IEEE Transactions on Consumer Electronics, Vol. 53, No. 4, NOVEMBER 2007
1348
Robust Timing and Frequency Synchronization Scheme for DTMB System Jianming Wu, Yun Chen, Xiaoyang Zeng, Member, IEEE, and Hao Min, Member, IEEE Abstract — In this paper a robust timing and frequency synchronization algorithm for DTMB system is developed. A two-step frequency offset estimation and compensation process is proposed to perform carrier recovery. Firstly coarse frequency estimation is achieved by utilizing the shiftand-add property of m-sequence. The second step finds the frame start position and the remaining frequency offset simultaneously. Meanwhile a timing tracking strategy is proposed to effectively track the dynamic changes in mobile environment. Thus the proposed scheme can resist large frequency offset and achieve accurate timing and frequency estimation. Simulation results under different channel situations verify the performance of the proposed scheme1. Index Terms — Timing synchronization, Frequency synchronization, orthogonal frequency division multiplexing (OFDM).
I. INTRODUCTION OFDM (Orthogonal Frequency Division Multiplexing) has become the first core technology choice in most wireless transmission because of its natural robustness to multipath fading channels. Recently time domain synchronous OFDM (TDS-OFDM) based digital television terrestrial broadcasting (DTTB) system has been adopted into the China DTV standard, named as Digital Terrestrial Multimedia Broadcasting (DTMB) [1]. Conventionally, a cyclic prefix (CP) is used as guard interval to eliminate inter-symbol interference (ISI) due to multipath, such as DVB-T (Digital Terrestrial Video Broadcasting) standard. However, in TDS-OFDM system, Pseudo Noise (PN) sequences are inserted as guard interval, which also serve as time domain pilots. The usage of PN-sequence as prefix can reduce transmission overhead and provide higher spectrum efficiency. Due to delta-like auto-correlation property of m-sequence, much faster synchronization can be achieved in TDS-OFDM system [2]. In [3], a timing and frequency offset estimation is proposed based on short repeated PN sequence. In [4], a combined code acquisition (CA) and symbol timing algorithm is presented based on searching and tracking of the correlation peaks. But the CA algorithm in [4] could not work properly in the presence of large carrier frequency offset (CFO) [5]. A timing recovery scheme is proposed in [6] based on the assumption that each signal frame has fixed Frame Head 1 The authors are with the State Key Lab of ASIC & System, Fudan University, Shanghai 201203, P.R. China(e-mail:
[email protected]).
Contributed Paper Manuscript received October 12, 2007
(FH) configuration, but in DTMB standard a rotated FH structure is utilized. Also division operation is unavoidable in [6] to find the correct timing, which increases the hardware realization complexity. In this paper a robust timing and frequency estimation strategy is proposed to simultaneously find the frame start position and estimate frequency offset, especially when a large frequency offset exists between transmitter and receiver. The remainder of this paper is organized as follows. Section II describes the TDS-OFDM-based DTMB system structure briefly. In section III, robust timing and frequency joint estimation algorithm is presented. Performance analysis and evaluation is given in section IV and section V concludes the paper. II. DTMB S YSTEM M ODEL As shown in Fig.1, each signal frame in DTMB system consists of two parts, the Frame Head and the Frame Body. The Frame Head can be configured as 420 or 945 symbols in multi-carrier mode specified in DTMB standard. In this paper Frame Head with 420 symbols (FH420) is considered, which comprises a pre-amble, a length 255 PN sequence and a post-amble. The pre-amble and post-amble are cyclical extensions of the PN sequence. The length of pre-amble and post-amble is defined as 82 and 83, respectively. Frame Head is BPSK modulated for robust synchronization and channel estimation. Frame Body has 3780 symbols and takes OFDM modulation scheme. The symbol rate 1/T is same for both Frame Head and Frame Body, which is 7.56MS/s. The basic transmitter and receiver structure of the DTMB system is shown in Fig.2. The complex data symbols s(k) are modulated by means of inverse fast Fourier transform (IFFT) to form frame body. Frame Head is appended in front of each OFDM Frame Body to form one signal frame. A square root raised cosine filter (SRRCF) with roll-off factor 0.05 is used to limit the transmitted signals to 8MHz bandwidth. At the receiver side, the received signals are fed in parallel to the synchronizer and channel estimator, respectively. Synchronization and channel estimation are performed based on correlation between the received signal and a local generated PN sequence. Upon the obtained channel impulse response (CIR), cyclic convolution between Frame Body and CIR is reconstructed by removing PN tailing effect. After PN removal, the reconstructed signal is send to FFT block to obtain demodulated data r(k).
