sequences for single-carrier block transmission in 60-GHz
systems,” Sep. 2005.
WPAN,” in Proc. 2009 IEEE WCNC, pp. 1–5.
[3] Z. Wang, X. Ma, and G. B. Giannakis, “OFDM or single-carrier block
transmissions,” IEEE Trans. Commun., vol. 52, no. 3, pp. 380–394, Mar.
2004.
Ying-Tsung Lin received his B.S. and M.S. degrees
[4] D. Falconer, S. L. Ariyavisitakul, A. B. Seeyar, and B. Eidson, “Fre-
in electrical engineering from National Sun Yat-
quency domain equalization for signle-carrier broadband wireless sys-
Sen University and National Tsing-Hua University,
tems,” IEEE Commun. Mag., pp. 58-66, Apr. 2002.
Taiwan, in 2001 and 2003, respectively. In 2007,
[5] H. Sari, G. Karam, and I. Jeanclaud, “Frequency-domain equalization
he received another master of engineering degree
of mobile radio and terrestrial broadcast channels,” in Proc. 1994 IEEE
in electrical and computer engineering from Cor-
GLOBECOM, vol. 1, pp. 1-5.
nell University, NY, USA. Currently, he is working
[6] P. H. Moose, “A technique to orthogonal frequency division multiplexing
toward his PhD degree in National Chiao-Tung
frequency offset correction,” IEEE Trans. Commun., vol. 42, no. 10, pp.
University. His research interests include digital
2908–2914, Oct. 1994.
communication, digital signal processing, and VLSI
[7] K. Shi, E. Serpedin, and P. Ciblat, “Decision-directed fine synchronization
signal processing.
in OFDM systems,” IEEE Trans. Commun., vol. 53, no. 3, pp. 408–412,
Mar. 2005.
[8] P. Y. Tsai, H. Y. Kang, and T. D. Chiueh, “Joint weighted least-
square estimation of carrier-frequency offset and timing offset for OFDM Sau-Gee Chen received his B.S. degree from Na-
systems over multipath fading channels,” IEEE Trans. Veh. Technol., vol. tional Tsing Hua University, Taiwan, in 1978, M.S.
54, no. 1, pp. 211–223, Jan. 2005. degree and Ph.D. degree in electrical engineering,
[9] P. H. Chiang, D. B. Lin, H. J. Li, and G. L. Stuber, “Joint estimation from the State University of New York at Buffalo,
of carrier-frequency and sampling-frequency offsets for SC-FDE systems NY, in 1984 and 1988, respectively. Currently, he
on multipath fading channels,” IEEE Trans. Commun., vol. 56, no. 8, pp. is the chairman and a professor at the Department
1231–1235, Aug. 2008. of Electronics Engineering, National Chiao Tung
[10] H. Yu, M. S. Kim, and J. Y. Ahn, “Carrier frequency and timing offset University, Taiwan. He also serves as the Chair,
tracking scheme for SC-FDE systems,” in Proc. 2003 IEEE PIMPC, no. IEEE Vehicular Technology Society, Taipei Chapter,
1, pp. 1–5. starting from 2012. He was Director of Honors Pro-
[11] A. J. Coulson, “Maximum likelihood synchronization for OFDM usinga gram, College of Electrical & Computer Engineer-
pilot: algorithms,” IEEE J. Sel. Areas Commun., vol. 19, no. 12, pp. 2486- ing/College of Computer Science from 2011-2012; he was Associate Dean,
2494, Dec. 2001. Office of International Affairs, during March-July, 2011, and was the director
[12] M. Morelli and M. Moretti, “Fine carrier and sampling freqeuncy of Institute of Electronics from 2003 to 2006, all at the same organization.
synchronization in OFDM systems,” IEEE Trans. Wireless Commun., vol. During 2004-2006, he served as an associate editor of IEEE T RANSACTIONS
9, no. 4, pp. 1514–1524, Apr. 2010. ON C IRCUITS AND S YSTEMS I. His research interests include digital commu-
[13] C. L. Wang and H. C. Wang, “Optimized joint fine timing synchroniza- nication, multi-media computing, digital signal processing, and VLSI signal
tion and channel estimation for MIMO systems,” IEEE Trans. Commun., processing. He has published more than 100 conference and journal papers,
vol. 59, no. 4, pp. 1089–1098. Apr. 2011. and holds 14 US and Taiwan patents.