0098 3063/07/$20.00 © 2007 IEEE
J. Wu et al.: Robust Timing and Frequency Synchronization Scheme for DTMB System
1349
Frame Head
Pea-amble
Frame Body
PN Sequence Post-amble Fig. 1 Signal frame format for DTMB system.
s(k)
Time & Frequency Synchronization
IFFT PN Generator
P s(n) r(n) Channel / S
Frame Number
Remove PN
S / P
FFT
r(k)
Channel Estimation Fig. 2 Block diagram of DTMB system.
300
Amplitude
250 200 150 100 50 0 0
100
200
300
400 500 600 Time,samples
700
800
900
1000
700
800
900
1000
(a) 100
Amplitude
80 60 40 20 0 0
100
200
300
400 500 600 Time,samples
(b) Fig. 3 Correlation of r(n) with local PN sequence in the presence of different carrier frequency offset. (a) Δf = 0 (b) Δf = 100 KHz
III. FREQUENCY OFFSET ESTIMATION AND DYNAMIC TRACKING STRATEGY The TDS-OFDM symbol can be expressed as PN ( n ) n ∈ [0, N g − 1) ⎪⎧ (1) s ( n ) = ⎨ N c −1 j 2π ( n − N g ) k / N c n [N g , N g + Nc ) ∈ ⎪⎩ ∑ n = 0 s ( k ) e where Ng and Nc stands for the length of Frame Head and Frame Body, respectively. In the following analysis, we assume that the transmitted signal is only affected by a complex additive white Gaussian noise (AWGN) w(n). However, we will evaluate our estimator performance for both the AWGN channel and frequency selective channel.
In order to correctly find the FFT window, code acquisition (CA) must be performed to capture the PN sequence inserted in each signal frame. The CA process is done by executing correlation between the received signal and a Local PN sequence [4]. Precise symbol boundary can be found due to the delta-like PN auto-correlation property. However, this property will be severely degraded when a large carrier frequency offset exists [5], thus causing ambiguity in finding correct frame start point. Fig. 3(a) and Fig. 3(b) show the correlation result without and with 100KHz carrier frequency offset, respectively. It can be seen that PN peak correlation property will be greatly deteriorated by the carrier offset. Thus coarse frequency estimation and compensation must be finished prior to CA process to recover the auto-correlation property of PN sequence. A. Coarse Frequency Estimation Considering the unknown symbol arrival time θ and the carrier frequency offset (CFO) f, the received signal with additive noise w(n) is given by r ( n ) = s ( n − θ ) e j 2 π nfT + w ( n )
(2)
Due to the CFO, correlation between the received signal r(n) and the local PN sequence PN(n) would result in an attenuated correlation peak. For correct detection of correlation peak, the initial CFO is limited to ±15 KHz in FH420 mode in DTMB standard [5]. Here we propose a novel timing metric to recover the correlation peak in the presence of large CFO, which is defined as R1 ( n ) =
L −1
∑r i=0
*
( n + i ) ⋅ r ( n + i + D ) ⋅ PN ( i − m )
(3)
f coa =
(4) n = n0
where arg( ) denotes argument of complex number measured in radians, n0 represents the correlation peak point. The frequency estimation range given by (4) is limited to 1/(2DT). To ensure a relatively large frequency offset acquisition range, the delay parameter D is chosen to be 10 to perform first stage estimation. After the first stage correction, the frequency offset of the input signal is less than 10KHz with high probability. The delay parameter D, is then set as 50 to achieve better estimation accuracy. After the second stage correction, the residual frequency offset is less than 1KHz, and a fine frequency tracking loop is switched on to acquire and track the residual frequency offset. B. Fine Frequency Estimation and Dynamic Tracking After the coarse carrier frequency offset is compensated, correlation peak can be correctly detected thus CA process can be achieved employing the method mentioned in [4]. Simultaneously the residual frequency offset can also be retrieved from the correlation peak value of two adjacent signal frames. The linear correlation between received signal and local PN sequence can be computed as R2 ( n ) =
L −1
350
300
300
250
250
∑ r ( n + i ) ⋅ PN (i − m )
(5)
i=0
and the residual frequency offset is expressed as arg ( R 2* ( n , l ) ⋅ R 2 ( n , l + 1) ) (6) fε = 2π NT where N represents the OFDM signal frame length N = Nc+Ng, and R2(n, l) and R2(n, l+1) denotes the correlation peak value in the l-th and (l+1)-th OFDM symbol, respectively. The estimated frequency offset is then fed to a first-order DPLL (Digital Phase Locked Loop) to track the residual carrier frequency offset. In real environment, multipath fading will result in a correlation peak for each of the resolvable paths. The amplitude and phase of each path change randomly thus the correlation peak profile will also be time varying. Dynamically tracking with the strongest multipath component
Amplitude
400
350
200 150
0 0
200 150 100
50
50 100
200
300 400 500 Time,samples
600
0 0
700
100
200
300 400 500 Time,samples
600
700
100
200
300 400 500 Time,samples
600
700
(a) 400
350
350
300
300
250
250
Amplitude
400
200 150 100
200 150 100
50 0 0
arg ( R1 ( n ) ) 2π DT
400
100
Amplitude
where the superscript * stands for complex conjugation, L represents the length of PN correlation , m denotes the slip between received PN sequence and the locally generated PN. D is the number of delay elements. Every m-sequence has a cyclic shift-and-add property such that, if the original msequence is rotated and added to itself in modulo 2, the resulting sequence is also a rotated version of the original msequence. Upon this property and the quasi-cyclic structure in the 420 length guard interval, we can see that the product PN*(i)·PN (i+D) in (3) is a cyclic shift of the original PN sequence with constant phase 2πfDT. When the product sequence aligns with the local PN sequence, a correlation peak will be obtained, meanwhile the coarse carrier frequency offset can be estimated as
Amplitude
IEEE Transactions on Consumer Electronics, Vol. 53, No. 4, NOVEMBER 2007
1350
50 100
200
300 400 500 Time,samples
600
700
0 0
(b) Fig. 4 Dynamic tracking under three-path fading channel. (a) Tracking with the strongest multipath component (b) Dynamic adjustment of correlation threshold
is needed in order to find the FFT window correctly. Fig. 4 shows the dynamic tracking strategy under three-path fading channel. The highest correlation peak position is monitored and correlation threshold is dynamically adjusted to match the level of input signal. As shown in Fig. 4(a), two observation windows, called forward and backward search window (FSW and BSW) respectively, are assumed to monitor neighboring paths around the main path. Once stronger correlation peak appears in the search window, the synchronizer will switch to the corresponding path and track it. Furthermore, as the path gain does not remain constant over time, the correlation threshold will be adaptively adjusted according to the detected path amplitude. Fig. 4(b) depicts the situation that all multipath components experience a simultaneous deep fading. In this case the threshold should be reduced accordingly. IV. SIMULATION AND PERFORMANCE ANALYSIS In this section the performance of the proposed timing and frequency synchronization algorithm on the AWGN channel and multipath channels will be evaluated by conducting computer simulations. The simulation parameters are listed in Table I. The multipath channel models are introduced from the DTV test report in Brazil [8]. A. Timing Estimation Performance The performance of code acquisition is evaluated by expressing the probability of synchronization versus the probability of false detection. We will first theoretically analyze the performance under AWGN channel and then give simulation results for multipath channels. For simplicity, we define the amplitude of the normalized timing signal as (7) R ( n ) = R 2 ( n ) Lσ s where σs denotes the standard deviation of the complex input sample. In the case of correct timing under AWGN channel,
J. Wu et al.: Robust Timing and Frequency Synchronization Scheme for DTMB System
1351 10
Symbol Rate 1/T Sample Rate 1/Ts Signal Constellation PN Sequence Length Frame Head Length FFT Size Frame Body Duration Sub-carrier spacing Pulse Shaping Filter
Root Mean Square Error(KHz)
TABLE I SIMULATION PARAMETERS 7.56MS/s 30.4MS/s 64QAM 255 420 3780 500µs 2KHz SRRCF, α=0.05
1
10
10
10
1
AWGN Brazil A Brazil B Brazil C Brazil D Brazil E
0
-1
-2
0
5
10
0.8 Correct Timing,SNR=10dB Correct Timing,SNR=0dB Wrong Timing,SNR=10dB Wrong Timing,SNR=0dB
Root Mean Square Error(KHz)
CDF
0.6
10
0.4
0.2
0 0
0.2
0.4 0.6 Normalized Timing
0.8
1
Fig. 5 CDFs of the timing signal, AWGN channel, Δf=10KHz.
Missed sync probability
10 10 10 10 10 10 10
10
10
-1
10
10
-3
20
Brazil A, Fixed threshold Brazil B, Fixed threshold Brazil E, Fixed threshold Brazil A, Dynamic threshold Brazil B, Dynamic threshold Brazil E, Dynamic threshold
-6
-7
-8
10
-4
10
-3
-2
-1
10 10 False detection probability
0
10
Fig. 6 Missed sync probability vs. false detection probability under different channel situations.
and variance var[| R (0) |] = 1 ( L ⋅ SNR ) where SNR =σ s2 / σ n2 .
10
10
R(n) can be approximated as a Gaussian variable with expectation E [| R (0) |] = sin(π fTL ) L ⋅ sin(π fT )
10
(8) (9)
When the timing is wrong, R(n) can be regarded as Rayleigh distributed with probability distribution function
30
AWGN Brazil A Brazil B Brazil C Brazil D Brazil E
0
-1
-2
-3
0
5
10
15 SNR(dB)
20
25
30
(b)
-1
AWGN Brazil A Brazil B Brazil C Brazil D Brazil E
-4
-5
25
(a)
1
-2
Root Mean Square Error(KHz)
10
10
15 SNR(dB)
-2
-3
-4
0
5
10
15 SNR(dB)
20
25
30
(c) Fig. 7 RMSE of the proposed frequency estimator versus SNR under different channel. (a) CFE (D = 10, Δfini = 200 KHz) (b) CFE (D = 50, Δfini =10 KHz)) (c) FFE
⎛ Lr Lr 2 ⎞ (10) exp ⎜ − 2 2 2 ⎟ σ /σ s σ σ 2 / n s ⎠ ⎝ In Fig.5, the cumulative density function (CDF) of the timing metric is presented for the case of correct and wrong timing, AWGN channel and 10KHz frequency offset. It is f (r ) =
2 n
IEEE Transactions on Consumer Electronics, Vol. 53, No. 4, NOVEMBER 2007
1352
obvious that high detection probability and low false alarm rate can be achieved simultaneously due to the processing gain of correlation. Fig.6 plots the missed synchronization probability versus false detection probability under fast fading channel with 20dB average SNR and maximal Doppler frequency fd =100Hz. The Doppler frequency shift makes the channel time varying, which results in time fluctuation of correlation peaks. If the correlation threshold is set to a fixed value, there exists a certain probability that the correlation peak of correct timing is below the fixed threshold. By adjusting the threshold adaptively to the channel situation and tracking the strongest multipath component, the miss synchronization probability can be remarkably reduced at the same false alarm rate. It can be seen that the proposed dynamic tracking strategy offers better performance than fixed threshold method. B. Frequency Estimation Performance Fig.7 depicts root mean square error (RMSE) of the proposed frequency estimator under different channel model. It is evident from Fig.7 (a) and (b) that the RMSE is well less than 10KHz and 1KHz under 20dB SNR, for parameter D chosen as 10 and 50, initial frequency offset set as 200KHz and 10KHz respectively. Fig.7(c) shows that the FFE estimation can achieve RMSE of about 1Hz under 20dB SNR. From the derivation given in [9], the SNR loss due to residual frequency offset is bounded by 1 + 0.5947 SNR (sin πε ) 2 (11) SNRloss ≤ 10 * log 10 (sin πε /(πε )) 2 where ε denotes the frequency offset normalized to the subcarrier spacing. From (11) it can be easily calculated that the SNR degradation induced by residual frequency offset is less than 0.001dB under 1Hz CFO for 20dB input SNR, which results in negligible performance loss. Hence, the proposed two-stage frequency estimation method can achieve very high estimation accuracy. V. CONCLUSION In this paper, a robust timing and frequency synchronization method is proposed for TDS-OFDM system. The coarse frequency estimation is achieved by exploiting the shift-and-add property of m-sequence. Fine frequency estimation is based on the auto-correlation property of PN sequence. To effectively track fading channel variation, a dynamic timing tracking algorithm is presented to detect the strongest path. In this way the proposed scheme can attain good performance in high mobility environment. Simulation results under different channel situations prove effectiveness of the proposed approach.
REFERENCES [1] [2]
[3] [4] [5] [6] [7] [8] [9]
Framing Structure, Channel Coding and Modulation for Digital Television Terrestrial Broadcasting System, Chinese National Standard GB 20600-2006. J. Song, Z. X. Yang, L. Yang, K. Gong, C. Y. Pan, J. Wang and Y. S. Wu, “Technical Review on Chinese Digital Terrestrial Television Broadcasting Standard and Measurements on Some Working Modes”, IEEE Trans. Broadcast., vol. 53, no. 1, pp. 1-7, March. 2007. F. Tufvesson, M. Faulkner, and O. Edfords, “Timing and frequency synchronization for OFDM using PN-sequence preambles”, Proc. IEEE VTC, pp.2203-2207, Sept. 1999. J. Wang, Z. X. Yang, C. Y. Pan, M. Han and L. Yang, “A combined code acquisition and symbol timing recovery method for TDS-OFDM”, IEEE Trans. Broadcast., vol. 49, no. 3, pp. 304-308, Sept. 2003. G. H. Liu, “Comments on ‘A combined code acquisition and symbol timing recovery method for TDS-OFDM’”, IEEE Trans. Broadcast., vol. 52, no.4, Dec. 2006. Z. W. Zheng, “Robust Timing Recovery for TDS-OFDM-Based Digital Television Terrestrial Broadcast Systems”, IEEE Trans. Consumer Electronics, vol. 52, no. 3, Aug. 2006. H. Meyr, M. Moeneclaey, and S.A.Fechtel, Digital Communications Receivers: Synchronization, Channel Estimation, and Signal Processing, New York: Wiley, 1997. F. Pollara, Digital Television Systems Brazilian Tests Final Report. SET/ABERT. [Online] Available: http://www.set.com.br/testing.pdf P.H. Moose, “A Technique for Orthogonal Frequency Division Multiplexing Frequency Offset Correction”, IEEE Trans. Comm., vol. 42, no. 10, Oct. 1994.
Jianming Wu received the B.S. degree in microelectronics from Fudan University, Shanghai, P.R.China, in 2003. He is currently working toward the Ph.D. degree in microelectronics at the ASIC & System State Key Lab of Fudan University. His research interests include VLSI architectures, SoC designs, and wireless communication systems. Yun Chen received the B.S. degree from UESTC, China in 2001. She is currently working toward the Ph.D. degree in microelectronics at the ASIC & System State Key Lab of Fudan University. Her research interests include VLSI architectures, SoC designs, and wireless communication systems. Xiaoyang Zeng received the B.S. degree from Xiangtan University, China in 1992, and the Ph.D. degree from Changchun Institute of Optics and Fine Mechanics, Chinese Academy of Sciences in 2001. From 2001 to 2003, he worked as a post-doctor researcher at the State-Key Lab of ASIC & System, Fudan University, P.R. China. Then he joined the faculty of Department of Micro-electronics at Fudan University as an associate professor. His research interests include information security chip design, VLSI signal processing, and communication systems. Prof. Zeng is the Chair of Design-Contest of ASP-DAC 2004 and 2005, also the TPC member of several international conferences such as ASCON 2005 and A-SSCC 2006, etc. Hao Min received the B.S. and M.S. degree in electrical engineering from Fudan University, Shanghai, and P.R.China in 1985 and 1988 respectively. In 1991 he received the Ph.D. degree in material science from Fudan. From 1996 to 1998 he was a visiting scholar in Stanford University, CA. He is now the general manager of Shanghai Huahong IC Design Co., Ltd as well as the professor of Fudan University in the ASIC & System State-Key Laboratory. His research interests include RF VLSI design, mixed signal VLSI design and digital signal processing.
